BISHENG is an open LLM application devops platform, focusing on enterprise scenarios. It has been used by a large number of industry leading organizations and Fortune 500 companies.
"Bi Sheng" was the inventor of movable type printing, which played a vital role in promoting the transmission of human knowledge. We hope that BISHENG can also provide strong support for the widespread implementation of intelligent applications. Everyone is welcome to participate.
## Features
1. **Lingsight, a general-purpose agent with expert-level taste**: Through the [AGL](https://github.com/dataelement/AgentGuidanceLanguage)(Agent Guidance Language) framework, we embed domain experts’ preferences, experience, and business logic into the AI, enabling the agent to exhibit “expert-level understanding” when handling tasks.
2. **Unique [BISHENG Workflow](https://dataelem.feishu.cn/wiki/R7HZwH5ZGiJUDrkHZXicA9pInif)**
- 🧩 **Independent and comprehensive application orchestration framework**: Enables the execution of various tasks within a single framework (while similar products rely on bot invocation or separate chatflow and workflow modules for different tasks).
- 🔄 **Human in the loop**: Allows users to intervene and provide feedback during the execution of workflows (including multi-turn conversations), whereas similar products can only execute workflows from start to finish without intervention.
- 💥 **Powerful**: Supports loops, parallelism, batch processing, conditional logic, and free combination of all logic components. It also handles complex scenarios such as multi-type input/output, report generation, content review, and more.
- 🖐️ **User-friendly and intuitive**: Operations like loops, parallelism, and batch processing, which require specialized components in similar products, can be easily visualized in BISHENG as a "flowchart" (drawing a loop forms a loop, aligning elements creates parallelism, and selecting multiple items enables batch processing).
3. Designed for Enterprise Applications: Document review, fixed-layout report generation, multi-agent collaboration, policy update comparison, support ticket assistance, customer service assistance, meeting minutes generation, resume screening, call record analysis, unstructured data governance, knowledge mining, data analysis, and more.
The platform supports the construction of highly complex enterprise application scenarios and offers deep optimization with hundreds of components and thousands of parameters.
4. Enterprise-grade features are the fundamental guarantee for application implementation: security review, RBAC, user group management, traffic control by group, SSO/LDAP, vulnerability scanning and patching, high availability deployment solutions, monitoring, statistics, and more.
5. High-Precision Document Parsing: Our high-precision document parsing model is trained on a vast amount of high-quality data accumulated over past 5 years. It includes high-precision printed text, handwritten text, and rare character recognition models, table recognition models, layout analysis models, and seal models., table recognition models, layout analysis models, and seal models. You can deploy it privately for free.
6. A community for sharing best practices across various enterprise scenarios: An open repository of application cases and best practices.
## Quick start
Please ensure the following conditions are met before installing BISHENG:
- CPU >= 4 Virtual Cores
- RAM >= 16 GB
- Docker 19.03.9+
- Docker Compose 1.25.1+
> Recommended hardware condition: 18 virtual cores, 48G. In addition to installing BISHENG, we will also install the following third-party components by default: ES, Milvus, and Onlyoffice.
Download BISHENG
```bash
git clone https://github.com/dataelement/bisheng.git
# Enter the installation directory
cd bisheng/docker
# If the system does not have the git command, you can download the BISHENG code as a zip file.
wget https://github.com/dataelement/bisheng/archive/refs/heads/main.zip
# Unzip and enter the installation directory
unzip main.zip && cd bisheng-main/docker
```
Start BISHENG
```bash
docker compose -f docker-compose.yml -p bisheng up -d
```
After the startup is complete, access http://IP:3001 in the browser. The login page will appear, proceed with user registration.
By default, the first registered user will become the system admin.
For more installation and deployment issues, refer to:：[Self-hosting](https://dataelem.feishu.cn/wiki/BSCcwKd4Yiot3IkOEC8cxGW7nPc)
## Acknowledgement
This repo benefits from [langchain](https://github.com/langchain-ai/langchain) [langflow](https://github.com/logspace-ai/langflow) [unstructured](https://github.com/Unstructured-IO/unstructured) and [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory) . Thanks for their wonderful works.
Thank you to our contributors：
## Community & contact
Welcome to join our discussion group
================================================
FILE: README_CN.md
================================================
6. 様々なエンタープライズシナリオにおけるベストプラクティスを共有するコミュニティ: オープンなアプリケーションケースとベストプラクティスのリポジトリ。
## クイックスタート
BISHENGをインストールする前に、以下の条件を満たしていることを確認してください:
- CPU >= 8 コア
- RAM >= 32 GB
- Docker 19.03.9以上
- Docker Compose 1.25.1以上
> BISHENGをインストールする際、デフォルトで以下のサードパーティコンポーネントもインストールされます: ES, Milvus, Onlyoffice。
BISHENGのダウンロード
```bash
git clone https://github.com/dataelement/bisheng.git
# インストールディレクトリに移動
cd bisheng/docker
# システムにgitコマンドがない場合は、BISHENGのコードをzipファイルとしてダウンロードできます。
wget https://github.com/dataelement/bisheng/archive/refs/heads/main.zip
# 解凍してインストールディレクトリに移動
unzip main.zip && cd bisheng-main/docker
```
BISHENGの起動
```bash
docker compose -f docker-compose.yml -p bisheng up -d
```
起動完了後、ブラウザでhttp://IP:3001にアクセスします。ログインページが表示されるので、ユーザー登録を行います。
デフォルトでは、最初に登録されたユーザーがシステム管理者となります。
詳細なインストールおよびデプロイに関する問題は、こちらを参照してください：[私有化部署](https://dataelem.feishu.cn/wiki/BSCcwKd4Yiot3IkOEC8cxGW7nPc)
## 謝辞
このリポジトリは [langchain](https://github.com/langchain-ai/langchain) [langflow](https://github.com/logspace-ai/langflow) [unstructured](https://github.com/Unstructured-IO/unstructured) および [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory) の恩恵を受けています。素晴らしい作品に感謝します。
**貢献者に感謝します：**
## コミュニティと連絡先
ディスカッショングループへの参加を歓迎します。
---
この翻訳を使用して、Markdownファイルを作成できます。
================================================
FILE: SECURITY.md
================================================
# Security Policy
## Reporting Security Issues
We take the security of our project seriously. If you believe you have found a security vulnerability, please report it to us privately. **Please do not report security vulnerabilities through public GitHub issues, discussions, or pull requests.**
> **Important Note**: Any code within the `classic/` folder is considered legacy, unsupported, and out of scope for security reports. We will not address security vulnerabilities in this deprecated code.
Instead, please report them via:
- [GitHub Security Advisory](https://github.com/dataelement/bisheng/security/advisories/new)
### Reporting Process
1. **Submit Report**: Use one of the above channels to submit your report
2. **Response Time**: Our team will acknowledge receipt of your report within 14 business days.
3. **Collaboration**: We will collaborate with you to understand and validate the issue
4. **Resolution**: We will work on a fix and coordinate the release process
### Disclosure Policy
- Please provide detailed reports with reproducible steps
- Include the version/commit hash where you discovered the vulnerability
- Allow us a 90-day security fix window before any public disclosure
- Share any potential mitigations or workarounds if known
## Supported Versions
Only the following versions are eligible for security updates:
| Version | Supported |
|---------|-----------|
| Latest release on master branch | ✅ |
| Development commits (pre-master) | ✅ |
| Classic folder (deprecated) | ❌ |
| All other versions | ❌ |
---
Last updated: November 2024
================================================
FILE: docker/bisheng/config/config.yaml
================================================
# 数据库配置， 当前加密串的密码是1234，
# 密码加密参考 https://dataelem.feishu.cn/wiki/BSCcwKd4Yiot3IkOEC8cxGW7nPc#Gxitd1xEeof1TzxdhINcGS6JnXd
database_url:
"mysql+pymysql://root:gAAAAABlp4b4c59FeVGF_OQRVf6NOUIGdxq8246EBD-b0hdK_jVKRs1x4PoAn0A6C5S6IiFKmWn0Nm5eBUWu-7jxcqw6TiVjQA==@mysql:3306/bisheng?charset=utf8mb4"
# 缓存配置 redis://[[username]:[password]]@localhost:6379/0
# 如果设置了密码，需要参考MySQL密码的加密逻辑对密码进行加密。eg: redis://root:gAAAAABlp4b4c59FeVGF_OQRVf6NOUIGdxq8246EBD-b0hdK_jVKRs1x4PoAn0A6C5S6IiFKmWn0Nm5eBUWu-7jxcqw6TiVjQA==@redis:6379/0
# 普通模式:
redis_url: "redis://redis:6379/1"
# 集群模式或者哨兵模式（只能选其一）:
# redis_url:
# mode: "cluster"
# startup_nodes:
# - {"host": "192.168.106.115", "port": 6002}
# password: encrypt(gAAAAABlp4b4c59FeVGF_OQRVf6NOUIGdxq8246EBD-b0hdK_jVKRs1x4PoAn0A6C5S6IiFKmWn0Nm5eBUWu-7jxcqw6TiVjQA==)
# #sentinel
# mode: "sentinel"
# sentinel_hosts: [("redis", 6379)]
# sentinel_master: "mymaster"
# sentinel_password: encrypt(gAAAAABlp4b4c59FeVGF_OQRVf6NOUIGdxq8246EBD-b0hdK_jVKRs1x4PoAn0A6C5S6IiFKmWn0Nm5eBUWu-7jxcqw6TiVjQA==)
# db: 1
# celery的broken地址
celery_redis_url: "redis://redis:6379/2"
celery_task:
# 对celery熟悉的用户可以自定义配置任务的路由，启动不同类型的worker处理不同类型的异步任务。注意工作流的执行只能在一个进程内！！！
task_routers:
bisheng.worker.knowledge.*: # 知识库文件处理相关任务
queue: knowledge_celery
bisheng.worker.workflow.*: # 工作流相关任务
queue: workflow_celery
# 知识库的milvus和es配置 支持使用 !env ${PATH} 填写环境变量的值, 若环境变量不存在则会报错
vector_stores:
milvus:
connection_args: !env ${BS_MILVUS_CONNECTION_ARGS}
is_partition: !env ${BS_MILVUS_IS_PARTITION}
partition_suffix: !env ${BS_MILVUS_PARTITION_SUFFIX}
elasticsearch:
url: !env ${BS_ELASTICSEARCH_URL}
ssl_verify: !env ${BS_ELASTICSEARCH_SSL_VERIFY}
# 对象存储， 目前只支持minio
object_storage:
type: minio
minio:
schema: !env ${BS_MINIO_SCHEMA}
cert_check: !env ${BS_MINIO_CERT_CHECK}
endpoint: !env ${BS_MINIO_ENDPOINT}
sharepoint: !env ${BS_MINIO_SHAREPOINT}
access_key: !env ${BS_MINIO_ACCESS_KEY}
secret_key: !env ${BS_MINIO_SECRET_KEY}
public_bucket: 'bisheng' # 公共bucket，存储平台上一些需要持久化的文件。会设置为可公开访问
tmp_bucket: 'tmp-dir' # 临时bucket，会对传到此bucket内的文件设置有效期
environment:
env: dev
uns_support: ['png','jpg','jpeg','bmp','doc', 'docx', 'ppt', 'pptx', 'xls', 'xlsx', 'txt', 'md', 'html', 'pdf', 'csv', 'tiff']
# 可根据loguru的文档配置不同 handlers
logger_conf:
# 默认输出到sys.stdout的日志级别, 大于等于此级别都会输出
level: DEBUG
# 默认输出格式
format: '[{time:YYYY-MM-DD HH:mm:ss.SSSSSS}] [{level.name} process-{process.id}-{thread.id} {name}:{line}] - trace={extra[trace_id]} {message}'
# 参考loguru.add()中的参数可以配置多个handler
handlers:
# 文件路径，支持插入一些系统环境变量，若环境变量不存在则置空。例如 HOSTNAME: 主机名。后端会处理环境变量的替换
- sink: "/app/data/bisheng.log"
# 日志级别
level: INFO
# 日志格式化函数，extra内支持trace_id
format: '[{time:YYYY-MM-DD HH:mm:ss.SSSSSS}] [{level.name} process-{process.id}-{thread.id} {name}:{line}] - trace={extra[trace_id]} {message}'
# 每天的几点进行切割
rotation: "00:00"
retention: "3 Days"
enqueue: ture
- sink: "/app/data/statistic.log"
level: INFO
# 和原生不一样，后端会将配置使用eval()执行转为函数用来过滤特定日志级别。推荐lambda
filter: "lambda record: record['level'].name == 'INFO' and record['message'].startswith('k=s')"
format: "[{time:YYYY-MM-DD HH:mm:ss.SSSSSS}]|{level}|BISHENG|{extra[trace_id]}||{process.id}|{thread.id}|||#EX_ERR:POS={name},line {line},ERR=500,EMSG={message}"
rotation: "00:00"
retention: "3 Days"
enqueue: ture
================================================
FILE: docker/bisheng/entrypoint.sh
================================================
#!/bin/bash
set -xe
export PYTHONPATH="./"
start_mode=${1:-api}
start_knowledge(){
# 知识库解析的celery worker
celery -A bisheng.worker.main worker -l info -c 20 -P threads -Q knowledge_celery -n knowledge@%h
}
start_workflow(){
# 工作流相关的celery worker
celery -A bisheng.worker.main worker -l info -c 100 -P threads -Q workflow_celery -n workflow@%h
}
start_beat(){
# 定时任务调度
celery -A bisheng.worker.main beat -l info
}
start_linsight(){
# 灵思后台任务worker
python bisheng/linsight/worker.py --worker_num 4 --max_concurrency 5
}
start_default(){
# 默认其他任务的执行worker，目前是定时统计埋点数据
celery -A bisheng.worker.main worker -l info -c 100 -P threads -Q celery -n celery@%h
}
if [ "$start_mode" = "api" ]; then
echo "Starting API server..."
uvicorn bisheng.main:app --host 0.0.0.0 --port 7860 --no-access-log --workers 8
elif [ "$start_mode" = "knowledge" ]; then
echo "Starting Knowledge Celery worker..."
start_knowledge
elif [ "$start_mode" = "workflow" ]; then
echo "Starting Workflow Celery worker..."
start_workflow
elif [ "$start_mode" = "beat" ]; then
echo "Starting Celery beat..."
start_beat
elif [ "$start_mode" = "default" ]; then
echo "Starting default celery worker..."
start_default
elif [ "$start_mode" = "linsight" ]; then
echo "Starting LinSight worker..."
start_linsight
elif [ "$start_mode" = "worker" ]; then
echo "Starting All worker..."
# 处理知识库相关任务的worker
start_knowledge &
# 处理工作流相关任务的worker
start_workflow &
# 处理linsight相关任务的worker
start_linsight &
# 默认其他任务的执行worker，目前是定时统计埋点数据
start_default &
start_beat
echo "All workers started successfully."
else
echo "Invalid start mode. Use api、worker、knowledge、workflow、beat、default、linsight."
exit 1
fi
================================================
FILE: docker/bisheng-ft/config.yaml
================================================
# Celery 的broker配置。存储ft指令执行结果。
# 密码加密规则和backend一致, 暂不支持集群redis
redis_url: "redis://bisheng-redis:6379/5"
================================================
FILE: docker/bisheng-uns/config.yaml
================================================
# 可根据loguru的文档配置不同 handlers
logger_conf:
# 默认输出到控制台的日志级别, 大于等于此级别都会输出
level: DEBUG
# 默认输出格式
format: '[{time:YYYY-MM-DD HH:mm:ss.SSSSSS}] [{level.name} process-{process.id}-{thread.id} {name}:{line}] - trace={extra[trace_id]} {message}'
# 参考loguru.add()中的参数可以配置多个handler
handlers:
# 文件路径，支持插入一些系统环境变量，若环境变量不存在则置空。例如 HOSTNAME: 主机名。后端会处理环境变量的替换
- sink: "/app/logs/bisheng_uns.log"
# 日志级别
level: INFO
# 和原生不一样，后端会将配置使用eval()执行转为函数用来过滤特定日志级别。推荐lambda
# filter: "lambda record: record['level'].name == 'INFO'"
# 日志格式化函数，extra内支持trace_id
format: "[{time:YYYY-MM-DD HH:mm:ss.SSSSSS}]|{level}|BISHENG|{extra[trace_id]}|{process.id}|{thread.id}|{message}"
# 每天的几点进行切割
rotation: "00:00"
retention: "3 Days"
- sink: "/app/logs/err-v0-BISHENG-UNS-{HOSTNAME}.log"
level: ERROR
filter: "lambda record: record['level'].name == 'ERROR'"
format: "[{time:YYYY-MM-DD HH:mm:ss.SSSSSS}]|{level}|BISHENG|{extra[trace_id]}||{process.id}|{thread.id}|||#EX_ERR:POS={name},line {line},ERR=500,EMSG={message}"
rotation: "00:00"
retention: "3 Days"
# pdf解析需要用到的模型配置, 配置了rt_server环境变量的话会替换为对应的地址
pdf_model_params:
layout_ep: "http://192.168.106.12:9001/v2.1/models/elem_layout_v1/infer"
cell_model_ep: "http://192.168.106.12:9001/v2.1/models/elem_table_cell_detect_v1/infer"
rowcol_model_ep: "http://192.168.106.12:9001/v2.1/models/elem_table_rowcol_detect_v1/infer"
table_model_ep: "http://192.168.106.12:9001/v2.1/models/elem_table_detect_v1/infer"
ocr_model_ep: "http://192.168.106.12:9001/v2.1/models/elem_ocr_collection_v3/infer"
# 是否全部走ocr识别, false的话则由代码逻辑判断是否需要走ocr识别
is_all_ocr: false
# ocr识别需要的配置项
ocr_conf:
params:
sort_filter_boxes: true,
enable_huarong_box_adjust: true,
rotateupright: false,
support_long_image_segment: true,
split_long_sentence_blank: true
scene_mapping:
print:
det: general_text_det_mrcnn_v2.0
recog: transformer-blank-v0.2-faster
hand:
det: general_text_det_mrcnn_v2.0
recog: transformer-hand-v1.16-faster
print_recog:
recog: transformer-blank-v0.2-faster
hand_recog:
recog: transformer-hand-v1.16-faster
det:
det: general_text_det_mrcnn_v2.0
================================================
FILE: docker/docker-compose-ft.yml
================================================
services:
ft_server:
container_name: bisheng-ft-server
image: dataelement/bisheng-ft:v0.5.0
shm_size: "4g"
ports:
- "8000:8000"
environment:
TZ: Asia/Shanghai
volumes:
- ${DOCKER_VOLUME_DIRECTORY:-.}/bisheng-ft/config.yaml:/opt/bisheng-ft/sft_server/config.yaml # 服务启动所需的配置文件地址，默认不用改
- ${DOCKER_VOLUME_DIRECTORY:-.}/data/llm:/opt/bisheng-ft/models/model_repository # 这个是存放基座模型的目录，挂载到本机目录
- ${DOCKER_VOLUME_DIRECTORY:-.}/data/finetune_output:/opt/bisheng-ft/finetune_output # 这个是存放微调过程的中间日志和微调训练后模型的目录，挂载到本机目录，不能与存放基座模型的目录相同
security_opt:
- seccomp:unconfined
command: bash start-sft-server.sh # 启动服务
restart: on-failure
healthcheck:
test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
start_period: 30s
interval: 90s
timeout: 30s
retries: 3
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]
================================================
FILE: docker/docker-compose-office.yml
================================================
services:
office:
container_name: bisheng-office
image: onlyoffice/documentserver:7.1.1
ports:
- "8701:80"
environment:
TZ: Asia/Shanghai
JWT_ENABLED: "false"
volumes:
- ${DOCKER_VOLUME_DIRECTORY:-.}/office/bisheng:/var/www/onlyoffice/documentserver/sdkjs-plugins/bisheng
command: bash -c "supervisorctl restart all"
restart: on-failure
================================================
FILE: docker/docker-compose-uns.yml
================================================
services:
bisheng-unstructured:
container_name: bisheng-unstructured
image: dataelement/bisheng-unstructured:v0.0.3.14
ports:
- "10001:10001"
environment:
# 填写ocr_sdk或rt服务的根地址
# server_address: bisheng-rt:9001
# 这里填 ocr_sdk 或 rt
# server_type: ocr_sdk
TZ: Asia/Shanghai
volumes:
- ${DOCKER_VOLUME_DIRECTORY:-.}/bisheng-uns/config.yaml:/opt/bisheng-unstructured/bisheng_unstructured/config/config.yaml
healthcheck:
test: ["CMD", "curl", "-f", "http://localhost:10001/health"]
interval: 30s
timeout: 20s
retries: 3
restart: on-failure
================================================
FILE: docker/docker-compose.yml
================================================
services:
mysql:
container_name: bisheng-mysql
image: mysql:8.0
ports:
- "3306:3306"
environment:
MYSQL_ROOT_PASSWORD: "1234" # 数据库密码，如果修改需要同步修改bisheng/congfig/config.yaml配置database_url的mysql连接密码
MYSQL_DATABASE: bisheng
TZ: Asia/Shanghai
volumes:
- ${DOCKER_VOLUME_DIRECTORY:-.}/mysql/conf/my.cnf:/etc/mysql/my.cnf
- ${DOCKER_VOLUME_DIRECTORY:-.}/mysql/data:/var/lib/mysql
healthcheck:
test: ["CMD-SHELL", "exit | mysql -u root -p$$MYSQL_ROOT_PASSWORD"]
start_period: 30s
interval: 20s
timeout: 10s
retries: 4
restart: on-failure
redis:
container_name: bisheng-redis
image: redis:7.0.4
ports:
- "6379:6379"
environment:
TZ: Asia/Shanghai
volumes:
- ${DOCKER_VOLUME_DIRECTORY:-.}/data/redis:/data
- ${DOCKER_VOLUME_DIRECTORY:-.}/redis/redis.conf:/etc/redis.conf
command: redis-server /etc/redis.conf
healthcheck:
test: ["CMD-SHELL", 'redis-cli ping|grep -e "PONG\|NOAUTH"']
interval: 10s
timeout: 5s
retries: 3
restart: on-failure
backend:
container_name: bisheng-backend
image: dataelement/bisheng-backend:v2.4.0-beta1-fix
ports:
- "7860:7860"
environment:
TZ: Asia/Shanghai
BS_MILVUS_CONNECTION_ARGS: '{"host":"milvus","port":"19530","user":"","password":"","secure":false}'
BS_MILVUS_IS_PARTITION: 'true'
BS_MILVUS_PARTITION_SUFFIX: '1'
BS_ELASTICSEARCH_URL: 'http://elasticsearch:9200'
BS_ELASTICSEARCH_SSL_VERIFY: '{}' # 可根据自己部署的密码进行配置 '{"basic_auth": ("elastic", "elastic")}'
BS_MINIO_SCHEMA: 'false'
BS_MINIO_CERT_CHECK: 'false'
BS_MINIO_ENDPOINT: 'minio:9000'
BS_MINIO_SHAREPOINT: 'minio:9000'
BS_MINIO_ACCESS_KEY: 'minioadmin'
BS_MINIO_SECRET_KEY: 'minioadmin'
volumes:
- ${DOCKER_VOLUME_DIRECTORY:-.}/bisheng/config/config.yaml:/app/bisheng/config.yaml
- ${DOCKER_VOLUME_DIRECTORY:-.}/bisheng/entrypoint.sh:/app/entrypoint.sh
- ${DOCKER_VOLUME_DIRECTORY:-.}/data/bisheng:/app/data
security_opt:
- seccomp:unconfined
command: sh entrypoint.sh api # 启动api服务
restart: on-failure
healthcheck:
test: ["CMD", "curl", "-f", "http://localhost:7860/health"]
start_period: 30s
interval: 90s
timeout: 30s
retries: 3
depends_on:
mysql:
condition: service_healthy
redis:
condition: service_healthy
backend_worker:
container_name: bisheng-backend-worker
image: dataelement/bisheng-backend:v2.4.0-beta1-fix
environment:
TZ: Asia/Shanghai
BS_MILVUS_CONNECTION_ARGS: '{"host":"milvus","port":"19530","user":"","password":"","secure":false}'
BS_MILVUS_IS_PARTITION: 'true'
BS_MILVUS_PARTITION_SUFFIX: '1'
BS_ELASTICSEARCH_URL: 'http://elasticsearch:9200'
BS_ELASTICSEARCH_SSL_VERIFY: '{}' # 可根据自己部署的密码进行配置 '{"basic_auth": ("elastic", "elastic")}'
BS_MINIO_SCHEMA: 'false'
BS_MINIO_CERT_CHECK: 'false'
BS_MINIO_ENDPOINT: 'minio:9000'
BS_MINIO_SHAREPOINT: 'minio:9000'
BS_MINIO_ACCESS_KEY: 'minioadmin'
BS_MINIO_SECRET_KEY: 'minioadmin'
volumes:
- ${DOCKER_VOLUME_DIRECTORY:-.}/bisheng/config/config.yaml:/app/bisheng/config.yaml
- ${DOCKER_VOLUME_DIRECTORY:-.}/bisheng/entrypoint.sh:/app/entrypoint.sh
- ${DOCKER_VOLUME_DIRECTORY:-.}/data/bisheng:/app/data
security_opt:
- seccomp:unconfined
command: sh entrypoint.sh worker # 启动celery的异步worker服务，用来处理一些耗时的任务
restart: on-failure
depends_on:
mysql:
condition: service_healthy
redis:
condition: service_healthy
frontend:
container_name: bisheng-frontend
image: dataelement/bisheng-frontend:v2.4.0-beta1-fix
ports:
- "3001:3001"
environment:
TZ: Asia/Shanghai
volumes:
- ${DOCKER_VOLUME_DIRECTORY:-.}/nginx/nginx.conf:/etc/nginx/nginx.conf
- ${DOCKER_VOLUME_DIRECTORY:-.}/nginx/conf.d:/etc/nginx/conf.d
restart: on-failure
depends_on:
- backend
elasticsearch:
container_name: bisheng-es
image: docker.io/bitnamilegacy/elasticsearch:8.12.0
user: root
ports:
- "9200:9200"
- "9300:9300"
environment:
TZ: Asia/Shanghai
volumes:
- ${DOCKER_VOLUME_DIRECTORY:-.}/data/es:/bitnami/elasticsearch/data
restart: on-failure
etcd:
container_name: bisheng-milvus-etcd
image: quay.io/coreos/etcd:v3.5.5
environment:
ETCD_AUTO_COMPACTION_MODE: revision
ETCD_AUTO_COMPACTION_RETENTION: "1000"
ETCD_QUOTA_BACKEND_BYTES: "4294967296"
ETCD_SNAPSHOT_COUNT: "50000"
TZ: Asia/Shanghai
volumes:
- ${DOCKER_VOLUME_DIRECTORY:-.}/data/milvus-etcd:/etcd
command: etcd -advertise-client-urls=http://127.0.0.1:2379 -listen-client-urls http://0.0.0.0:2379 --data-dir /etcd
restart: on-failure
healthcheck:
test: ["CMD", "etcdctl", "endpoint", "health"]
interval: 30s
timeout: 20s
retries: 3
minio:
container_name: bisheng-milvus-minio
image: minio/minio:RELEASE.2023-03-20T20-16-18Z
environment:
MINIO_ACCESS_KEY: minioadmin
MINIO_SECRET_KEY: minioadmin
ports:
- "9100:9000"
- "9101:9001"
volumes:
- /etc/localtime:/etc/localtime:ro
- ${DOCKER_VOLUME_DIRECTORY:-.}/data/milvus-minio:/minio_data
command: minio server /minio_data --console-address ":9001"
restart: on-failure
healthcheck:
test: ["CMD", "curl", "-f", "http://localhost:9000/minio/health/live"]
interval: 30s
timeout: 20s
retries: 3
milvus:
container_name: bisheng-milvus-standalone
image: milvusdb/milvus:v2.5.10
command: ["milvus", "run", "standalone"]
security_opt:
- seccomp:unconfined
environment:
ETCD_ENDPOINTS: etcd:2379
MINIO_ADDRESS: minio:9000
volumes:
- /etc/localtime:/etc/localtime:ro
- ${DOCKER_VOLUME_DIRECTORY:-.}/data/milvus:/var/lib/milvus
restart: on-failure
healthcheck:
test: ["CMD", "curl", "-f", "http://localhost:9091/healthz"]
start_period: 90s
interval: 30s
timeout: 20s
retries: 3
ports:
- "19530:19530"
- "9091:9091"
depends_on:
- etcd
- minio
================================================
FILE: docker/mysql/conf/my.cnf
================================================
[client]
default-character-set=utf8mb4
[mysql]
default-character-set=utf8mb4
[mysqld]
init_connect='SET collation_connection = utf8mb4_unicode_ci, NAMES utf8mb4'
character-set-server=utf8mb4
collation-server=utf8mb4_unicode_ci
# skip-character-set-client-handshake
sql_mode=STRICT_TRANS_TABLES,NO_ZERO_IN_DATE,NO_ZERO_DATE,ERROR_FOR_DIVISION_BY_ZERO,NO_ENGINE_SUBSTITUTION
================================================
FILE: docker/nginx/conf.d/default.conf
================================================
# 在http区域内一定要添加下面配置, 支持websocket
map $http_upgrade $connection_upgrade {
default upgrade;
'' close;
}
server {
gzip on;
gzip_comp_level 2;
gzip_min_length 1000;
gzip_types text/xml text/css;
gzip_http_version 1.1;
gzip_vary on;
gzip_disable "MSIE [4-6] \.";
listen 3001;
location / {
root /usr/share/nginx/html/platform;
index index.html index.htm;
# 禁止浏览器缓存 index.html
location = /index.html {
add_header Cache-Control "no-store, no-cache, must-revalidate, proxy-revalidate" always;
add_header Pragma "no-cache" always;
add_header Expires 0 always;
}
try_files $uri $uri/ /index.html =404;
add_header X-Frame-Options SAMEORIGIN;
}
location /workspace/ {
alias /usr/share/nginx/html/client/;
index index.html index.htm;
# 禁止浏览器缓存 /workspace/index.html
location = /workspace/index.html {
add_header Cache-Control "no-store, no-cache, must-revalidate, proxy-revalidate" always;
add_header Pragma "no-cache" always;
add_header Expires 0 always;
}
try_files $uri $uri/ /workspace/index.html;
}
location ~ ^(/workspace)?/api(/|$) {
rewrite ^/workspace(/.*)$ $1 break;
proxy_pass http://backend:7860;
proxy_read_timeout 300s;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_http_version 1.1;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection $connection_upgrade;
client_max_body_size 200m;
add_header Access-Control-Allow-Origin $host;
add_header X-Frame-Options SAMEORIGIN;
}
location ~ ^(/workspace)?/bisheng|/tmp-dir {
rewrite ^/workspace(/.*)$ $1 break;
proxy_pass http://minio:9000;
}
}
================================================
FILE: docker/nginx/conf.d/websocket.conf
================================================
# 在http区域内一定要添加下面配置, 支持websocket
map $http_upgrade $connection_upgrade {
default upgrade;
'' close;
}
server {
gzip on;
gzip_comp_level 2;
gzip_min_length 1000;
gzip_types text/xml text/css;
gzip_http_version 1.1;
gzip_vary on;
gzip_disable "MSIE [4-6] \.";
listen 8443;
location /api {
proxy_pass http://backend:7860;
proxy_read_timeout 300s;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_http_version 1.1;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection $connection_upgrade;
client_max_body_size 50m;
}
}
================================================
FILE: docker/nginx/nginx.conf
================================================
user nginx;
worker_processes auto;
error_log /var/log/nginx/error.log notice;
pid /var/run/nginx.pid;
events {
worker_connections 1024;
}
http {
include /etc/nginx/mime.types;
default_type application/octet-stream;
log_format main '$remote_addr - $remote_user [$time_local] "$request" '
'$status $body_bytes_sent "$http_referer" '
'"$http_user_agent" "$http_x_forwarded_for"';
access_log /var/log/nginx/access.log main;
sendfile on;
#tcp_nopush on;
keepalive_timeout 65;
#gzip on;
include /etc/nginx/conf.d/*.conf;
}
================================================
FILE: docker/office/bisheng/all.js
================================================
(function (window, undefined) {
let selectText = ''
window.Asc.plugin.init = function (e) {
selectText = e
}
window.Asc.plugin.event_onClick = function () {
selectText = ''
}
window.Asc.plugin.button = function (id) {
}
const EventMap = {
sendToParent (method, data) {
let params = {
type: 'onExternalFrameMessage',
method,
data
}
window.top.postMessage(JSON.stringify(params), location.origin)
},
focusInDocument (data) {
window.Asc.scope.field = {
id: data.id,
fieldFlag: data.fieldFlag,
$index: data.$index || 1
}
window.Asc.plugin.callCommand(function () {
let field = Asc.scope.field || {}
let index = field.$index ? field.$index : 1
let oDoc = Api.GetDocument()
let flag = `{{${field.fieldFlag}}}`
let oRange = oDoc.Search(flag)
let cur = 1
for (let i = 0; i < oRange.length; i++) {
if (oRange[i].GetText() === flag) {
if (cur === index) {
oRange[i].Select()
break
}
cur = cur + 1
}
}
})
},
focusTableInDoc (data) {
window.Asc.scope.marker = data.marker
window.Asc.plugin.callCommand(function () {
let flag = Asc.scope.marker || ''
let oDoc = Api.GetDocument()
let oRange = oDoc.GetBookmarkRange(flag)
oRange.Select()
})
},
addMarker (data) {
let flag = '{{' + data.fieldFlag + '}}'
window.Asc.plugin.executeMethod('PasteText', [flag])
},
addBookMarker (data) {
window.Asc.scope.value = data
window.Asc.plugin.callCommand(function () {
let oDoc = Api.GetDocument()
let range = oDoc.GetRangeBySelect()
let params = {
type: 'onExternalFrameMessage',
method: 'addBookMarker'
}
let marker = Asc.scope.value
let markers = []
if (range) {
let texts = range.GetText()
let pars = range.GetAllParagraphs() || []
let txtList = []
for (let i = 0; i < pars.length; i++) {
let text = pars[i].GetText()
txtList.push(text)
}
let table = pars[0] ? pars[0].GetParentTable() : null
let count = table ? table.GetRowsCount() : 0
for (let i = 0; i < count; i++) {
let row = table.GetRow(i)
let firstCell = row.GetCell(0)
let cellText = firstCell ? firstCell.GetContent().GetElement(0).GetText() : ''
// 序号
let isNumbering = false
if (firstCell.GetContent().GetElement(0).GetNumbering()) {
isNumbering = true
let cellCount = row.GetCellsCount()
for (let j = 1; j < cellCount; j++) {
let cellItem = row.GetCell(j)
if (!cellItem.GetContent().GetElement(0).GetNumbering()) {
cellText = cellItem.GetContent().GetElement(0).GetText()
firstCell = cellItem
break
}
}
}
if (cellText && txtList.includes(cellText)) {
let cRange = firstCell.Search(cellText)[0]
cRange.AddBookmark(marker.key + i)
markers.push(marker.key + i)
}
}
// range.AddBookmark(Asc.scope.value.key)
params.data = Object.assign(marker, {
key: markers.join(','),
texts
})
} else {
params.data = false
}
window.top.postMessage(JSON.stringify(params), location.origin)
})
},
deleteBookMarker (data) {
window.Asc.scope.value = data
window.Asc.plugin.callCommand(function () {
let oDoc = Api.GetDocument()
let markers = Asc.scope.value || []
for (let i = 0; i < markers.length; i++) {
oDoc.DeleteBookmark(markers[i])
}
})
},
// 批量删除循环应用内标签
deleteLoopApp (list) {
window.Asc.scope.value = list
window.Asc.plugin.callCommand(function () {
let list = window.Asc.scope.value || []
let oDoc = Api.GetDocument()
list.forEach(row => {
if (row.loopType === 0) {
oDoc.SearchAndReplace({ searchString: `{{${row.startTag}}}`, replaceString: '' }, `{{${row.startTag}}}`, '')
oDoc.SearchAndReplace({ searchString: `{{${row.endTag}}}`, replaceString: '' }, `{{${row.endTag}}}`, '')
} else if (row.loopType === 1) {
oDoc.DeleteBookmark(row.bookmark)
}
})
})
},
// 更新占位符
replaceMarker (data) {
window.Asc.scope.st = '{{' + data.newValue + '}}'
// 原来的值
if (data.oldValue) {
window.Asc.scope.old = '{{' + data.oldValue + '}}'
} else {
this.addMarker(data)
return
}
window.Asc.plugin.callCommand(function () {
let oDocument = Api.GetDocument()
oDocument.SearchAndReplace({ searchString: Asc.scope.old, replaceString: Asc.scope.st }, Asc.scope.old, Asc.scope.st)
}, false)
},
// 查找并插入占位符
findAndInsertMarker (data) {
window.Asc.scope.st = '{{' + data.fieldName + '}}'
window.Asc.scope.searchStr = data.fieldValue
window.Asc.plugin.callCommand(function () {
let oDocument = Api.GetDocument()
oDocument.SearchAndReplace({ searchString: Asc.scope.searchStr, replaceString: Asc.scope.st }, Asc.scope.searchStr, Asc.scope.st)
}, false)
},
insertPosition (data) {
if (!selectText) {
let postData = {
text: selectText,
...data,
selected: false
}
this.sendToParent('addRange', postData)
return false
}
window.Asc.scope.postData = data
window.Asc.plugin.callCommand(function() {
let postData = Asc.scope.postData || {}
let oDoc = Api.GetDocument()
let oRange = oDoc.GetRangeBySelect()
let selectText = oRange.GetText()
let oAllPar = oRange.GetAllParagraphs()
let oPar = oAllPar[oAllPar.length - 1]
let parText = oPar.GetText()
if (oAllPar.length > 1) {
oRange.AddText(`{{${postData.start}}}`, 'before')
if (selectText.includes(parText)) {
let newRange = oPar.GetRange(0, parText.length - 1)
newRange.AddText(`{{${postData.end}}}`, 'after')
} else {
oRange.AddText(`{{${postData.end}}}`, 'after')
}
} else {
let isEnd = parText.substr(0 - selectText.length) === selectText
isEnd = isEnd || selectText.includes(parText)
console.log('end = ', isEnd)
let start = Math.max(parText.indexOf(selectText), 0)
let end = start + Math.min(parText.length, selectText.length) - 1
let newRange = oPar.GetRange(start, end)
oRange.AddText(`{{${postData.start}}}`, 'before')
newRange.AddText(`{{${postData.end}}}`, 'after')
}
postData.selected = true
postData.text = selectText
let params = {
type: 'onExternalFrameMessage',
method: 'addRange',
data: postData
}
window.top.postMessage(JSON.stringify(params), location.origin)
})
},
deletePosition (data) {
window.Asc.scope.range = data
window.Asc.plugin.callCommand(function () {
let oDocument = Api.GetDocument()
let { start, end } = Asc.scope.range
let markers = [`{{${start}}}`, `{{${end}}}`]
for (let j = 0; j < markers.length; j++) {
oDocument.SearchAndReplace({ searchString: markers[j], replaceString: '' }, markers[j], '')
}
})
},
deletePositionMarker (data) {
window.Asc.scope.data = data
window.Asc.plugin.callCommand(function () {
let oDocument = Api.GetDocument()
let markers = Asc.scope.data || []
for (let j = 0; j < markers.length; j++) {
oDocument.SearchAndReplace({ searchString: markers[j], replaceString: '' }, markers[j], '')
}
})
},
deletePositionArray (data) {
window.Asc.scope.data = data
window.Asc.plugin.callCommand(function () {
let oDocument = Api.GetDocument()
let markers = Asc.scope.data || []
for (let j = 0; j < markers.length; j++) {
oDocument.SearchAndReplace({ searchString: markers[j], replaceString: '' }, markers[j], '')
}
})
},
replaceRangePosition (data) {
window.Asc.scope.list = data
window.Asc.plugin.callCommand(function () {
let list = Asc.scope.list || []
let oDocument = Api.GetDocument()
list.forEach(row => {
oDocument.SearchAndReplace({ searchString: row.str, replaceString: row.newStr }, row.str, row.newStr)
})
})
},
delMarker (data) {
let fields = []
data.forEach(item => {
fields.push({
text: '{{' + item.fieldFlag + '}}',
type: 'field'
})
})
window.Asc.scope.st = fields
window.Asc.plugin.callCommand(function () {
let oDocument = Api.GetDocument()
let markers = Asc.scope.st.slice(0)
for (let j = 0; j < markers.length; j++) {
let marker = markers[j]
if (marker.type === 'field') {
oDocument.SearchAndReplace({ searchString: marker.text, replaceString: '' })
}
}
})
},
delMarkerGroup (data) {
this.delMarker(data.fields)
},
// excel
insertCellName (field) {
window.Asc.scope.field = field
window.Asc.plugin.callCommand(function () {
let fieldItem = Asc.scope.field
let sheetObj = Api.GetActiveSheet()
let sheetName = sheetObj.GetName()
let oRange = Api.GetSelection()
let oCount = oRange.GetCount()
let params = {
type: 'onExternalFrameMessage',
method: 'addCellName'
}
if (oCount !== 1) {
params.data = false
} else {
let oAddr = oRange.GetAddress(true, true, '', false)
let sheetFlag = `${sheetName}!${oAddr}`
// let name = [fieldItem.fieldName, 'DEF', fieldItem.id].join('')
// let nameObj = sheetObj.GetDefName(name)
// console.log('inser cell before = ', name, sheetFlag, nameObj)
// let result = sheetObj.AddDefName(name, sheetFlag)
// console.log('insert cell ', name, sheetFlag, result)
fieldItem.fieldFlag = sheetFlag
fieldItem.$success = oAddr !== ''
params.data = fieldItem
}
window.top.postMessage(JSON.stringify(params), location.origin)
})
},
getFocusedCell () {
window.Asc.plugin.callCommand(function () {
let sheetObj = Api.GetActiveSheet()
let sheetName = sheetObj.GetName()
let oRange = Api.GetSelection()
let params = {
type: 'onExternalFrameMessage',
method: 'getFocusedCell'
}
let oAddr = oRange.GetAddress(true, true, '', false)
let sheetFlag = `${sheetName}!${oAddr}`
params.data = sheetFlag
window.top.postMessage(JSON.stringify(params), location.origin)
})
},
loadFileFlags (data) {
window.Asc.scope.list = data
window.Asc.plugin.callCommand(function () {
let list = Asc.scope.list || []
let oDocument = Api.GetDocument()
let oParCount = oDocument.GetElementsCount()
let dataMap = {}
for (let i = 0; i < oParCount; i++) {
let oPar = oDocument.GetElement(i)
let ctype = oPar.GetClassType()
if (ctype === 'table') {
list.forEach(row => {
let flag = `{{${row.fieldFlag}}}`
let rs = oPar.Search(flag)
for (let i = 0; i < rs.length; i++) {
let oRange = rs[i]
if (oRange && oRange.GetText() === flag) {
if (dataMap[row.id]) {
dataMap[row.id] = dataMap[row.id] + 1
} else {
dataMap[row.id] = 1
}
}
}
})
} else if (ctype === 'paragraph') {
let oParText = oPar.GetText()
list.forEach(row => {
let count = oParText.split(`{{${row.fieldFlag}}}`).length - 1
if (dataMap[row.id]) {
dataMap[row.id] = dataMap[row.id] + count
} else {
dataMap[row.id] = count
}
})
}
}
let params = {
type: 'onExternalFrameMessage',
method: 'loadFieldFlagCount',
data: dataMap
}
window.top.postMessage(JSON.stringify(params), location.origin)
})
},
/**
* data.sheetName: 要聚焦的sheet名称
* data.cellName: 要聚焦的cell名称，如C1, D3等
*/
focusCell (data) {
window.Asc.scope.data = data
window.Asc.plugin.callCommand(function () {
const theData = Asc.scope.data
const theSheet = Api.GetSheet(theData.sheetName || '')
if (theSheet) {
theSheet.SetActive()
const theCell = theSheet.GetRange(theData.cellName || '')
if (theCell) {
theCell.Select()
}
}
})
},
getSelectedText (data) {
window.Asc.scope.data = data
window.Asc.plugin.callCommand(function () {
const theData = Asc.scope.data
const oDoc = Api.GetDocument()
const oRange = oDoc.GetRangeBySelect()
if (oRange) {
const oParas = oRange.GetAllParagraphs()
// 只能选择一个段落，否则认为不成功
if (oParas.length === 1) {
const params = {
type: 'onExternalFrameMessage',
method: 'getSelectedText',
data: {
id: theData.id,
text: oRange.GetText()
}
}
window.top.postMessage(JSON.stringify(params), location.origin)
}
}
})
}
}
function receiveMessage (e) {
let data = e.data ? JSON.parse(e.data) : {}
if (data.type === 'onExternalPluginMessage') {
switch (data.method) {
case 'focus':
EventMap.focusInDocument(data.data)
break
case 'focusTable':
EventMap.focusTableInDoc(data.data)
break
case 'insert':
EventMap.addMarker(data.data)
break
case 'addBookMarker':
EventMap.addBookMarker(data.data)
break
case 'delBookMarker':
EventMap.deleteBookMarker(data.data)
break
case 'delLoopApp':
EventMap.deleteLoopApp(data.data)
break
case 'update':
EventMap.replaceMarker(data.data)
break
case 'findAndInsertMarker':
EventMap.findAndInsertMarker(data.data)
break
case 'addRange':
EventMap.insertPosition(data.data)
break
case 'updateRange':
EventMap.replaceRangePosition(data.data)
break
case 'delRange':
EventMap.deletePosition(data.data)
break
case 'delRangeArray':
EventMap.deletePositionArray(data.data)
break
case 'delQuoteGroup':
EventMap.deletePositionMarker(data.data)
break
case 'remove':
EventMap.delMarker([ data.data ])
break
case 'removeQuestion':
EventMap.delMarkerGroup(data.data)
break
// excel
case 'addCellName':
EventMap.insertCellName(data.data)
break
case 'loadFieldFlagCount':
EventMap.loadFileFlags(data.data)
break
// 聚焦到某个单元格
case 'focusCell':
EventMap.focusCell(data.data)
break
// 获取当前选中的单元格
case 'getFocusedCell':
EventMap.getFocusedCell()
break
// 获取当前选中的文字
case 'getSelectedText':
EventMap.getSelectedText(data.data)
break
}
}
}
window.addEventListener('message', receiveMessage, false)
})(window, undefined)
================================================
FILE: docker/office/bisheng/bisheng.js
================================================
(function () {
window.Asc.plugin.init = function (e) {}
window.Asc.plugin.event_onClick = function () {}
window.Asc.plugin.button = function (id) {}
function onMessage(e) {
var data = e.data ? JSON.parse(e.data) : {}
if (data.action === 'insetMarker') {
const flag = '{{' + data.data + '}}'
window.Asc.plugin.executeMethod('PasteText', [flag])
}
}
window.addEventListener('message', onMessage, false)
})()
================================================
FILE: docker/office/bisheng/config.json
================================================
{
"name": "文档自动化",
"guid": "asc.{D2A0F3BE-CC8D-4956-BCD9-6CBEA6E8960E}",
"variations": [
{
"description": "插入label-配置",
"url": "index.html",
"icons": [
"icon.png",
"icon.png",
"icon.png",
"icon.png"
],
"EditorsSupport": [
"word",
"cell",
"slide"
],
"isViewer": false,
"isVisual": false,
"isModal": true,
"isInsideMode": false,
"isSystem": false,
"initOnSelectionChanged": true,
"hideClose": true,
"initDataType": "text",
"isDisplayedInViewer": true,
"isUpdateOleOnResize": true,
"events": [
"onClick",
"onTargetPositionChanged"
]
}
]
}
================================================
FILE: docker/office/bisheng/index.html
================================================
================================================
FILE: docker/redis/redis.conf
================================================
# Redis configuration file example.
#
# Note that in order to read the configuration file, Redis must be
# started with the file path as first argument:
#
# ./redis-server /path/to/redis.conf
# Note on units: when memory size is needed, it is possible to specify
# it in the usual form of 1k 5GB 4M and so forth:
#
# 1k => 1000 bytes
# 1kb => 1024 bytes
# 1m => 1000000 bytes
# 1mb => 1024*1024 bytes
# 1g => 1000000000 bytes
# 1gb => 1024*1024*1024 bytes
#
# units are case insensitive so 1GB 1Gb 1gB are all the same.
################################## INCLUDES ###################################
# Include one or more other config files here. This is useful if you
# have a standard template that goes to all Redis servers but also need
# to customize a few per-server settings. Include files can include
# other files, so use this wisely.
#
# Note that option "include" won't be rewritten by command "CONFIG REWRITE"
# from admin or Redis Sentinel. Since Redis always uses the last processed
# line as value of a configuration directive, you'd better put includes
# at the beginning of this file to avoid overwriting config change at runtime.
#
# If instead you are interested in using includes to override configuration
# options, it is better to use include as the last line.
#
# Included paths may contain wildcards. All files matching the wildcards will
# be included in alphabetical order.
# Note that if an include path contains a wildcards but no files match it when
# the server is started, the include statement will be ignored and no error will
# be emitted. It is safe, therefore, to include wildcard files from empty
# directories.
#
# include /path/to/local.conf
# include /path/to/other.conf
# include /path/to/fragments/*.conf
#
################################## MODULES #####################################
# Load modules at startup. If the server is not able to load modules
# it will abort. It is possible to use multiple loadmodule directives.
#
# loadmodule /path/to/my_module.so
# loadmodule /path/to/other_module.so
################################## NETWORK #####################################
# By default, if no "bind" configuration directive is specified, Redis listens
# for connections from all available network interfaces on the host machine.
# It is possible to listen to just one or multiple selected interfaces using
# the "bind" configuration directive, followed by one or more IP addresses.
# Each address can be prefixed by "-", which means that redis will not fail to
# start if the address is not available. Being not available only refers to
# addresses that does not correspond to any network interface. Addresses that
# are already in use will always fail, and unsupported protocols will always BE
# silently skipped.
#
# Examples:
#
# bind 192.168.1.100 10.0.0.1 # listens on two specific IPv4 addresses
# bind 127.0.0.1 ::1 # listens on loopback IPv4 and IPv6
# bind * -::* # like the default, all available interfaces
#
# ~~~ WARNING ~~~ If the computer running Redis is directly exposed to the
# internet, binding to all the interfaces is dangerous and will expose the
# instance to everybody on the internet. So by default we uncomment the
# following bind directive, that will force Redis to listen only on the
# IPv4 and IPv6 (if available) loopback interface addresses (this means Redis
# will only be able to accept client connections from the same host that it is
# running on).
#
# IF YOU ARE SURE YOU WANT YOUR INSTANCE TO LISTEN TO ALL THE INTERFACES
# COMMENT OUT THE FOLLOWING LINE.
#
# You will also need to set a password unless you explicitly disable protected
# mode.
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# bind 127.0.0.1 -::1
# By default, outgoing connections (from replica to master, from Sentinel to
# instances, cluster bus, etc.) are not bound to a specific local address. In
# most cases, this means the operating system will handle that based on routing
# and the interface through which the connection goes out.
#
# Using bind-source-addr it is possible to configure a specific address to bind
# to, which may also affect how the connection gets routed.
#
# Example:
#
# bind-source-addr 10.0.0.1
# Protected mode is a layer of security protection, in order to avoid that
# Redis instances left open on the internet are accessed and exploited.
#
# When protected mode is on and the default user has no password, the server
# only accepts local connections from the IPv4 address (127.0.0.1), IPv6 address
# (::1) or Unix domain sockets.
#
# By default protected mode is enabled. You should disable it only if
# you are sure you want clients from other hosts to connect to Redis
# even if no authentication is configured.
protected-mode no
# Redis uses default hardened security configuration directives to reduce the
# attack surface on innocent users. Therefore, several sensitive configuration
# directives are immutable, and some potentially-dangerous commands are blocked.
#
# Configuration directives that control files that Redis writes to (e.g., 'dir'
# and 'dbfilename') and that aren't usually modified during runtime
# are protected by making them immutable.
#
# Commands that can increase the attack surface of Redis and that aren't usually
# called by users are blocked by default.
#
# These can be exposed to either all connections or just local ones by setting
# each of the configs listed below to either of these values:
#
# no - Block for any connection (remain immutable)
# yes - Allow for any connection (no protection)
# local - Allow only for local connections. Ones originating from the
# IPv4 address (127.0.0.1), IPv6 address (::1) or Unix domain sockets.
#
# enable-protected-configs no
# enable-debug-command no
# enable-module-command no
# Accept connections on the specified port, default is 6379 (IANA #815344).
# If port 0 is specified Redis will not listen on a TCP socket.
port 6379
# TCP listen() backlog.
#
# In high requests-per-second environments you need a high backlog in order
# to avoid slow clients connection issues. Note that the Linux kernel
# will silently truncate it to the value of /proc/sys/net/core/somaxconn so
# make sure to raise both the value of somaxconn and tcp_max_syn_backlog
# in order to get the desired effect.
tcp-backlog 511
# Unix socket.
#
# Specify the path for the Unix socket that will be used to listen for
# incoming connections. There is no default, so Redis will not listen
# on a unix socket when not specified.
#
# unixsocket /run/redis.sock
# unixsocketperm 700
# Close the connection after a client is idle for N seconds (0 to disable)
timeout 0
# TCP keepalive.
#
# If non-zero, use SO_KEEPALIVE to send TCP ACKs to clients in absence
# of communication. This is useful for two reasons:
#
# 1) Detect dead peers.
# 2) Force network equipment in the middle to consider the connection to be
# alive.
#
# On Linux, the specified value (in seconds) is the period used to send ACKs.
# Note that to close the connection the double of the time is needed.
# On other kernels the period depends on the kernel configuration.
#
# A reasonable value for this option is 300 seconds, which is the new
# Redis default starting with Redis 3.2.1.
tcp-keepalive 300
# Apply OS-specific mechanism to mark the listening socket with the specified
# ID, to support advanced routing and filtering capabilities.
#
# On Linux, the ID represents a connection mark.
# On FreeBSD, the ID represents a socket cookie ID.
# On OpenBSD, the ID represents a route table ID.
#
# The default value is 0, which implies no marking is required.
# socket-mark-id 0
################################# TLS/SSL #####################################
# By default, TLS/SSL is disabled. To enable it, the "tls-port" configuration
# directive can be used to define TLS-listening ports. To enable TLS on the
# default port, use:
#
# port 0
# tls-port 6379
# Configure a X.509 certificate and private key to use for authenticating the
# server to connected clients, masters or cluster peers. These files should be
# PEM formatted.
#
# tls-cert-file redis.crt
# tls-key-file redis.key
#
# If the key file is encrypted using a passphrase, it can be included here
# as well.
#
# tls-key-file-pass secret
# Normally Redis uses the same certificate for both server functions (accepting
# connections) and client functions (replicating from a master, establishing
# cluster bus connections, etc.).
#
# Sometimes certificates are issued with attributes that designate them as
# client-only or server-only certificates. In that case it may be desired to use
# different certificates for incoming (server) and outgoing (client)
# connections. To do that, use the following directives:
#
# tls-client-cert-file client.crt
# tls-client-key-file client.key
#
# If the key file is encrypted using a passphrase, it can be included here
# as well.
#
# tls-client-key-file-pass secret
# Configure a DH parameters file to enable Diffie-Hellman (DH) key exchange,
# required by older versions of OpenSSL (<3.0). Newer versions do not require
# this configuration and recommend against it.
#
# tls-dh-params-file redis.dh
# Configure a CA certificate(s) bundle or directory to authenticate TLS/SSL
# clients and peers. Redis requires an explicit configuration of at least one
# of these, and will not implicitly use the system wide configuration.
#
# tls-ca-cert-file ca.crt
# tls-ca-cert-dir /etc/ssl/certs
# By default, clients (including replica servers) on a TLS port are required
# to authenticate using valid client side certificates.
#
# If "no" is specified, client certificates are not required and not accepted.
# If "optional" is specified, client certificates are accepted and must be
# valid if provided, but are not required.
#
# tls-auth-clients no
# tls-auth-clients optional
# By default, a Redis replica does not attempt to establish a TLS connection
# with its master.
#
# Use the following directive to enable TLS on replication links.
#
# tls-replication yes
# By default, the Redis Cluster bus uses a plain TCP connection. To enable
# TLS for the bus protocol, use the following directive:
#
# tls-cluster yes
# By default, only TLSv1.2 and TLSv1.3 are enabled and it is highly recommended
# that older formally deprecated versions are kept disabled to reduce the attack surface.
# You can explicitly specify TLS versions to support.
# Allowed values are case insensitive and include "TLSv1", "TLSv1.1", "TLSv1.2",
# "TLSv1.3" (OpenSSL >= 1.1.1) or any combination.
# To enable only TLSv1.2 and TLSv1.3, use:
#
# tls-protocols "TLSv1.2 TLSv1.3"
# Configure allowed ciphers. See the ciphers(1ssl) manpage for more information
# about the syntax of this string.
#
# Note: this configuration applies only to <= TLSv1.2.
#
# tls-ciphers DEFAULT:!MEDIUM
# Configure allowed TLSv1.3 ciphersuites. See the ciphers(1ssl) manpage for more
# information about the syntax of this string, and specifically for TLSv1.3
# ciphersuites.
#
# tls-ciphersuites TLS_CHACHA20_POLY1305_SHA256
# When choosing a cipher, use the server's preference instead of the client
# preference. By default, the server follows the client's preference.
#
# tls-prefer-server-ciphers yes
# By default, TLS session caching is enabled to allow faster and less expensive
# reconnections by clients that support it. Use the following directive to disable
# caching.
#
# tls-session-caching no
# Change the default number of TLS sessions cached. A zero value sets the cache
# to unlimited size. The default size is 20480.
#
# tls-session-cache-size 5000
# Change the default timeout of cached TLS sessions. The default timeout is 300
# seconds.
#
# tls-session-cache-timeout 60
################################# GENERAL #####################################
# By default Redis does not run as a daemon. Use 'yes' if you need it.
# Note that Redis will write a pid file in /var/run/redis.pid when daemonized.
# When Redis is supervised by upstart or systemd, this parameter has no impact.
daemonize no
# If you run Redis from upstart or systemd, Redis can interact with your
# supervision tree. Options:
# supervised no - no supervision interaction
# supervised upstart - signal upstart by putting Redis into SIGSTOP mode
# requires "expect stop" in your upstart job config
# supervised systemd - signal systemd by writing READY=1 to $NOTIFY_SOCKET
# on startup, and updating Redis status on a regular
# basis.
# supervised auto - detect upstart or systemd method based on
# UPSTART_JOB or NOTIFY_SOCKET environment variables
# Note: these supervision methods only signal "process is ready."
# They do not enable continuous pings back to your supervisor.
#
# The default is "no". To run under upstart/systemd, you can simply uncomment
# the line below:
#
# supervised auto
# If a pid file is specified, Redis writes it where specified at startup
# and removes it at exit.
#
# When the server runs non daemonized, no pid file is created if none is
# specified in the configuration. When the server is daemonized, the pid file
# is used even if not specified, defaulting to "/var/run/redis.pid".
#
# Creating a pid file is best effort: if Redis is not able to create it
# nothing bad happens, the server will start and run normally.
#
# Note that on modern Linux systems "/run/redis.pid" is more conforming
# and should be used instead.
pidfile /var/run/redis_6379.pid
# Specify the server verbosity level.
# This can be one of:
# debug (a lot of information, useful for development/testing)
# verbose (many rarely useful info, but not a mess like the debug level)
# notice (moderately verbose, what you want in production probably)
# warning (only very important / critical messages are logged)
loglevel notice
# Specify the log file name. Also the empty string can be used to force
# Redis to log on the standard output. Note that if you use standard
# output for logging but daemonize, logs will be sent to /dev/null
logfile ""
# To enable logging to the system logger, just set 'syslog-enabled' to yes,
# and optionally update the other syslog parameters to suit your needs.
# syslog-enabled no
# Specify the syslog identity.
# syslog-ident redis
# Specify the syslog facility. Must be USER or between LOCAL0-LOCAL7.
# syslog-facility local0
# To disable the built in crash log, which will possibly produce cleaner core
# dumps when they are needed, uncomment the following:
#
# crash-log-enabled no
# To disable the fast memory check that's run as part of the crash log, which
# will possibly let redis terminate sooner, uncomment the following:
#
# crash-memcheck-enabled no
# Set the number of databases. The default database is DB 0, you can select
# a different one on a per-connection basis using SELECT where
# dbid is a number between 0 and 'databases'-1
databases 16
# By default Redis shows an ASCII art logo only when started to log to the
# standard output and if the standard output is a TTY and syslog logging is
# disabled. Basically this means that normally a logo is displayed only in
# interactive sessions.
#
# However it is possible to force the pre-4.0 behavior and always show a
# ASCII art logo in startup logs by setting the following option to yes.
always-show-logo no
# By default, Redis modifies the process title (as seen in 'top' and 'ps') to
# provide some runtime information. It is possible to disable this and leave
# the process name as executed by setting the following to no.
set-proc-title yes
# When changing the process title, Redis uses the following template to construct
# the modified title.
#
# Template variables are specified in curly brackets. The following variables are
# supported:
#
# {title} Name of process as executed if parent, or type of child process.
# {listen-addr} Bind address or '*' followed by TCP or TLS port listening on, or
# Unix socket if only that's available.
# {server-mode} Special mode, i.e. "[sentinel]" or "[cluster]".
# {port} TCP port listening on, or 0.
# {tls-port} TLS port listening on, or 0.
# {unixsocket} Unix domain socket listening on, or "".
# {config-file} Name of configuration file used.
#
proc-title-template "{title} {listen-addr} {server-mode}"
################################ SNAPSHOTTING ################################
# Save the DB to disk.
#
# save [ ...]
#
# Redis will save the DB if the given number of seconds elapsed and it
# surpassed the given number of write operations against the DB.
#
# Snapshotting can be completely disabled with a single empty string argument
# as in following example:
#
# save ""
#
# Unless specified otherwise, by default Redis will save the DB:
# * After 3600 seconds (an hour) if at least 1 change was performed
# * After 300 seconds (5 minutes) if at least 100 changes were performed
# * After 60 seconds if at least 10000 changes were performed
#
# You can set these explicitly by uncommenting the following line.
#
# save 3600 1 300 100 60 10000
# By default Redis will stop accepting writes if RDB snapshots are enabled
# (at least one save point) and the latest background save failed.
# This will make the user aware (in a hard way) that data is not persisting
# on disk properly, otherwise chances are that no one will notice and some
# disaster will happen.
#
# If the background saving process will start working again Redis will
# automatically allow writes again.
#
# However if you have setup your proper monitoring of the Redis server
# and persistence, you may want to disable this feature so that Redis will
# continue to work as usual even if there are problems with disk,
# permissions, and so forth.
stop-writes-on-bgsave-error yes
# Compress string objects using LZF when dump .rdb databases?
# By default compression is enabled as it's almost always a win.
# If you want to save some CPU in the saving child set it to 'no' but
# the dataset will likely be bigger if you have compressible values or keys.
rdbcompression yes
# Since version 5 of RDB a CRC64 checksum is placed at the end of the file.
# This makes the format more resistant to corruption but there is a performance
# hit to pay (around 10%) when saving and loading RDB files, so you can disable it
# for maximum performances.
#
# RDB files created with checksum disabled have a checksum of zero that will
# tell the loading code to skip the check.
rdbchecksum yes
# Enables or disables full sanitization checks for ziplist and listpack etc when
# loading an RDB or RESTORE payload. This reduces the chances of a assertion or
# crash later on while processing commands.
# Options:
# no - Never perform full sanitization
# yes - Always perform full sanitization
# clients - Perform full sanitization only for user connections.
# Excludes: RDB files, RESTORE commands received from the master
# connection, and client connections which have the
# skip-sanitize-payload ACL flag.
# The default should be 'clients' but since it currently affects cluster
# resharding via MIGRATE, it is temporarily set to 'no' by default.
#
# sanitize-dump-payload no
# The filename where to dump the DB
dbfilename dump.rdb
# Remove RDB files used by replication in instances without persistence
# enabled. By default this option is disabled, however there are environments
# where for regulations or other security concerns, RDB files persisted on
# disk by masters in order to feed replicas, or stored on disk by replicas
# in order to load them for the initial synchronization, should be deleted
# ASAP. Note that this option ONLY WORKS in instances that have both AOF
# and RDB persistence disabled, otherwise is completely ignored.
#
# An alternative (and sometimes better) way to obtain the same effect is
# to use diskless replication on both master and replicas instances. However
# in the case of replicas, diskless is not always an option.
rdb-del-sync-files no
# The working directory.
#
# The DB will be written inside this directory, with the filename specified
# above using the 'dbfilename' configuration directive.
#
# The Append Only File will also be created inside this directory.
#
# Note that you must specify a directory here, not a file name.
dir ./
################################# REPLICATION #################################
# Master-Replica replication. Use replicaof to make a Redis instance a copy of
# another Redis server. A few things to understand ASAP about Redis replication.
#
# +------------------+ +---------------+
# | Master | ---> | Replica |
# | (receive writes) | | (exact copy) |
# +------------------+ +---------------+
#
# 1) Redis replication is asynchronous, but you can configure a master to
# stop accepting writes if it appears to be not connected with at least
# a given number of replicas.
# 2) Redis replicas are able to perform a partial resynchronization with the
# master if the replication link is lost for a relatively small amount of
# time. You may want to configure the replication backlog size (see the next
# sections of this file) with a sensible value depending on your needs.
# 3) Replication is automatic and does not need user intervention. After a
# network partition replicas automatically try to reconnect to masters
# and resynchronize with them.
#
# replicaof
# If the master is password protected (using the "requirepass" configuration
# directive below) it is possible to tell the replica to authenticate before
# starting the replication synchronization process, otherwise the master will
# refuse the replica request.
#
# masterauth
#
# However this is not enough if you are using Redis ACLs (for Redis version
# 6 or greater), and the default user is not capable of running the PSYNC
# command and/or other commands needed for replication. In this case it's
# better to configure a special user to use with replication, and specify the
# masteruser configuration as such:
#
# masteruser
#
# When masteruser is specified, the replica will authenticate against its
# master using the new AUTH form: AUTH .
# When a replica loses its connection with the master, or when the replication
# is still in progress, the replica can act in two different ways:
#
# 1) if replica-serve-stale-data is set to 'yes' (the default) the replica will
# still reply to client requests, possibly with out of date data, or the
# data set may just be empty if this is the first synchronization.
#
# 2) If replica-serve-stale-data is set to 'no' the replica will reply with error
# "MASTERDOWN Link with MASTER is down and replica-serve-stale-data is set to 'no'"
# to all data access commands, excluding commands such as:
# INFO, REPLICAOF, AUTH, SHUTDOWN, REPLCONF, ROLE, CONFIG, SUBSCRIBE,
# UNSUBSCRIBE, PSUBSCRIBE, PUNSUBSCRIBE, PUBLISH, PUBSUB, COMMAND, POST,
# HOST and LATENCY.
#
replica-serve-stale-data yes
# You can configure a replica instance to accept writes or not. Writing against
# a replica instance may be useful to store some ephemeral data (because data
# written on a replica will be easily deleted after resync with the master) but
# may also cause problems if clients are writing to it because of a
# misconfiguration.
#
# Since Redis 2.6 by default replicas are read-only.
#
# Note: read only replicas are not designed to be exposed to untrusted clients
# on the internet. It's just a protection layer against misuse of the instance.
# Still a read only replica exports by default all the administrative commands
# such as CONFIG, DEBUG, and so forth. To a limited extent you can improve
# security of read only replicas using 'rename-command' to shadow all the
# administrative / dangerous commands.
replica-read-only yes
# Replication SYNC strategy: disk or socket.
#
# New replicas and reconnecting replicas that are not able to continue the
# replication process just receiving differences, need to do what is called a
# "full synchronization". An RDB file is transmitted from the master to the
# replicas.
#
# The transmission can happen in two different ways:
#
# 1) Disk-backed: The Redis master creates a new process that writes the RDB
# file on disk. Later the file is transferred by the parent
# process to the replicas incrementally.
# 2) Diskless: The Redis master creates a new process that directly writes the
# RDB file to replica sockets, without touching the disk at all.
#
# With disk-backed replication, while the RDB file is generated, more replicas
# can be queued and served with the RDB file as soon as the current child
# producing the RDB file finishes its work. With diskless replication instead
# once the transfer starts, new replicas arriving will be queued and a new
# transfer will start when the current one terminates.
#
# When diskless replication is used, the master waits a configurable amount of
# time (in seconds) before starting the transfer in the hope that multiple
# replicas will arrive and the transfer can be parallelized.
#
# With slow disks and fast (large bandwidth) networks, diskless replication
# works better.
repl-diskless-sync yes
# When diskless replication is enabled, it is possible to configure the delay
# the server waits in order to spawn the child that transfers the RDB via socket
# to the replicas.
#
# This is important since once the transfer starts, it is not possible to serve
# new replicas arriving, that will be queued for the next RDB transfer, so the
# server waits a delay in order to let more replicas arrive.
#
# The delay is specified in seconds, and by default is 5 seconds. To disable
# it entirely just set it to 0 seconds and the transfer will start ASAP.
repl-diskless-sync-delay 5
# When diskless replication is enabled with a delay, it is possible to let
# the replication start before the maximum delay is reached if the maximum
# number of replicas expected have connected. Default of 0 means that the
# maximum is not defined and Redis will wait the full delay.
repl-diskless-sync-max-replicas 0
# -----------------------------------------------------------------------------
# WARNING: RDB diskless load is experimental. Since in this setup the replica
# does not immediately store an RDB on disk, it may cause data loss during
# failovers. RDB diskless load + Redis modules not handling I/O reads may also
# cause Redis to abort in case of I/O errors during the initial synchronization
# stage with the master. Use only if you know what you are doing.
# -----------------------------------------------------------------------------
#
# Replica can load the RDB it reads from the replication link directly from the
# socket, or store the RDB to a file and read that file after it was completely
# received from the master.
#
# In many cases the disk is slower than the network, and storing and loading
# the RDB file may increase replication time (and even increase the master's
# Copy on Write memory and replica buffers).
# However, parsing the RDB file directly from the socket may mean that we have
# to flush the contents of the current database before the full rdb was
# received. For this reason we have the following options:
#
# "disabled" - Don't use diskless load (store the rdb file to the disk first)
# "on-empty-db" - Use diskless load only when it is completely safe.
# "swapdb" - Keep current db contents in RAM while parsing the data directly
# from the socket. Replicas in this mode can keep serving current
# data set while replication is in progress, except for cases where
# they can't recognize master as having a data set from same
# replication history.
# Note that this requires sufficient memory, if you don't have it,
# you risk an OOM kill.
repl-diskless-load disabled
# Master send PINGs to its replicas in a predefined interval. It's possible to
# change this interval with the repl_ping_replica_period option. The default
# value is 10 seconds.
#
# repl-ping-replica-period 10
# The following option sets the replication timeout for:
#
# 1) Bulk transfer I/O during SYNC, from the point of view of replica.
# 2) Master timeout from the point of view of replicas (data, pings).
# 3) Replica timeout from the point of view of masters (REPLCONF ACK pings).
#
# It is important to make sure that this value is greater than the value
# specified for repl-ping-replica-period otherwise a timeout will be detected
# every time there is low traffic between the master and the replica. The default
# value is 60 seconds.
#
# repl-timeout 60
# Disable TCP_NODELAY on the replica socket after SYNC?
#
# If you select "yes" Redis will use a smaller number of TCP packets and
# less bandwidth to send data to replicas. But this can add a delay for
# the data to appear on the replica side, up to 40 milliseconds with
# Linux kernels using a default configuration.
#
# If you select "no" the delay for data to appear on the replica side will
# be reduced but more bandwidth will be used for replication.
#
# By default we optimize for low latency, but in very high traffic conditions
# or when the master and replicas are many hops away, turning this to "yes" may
# be a good idea.
repl-disable-tcp-nodelay no
# Set the replication backlog size. The backlog is a buffer that accumulates
# replica data when replicas are disconnected for some time, so that when a
# replica wants to reconnect again, often a full resync is not needed, but a
# partial resync is enough, just passing the portion of data the replica
# missed while disconnected.
#
# The bigger the replication backlog, the longer the replica can endure the
# disconnect and later be able to perform a partial resynchronization.
#
# The backlog is only allocated if there is at least one replica connected.
#
# repl-backlog-size 1mb
# After a master has no connected replicas for some time, the backlog will be
# freed. The following option configures the amount of seconds that need to
# elapse, starting from the time the last replica disconnected, for the backlog
# buffer to be freed.
#
# Note that replicas never free the backlog for timeout, since they may be
# promoted to masters later, and should be able to correctly "partially
# resynchronize" with other replicas: hence they should always accumulate backlog.
#
# A value of 0 means to never release the backlog.
#
# repl-backlog-ttl 3600
# The replica priority is an integer number published by Redis in the INFO
# output. It is used by Redis Sentinel in order to select a replica to promote
# into a master if the master is no longer working correctly.
#
# A replica with a low priority number is considered better for promotion, so
# for instance if there are three replicas with priority 10, 100, 25 Sentinel
# will pick the one with priority 10, that is the lowest.
#
# However a special priority of 0 marks the replica as not able to perform the
# role of master, so a replica with priority of 0 will never be selected by
# Redis Sentinel for promotion.
#
# By default the priority is 100.
replica-priority 100
# The propagation error behavior controls how Redis will behave when it is
# unable to handle a command being processed in the replication stream from a master
# or processed while reading from an AOF file. Errors that occur during propagation
# are unexpected, and can cause data inconsistency. However, there are edge cases
# in earlier versions of Redis where it was possible for the server to replicate or persist
# commands that would fail on future versions. For this reason the default behavior
# is to ignore such errors and continue processing commands.
#
# If an application wants to ensure there is no data divergence, this configuration
# should be set to 'panic' instead. The value can also be set to 'panic-on-replicas'
# to only panic when a replica encounters an error on the replication stream. One of
# these two panic values will become the default value in the future once there are
# sufficient safety mechanisms in place to prevent false positive crashes.
#
# propagation-error-behavior ignore
# Replica ignore disk write errors controls the behavior of a replica when it is
# unable to persist a write command received from its master to disk. By default,
# this configuration is set to 'no' and will crash the replica in this condition.
# It is not recommended to change this default, however in order to be compatible
# with older versions of Redis this config can be toggled to 'yes' which will just
# log a warning and execute the write command it got from the master.
#
# replica-ignore-disk-write-errors no
# -----------------------------------------------------------------------------
# By default, Redis Sentinel includes all replicas in its reports. A replica
# can be excluded from Redis Sentinel's announcements. An unannounced replica
# will be ignored by the 'sentinel replicas ' command and won't be
# exposed to Redis Sentinel's clients.
#
# This option does not change the behavior of replica-priority. Even with
# replica-announced set to 'no', the replica can be promoted to master. To
# prevent this behavior, set replica-priority to 0.
#
# replica-announced yes
# It is possible for a master to stop accepting writes if there are less than
# N replicas connected, having a lag less or equal than M seconds.
#
# The N replicas need to be in "online" state.
#
# The lag in seconds, that must be <= the specified value, is calculated from
# the last ping received from the replica, that is usually sent every second.
#
# This option does not GUARANTEE that N replicas will accept the write, but
# will limit the window of exposure for lost writes in case not enough replicas
# are available, to the specified number of seconds.
#
# For example to require at least 3 replicas with a lag <= 10 seconds use:
#
# min-replicas-to-write 3
# min-replicas-max-lag 10
#
# Setting one or the other to 0 disables the feature.
#
# By default min-replicas-to-write is set to 0 (feature disabled) and
# min-replicas-max-lag is set to 10.
# A Redis master is able to list the address and port of the attached
# replicas in different ways. For example the "INFO replication" section
# offers this information, which is used, among other tools, by
# Redis Sentinel in order to discover replica instances.
# Another place where this info is available is in the output of the
# "ROLE" command of a master.
#
# The listed IP address and port normally reported by a replica is
# obtained in the following way:
#
# IP: The address is auto detected by checking the peer address
# of the socket used by the replica to connect with the master.
#
# Port: The port is communicated by the replica during the replication
# handshake, and is normally the port that the replica is using to
# listen for connections.
#
# However when port forwarding or Network Address Translation (NAT) is
# used, the replica may actually be reachable via different IP and port
# pairs. The following two options can be used by a replica in order to
# report to its master a specific set of IP and port, so that both INFO
# and ROLE will report those values.
#
# There is no need to use both the options if you need to override just
# the port or the IP address.
#
# replica-announce-ip 5.5.5.5
# replica-announce-port 1234
############################### KEYS TRACKING #################################
# Redis implements server assisted support for client side caching of values.
# This is implemented using an invalidation table that remembers, using
# a radix key indexed by key name, what clients have which keys. In turn
# this is used in order to send invalidation messages to clients. Please
# check this page to understand more about the feature:
#
# https://redis.io/topics/client-side-caching
#
# When tracking is enabled for a client, all the read only queries are assumed
# to be cached: this will force Redis to store information in the invalidation
# table. When keys are modified, such information is flushed away, and
# invalidation messages are sent to the clients. However if the workload is
# heavily dominated by reads, Redis could use more and more memory in order
# to track the keys fetched by many clients.
#
# For this reason it is possible to configure a maximum fill value for the
# invalidation table. By default it is set to 1M of keys, and once this limit
# is reached, Redis will start to evict keys in the invalidation table
# even if they were not modified, just to reclaim memory: this will in turn
# force the clients to invalidate the cached values. Basically the table
# maximum size is a trade off between the memory you want to spend server
# side to track information about who cached what, and the ability of clients
# to retain cached objects in memory.
#
# If you set the value to 0, it means there are no limits, and Redis will
# retain as many keys as needed in the invalidation table.
# In the "stats" INFO section, you can find information about the number of
# keys in the invalidation table at every given moment.
#
# Note: when key tracking is used in broadcasting mode, no memory is used
# in the server side so this setting is useless.
#
# tracking-table-max-keys 1000000
################################## SECURITY ###################################
# Warning: since Redis is pretty fast, an outside user can try up to
# 1 million passwords per second against a modern box. This means that you
# should use very strong passwords, otherwise they will be very easy to break.
# Note that because the password is really a shared secret between the client
# and the server, and should not be memorized by any human, the password
# can be easily a long string from /dev/urandom or whatever, so by using a
# long and unguessable password no brute force attack will be possible.
# Redis ACL users are defined in the following format:
#
# user ... acl rules ...
#
# For example:
#
# user worker +@list +@connection ~jobs:* on >ffa9203c493aa99
#
# The special username "default" is used for new connections. If this user
# has the "nopass" rule, then new connections will be immediately authenticated
# as the "default" user without the need of any password provided via the
# AUTH command. Otherwise if the "default" user is not flagged with "nopass"
# the connections will start in not authenticated state, and will require
# AUTH (or the HELLO command AUTH option) in order to be authenticated and
# start to work.
#
# The ACL rules that describe what a user can do are the following:
#
# on Enable the user: it is possible to authenticate as this user.
# off Disable the user: it's no longer possible to authenticate
# with this user, however the already authenticated connections
# will still work.
# skip-sanitize-payload RESTORE dump-payload sanitization is skipped.
# sanitize-payload RESTORE dump-payload is sanitized (default).
# + Allow the execution of that command.
# May be used with `|` for allowing subcommands (e.g "+config|get")
# - Disallow the execution of that command.
# May be used with `|` for blocking subcommands (e.g "-config|set")
# +@ Allow the execution of all the commands in such category
# with valid categories are like @admin, @set, @sortedset, ...
# and so forth, see the full list in the server.c file where
# the Redis command table is described and defined.
# The special category @all means all the commands, but currently
# present in the server, and that will be loaded in the future
# via modules.
# +|first-arg Allow a specific first argument of an otherwise
# disabled command. It is only supported on commands with
# no sub-commands, and is not allowed as negative form
# like -SELECT|1, only additive starting with "+". This
# feature is deprecated and may be removed in the future.
# allcommands Alias for +@all. Note that it implies the ability to execute
# all the future commands loaded via the modules system.
# nocommands Alias for -@all.
# ~ Add a pattern of keys that can be mentioned as part of
# commands. For instance ~* allows all the keys. The pattern
# is a glob-style pattern like the one of KEYS.
# It is possible to specify multiple patterns.
# %R~ Add key read pattern that specifies which keys can be read
# from.
# %W~ Add key write pattern that specifies which keys can be
# written to.
# allkeys Alias for ~*
# resetkeys Flush the list of allowed keys patterns.
# & Add a glob-style pattern of Pub/Sub channels that can be
# accessed by the user. It is possible to specify multiple channel
# patterns.
# allchannels Alias for &*
# resetchannels Flush the list of allowed channel patterns.
# > Add this password to the list of valid password for the user.
# For example >mypass will add "mypass" to the list.
# This directive clears the "nopass" flag (see later).
# < Remove this password from the list of valid passwords.
# nopass All the set passwords of the user are removed, and the user
# is flagged as requiring no password: it means that every
# password will work against this user. If this directive is
# used for the default user, every new connection will be
# immediately authenticated with the default user without
# any explicit AUTH command required. Note that the "resetpass"
# directive will clear this condition.
# resetpass Flush the list of allowed passwords. Moreover removes the
# "nopass" status. After "resetpass" the user has no associated
# passwords and there is no way to authenticate without adding
# some password (or setting it as "nopass" later).
# reset Performs the following actions: resetpass, resetkeys, off,
# -@all. The user returns to the same state it has immediately
# after its creation.
# () Create a new selector with the options specified within the
# parentheses and attach it to the user. Each option should be
# space separated. The first character must be ( and the last
# character must be ).
# clearselectors Remove all of the currently attached selectors.
# Note this does not change the "root" user permissions,
# which are the permissions directly applied onto the
# user (outside the parentheses).
#
# ACL rules can be specified in any order: for instance you can start with
# passwords, then flags, or key patterns. However note that the additive
# and subtractive rules will CHANGE MEANING depending on the ordering.
# For instance see the following example:
#
# user alice on +@all -DEBUG ~* >somepassword
#
# This will allow "alice" to use all the commands with the exception of the
# DEBUG command, since +@all added all the commands to the set of the commands
# alice can use, and later DEBUG was removed. However if we invert the order
# of two ACL rules the result will be different:
#
# user alice on -DEBUG +@all ~* >somepassword
#
# Now DEBUG was removed when alice had yet no commands in the set of allowed
# commands, later all the commands are added, so the user will be able to
# execute everything.
#
# Basically ACL rules are processed left-to-right.
#
# The following is a list of command categories and their meanings:
# * keyspace - Writing or reading from keys, databases, or their metadata
# in a type agnostic way. Includes DEL, RESTORE, DUMP, RENAME, EXISTS, DBSIZE,
# KEYS, EXPIRE, TTL, FLUSHALL, etc. Commands that may modify the keyspace,
# key or metadata will also have `write` category. Commands that only read
# the keyspace, key or metadata will have the `read` category.
# * read - Reading from keys (values or metadata). Note that commands that don't
# interact with keys, will not have either `read` or `write`.
# * write - Writing to keys (values or metadata)
# * admin - Administrative commands. Normal applications will never need to use
# these. Includes REPLICAOF, CONFIG, DEBUG, SAVE, MONITOR, ACL, SHUTDOWN, etc.
# * dangerous - Potentially dangerous (each should be considered with care for
# various reasons). This includes FLUSHALL, MIGRATE, RESTORE, SORT, KEYS,
# CLIENT, DEBUG, INFO, CONFIG, SAVE, REPLICAOF, etc.
# * connection - Commands affecting the connection or other connections.
# This includes AUTH, SELECT, COMMAND, CLIENT, ECHO, PING, etc.
# * blocking - Potentially blocking the connection until released by another
# command.
# * fast - Fast O(1) commands. May loop on the number of arguments, but not the
# number of elements in the key.
# * slow - All commands that are not Fast.
# * pubsub - PUBLISH / SUBSCRIBE related
# * transaction - WATCH / MULTI / EXEC related commands.
# * scripting - Scripting related.
# * set - Data type: sets related.
# * sortedset - Data type: zsets related.
# * list - Data type: lists related.
# * hash - Data type: hashes related.
# * string - Data type: strings related.
# * bitmap - Data type: bitmaps related.
# * hyperloglog - Data type: hyperloglog related.
# * geo - Data type: geo related.
# * stream - Data type: streams related.
#
# For more information about ACL configuration please refer to
# the Redis web site at https://redis.io/topics/acl
# ACL LOG
#
# The ACL Log tracks failed commands and authentication events associated
# with ACLs. The ACL Log is useful to troubleshoot failed commands blocked
# by ACLs. The ACL Log is stored in memory. You can reclaim memory with
# ACL LOG RESET. Define the maximum entry length of the ACL Log below.
acllog-max-len 128
# Using an external ACL file
#
# Instead of configuring users here in this file, it is possible to use
# a stand-alone file just listing users. The two methods cannot be mixed:
# if you configure users here and at the same time you activate the external
# ACL file, the server will refuse to start.
#
# The format of the external ACL user file is exactly the same as the
# format that is used inside redis.conf to describe users.
#
# aclfile /etc/redis/users.acl
# IMPORTANT NOTE: starting with Redis 6 "requirepass" is just a compatibility
# layer on top of the new ACL system. The option effect will be just setting
# the password for the default user. Clients will still authenticate using
# AUTH as usually, or more explicitly with AUTH default
# if they follow the new protocol: both will work.
#
# The requirepass is not compatible with aclfile option and the ACL LOAD
# command, these will cause requirepass to be ignored.
#
# requirepass foobared
# New users are initialized with restrictive permissions by default, via the
# equivalent of this ACL rule 'off resetkeys -@all'. Starting with Redis 6.2, it
# is possible to manage access to Pub/Sub channels with ACL rules as well. The
# default Pub/Sub channels permission if new users is controlled by the
# acl-pubsub-default configuration directive, which accepts one of these values:
#
# allchannels: grants access to all Pub/Sub channels
# resetchannels: revokes access to all Pub/Sub channels
#
# From Redis 7.0, acl-pubsub-default defaults to 'resetchannels' permission.
#
# acl-pubsub-default resetchannels
# Command renaming (DEPRECATED).
#
# ------------------------------------------------------------------------
# WARNING: avoid using this option if possible. Instead use ACLs to remove
# commands from the default user, and put them only in some admin user you
# create for administrative purposes.
# ------------------------------------------------------------------------
#
# It is possible to change the name of dangerous commands in a shared
# environment. For instance the CONFIG command may be renamed into something
# hard to guess so that it will still be available for internal-use tools
# but not available for general clients.
#
# Example:
#
# rename-command CONFIG b840fc02d524045429941cc15f59e41cb7be6c52
#
# It is also possible to completely kill a command by renaming it into
# an empty string:
#
# rename-command CONFIG ""
#
# Please note that changing the name of commands that are logged into the
# AOF file or transmitted to replicas may cause problems.
################################### CLIENTS ####################################
# Set the max number of connected clients at the same time. By default
# this limit is set to 10000 clients, however if the Redis server is not
# able to configure the process file limit to allow for the specified limit
# the max number of allowed clients is set to the current file limit
# minus 32 (as Redis reserves a few file descriptors for internal uses).
#
# Once the limit is reached Redis will close all the new connections sending
# an error 'max number of clients reached'.
#
# IMPORTANT: When Redis Cluster is used, the max number of connections is also
# shared with the cluster bus: every node in the cluster will use two
# connections, one incoming and another outgoing. It is important to size the
# limit accordingly in case of very large clusters.
#
# maxclients 10000
############################## MEMORY MANAGEMENT ################################
# Set a memory usage limit to the specified amount of bytes.
# When the memory limit is reached Redis will try to remove keys
# according to the eviction policy selected (see maxmemory-policy).
#
# If Redis can't remove keys according to the policy, or if the policy is
# set to 'noeviction', Redis will start to reply with errors to commands
# that would use more memory, like SET, LPUSH, and so on, and will continue
# to reply to read-only commands like GET.
#
# This option is usually useful when using Redis as an LRU or LFU cache, or to
# set a hard memory limit for an instance (using the 'noeviction' policy).
#
# WARNING: If you have replicas attached to an instance with maxmemory on,
# the size of the output buffers needed to feed the replicas are subtracted
# from the used memory count, so that network problems / resyncs will
# not trigger a loop where keys are evicted, and in turn the output
# buffer of replicas is full with DELs of keys evicted triggering the deletion
# of more keys, and so forth until the database is completely emptied.
#
# In short... if you have replicas attached it is suggested that you set a lower
# limit for maxmemory so that there is some free RAM on the system for replica
# output buffers (but this is not needed if the policy is 'noeviction').
#
# maxmemory
# MAXMEMORY POLICY: how Redis will select what to remove when maxmemory
# is reached. You can select one from the following behaviors:
#
# volatile-lru -> Evict using approximated LRU, only keys with an expire set.
# allkeys-lru -> Evict any key using approximated LRU.
# volatile-lfu -> Evict using approximated LFU, only keys with an expire set.
# allkeys-lfu -> Evict any key using approximated LFU.
# volatile-random -> Remove a random key having an expire set.
# allkeys-random -> Remove a random key, any key.
# volatile-ttl -> Remove the key with the nearest expire time (minor TTL)
# noeviction -> Don't evict anything, just return an error on write operations.
#
# LRU means Least Recently Used
# LFU means Least Frequently Used
#
# Both LRU, LFU and volatile-ttl are implemented using approximated
# randomized algorithms.
#
# Note: with any of the above policies, when there are no suitable keys for
# eviction, Redis will return an error on write operations that require
# more memory. These are usually commands that create new keys, add data or
# modify existing keys. A few examples are: SET, INCR, HSET, LPUSH, SUNIONSTORE,
# SORT (due to the STORE argument), and EXEC (if the transaction includes any
# command that requires memory).
#
# The default is:
#
# maxmemory-policy noeviction
# LRU, LFU and minimal TTL algorithms are not precise algorithms but approximated
# algorithms (in order to save memory), so you can tune it for speed or
# accuracy. By default Redis will check five keys and pick the one that was
# used least recently, you can change the sample size using the following
# configuration directive.
#
# The default of 5 produces good enough results. 10 Approximates very closely
# true LRU but costs more CPU. 3 is faster but not very accurate.
#
# maxmemory-samples 5
# Eviction processing is designed to function well with the default setting.
# If there is an unusually large amount of write traffic, this value may need to
# be increased. Decreasing this value may reduce latency at the risk of
# eviction processing effectiveness
# 0 = minimum latency, 10 = default, 100 = process without regard to latency
#
# maxmemory-eviction-tenacity 10
# Starting from Redis 5, by default a replica will ignore its maxmemory setting
# (unless it is promoted to master after a failover or manually). It means
# that the eviction of keys will be just handled by the master, sending the
# DEL commands to the replica as keys evict in the master side.
#
# This behavior ensures that masters and replicas stay consistent, and is usually
# what you want, however if your replica is writable, or you want the replica
# to have a different memory setting, and you are sure all the writes performed
# to the replica are idempotent, then you may change this default (but be sure
# to understand what you are doing).
#
# Note that since the replica by default does not evict, it may end using more
# memory than the one set via maxmemory (there are certain buffers that may
# be larger on the replica, or data structures may sometimes take more memory
# and so forth). So make sure you monitor your replicas and make sure they
# have enough memory to never hit a real out-of-memory condition before the
# master hits the configured maxmemory setting.
#
# replica-ignore-maxmemory yes
# Redis reclaims expired keys in two ways: upon access when those keys are
# found to be expired, and also in background, in what is called the
# "active expire key". The key space is slowly and interactively scanned
# looking for expired keys to reclaim, so that it is possible to free memory
# of keys that are expired and will never be accessed again in a short time.
#
# The default effort of the expire cycle will try to avoid having more than
# ten percent of expired keys still in memory, and will try to avoid consuming
# more than 25% of total memory and to add latency to the system. However
# it is possible to increase the expire "effort" that is normally set to
# "1", to a greater value, up to the value "10". At its maximum value the
# system will use more CPU, longer cycles (and technically may introduce
# more latency), and will tolerate less already expired keys still present
# in the system. It's a tradeoff between memory, CPU and latency.
#
# active-expire-effort 1
############################# LAZY FREEING ####################################
# Redis has two primitives to delete keys. One is called DEL and is a blocking
# deletion of the object. It means that the server stops processing new commands
# in order to reclaim all the memory associated with an object in a synchronous
# way. If the key deleted is associated with a small object, the time needed
# in order to execute the DEL command is very small and comparable to most other
# O(1) or O(log_N) commands in Redis. However if the key is associated with an
# aggregated value containing millions of elements, the server can block for
# a long time (even seconds) in order to complete the operation.
#
# For the above reasons Redis also offers non blocking deletion primitives
# such as UNLINK (non blocking DEL) and the ASYNC option of FLUSHALL and
# FLUSHDB commands, in order to reclaim memory in background. Those commands
# are executed in constant time. Another thread will incrementally free the
# object in the background as fast as possible.
#
# DEL, UNLINK and ASYNC option of FLUSHALL and FLUSHDB are user-controlled.
# It's up to the design of the application to understand when it is a good
# idea to use one or the other. However the Redis server sometimes has to
# delete keys or flush the whole database as a side effect of other operations.
# Specifically Redis deletes objects independently of a user call in the
# following scenarios:
#
# 1) On eviction, because of the maxmemory and maxmemory policy configurations,
# in order to make room for new data, without going over the specified
# memory limit.
# 2) Because of expire: when a key with an associated time to live (see the
# EXPIRE command) must be deleted from memory.
# 3) Because of a side effect of a command that stores data on a key that may
# already exist. For example the RENAME command may delete the old key
# content when it is replaced with another one. Similarly SUNIONSTORE
# or SORT with STORE option may delete existing keys. The SET command
# itself removes any old content of the specified key in order to replace
# it with the specified string.
# 4) During replication, when a replica performs a full resynchronization with
# its master, the content of the whole database is removed in order to
# load the RDB file just transferred.
#
# In all the above cases the default is to delete objects in a blocking way,
# like if DEL was called. However you can configure each case specifically
# in order to instead release memory in a non-blocking way like if UNLINK
# was called, using the following configuration directives.
lazyfree-lazy-eviction no
lazyfree-lazy-expire no
lazyfree-lazy-server-del no
replica-lazy-flush no
# It is also possible, for the case when to replace the user code DEL calls
# with UNLINK calls is not easy, to modify the default behavior of the DEL
# command to act exactly like UNLINK, using the following configuration
# directive:
lazyfree-lazy-user-del no
# FLUSHDB, FLUSHALL, SCRIPT FLUSH and FUNCTION FLUSH support both asynchronous and synchronous
# deletion, which can be controlled by passing the [SYNC|ASYNC] flags into the
# commands. When neither flag is passed, this directive will be used to determine
# if the data should be deleted asynchronously.
lazyfree-lazy-user-flush no
################################ THREADED I/O #################################
# Redis is mostly single threaded, however there are certain threaded
# operations such as UNLINK, slow I/O accesses and other things that are
# performed on side threads.
#
# Now it is also possible to handle Redis clients socket reads and writes
# in different I/O threads. Since especially writing is so slow, normally
# Redis users use pipelining in order to speed up the Redis performances per
# core, and spawn multiple instances in order to scale more. Using I/O
# threads it is possible to easily speedup two times Redis without resorting
# to pipelining nor sharding of the instance.
#
# By default threading is disabled, we suggest enabling it only in machines
# that have at least 4 or more cores, leaving at least one spare core.
# Using more than 8 threads is unlikely to help much. We also recommend using
# threaded I/O only if you actually have performance problems, with Redis
# instances being able to use a quite big percentage of CPU time, otherwise
# there is no point in using this feature.
#
# So for instance if you have a four cores boxes, try to use 2 or 3 I/O
# threads, if you have a 8 cores, try to use 6 threads. In order to
# enable I/O threads use the following configuration directive:
#
# io-threads 4
#
# Setting io-threads to 1 will just use the main thread as usual.
# When I/O threads are enabled, we only use threads for writes, that is
# to thread the write(2) syscall and transfer the client buffers to the
# socket. However it is also possible to enable threading of reads and
# protocol parsing using the following configuration directive, by setting
# it to yes:
#
# io-threads-do-reads no
#
# Usually threading reads doesn't help much.
#
# NOTE 1: This configuration directive cannot be changed at runtime via
# CONFIG SET. Also, this feature currently does not work when SSL is
# enabled.
#
# NOTE 2: If you want to test the Redis speedup using redis-benchmark, make
# sure you also run the benchmark itself in threaded mode, using the
# --threads option to match the number of Redis threads, otherwise you'll not
# be able to notice the improvements.
############################ KERNEL OOM CONTROL ##############################
# On Linux, it is possible to hint the kernel OOM killer on what processes
# should be killed first when out of memory.
#
# Enabling this feature makes Redis actively control the oom_score_adj value
# for all its processes, depending on their role. The default scores will
# attempt to have background child processes killed before all others, and
# replicas killed before masters.
#
# Redis supports these options:
#
# no: Don't make changes to oom-score-adj (default).
# yes: Alias to "relative" see below.
# absolute: Values in oom-score-adj-values are written as is to the kernel.
# relative: Values are used relative to the initial value of oom_score_adj when
# the server starts and are then clamped to a range of -1000 to 1000.
# Because typically the initial value is 0, they will often match the
# absolute values.
oom-score-adj no
# When oom-score-adj is used, this directive controls the specific values used
# for master, replica and background child processes. Values range -2000 to
# 2000 (higher means more likely to be killed).
#
# Unprivileged processes (not root, and without CAP_SYS_RESOURCE capabilities)
# can freely increase their value, but not decrease it below its initial
# settings. This means that setting oom-score-adj to "relative" and setting the
# oom-score-adj-values to positive values will always succeed.
oom-score-adj-values 0 200 800
#################### KERNEL transparent hugepage CONTROL ######################
# Usually the kernel Transparent Huge Pages control is set to "madvise" or
# or "never" by default (/sys/kernel/mm/transparent_hugepage/enabled), in which
# case this config has no effect. On systems in which it is set to "always",
# redis will attempt to disable it specifically for the redis process in order
# to avoid latency problems specifically with fork(2) and CoW.
# If for some reason you prefer to keep it enabled, you can set this config to
# "no" and the kernel global to "always".
disable-thp yes
############################## APPEND ONLY MODE ###############################
# By default Redis asynchronously dumps the dataset on disk. This mode is
# good enough in many applications, but an issue with the Redis process or
# a power outage may result into a few minutes of writes lost (depending on
# the configured save points).
#
# The Append Only File is an alternative persistence mode that provides
# much better durability. For instance using the default data fsync policy
# (see later in the config file) Redis can lose just one second of writes in a
# dramatic event like a server power outage, or a single write if something
# wrong with the Redis process itself happens, but the operating system is
# still running correctly.
#
# AOF and RDB persistence can be enabled at the same time without problems.
# If the AOF is enabled on startup Redis will load the AOF, that is the file
# with the better durability guarantees.
#
# Please check https://redis.io/topics/persistence for more information.
appendonly no
# The base name of the append only file.
#
# Redis 7 and newer use a set of append-only files to persist the dataset
# and changes applied to it. There are two basic types of files in use:
#
# - Base files, which are a snapshot representing the complete state of the
# dataset at the time the file was created. Base files can be either in
# the form of RDB (binary serialized) or AOF (textual commands).
# - Incremental files, which contain additional commands that were applied
# to the dataset following the previous file.
#
# In addition, manifest files are used to track the files and the order in
# which they were created and should be applied.
#
# Append-only file names are created by Redis following a specific pattern.
# The file name's prefix is based on the 'appendfilename' configuration
# parameter, followed by additional information about the sequence and type.
#
# For example, if appendfilename is set to appendonly.aof, the following file
# names could be derived:
#
# - appendonly.aof.1.base.rdb as a base file.
# - appendonly.aof.1.incr.aof, appendonly.aof.2.incr.aof as incremental files.
# - appendonly.aof.manifest as a manifest file.
appendfilename "appendonly.aof"
# For convenience, Redis stores all persistent append-only files in a dedicated
# directory. The name of the directory is determined by the appenddirname
# configuration parameter.
appenddirname "appendonlydir"
# The fsync() call tells the Operating System to actually write data on disk
# instead of waiting for more data in the output buffer. Some OS will really flush
# data on disk, some other OS will just try to do it ASAP.
#
# Redis supports three different modes:
#
# no: don't fsync, just let the OS flush the data when it wants. Faster.
# always: fsync after every write to the append only log. Slow, Safest.
# everysec: fsync only one time every second. Compromise.
#
# The default is "everysec", as that's usually the right compromise between
# speed and data safety. It's up to you to understand if you can relax this to
# "no" that will let the operating system flush the output buffer when
# it wants, for better performances (but if you can live with the idea of
# some data loss consider the default persistence mode that's snapshotting),
# or on the contrary, use "always" that's very slow but a bit safer than
# everysec.
#
# More details please check the following article:
# http://antirez.com/post/redis-persistence-demystified.html
#
# If unsure, use "everysec".
# appendfsync always
appendfsync everysec
# appendfsync no
# When the AOF fsync policy is set to always or everysec, and a background
# saving process (a background save or AOF log background rewriting) is
# performing a lot of I/O against the disk, in some Linux configurations
# Redis may block too long on the fsync() call. Note that there is no fix for
# this currently, as even performing fsync in a different thread will block
# our synchronous write(2) call.
#
# In order to mitigate this problem it's possible to use the following option
# that will prevent fsync() from being called in the main process while a
# BGSAVE or BGREWRITEAOF is in progress.
#
# This means that while another child is saving, the durability of Redis is
# the same as "appendfsync no". In practical terms, this means that it is
# possible to lose up to 30 seconds of log in the worst scenario (with the
# default Linux settings).
#
# If you have latency problems turn this to "yes". Otherwise leave it as
# "no" that is the safest pick from the point of view of durability.
no-appendfsync-on-rewrite no
# Automatic rewrite of the append only file.
# Redis is able to automatically rewrite the log file implicitly calling
# BGREWRITEAOF when the AOF log size grows by the specified percentage.
#
# This is how it works: Redis remembers the size of the AOF file after the
# latest rewrite (if no rewrite has happened since the restart, the size of
# the AOF at startup is used).
#
# This base size is compared to the current size. If the current size is
# bigger than the specified percentage, the rewrite is triggered. Also
# you need to specify a minimal size for the AOF file to be rewritten, this
# is useful to avoid rewriting the AOF file even if the percentage increase
# is reached but it is still pretty small.
#
# Specify a percentage of zero in order to disable the automatic AOF
# rewrite feature.
auto-aof-rewrite-percentage 100
auto-aof-rewrite-min-size 64mb
# An AOF file may be found to be truncated at the end during the Redis
# startup process, when the AOF data gets loaded back into memory.
# This may happen when the system where Redis is running
# crashes, especially when an ext4 filesystem is mounted without the
# data=ordered option (however this can't happen when Redis itself
# crashes or aborts but the operating system still works correctly).
#
# Redis can either exit with an error when this happens, or load as much
# data as possible (the default now) and start if the AOF file is found
# to be truncated at the end. The following option controls this behavior.
#
# If aof-load-truncated is set to yes, a truncated AOF file is loaded and
# the Redis server starts emitting a log to inform the user of the event.
# Otherwise if the option is set to no, the server aborts with an error
# and refuses to start. When the option is set to no, the user requires
# to fix the AOF file using the "redis-check-aof" utility before to restart
# the server.
#
# Note that if the AOF file will be found to be corrupted in the middle
# the server will still exit with an error. This option only applies when
# Redis will try to read more data from the AOF file but not enough bytes
# will be found.
aof-load-truncated yes
# Redis can create append-only base files in either RDB or AOF formats. Using
# the RDB format is always faster and more efficient, and disabling it is only
# supported for backward compatibility purposes.
aof-use-rdb-preamble yes
# Redis supports recording timestamp annotations in the AOF to support restoring
# the data from a specific point-in-time. However, using this capability changes
# the AOF format in a way that may not be compatible with existing AOF parsers.
aof-timestamp-enabled no
################################ SHUTDOWN #####################################
# Maximum time to wait for replicas when shutting down, in seconds.
#
# During shut down, a grace period allows any lagging replicas to catch up with
# the latest replication offset before the master exists. This period can
# prevent data loss, especially for deployments without configured disk backups.
#
# The 'shutdown-timeout' value is the grace period's duration in seconds. It is
# only applicable when the instance has replicas. To disable the feature, set
# the value to 0.
#
# shutdown-timeout 10
# When Redis receives a SIGINT or SIGTERM, shutdown is initiated and by default
# an RDB snapshot is written to disk in a blocking operation if save points are configured.
# The options used on signaled shutdown can include the following values:
# default: Saves RDB snapshot only if save points are configured.
# Waits for lagging replicas to catch up.
# save: Forces a DB saving operation even if no save points are configured.
# nosave: Prevents DB saving operation even if one or more save points are configured.
# now: Skips waiting for lagging replicas.
# force: Ignores any errors that would normally prevent the server from exiting.
#
# Any combination of values is allowed as long as "save" and "nosave" are not set simultaneously.
# Example: "nosave force now"
#
# shutdown-on-sigint default
# shutdown-on-sigterm default
################ NON-DETERMINISTIC LONG BLOCKING COMMANDS #####################
# Maximum time in milliseconds for EVAL scripts, functions and in some cases
# modules' commands before Redis can start processing or rejecting other clients.
#
# If the maximum execution time is reached Redis will start to reply to most
# commands with a BUSY error.
#
# In this state Redis will only allow a handful of commands to be executed.
# For instance, SCRIPT KILL, FUNCTION KILL, SHUTDOWN NOSAVE and possibly some
# module specific 'allow-busy' commands.
#
# SCRIPT KILL and FUNCTION KILL will only be able to stop a script that did not
# yet call any write commands, so SHUTDOWN NOSAVE may be the only way to stop
# the server in the case a write command was already issued by the script when
# the user doesn't want to wait for the natural termination of the script.
#
# The default is 5 seconds. It is possible to set it to 0 or a negative value
# to disable this mechanism (uninterrupted execution). Note that in the past
# this config had a different name, which is now an alias, so both of these do
# the same:
# lua-time-limit 5000
# busy-reply-threshold 5000
################################ REDIS CLUSTER ###############################
# Normal Redis instances can't be part of a Redis Cluster; only nodes that are
# started as cluster nodes can. In order to start a Redis instance as a
# cluster node enable the cluster support uncommenting the following:
#
# cluster-enabled yes
# Every cluster node has a cluster configuration file. This file is not
# intended to be edited by hand. It is created and updated by Redis nodes.
# Every Redis Cluster node requires a different cluster configuration file.
# Make sure that instances running in the same system do not have
# overlapping cluster configuration file names.
#
# cluster-config-file nodes-6379.conf
# Cluster node timeout is the amount of milliseconds a node must be unreachable
# for it to be considered in failure state.
# Most other internal time limits are a multiple of the node timeout.
#
# cluster-node-timeout 15000
# The cluster port is the port that the cluster bus will listen for inbound connections on. When set
# to the default value, 0, it will be bound to the command port + 10000. Setting this value requires
# you to specify the cluster bus port when executing cluster meet.
# cluster-port 0
# A replica of a failing master will avoid to start a failover if its data
# looks too old.
#
# There is no simple way for a replica to actually have an exact measure of
# its "data age", so the following two checks are performed:
#
# 1) If there are multiple replicas able to failover, they exchange messages
# in order to try to give an advantage to the replica with the best
# replication offset (more data from the master processed).
# Replicas will try to get their rank by offset, and apply to the start
# of the failover a delay proportional to their rank.
#
# 2) Every single replica computes the time of the last interaction with
# its master. This can be the last ping or command received (if the master
# is still in the "connected" state), or the time that elapsed since the
# disconnection with the master (if the replication link is currently down).
# If the last interaction is too old, the replica will not try to failover
# at all.
#
# The point "2" can be tuned by user. Specifically a replica will not perform
# the failover if, since the last interaction with the master, the time
# elapsed is greater than:
#
# (node-timeout * cluster-replica-validity-factor) + repl-ping-replica-period
#
# So for example if node-timeout is 30 seconds, and the cluster-replica-validity-factor
# is 10, and assuming a default repl-ping-replica-period of 10 seconds, the
# replica will not try to failover if it was not able to talk with the master
# for longer than 310 seconds.
#
# A large cluster-replica-validity-factor may allow replicas with too old data to failover
# a master, while a too small value may prevent the cluster from being able to
# elect a replica at all.
#
# For maximum availability, it is possible to set the cluster-replica-validity-factor
# to a value of 0, which means, that replicas will always try to failover the
# master regardless of the last time they interacted with the master.
# (However they'll always try to apply a delay proportional to their
# offset rank).
#
# Zero is the only value able to guarantee that when all the partitions heal
# the cluster will always be able to continue.
#
# cluster-replica-validity-factor 10
# Cluster replicas are able to migrate to orphaned masters, that are masters
# that are left without working replicas. This improves the cluster ability
# to resist to failures as otherwise an orphaned master can't be failed over
# in case of failure if it has no working replicas.
#
# Replicas migrate to orphaned masters only if there are still at least a
# given number of other working replicas for their old master. This number
# is the "migration barrier". A migration barrier of 1 means that a replica
# will migrate only if there is at least 1 other working replica for its master
# and so forth. It usually reflects the number of replicas you want for every
# master in your cluster.
#
# Default is 1 (replicas migrate only if their masters remain with at least
# one replica). To disable migration just set it to a very large value or
# set cluster-allow-replica-migration to 'no'.
# A value of 0 can be set but is useful only for debugging and dangerous
# in production.
#
# cluster-migration-barrier 1
# Turning off this option allows to use less automatic cluster configuration.
# It both disables migration to orphaned masters and migration from masters
# that became empty.
#
# Default is 'yes' (allow automatic migrations).
#
# cluster-allow-replica-migration yes
# By default Redis Cluster nodes stop accepting queries if they detect there
# is at least a hash slot uncovered (no available node is serving it).
# This way if the cluster is partially down (for example a range of hash slots
# are no longer covered) all the cluster becomes, eventually, unavailable.
# It automatically returns available as soon as all the slots are covered again.
#
# However sometimes you want the subset of the cluster which is working,
# to continue to accept queries for the part of the key space that is still
# covered. In order to do so, just set the cluster-require-full-coverage
# option to no.
#
# cluster-require-full-coverage yes
# This option, when set to yes, prevents replicas from trying to failover its
# master during master failures. However the replica can still perform a
# manual failover, if forced to do so.
#
# This is useful in different scenarios, especially in the case of multiple
# data center operations, where we want one side to never be promoted if not
# in the case of a total DC failure.
#
# cluster-replica-no-failover no
# This option, when set to yes, allows nodes to serve read traffic while the
# cluster is in a down state, as long as it believes it owns the slots.
#
# This is useful for two cases. The first case is for when an application
# doesn't require consistency of data during node failures or network partitions.
# One example of this is a cache, where as long as the node has the data it
# should be able to serve it.
#
# The second use case is for configurations that don't meet the recommended
# three shards but want to enable cluster mode and scale later. A
# master outage in a 1 or 2 shard configuration causes a read/write outage to the
# entire cluster without this option set, with it set there is only a write outage.
# Without a quorum of masters, slot ownership will not change automatically.
#
# cluster-allow-reads-when-down no
# This option, when set to yes, allows nodes to serve pubsub shard traffic while
# the cluster is in a down state, as long as it believes it owns the slots.
#
# This is useful if the application would like to use the pubsub feature even when
# the cluster global stable state is not OK. If the application wants to make sure only
# one shard is serving a given channel, this feature should be kept as yes.
#
# cluster-allow-pubsubshard-when-down yes
# Cluster link send buffer limit is the limit on the memory usage of an individual
# cluster bus link's send buffer in bytes. Cluster links would be freed if they exceed
# this limit. This is to primarily prevent send buffers from growing unbounded on links
# toward slow peers (E.g. PubSub messages being piled up).
# This limit is disabled by default. Enable this limit when 'mem_cluster_links' INFO field
# and/or 'send-buffer-allocated' entries in the 'CLUSTER LINKS` command output continuously increase.
# Minimum limit of 1gb is recommended so that cluster link buffer can fit in at least a single
# PubSub message by default. (client-query-buffer-limit default value is 1gb)
#
# cluster-link-sendbuf-limit 0
# Clusters can configure their announced hostname using this config. This is a common use case for
# applications that need to use TLS Server Name Indication (SNI) or dealing with DNS based
# routing. By default this value is only shown as additional metadata in the CLUSTER SLOTS
# command, but can be changed using 'cluster-preferred-endpoint-type' config. This value is
# communicated along the clusterbus to all nodes, setting it to an empty string will remove
# the hostname and also propagate the removal.
#
# cluster-announce-hostname ""
# Clusters can advertise how clients should connect to them using either their IP address,
# a user defined hostname, or by declaring they have no endpoint. Which endpoint is
# shown as the preferred endpoint is set by using the cluster-preferred-endpoint-type
# config with values 'ip', 'hostname', or 'unknown-endpoint'. This value controls how
# the endpoint returned for MOVED/ASKING requests as well as the first field of CLUSTER SLOTS.
# If the preferred endpoint type is set to hostname, but no announced hostname is set, a '?'
# will be returned instead.
#
# When a cluster advertises itself as having an unknown endpoint, it's indicating that
# the server doesn't know how clients can reach the cluster. This can happen in certain
# networking situations where there are multiple possible routes to the node, and the
# server doesn't know which one the client took. In this case, the server is expecting
# the client to reach out on the same endpoint it used for making the last request, but use
# the port provided in the response.
#
# cluster-preferred-endpoint-type ip
# In order to setup your cluster make sure to read the documentation
# available at https://redis.io web site.
########################## CLUSTER DOCKER/NAT support ########################
# In certain deployments, Redis Cluster nodes address discovery fails, because
# addresses are NAT-ted or because ports are forwarded (the typical case is
# Docker and other containers).
#
# In order to make Redis Cluster working in such environments, a static
# configuration where each node knows its public address is needed. The
# following four options are used for this scope, and are:
#
# * cluster-announce-ip
# * cluster-announce-port
# * cluster-announce-tls-port
# * cluster-announce-bus-port
#
# Each instructs the node about its address, client ports (for connections
# without and with TLS) and cluster message bus port. The information is then
# published in the header of the bus packets so that other nodes will be able to
# correctly map the address of the node publishing the information.
#
# If cluster-tls is set to yes and cluster-announce-tls-port is omitted or set
# to zero, then cluster-announce-port refers to the TLS port. Note also that
# cluster-announce-tls-port has no effect if cluster-tls is set to no.
#
# If the above options are not used, the normal Redis Cluster auto-detection
# will be used instead.
#
# Note that when remapped, the bus port may not be at the fixed offset of
# clients port + 10000, so you can specify any port and bus-port depending
# on how they get remapped. If the bus-port is not set, a fixed offset of
# 10000 will be used as usual.
#
# Example:
#
# cluster-announce-ip 10.1.1.5
# cluster-announce-tls-port 6379
# cluster-announce-port 0
# cluster-announce-bus-port 6380
################################## SLOW LOG ###################################
# The Redis Slow Log is a system to log queries that exceeded a specified
# execution time. The execution time does not include the I/O operations
# like talking with the client, sending the reply and so forth,
# but just the time needed to actually execute the command (this is the only
# stage of command execution where the thread is blocked and can not serve
# other requests in the meantime).
#
# You can configure the slow log with two parameters: one tells Redis
# what is the execution time, in microseconds, to exceed in order for the
# command to get logged, and the other parameter is the length of the
# slow log. When a new command is logged the oldest one is removed from the
# queue of logged commands.
# The following time is expressed in microseconds, so 1000000 is equivalent
# to one second. Note that a negative number disables the slow log, while
# a value of zero forces the logging of every command.
slowlog-log-slower-than 10000
# There is no limit to this length. Just be aware that it will consume memory.
# You can reclaim memory used by the slow log with SLOWLOG RESET.
slowlog-max-len 128
################################ LATENCY MONITOR ##############################
# The Redis latency monitoring subsystem samples different operations
# at runtime in order to collect data related to possible sources of
# latency of a Redis instance.
#
# Via the LATENCY command this information is available to the user that can
# print graphs and obtain reports.
#
# The system only logs operations that were performed in a time equal or
# greater than the amount of milliseconds specified via the
# latency-monitor-threshold configuration directive. When its value is set
# to zero, the latency monitor is turned off.
#
# By default latency monitoring is disabled since it is mostly not needed
# if you don't have latency issues, and collecting data has a performance
# impact, that while very small, can be measured under big load. Latency
# monitoring can easily be enabled at runtime using the command
# "CONFIG SET latency-monitor-threshold " if needed.
latency-monitor-threshold 0
################################ LATENCY TRACKING ##############################
# The Redis extended latency monitoring tracks the per command latencies and enables
# exporting the percentile distribution via the INFO latencystats command,
# and cumulative latency distributions (histograms) via the LATENCY command.
#
# By default, the extended latency monitoring is enabled since the overhead
# of keeping track of the command latency is very small.
# latency-tracking yes
# By default the exported latency percentiles via the INFO latencystats command
# are the p50, p99, and p999.
# latency-tracking-info-percentiles 50 99 99.9
############################# EVENT NOTIFICATION ##############################
# Redis can notify Pub/Sub clients about events happening in the key space.
# This feature is documented at https://redis.io/topics/notifications
#
# For instance if keyspace events notification is enabled, and a client
# performs a DEL operation on key "foo" stored in the Database 0, two
# messages will be published via Pub/Sub:
#
# PUBLISH __keyspace@0__:foo del
# PUBLISH __keyevent@0__:del foo
#
# It is possible to select the events that Redis will notify among a set
# of classes. Every class is identified by a single character:
#
# K Keyspace events, published with __keyspace@__ prefix.
# E Keyevent events, published with __keyevent@__ prefix.
# g Generic commands (non-type specific) like DEL, EXPIRE, RENAME, ...
# $ String commands
# l List commands
# s Set commands
# h Hash commands
# z Sorted set commands
# x Expired events (events generated every time a key expires)
# e Evicted events (events generated when a key is evicted for maxmemory)
# n New key events (Note: not included in the 'A' class)
# t Stream commands
# d Module key type events
# m Key-miss events (Note: It is not included in the 'A' class)
# A Alias for g$lshzxetd, so that the "AKE" string means all the events
# (Except key-miss events which are excluded from 'A' due to their
# unique nature).
#
# The "notify-keyspace-events" takes as argument a string that is composed
# of zero or multiple characters. The empty string means that notifications
# are disabled.
#
# Example: to enable list and generic events, from the point of view of the
# event name, use:
#
# notify-keyspace-events Elg
#
# Example 2: to get the stream of the expired keys subscribing to channel
# name __keyevent@0__:expired use:
#
# notify-keyspace-events Ex
#
# By default all notifications are disabled because most users don't need
# this feature and the feature has some overhead. Note that if you don't
# specify at least one of K or E, no events will be delivered.
notify-keyspace-events ""
############################### ADVANCED CONFIG ###############################
# Hashes are encoded using a memory efficient data structure when they have a
# small number of entries, and the biggest entry does not exceed a given
# threshold. These thresholds can be configured using the following directives.
hash-max-listpack-entries 512
hash-max-listpack-value 64
# Lists are also encoded in a special way to save a lot of space.
# The number of entries allowed per internal list node can be specified
# as a fixed maximum size or a maximum number of elements.
# For a fixed maximum size, use -5 through -1, meaning:
# -5: max size: 64 Kb <-- not recommended for normal workloads
# -4: max size: 32 Kb <-- not recommended
# -3: max size: 16 Kb <-- probably not recommended
# -2: max size: 8 Kb <-- good
# -1: max size: 4 Kb <-- good
# Positive numbers mean store up to _exactly_ that number of elements
# per list node.
# The highest performing option is usually -2 (8 Kb size) or -1 (4 Kb size),
# but if your use case is unique, adjust the settings as necessary.
list-max-listpack-size -2
# Lists may also be compressed.
# Compress depth is the number of quicklist ziplist nodes from *each* side of
# the list to *exclude* from compression. The head and tail of the list
# are always uncompressed for fast push/pop operations. Settings are:
# 0: disable all list compression
# 1: depth 1 means "don't start compressing until after 1 node into the list,
# going from either the head or tail"
# So: [head]->node->node->...->node->[tail]
# [head], [tail] will always be uncompressed; inner nodes will compress.
# 2: [head]->[next]->node->node->...->node->[prev]->[tail]
# 2 here means: don't compress head or head->next or tail->prev or tail,
# but compress all nodes between them.
# 3: [head]->[next]->[next]->node->node->...->node->[prev]->[prev]->[tail]
# etc.
list-compress-depth 0
# Sets have a special encoding in just one case: when a set is composed
# of just strings that happen to be integers in radix 10 in the range
# of 64 bit signed integers.
# The following configuration setting sets the limit in the size of the
# set in order to use this special memory saving encoding.
set-max-intset-entries 512
# Similarly to hashes and lists, sorted sets are also specially encoded in
# order to save a lot of space. This encoding is only used when the length and
# elements of a sorted set are below the following limits:
zset-max-listpack-entries 128
zset-max-listpack-value 64
# HyperLogLog sparse representation bytes limit. The limit includes the
# 16 bytes header. When an HyperLogLog using the sparse representation crosses
# this limit, it is converted into the dense representation.
#
# A value greater than 16000 is totally useless, since at that point the
# dense representation is more memory efficient.
#
# The suggested value is ~ 3000 in order to have the benefits of
# the space efficient encoding without slowing down too much PFADD,
# which is O(N) with the sparse encoding. The value can be raised to
# ~ 10000 when CPU is not a concern, but space is, and the data set is
# composed of many HyperLogLogs with cardinality in the 0 - 15000 range.
hll-sparse-max-bytes 3000
# Streams macro node max size / items. The stream data structure is a radix
# tree of big nodes that encode multiple items inside. Using this configuration
# it is possible to configure how big a single node can be in bytes, and the
# maximum number of items it may contain before switching to a new node when
# appending new stream entries. If any of the following settings are set to
# zero, the limit is ignored, so for instance it is possible to set just a
# max entries limit by setting max-bytes to 0 and max-entries to the desired
# value.
stream-node-max-bytes 4096
stream-node-max-entries 100
# Active rehashing uses 1 millisecond every 100 milliseconds of CPU time in
# order to help rehashing the main Redis hash table (the one mapping top-level
# keys to values). The hash table implementation Redis uses (see dict.c)
# performs a lazy rehashing: the more operation you run into a hash table
# that is rehashing, the more rehashing "steps" are performed, so if the
# server is idle the rehashing is never complete and some more memory is used
# by the hash table.
#
# The default is to use this millisecond 10 times every second in order to
# actively rehash the main dictionaries, freeing memory when possible.
#
# If unsure:
# use "activerehashing no" if you have hard latency requirements and it is
# not a good thing in your environment that Redis can reply from time to time
# to queries with 2 milliseconds delay.
#
# use "activerehashing yes" if you don't have such hard requirements but
# want to free memory asap when possible.
activerehashing yes
# The client output buffer limits can be used to force disconnection of clients
# that are not reading data from the server fast enough for some reason (a
# common reason is that a Pub/Sub client can't consume messages as fast as the
# publisher can produce them).
#
# The limit can be set differently for the three different classes of clients:
#
# normal -> normal clients including MONITOR clients
# replica -> replica clients
# pubsub -> clients subscribed to at least one pubsub channel or pattern
#
# The syntax of every client-output-buffer-limit directive is the following:
#
# client-output-buffer-limit
#
# A client is immediately disconnected once the hard limit is reached, or if
# the soft limit is reached and remains reached for the specified number of
# seconds (continuously).
# So for instance if the hard limit is 32 megabytes and the soft limit is
# 16 megabytes / 10 seconds, the client will get disconnected immediately
# if the size of the output buffers reach 32 megabytes, but will also get
# disconnected if the client reaches 16 megabytes and continuously overcomes
# the limit for 10 seconds.
#
# By default normal clients are not limited because they don't receive data
# without asking (in a push way), but just after a request, so only
# asynchronous clients may create a scenario where data is requested faster
# than it can read.
#
# Instead there is a default limit for pubsub and replica clients, since
# subscribers and replicas receive data in a push fashion.
#
# Note that it doesn't make sense to set the replica clients output buffer
# limit lower than the repl-backlog-size config (partial sync will succeed
# and then replica will get disconnected).
# Such a configuration is ignored (the size of repl-backlog-size will be used).
# This doesn't have memory consumption implications since the replica client
# will share the backlog buffers memory.
#
# Both the hard or the soft limit can be disabled by setting them to zero.
client-output-buffer-limit normal 0 0 0
client-output-buffer-limit replica 256mb 64mb 60
client-output-buffer-limit pubsub 32mb 8mb 60
# Client query buffers accumulate new commands. They are limited to a fixed
# amount by default in order to avoid that a protocol desynchronization (for
# instance due to a bug in the client) will lead to unbound memory usage in
# the query buffer. However you can configure it here if you have very special
# needs, such us huge multi/exec requests or alike.
#
# client-query-buffer-limit 1gb
# In some scenarios client connections can hog up memory leading to OOM
# errors or data eviction. To avoid this we can cap the accumulated memory
# used by all client connections (all pubsub and normal clients). Once we
# reach that limit connections will be dropped by the server freeing up
# memory. The server will attempt to drop the connections using the most
# memory first. We call this mechanism "client eviction".
#
# Client eviction is configured using the maxmemory-clients setting as follows:
# 0 - client eviction is disabled (default)
#
# A memory value can be used for the client eviction threshold,
# for example:
# maxmemory-clients 1g
#
# A percentage value (between 1% and 100%) means the client eviction threshold
# is based on a percentage of the maxmemory setting. For example to set client
# eviction at 5% of maxmemory:
# maxmemory-clients 5%
# In the Redis protocol, bulk requests, that are, elements representing single
# strings, are normally limited to 512 mb. However you can change this limit
# here, but must be 1mb or greater
#
# proto-max-bulk-len 512mb
# Redis calls an internal function to perform many background tasks, like
# closing connections of clients in timeout, purging expired keys that are
# never requested, and so forth.
#
# Not all tasks are performed with the same frequency, but Redis checks for
# tasks to perform according to the specified "hz" value.
#
# By default "hz" is set to 10. Raising the value will use more CPU when
# Redis is idle, but at the same time will make Redis more responsive when
# there are many keys expiring at the same time, and timeouts may be
# handled with more precision.
#
# The range is between 1 and 500, however a value over 100 is usually not
# a good idea. Most users should use the default of 10 and raise this up to
# 100 only in environments where very low latency is required.
hz 10
# Normally it is useful to have an HZ value which is proportional to the
# number of clients connected. This is useful in order, for instance, to
# avoid too many clients are processed for each background task invocation
# in order to avoid latency spikes.
#
# Since the default HZ value by default is conservatively set to 10, Redis
# offers, and enables by default, the ability to use an adaptive HZ value
# which will temporarily raise when there are many connected clients.
#
# When dynamic HZ is enabled, the actual configured HZ will be used
# as a baseline, but multiples of the configured HZ value will be actually
# used as needed once more clients are connected. In this way an idle
# instance will use very little CPU time while a busy instance will be
# more responsive.
dynamic-hz yes
# When a child rewrites the AOF file, if the following option is enabled
# the file will be fsync-ed every 4 MB of data generated. This is useful
# in order to commit the file to the disk more incrementally and avoid
# big latency spikes.
aof-rewrite-incremental-fsync yes
# When redis saves RDB file, if the following option is enabled
# the file will be fsync-ed every 4 MB of data generated. This is useful
# in order to commit the file to the disk more incrementally and avoid
# big latency spikes.
rdb-save-incremental-fsync yes
# Redis LFU eviction (see maxmemory setting) can be tuned. However it is a good
# idea to start with the default settings and only change them after investigating
# how to improve the performances and how the keys LFU change over time, which
# is possible to inspect via the OBJECT FREQ command.
#
# There are two tunable parameters in the Redis LFU implementation: the
# counter logarithm factor and the counter decay time. It is important to
# understand what the two parameters mean before changing them.
#
# The LFU counter is just 8 bits per key, it's maximum value is 255, so Redis
# uses a probabilistic increment with logarithmic behavior. Given the value
# of the old counter, when a key is accessed, the counter is incremented in
# this way:
#
# 1. A random number R between 0 and 1 is extracted.
# 2. A probability P is calculated as 1/(old_value*lfu_log_factor+1).
# 3. The counter is incremented only if R < P.
#
# The default lfu-log-factor is 10. This is a table of how the frequency
# counter changes with a different number of accesses with different
# logarithmic factors:
#
# +--------+------------+------------+------------+------------+------------+
# | factor | 100 hits | 1000 hits | 100K hits | 1M hits | 10M hits |
# +--------+------------+------------+------------+------------+------------+
# | 0 | 104 | 255 | 255 | 255 | 255 |
# +--------+------------+------------+------------+------------+------------+
# | 1 | 18 | 49 | 255 | 255 | 255 |
# +--------+------------+------------+------------+------------+------------+
# | 10 | 10 | 18 | 142 | 255 | 255 |
# +--------+------------+------------+------------+------------+------------+
# | 100 | 8 | 11 | 49 | 143 | 255 |
# +--------+------------+------------+------------+------------+------------+
#
# NOTE: The above table was obtained by running the following commands:
#
# redis-benchmark -n 1000000 incr foo
# redis-cli object freq foo
#
# NOTE 2: The counter initial value is 5 in order to give new objects a chance
# to accumulate hits.
#
# The counter decay time is the time, in minutes, that must elapse in order
# for the key counter to be divided by two (or decremented if it has a value
# less <= 10).
#
# The default value for the lfu-decay-time is 1. A special value of 0 means to
# decay the counter every time it happens to be scanned.
#
# lfu-log-factor 10
# lfu-decay-time 1
########################### ACTIVE DEFRAGMENTATION #######################
#
# What is active defragmentation?
# -------------------------------
#
# Active (online) defragmentation allows a Redis server to compact the
# spaces left between small allocations and deallocations of data in memory,
# thus allowing to reclaim back memory.
#
# Fragmentation is a natural process that happens with every allocator (but
# less so with Jemalloc, fortunately) and certain workloads. Normally a server
# restart is needed in order to lower the fragmentation, or at least to flush
# away all the data and create it again. However thanks to this feature
# implemented by Oran Agra for Redis 4.0 this process can happen at runtime
# in a "hot" way, while the server is running.
#
# Basically when the fragmentation is over a certain level (see the
# configuration options below) Redis will start to create new copies of the
# values in contiguous memory regions by exploiting certain specific Jemalloc
# features (in order to understand if an allocation is causing fragmentation
# and to allocate it in a better place), and at the same time, will release the
# old copies of the data. This process, repeated incrementally for all the keys
# will cause the fragmentation to drop back to normal values.
#
# Important things to understand:
#
# 1. This feature is disabled by default, and only works if you compiled Redis
# to use the copy of Jemalloc we ship with the source code of Redis.
# This is the default with Linux builds.
#
# 2. You never need to enable this feature if you don't have fragmentation
# issues.
#
# 3. Once you experience fragmentation, you can enable this feature when
# needed with the command "CONFIG SET activedefrag yes".
#
# The configuration parameters are able to fine tune the behavior of the
# defragmentation process. If you are not sure about what they mean it is
# a good idea to leave the defaults untouched.
# Active defragmentation is disabled by default
# activedefrag no
# Minimum amount of fragmentation waste to start active defrag
# active-defrag-ignore-bytes 100mb
# Minimum percentage of fragmentation to start active defrag
# active-defrag-threshold-lower 10
# Maximum percentage of fragmentation at which we use maximum effort
# active-defrag-threshold-upper 100
# Minimal effort for defrag in CPU percentage, to be used when the lower
# threshold is reached
# active-defrag-cycle-min 1
# Maximal effort for defrag in CPU percentage, to be used when the upper
# threshold is reached
# active-defrag-cycle-max 25
# Maximum number of set/hash/zset/list fields that will be processed from
# the main dictionary scan
# active-defrag-max-scan-fields 1000
# Jemalloc background thread for purging will be enabled by default
jemalloc-bg-thread yes
# It is possible to pin different threads and processes of Redis to specific
# CPUs in your system, in order to maximize the performances of the server.
# This is useful both in order to pin different Redis threads in different
# CPUs, but also in order to make sure that multiple Redis instances running
# in the same host will be pinned to different CPUs.
#
# Normally you can do this using the "taskset" command, however it is also
# possible to this via Redis configuration directly, both in Linux and FreeBSD.
#
# You can pin the server/IO threads, bio threads, aof rewrite child process, and
# the bgsave child process. The syntax to specify the cpu list is the same as
# the taskset command:
#
# Set redis server/io threads to cpu affinity 0,2,4,6:
# server_cpulist 0-7:2
#
# Set bio threads to cpu affinity 1,3:
# bio_cpulist 1,3
#
# Set aof rewrite child process to cpu affinity 8,9,10,11:
# aof_rewrite_cpulist 8-11
#
# Set bgsave child process to cpu affinity 1,10,11
# bgsave_cpulist 1,10-11
# In some cases redis will emit warnings and even refuse to start if it detects
# that the system is in bad state, it is possible to suppress these warnings
# by setting the following config which takes a space delimited list of warnings
# to suppress
#
# ignore-warnings ARM64-COW-BUG
================================================
FILE: src/backend/Dockerfile
================================================
FROM dataelement/bisheng-backend:base.v8
WORKDIR /app
COPY ./ ./
# 生成并安装依赖
RUN uv pip compile pyproject.toml --upgrade --output-file requirements.txt && \
uv pip install -r requirements.txt --system --no-cache-dir && \
uv cache clean && \
rm -f requirements.txt
# patch langchain-openai lib. remove this when langchain-openai support reasoning_content
RUN patch -p1 < /app/bisheng/patches/langchain_openai.patch /usr/local/lib/python3.10/site-packages/langchain_openai/chat_models/base.py
CMD ["sh entrypoint.sh"]
================================================
FILE: src/backend/README.md
================================================
# 毕昇后端代码
* Dockerfile 使用 uv 进行 Python 依赖管理
================================================
FILE: src/backend/alembic.ini
================================================
# A generic, single database configuration.
[alembic]
# path to migration scripts.
# this is typically a path given in POSIX (e.g. forward slashes)
# format, relative to the token %(here)s which refers to the location of this
# ini file
script_location = ./bisheng/core/database/alembic
# template used to generate migration file names; The default value is %%(rev)s_%%(slug)s
# Uncomment the line below if you want the files to be prepended with date and time
# see https://alembic.sqlalchemy.org/en/latest/tutorial.html#editing-the-ini-file
# for all available tokens
file_template = %%(year)d_%%(month).2d_%%(day).2d_%%(hour).2d%%(minute).2d-%%(rev)s_%%(slug)s
# sys.path path, will be prepended to sys.path if present.
# defaults to the current working directory. for multiple paths, the path separator
# is defined by "path_separator" below.
prepend_sys_path = .
# timezone to use when rendering the date within the migration file
# as well as the filename.
# If specified, requires the tzdata library which can be installed by adding
# `alembic[tz]` to the pip requirements.
# string value is passed to ZoneInfo()
# leave blank for localtime
# timezone =
# max length of characters to apply to the "slug" field
# truncate_slug_length = 40
# set to 'true' to run the environment during
# the 'revision' command, regardless of autogenerate
# revision_environment = false
# set to 'true' to allow .pyc and .pyo files without
# a source .py file to be detected as revisions in the
# versions/ directory
# sourceless = false
# version location specification; This defaults
# to /versions. When using multiple version
# directories, initial revisions must be specified with --version-path.
# The path separator used here should be the separator specified by "path_separator"
# below.
# version_locations = %(here)s/bar:%(here)s/bat:%(here)s/alembic/versions
# path_separator; This indicates what character is used to split lists of file
# paths, including version_locations and prepend_sys_path within configparser
# files such as alembic.ini.
# The default rendered in new alembic.ini files is "os", which uses os.pathsep
# to provide os-dependent path splitting.
#
# Note that in order to support legacy alembic.ini files, this default does NOT
# take place if path_separator is not present in alembic.ini. If this
# option is omitted entirely, fallback logic is as follows:
#
# 1. Parsing of the version_locations option falls back to using the legacy
# "version_path_separator" key, which if absent then falls back to the legacy
# behavior of splitting on spaces and/or commas.
# 2. Parsing of the prepend_sys_path option falls back to the legacy
# behavior of splitting on spaces, commas, or colons.
#
# Valid values for path_separator are:
#
# path_separator = :
# path_separator = ;
# path_separator = space
# path_separator = newline
#
# Use os.pathsep. Default configuration used for new projects.
path_separator = os
# set to 'true' to search source files recursively
# in each "version_locations" directory
# new in Alembic version 1.10
# recursive_version_locations = false
# the output encoding used when revision files
# are written from script.py.mako
# output_encoding = utf-8
# database URL. This is consumed by the user-maintained env.py script only.
# other means of configuring database URLs may be customized within the env.py
# file.
# sqlalchemy.url =
[post_write_hooks]
# post_write_hooks defines scripts or Python functions that are run
# on newly generated revision scripts. See the documentation for further
# detail and examples
# format using "black" - use the console_scripts runner, against the "black" entrypoint
# hooks = black
# black.type = console_scripts
# black.entrypoint = black
# black.options = -l 79 REVISION_SCRIPT_FILENAME
# lint with attempts to fix using "ruff" - use the module runner, against the "ruff" module
# hooks = ruff
# ruff.type = module
# ruff.module = ruff
# ruff.options = check --fix REVISION_SCRIPT_FILENAME
# Alternatively, use the exec runner to execute a binary found on your PATH
# hooks = ruff
# ruff.type = exec
# ruff.executable = ruff
# ruff.options = check --fix REVISION_SCRIPT_FILENAME
# Logging configuration. This is also consumed by the user-maintained
# env.py script only.
[loggers]
keys = root,sqlalchemy,alembic
[handlers]
keys = console
[formatters]
keys = generic
[logger_root]
level = WARNING
handlers = console
qualname =
[logger_sqlalchemy]
level = WARNING
handlers =
qualname = sqlalchemy.engine
[logger_alembic]
level = INFO
handlers =
qualname = alembic
[handler_console]
class = StreamHandler
args = (sys.stderr,)
level = NOTSET
formatter = generic
[formatter_generic]
format = %(levelname)-5.5s [%(name)s] %(message)s
datefmt = %H:%M:%S
================================================
FILE: src/backend/base.Dockerfile
================================================
FROM python:3.10-slim
ARG PANDOC_ARCH=amd64
ENV PANDOC_ARCH=$PANDOC_ARCH
ENV PATH="${PATH}:/root/.local/bin"
WORKDIR /app
# 安装依赖（合并指令、清理缓存、禁用推荐包）
RUN apt-get update && \
apt-get install -y --no-install-recommends \
gcc g++ curl build-essential libreoffice \
wget procps vim fonts-wqy-zenhei \
libglib2.0-0 libsm6 libxrender1 libxext6 libgl1 \
&& rm -rf /var/lib/apt/lists/*
# 安装 FFmpeg
RUN apt-get update && apt-get install -y --no-install-recommends ffmpeg && rm -rf /var/lib/apt/lists/*
# 安装 pandoc
RUN mkdir -p /opt/pandoc && \
cd /opt/pandoc && \
wget https://github.com/jgm/pandoc/releases/download/3.6.4/pandoc-3.6.4-linux-${PANDOC_ARCH}.tar.gz && \
tar xvf pandoc-3.6.4-linux-${PANDOC_ARCH}.tar.gz && \
cp pandoc-3.6.4/bin/pandoc /usr/bin/ && \
rm -rf /opt/pandoc
# 安装 uv
RUN curl -LsSf https://astral.sh/uv/install.sh | sh
# 安装 Poetry
#RUN curl -sSL https://install.python-poetry.org | python3 - --version 1.8.2
# 拷贝项目依赖文件
COPY ./pyproject.toml ./
# 安装 Python 依赖
RUN python -m pip install --upgrade pip && \
uv pip compile pyproject.toml --output-file requirements.txt && \
uv pip install -r requirements.txt --system --no-cache-dir && \
uv cache clean
#RUN python -m pip install --upgrade pip && \
# pip install shapely==2.0.1 && \
# poetry config virtualenvs.create false && \
# poetry install --no-interaction --no-ansi --without dev
# 安装 NLTK 数据
RUN python -c "import nltk; nltk.download('punkt'); nltk.download('punkt_tab'); nltk.download('averaged_perceptron_tagger'); nltk.download('averaged_perceptron_tagger_eng')"
# 安装 playwright chromium
RUN playwright install chromium && playwright install-deps
COPY . .
CMD ["sh", "entrypoint.sh"]
================================================
FILE: src/backend/bisheng/.gitignore
================================================
config.yaml
================================================
FILE: src/backend/bisheng/__init__.py
================================================
from importlib import metadata
from bisheng.core.cache import cache_manager
from bisheng.interface.custom.custom_component import CustomComponent
# from bisheng.processing.process import load_flow_from_json # noqa: E402
try:
# SetujuciGo to automatic modification
__version__ = '2.4.0-beta1-fix'
except metadata.PackageNotFoundError:
# Case where package metadata is not available.
__version__ = ''
del metadata # optional, avoids polluting the results of dir(__package__)
__all__ = ['cache_manager', 'CustomComponent']
================================================
FILE: src/backend/bisheng/api/__init__.py
================================================
from bisheng.api.router import router, router_rpc
__all__ = ['router', 'router_rpc']
================================================
FILE: src/backend/bisheng/api/router.py
================================================
# Router for base api
from fastapi import APIRouter
from bisheng.api.v1 import (assistant_router, audit_router, chat_router, component_router,
endpoints_router, evaluation_router, flows_router,
group_router, mark_router,
report_router, skillcenter_router, tag_router,
user_router, validate_router, variable_router, workflow_router,
workstation_router, tool_router, invite_code_router)
from bisheng.chat_session.api.router import router as session_router
from bisheng.finetune.api.finetune import router as finetune_router
from bisheng.finetune.api.server import router as server_router
from bisheng.knowledge.api.router import qa_router, knowledge_router
from bisheng.llm.api.router import router as llm_router
from bisheng.open_endpoints.api.endpoints.llm import router as llm_router_rpc
from bisheng.open_endpoints.api.router import (assistant_router_rpc, chat_router_rpc, flow_router,
knowledge_router_rpc, workflow_router_rpc,
filelib_router_rpc)
from bisheng.share_link.api.router import router as share_link_router
from bisheng.linsight.api.router import router as linsight_router
from bisheng.telemetry_search.api.router import router as telemetry_search_router
router = APIRouter(prefix='/api/v1', )
router.include_router(chat_router)
router.include_router(endpoints_router)
router.include_router(validate_router)
router.include_router(flows_router)
router.include_router(skillcenter_router)
router.include_router(knowledge_router)
router.include_router(server_router)
router.include_router(user_router)
router.include_router(qa_router)
router.include_router(variable_router)
router.include_router(report_router)
router.include_router(finetune_router)
router.include_router(component_router)
router.include_router(assistant_router)
router.include_router(group_router)
router.include_router(audit_router)
router.include_router(evaluation_router)
router.include_router(tag_router)
router.include_router(llm_router)
router.include_router(workflow_router)
router.include_router(mark_router)
router.include_router(workstation_router)
router.include_router(linsight_router)
router.include_router(tool_router)
router.include_router(invite_code_router)
router.include_router(session_router)
router.include_router(share_link_router)
router.include_router(telemetry_search_router)
router_rpc = APIRouter(prefix='/api/v2', )
router_rpc.include_router(knowledge_router_rpc)
router_rpc.include_router(filelib_router_rpc)
router_rpc.include_router(chat_router_rpc)
router_rpc.include_router(flow_router)
router_rpc.include_router(assistant_router_rpc)
router_rpc.include_router(workflow_router_rpc)
router_rpc.include_router(llm_router_rpc)
================================================
FILE: src/backend/bisheng/api/services/__init__.py
================================================
================================================
FILE: src/backend/bisheng/api/services/assistant.py
================================================
from datetime import datetime
from typing import Any, List, Optional, Union
from fastapi import Request
from loguru import logger
from bisheng.api.services.assistant_agent import AssistantAgent
from bisheng.api.services.assistant_base import AssistantUtils
from bisheng.api.services.audit_log import AuditLogService
from bisheng.api.services.base import BaseService
from bisheng.api.v1.schemas import (AssistantInfo, AssistantSimpleInfo, AssistantUpdateReq,
StreamData)
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum, ApplicationTypeEnum
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.assistant import (AssistantInitError, AssistantNameRepeatError,
AssistantNotEditError, AssistantNotExistsError)
from bisheng.common.errcode.http_error import UnAuthorizedError
from bisheng.common.schemas.telemetry.event_data_schema import NewApplicationEventData
from bisheng.common.services import telemetry_service
from bisheng.core.cache import InMemoryCache
from bisheng.core.logger import trace_id_var
from bisheng.database.models.assistant import (Assistant, AssistantDao, AssistantLinkDao,
AssistantStatus)
from bisheng.database.models.flow import Flow, FlowDao, FlowType
from bisheng.database.models.group_resource import GroupResourceDao, GroupResource, ResourceTypeEnum
from bisheng.database.models.role_access import AccessType, RoleAccessDao
from bisheng.database.models.session import MessageSessionDao
from bisheng.database.models.tag import TagDao
from bisheng.database.models.user_group import UserGroupDao
from bisheng.knowledge.domain.models.knowledge import KnowledgeDao
from bisheng.llm.domain.services import LLMService
from bisheng.share_link.domain.models.share_link import ShareLink
from bisheng.tool.domain.models.gpts_tools import GptsToolsDao, GptsTools
from bisheng.user.domain.models.user import UserDao
from bisheng.user.domain.models.user_role import UserRoleDao
from bisheng.utils import get_request_ip
class AssistantService(BaseService, AssistantUtils):
UserCache: InMemoryCache = InMemoryCache()
@classmethod
def get_assistant(cls,
user: UserPayload,
name: str = None,
status: int | None = None,
tag_id: int | None = None,
page: int = 1,
limit: int = 20) -> (List[AssistantSimpleInfo], int):
"""
Get list of assistants
"""
assistant_ids = []
if tag_id:
ret = TagDao.get_resources_by_tags([tag_id], ResourceTypeEnum.ASSISTANT)
assistant_ids = [one.resource_id for one in ret]
if not assistant_ids:
return [], 0
data = []
if user.is_admin():
res, total = AssistantDao.get_all_assistants(name, page, limit, assistant_ids, status)
else:
# Permission management visible assistant information
assistant_ids_extra = []
user_role = UserRoleDao.get_user_roles(user.user_id)
if user_role:
role_ids = [role.role_id for role in user_role]
role_access = RoleAccessDao.get_role_access(role_ids, AccessType.ASSISTANT_READ)
if role_access:
assistant_ids_extra = [access.third_id for access in role_access]
res, total = AssistantDao.get_assistants(user.user_id, name, assistant_ids_extra, status, page, limit,
assistant_ids)
assistant_ids = [one.id for one in res]
# Query groups to which the assistant belongs
assistant_groups = GroupResourceDao.get_resources_group(ResourceTypeEnum.ASSISTANT, assistant_ids)
assistant_group_dict = {}
for one in assistant_groups:
if one.third_id not in assistant_group_dict:
assistant_group_dict[one.third_id] = []
assistant_group_dict[one.third_id].append(one.group_id)
# Get assistant-associatedtag
flow_tags = TagDao.get_tags_by_resource(ResourceTypeEnum.ASSISTANT, assistant_ids)
for one in res:
one.logo = cls.get_logo_share_link(one.logo)
simple_assistant = cls.return_simple_assistant_info(one)
if one.user_id == user.user_id or user.is_admin():
simple_assistant.write = True
simple_assistant.group_ids = assistant_group_dict.get(one.id, [])
simple_assistant.tags = flow_tags.get(one.id, [])
data.append(simple_assistant)
return data, total
@classmethod
def return_simple_assistant_info(cls, one: Assistant) -> AssistantSimpleInfo:
"""
Put the database's assistantmodelSimplified After processing, it returns to the front-end format
"""
simple_dict = one.model_dump(include={
'id', 'name', 'desc', 'logo', 'status', 'user_id', 'create_time', 'update_time'
})
simple_dict['user_name'] = cls.get_user_name(one.user_id)
return AssistantSimpleInfo(**simple_dict)
@classmethod
async def get_assistant_info(cls, assistant_id: str, login_user: UserPayload,
share_link: Union['ShareLink', None] = None) -> AssistantInfo:
assistant = await AssistantDao.aget_one_assistant(assistant_id)
if not assistant or assistant.is_delete:
raise AssistantNotExistsError()
# Check if you have permission to access the information
if not await login_user.async_access_check(assistant.user_id, assistant.id, AccessType.ASSISTANT_READ):
if (share_link is None
or share_link.meta_data is None
or share_link.meta_data.get("flowId") != assistant.id):
raise UnAuthorizedError()
tool_list = []
flow_list = []
knowledge_list = []
links = await AssistantLinkDao.get_assistant_link(assistant_id)
for one in links:
if one.tool_id:
tool_list.append(one.tool_id)
elif one.knowledge_id:
knowledge_list.append(one.knowledge_id)
elif one.flow_id:
flow_list.append(one.flow_id)
else:
logger.error(f'not expect link info: {one.model_dump()}')
tool_list, flow_list, knowledge_list = cls.get_link_info(tool_list, flow_list,
knowledge_list)
assistant.logo = await cls.get_logo_share_link_async(assistant.logo)
return AssistantInfo(**assistant.model_dump(),
tool_list=tool_list,
flow_list=flow_list,
knowledge_list=knowledge_list)
@classmethod
async def get_one_assistant(cls, assistant_id: str) -> Optional[Assistant]:
assistant = await AssistantDao.aget_one_assistant(assistant_id)
return assistant
# Create Assistant
@classmethod
async def create_assistant(cls, request: Request, login_user: UserPayload, assistant: Assistant) \
-> AssistantInfo:
# Check if there are any duplicate names under
if cls.judge_name_repeat(assistant.name, assistant.user_id):
raise AssistantNameRepeatError()
logger.info(f"assistant original prompt id: {assistant.id}, desc: {assistant.prompt}")
# Automatically replenish default model configurations
assistant_llm = await LLMService.get_assistant_llm()
if assistant_llm.llm_list:
for one in assistant_llm.llm_list:
if one.default:
assistant.model_name = str(one.model_id)
break
# Autogenerate Descriptions
assistant, _, _ = await cls.get_auto_info(assistant, login_user)
assistant = AssistantDao.create_assistant(assistant)
# RecordTelemetryJournal
await telemetry_service.log_event(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.NEW_APPLICATION,
trace_id=trace_id_var.get(),
event_data=NewApplicationEventData(
app_id=assistant.id,
app_name=assistant.name,
app_type=ApplicationTypeEnum.ASSISTANT
))
cls.create_assistant_hook(request, assistant, login_user)
return AssistantInfo(**assistant.model_dump(),
tool_list=[],
flow_list=[],
knowledge_list=[])
@classmethod
def create_assistant_hook(cls, request: Request, assistant: Assistant, user_payload: UserPayload) -> bool:
"""
After successful creation of the assistanthook, perform some other business logic
"""
# Query the user group the user belongs to under
user_group = UserGroupDao.get_user_group(user_payload.user_id)
if user_group:
# Batch Insert Assistant Resources into Correlation Table
batch_resource = []
for one in user_group:
batch_resource.append(GroupResource(
group_id=one.group_id,
third_id=assistant.id,
type=ResourceTypeEnum.ASSISTANT.value))
GroupResourceDao.insert_group_batch(batch_resource)
# Write Audit Log
AuditLogService.create_build_assistant(user_payload, get_request_ip(request), assistant.id)
# WritelogoCeacle
cls.get_logo_share_link(assistant.logo)
return True
# Delete Assistant
@classmethod
def delete_assistant(cls, request: Request, login_user: UserPayload, assistant_id: str) -> bool:
assistant = AssistantDao.get_one_assistant(assistant_id)
if not assistant:
raise AssistantNotExistsError()
# Judgment Authorization
if not login_user.access_check(assistant.user_id, assistant.id, AccessType.ASSISTANT_WRITE):
raise UnAuthorizedError()
AssistantDao.delete_assistant(assistant)
telemetry_service.log_event_sync(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.DELETE_APPLICATION,
trace_id=trace_id_var.get())
cls.delete_assistant_hook(request, login_user, assistant)
return True
@classmethod
def delete_assistant_hook(cls, request: Request, login_user: UserPayload, assistant: Assistant) -> bool:
""" Clean up associated assistant resources """
logger.info(f"delete_assistant_hook id: {assistant.id}, user: {login_user.user_id}")
# Write Audit Log
AuditLogService.delete_build_assistant(login_user, get_request_ip(request), assistant.id)
# Clean up associations with user groups
GroupResourceDao.delete_group_resource_by_third_id(assistant.id, ResourceTypeEnum.ASSISTANT)
# Update session information
MessageSessionDao.update_session_info_by_flow(assistant.name, assistant.desc, assistant.logo,
assistant.id, FlowType.ASSISTANT.value)
return True
@classmethod
async def auto_update_stream(cls, assistant_id: str, prompt: str, login_user: UserPayload):
""" Regenerate Assistant Prompts and Tool Selection, Only call the model capability without modifying the database data """
assistant = AssistantDao.get_one_assistant(assistant_id)
assistant.prompt = prompt
# Inisialisasillm
auto_agent = AssistantAgent(assistant, '', login_user.user_id)
await auto_agent.init_auto_update_llm()
# Streaming Generation Prompts
final_prompt = ''
async for one_prompt in auto_agent.optimize_assistant_prompt():
if one_prompt.content in ('```', 'markdown'):
continue
yield str(StreamData(event='message', data={'type': 'prompt', 'message': one_prompt.content}))
final_prompt += one_prompt.content
assistant.prompt = final_prompt
yield str(StreamData(event='message', data={'type': 'end', 'message': ""}))
# Generate opening remarks and opening questions
guide_info = auto_agent.generate_guide(assistant.prompt)
yield str(StreamData(event='message', data={'type': 'guide_word', 'message': guide_info['opening_lines']}))
yield str(StreamData(event='message', data={'type': 'end', 'message': ""}))
yield str(StreamData(event='message', data={'type': 'guide_question', 'message': guide_info['questions']}))
yield str(StreamData(event='message', data={'type': 'end', 'message': ""}))
# Automatic selection of tools and skills
tool_info = cls.get_auto_tool_info(assistant, auto_agent)
tool_info = [one.model_dump() for one in tool_info]
yield str(StreamData(event='message', data={'type': 'tool_list', 'message': tool_info}))
yield str(StreamData(event='message', data={'type': 'end', 'message': ""}))
flow_info = cls.get_auto_flow_info(assistant, auto_agent)
flow_info = [one.model_dump() for one in flow_info]
yield str(StreamData(event='message', data={'type': 'flow_list', 'message': flow_info}))
@classmethod
async def update_assistant(cls, request: Request, login_user: UserPayload, req: AssistantUpdateReq) \
-> AssistantInfo:
""" Update Assistant Information """
assistant = AssistantDao.get_one_assistant(req.id)
if not assistant:
raise AssistantNotExistsError()
cls.check_update_permission(assistant, login_user)
# Update Assistant Data
if req.name and req.name != assistant.name:
# Check if there are any duplicate names under
if cls.judge_name_repeat(req.name, assistant.user_id):
raise AssistantNameRepeatError()
assistant.name = req.name
assistant.desc = req.desc
assistant.logo = req.logo if req.logo else assistant.logo
assistant.prompt = req.prompt
assistant.guide_word = req.guide_word
assistant.guide_question = req.guide_question
assistant.model_name = req.model_name
assistant.temperature = req.temperature
assistant.update_time = datetime.now()
assistant.max_token = req.max_token
AssistantDao.update_assistant(assistant)
telemetry_service.log_event_sync(user_id=login_user.user_id, event_type=BaseTelemetryTypeEnum.EDIT_APPLICATION,
trace_id=trace_id_var.get())
# Update assistant association information
if req.tool_list is not None:
AssistantLinkDao.update_assistant_tool(assistant.id, tool_list=req.tool_list)
if req.flow_list is not None:
AssistantLinkDao.update_assistant_flow(assistant.id, flow_list=req.flow_list)
if req.knowledge_list is not None:
# Using Configuredflow Perform skill replenishment
AssistantLinkDao.update_assistant_knowledge(assistant.id,
knowledge_list=req.knowledge_list,
flow_id='')
tool_list, flow_list, knowledge_list = cls.get_link_info(req.tool_list, req.flow_list,
req.knowledge_list)
cls.update_assistant_hook(request, login_user, assistant)
return AssistantInfo(**assistant.model_dump(),
tool_list=tool_list,
flow_list=flow_list,
knowledge_list=knowledge_list)
@classmethod
def update_assistant_hook(cls, request: Request, login_user: UserPayload, assistant: Assistant) -> bool:
""" Update Assistant's Hook """
logger.info(f"delete_assistant_hook id: {assistant.id}, user: {login_user.user_id}")
# Write Audit Log
AuditLogService.update_build_assistant(login_user, get_request_ip(request), assistant.id)
# Write cache
cls.get_logo_share_link(assistant.logo)
return True
@classmethod
async def update_status(cls, request: Request, login_user: UserPayload, assistant_id: str,
status: int) -> bool:
""" Update Assistant Status """
assistant = AssistantDao.get_one_assistant(assistant_id)
if not assistant:
raise AssistantNotExistsError()
# Determine permissions
if not login_user.access_check(assistant.user_id, assistant.id, AccessType.ASSISTANT_WRITE):
raise UnAuthorizedError()
# Equal status without modification
if assistant.status == status:
return True
# Try to initializeagent, go online if initialization is successful, otherwise not go online
if status == AssistantStatus.ONLINE.value:
tmp_agent = AssistantAgent(assistant, '', login_user.user_id)
try:
await tmp_agent.init_assistant()
except Exception as e:
logger.exception('online agent init failed')
raise AssistantInitError(exception=e)
assistant.status = status
AssistantDao.update_assistant(assistant)
telemetry_service.log_event_sync(user_id=login_user.user_id, event_type=BaseTelemetryTypeEnum.EDIT_APPLICATION,
trace_id=trace_id_var.get())
cls.update_assistant_hook(request, login_user, assistant)
return True
@classmethod
def update_prompt(cls, assistant_id: str, prompt: str, user_payload: UserPayload) -> bool:
""" Update assistant prompts """
assistant = AssistantDao.get_one_assistant(assistant_id)
if not assistant:
raise AssistantNotExistsError()
cls.check_update_permission(assistant, user_payload)
assistant.prompt = prompt
AssistantDao.update_assistant(assistant)
telemetry_service.log_event_sync(user_id=user_payload.user_id,
event_type=BaseTelemetryTypeEnum.EDIT_APPLICATION,
trace_id=trace_id_var.get())
return True
@classmethod
def update_flow_list(cls, assistant_id: str, flow_list: List[str],
user_payload: UserPayload) -> bool:
""" Update Assistant Skills List """
assistant = AssistantDao.get_one_assistant(assistant_id)
if not assistant:
raise AssistantNotExistsError()
cls.check_update_permission(assistant, user_payload)
AssistantLinkDao.update_assistant_flow(assistant_id, flow_list=flow_list)
return True
@classmethod
def update_tool_list(cls, assistant_id: str, tool_list: List[int],
user_payload: UserPayload) -> bool:
""" Update Assistant Tool List """
assistant = AssistantDao.get_one_assistant(assistant_id)
if not assistant:
raise AssistantNotExistsError()
cls.check_update_permission(assistant, user_payload)
AssistantLinkDao.update_assistant_tool(assistant_id, tool_list=tool_list)
return True
@classmethod
def check_update_permission(cls, assistant: Assistant, user_payload: UserPayload) -> Any:
# Determine permissions
if not user_payload.access_check(assistant.user_id, assistant.id, AccessType.ASSISTANT_WRITE):
raise UnAuthorizedError()
# Changes are not allowed when online
if assistant.status == AssistantStatus.ONLINE.value:
raise AssistantNotEditError()
return None
@classmethod
def get_link_info(cls,
tool_list: List[int],
flow_list: List[str],
knowledge_list: List[int] = None):
tool_list = GptsToolsDao.get_list_by_ids(tool_list) if tool_list else []
flow_list = FlowDao.get_flow_by_ids(flow_list) if flow_list else []
knowledge_list = KnowledgeDao.get_list_by_ids(knowledge_list) if knowledge_list else []
return tool_list, flow_list, knowledge_list
@classmethod
def get_user_name(cls, user_id: int):
if not user_id:
return 'system'
user = cls.UserCache.get(user_id)
if user:
return user.user_name
user = UserDao.get_user(user_id)
if not user:
return f'{user_id}'
cls.UserCache.set(user_id, user)
return user.user_name
@classmethod
def judge_name_repeat(cls, name: str, user_id: int) -> bool:
""" Determine if the assistant name is a duplicate """
assistant = AssistantDao.get_assistant_by_name_user_id(name, user_id)
if assistant:
return True
return False
@classmethod
async def get_auto_info(cls, assistant: Assistant, login_user: UserPayload) -> (Assistant, List[int], List[int]):
"""
Auto Generate Assistant'sprompt, Automatically select tools and skills
return: Assistant Information, ToolsIDList, SkillsIDVertical
"""
# Inisialisasiagent
auto_agent = AssistantAgent(assistant, '', login_user.user_id)
await auto_agent.init_auto_update_llm()
# Autogenerate Descriptions
assistant.desc = auto_agent.generate_description(assistant.prompt)
return assistant, [], []
@classmethod
def get_auto_tool_info(cls, assistant: Assistant, auto_agent: AssistantAgent) -> List[GptsTools]:
# Pagination Auto-Select Tool
res = []
page = 1
page_num = 50
while True:
all_tool = GptsToolsDao.get_list_by_user(assistant.user_id, page, page_num)
if len(all_tool) == 0:
break
logger.info(f"auto select tools: page: {page}, number: {len(all_tool)}")
tool_list = []
all_tool_dict = {}
for one in all_tool:
all_tool_dict[one.name] = one
tool_list.append({
'name': one.name,
'description': one.desc if one.desc else '',
})
tool_info = []
tool_list = auto_agent.choose_tools(tool_list, assistant.prompt)
for one in tool_list:
if all_tool_dict.get(one):
tool_info.append(all_tool_dict[one])
res += tool_info
page += 1
return res
@classmethod
def get_auto_flow_info(cls, assistant: Assistant, auto_agent: AssistantAgent) -> List[Flow]:
# Automatically select skills, Before picking50skills to make automatic selections
all_flow = FlowDao.get_user_access_online_flows(assistant.user_id, 1, 50)
flow_dict = {}
flow_list = []
for one in all_flow:
flow_dict[one.name] = one
flow_list.append({
'name': one.name,
'description': one.description if one.description else '',
})
flow_list = auto_agent.choose_tools(flow_list, assistant.prompt)
flow_info = []
for one in flow_list:
if flow_dict.get(one):
flow_info.append(flow_dict[one])
return flow_info
================================================
FILE: src/backend/bisheng/api/services/assistant_agent.py
================================================
import json
import os
import time
import uuid
from typing import Any, Dict, List
from langchain_core.callbacks import Callbacks
from langchain_core.language_models import BaseLanguageModel
from langchain_core.messages import AIMessage, HumanMessage, BaseMessage
from langchain_core.runnables import RunnableConfig
from langchain_core.tools import BaseTool, Tool
from langchain_core.utils.function_calling import format_tool_to_openai_tool
from langgraph.prebuilt import create_react_agent
from loguru import logger
from bisheng.api.services.assistant_base import AssistantUtils
from bisheng.api.utils import build_flow_no_yield
from bisheng.api.v1.schemas import InputRequest
from bisheng.common.constants.enums.telemetry import ApplicationTypeEnum
from bisheng.common.errcode.assistant import AssistantModelEmptyError, AssistantModelNotConfigError, \
AssistantAutoLLMError
from bisheng.database.models.assistant import Assistant, AssistantLink, AssistantLinkDao
from bisheng.database.models.flow import FlowDao, FlowStatus
from bisheng.llm.domain.services import LLMService
from bisheng.tool.domain.services.executor import ToolExecutor
from bisheng_langchain.gpts.assistant import ConfigurableAssistant
from bisheng_langchain.gpts.auto_optimization import (generate_breif_description,
generate_opening_dialog,
optimize_assistant_prompt)
from bisheng_langchain.gpts.auto_tool_selected import ToolInfo, ToolSelector
from bisheng_langchain.gpts.prompts import ASSISTANT_PROMPT_OPT
class AssistantAgent(AssistantUtils):
# cohereThe special needs of the model prompt
ASSISTANT_PROMPT_COHERE = """{preamble}||Carefully perform the following instructions, in order, starting each with a new line.
Firstly, You may need to use complex and advanced reasoning to complete your task and answer the question. Think about how you can use the provided tools to answer the question and come up with a high level plan you will execute.
Write 'Plan:' followed by an initial high level plan of how you will solve the problem including the tools and steps required.
Secondly, Carry out your plan by repeatedly using actions, reasoning over the results, and re-evaluating your plan. Perform Action, Observation, Reflection steps with the following format. Write 'Action:' followed by a json formatted action containing the "tool_name" and "parameters"
Next you will analyze the 'Observation:', this is the result of the action.
After that you should always think about what to do next. Write 'Reflection:' followed by what you've figured out so far, any changes you need to make to your plan, and what you will do next including if you know the answer to the question.
... (this Action/Observation/Reflection can repeat N times)
Thirdly, Decide which of the retrieved documents are relevant to the user's last input by writing 'Relevant Documents:' followed by comma-separated list of document numbers. If none are relevant, you should instead write 'None'.
Fourthly, Decide which of the retrieved documents contain facts that should be cited in a good answer to the user's last input by writing 'Cited Documents:' followed a comma-separated list of document numbers. If you dont want to cite any of them, you should instead write 'None'.
Fifthly, Write 'Answer:' followed by a response to the user's last input. Use the retrieved documents to help you. Do not insert any citations or grounding markup.
Finally, Write 'Grounded answer:' followed by a response to the user's last input in high quality natural english. Use the symbols and to indicate when a fact comes from a document in the search result, e.g my fact for a fact from document 4.
Additional instructions to note:
- If the user's question is in Chinese, please answer it in Chinese.
- When there is time information involved in a question, such as recently6Months, yesterday, last year, etc., you need to use the time tool to query the time information.
""" # noqa
def __init__(self, assistant_info: Assistant, chat_id: str, invoke_user_id: int):
self.assistant = assistant_info
# To record the data tracking points
self.invoke_user_id = invoke_user_id
self.chat_id = chat_id
self.tools: List[BaseTool] = []
self.offline_flows = []
self.agent: ConfigurableAssistant | None = None
self.agent_executor_dict = {
'ReAct': 'get_react_agent_executor',
'function call': 'get_openai_functions_agent_executor',
}
self.current_agent_executor = None
self.llm: BaseLanguageModel | None = None
self.llm_agent_executor = None
# Knowledge Base Retrieval Related Parameters
self.knowledge_retriever = {'max_content': 15000, 'sort_by_source_and_index': False}
async def init_assistant(self, callbacks: Callbacks = None):
await self.init_llm()
await self.init_tools(callbacks)
await self.init_agent()
async def init_llm(self):
# Get a list of configured helper models
assistant_llm = await LLMService.get_assistant_llm()
if not assistant_llm.llm_list:
raise AssistantModelEmptyError()
default_llm = None
for one in assistant_llm.llm_list:
if str(one.model_id) == self.assistant.model_name:
default_llm = one
break
elif not default_llm and one.default:
default_llm = one
if not default_llm:
raise AssistantModelNotConfigError()
self.llm_agent_executor = default_llm.agent_executor_type
self.knowledge_retriever = {
'max_content': default_llm.knowledge_max_content,
'sort_by_source_and_index': default_llm.knowledge_sort_index
}
# Inisialisasillm
self.llm = await LLMService.get_bisheng_llm(model_id=default_llm.model_id,
temperature=self.assistant.temperature,
streaming=default_llm.streaming,
app_id=self.assistant.id,
app_name=self.assistant.name,
app_type=ApplicationTypeEnum.ASSISTANT,
user_id=self.invoke_user_id)
async def init_auto_update_llm(self):
""" Initialize Automatic Optimization prompt and other information.llmInstances """
assistant_llm = await LLMService.get_assistant_llm()
if not assistant_llm.auto_llm:
raise AssistantAutoLLMError()
self.llm = await LLMService.get_bisheng_llm(model_id=assistant_llm.auto_llm.model_id,
temperature=self.assistant.temperature,
streaming=assistant_llm.auto_llm.streaming,
app_id=self.assistant.id,
app_name=self.assistant.name,
app_type=ApplicationTypeEnum.ASSISTANT,
user_id=self.invoke_user_id)
async def init_tools(self, callbacks: Callbacks = None):
"""Get by nametool Vertical
tools_name_param:: {name: params}
"""
links: List[AssistantLink] = await AssistantLinkDao.get_assistant_link(
assistant_id=self.assistant.id)
# tool
tools: List[BaseTool] = []
tool_ids = []
flow_links = []
for link in links:
if link.tool_id:
tool_ids.append(link.tool_id)
else:
flow_links.append(link)
if tool_ids:
tools = await ToolExecutor.init_by_tool_ids(tool_ids,
app_id=self.assistant.id,
app_name=self.assistant.name,
app_type=ApplicationTypeEnum.ASSISTANT,
user_id=self.invoke_user_id,
llm=self.llm,
callbacks=callbacks)
# flow + knowledge
flow_data = FlowDao.get_flow_by_ids([link.flow_id for link in flow_links if link.flow_id])
flow_id2data = {flow.id: flow for flow in flow_data}
for link in flow_links:
knowledge_id = link.knowledge_id
if knowledge_id:
knowledge_tool = await ToolExecutor.init_knowledge_tool(self.invoke_user_id, knowledge_id, llm=self.llm,
callbacks=callbacks,
**self.knowledge_retriever)
tools.append(knowledge_tool)
else:
tmp_flow_id = link.flow_id
one_flow_data = flow_id2data.get(link.flow_id)
tool_name = f'flow_{link.flow_id}'
if not one_flow_data:
logger.warning('act=init_tools not find flow_id: {}', link.flow_id)
continue
if one_flow_data.status != FlowStatus.ONLINE.value:
self.offline_flows.append(tool_name)
logger.warning('act=init_tools not online flow_id: {}', link.flow_id)
continue
flow_graph_data = one_flow_data.data
tool_description = f'{one_flow_data.name}:{one_flow_data.description}'
try:
artifacts = {}
graph = await build_flow_no_yield(graph_data=flow_graph_data,
artifacts=artifacts,
process_file=True,
flow_id=tmp_flow_id,
chat_id=self.assistant.id)
built_object = await graph.abuild()
logger.info('act=init_flow_tool build_end')
flow_tool = Tool(name=tool_name,
func=built_object,
coroutine=built_object.ainvoke,
description=tool_description,
args_schema=InputRequest,
callbacks=callbacks)
tools.append(flow_tool)
except Exception as exc:
logger.error(f'Error processing {tmp_flow_id} tweaks: {exc}')
raise Exception(f'Flow Build Error: {exc}')
self.tools = tools
async def init_agent(self):
"""
Initialize agentagent
"""
# Introductionagentexecution parameter
agent_executor_type = self.llm_agent_executor
self.current_agent_executor = agent_executor_type
# Do the Conversion
agent_executor_type = self.agent_executor_dict.get(agent_executor_type,
agent_executor_type)
prompt = self.assistant.prompt
if getattr(self.llm, 'model_name', '').startswith('command-r'):
prompt = self.ASSISTANT_PROMPT_COHERE.format(preamble=prompt)
if self.current_agent_executor == 'ReAct':
# Inisialisasiagent
self.agent = ConfigurableAssistant(agent_executor_type=agent_executor_type,
tools=self.tools,
llm=self.llm,
assistant_message=prompt)
else:
# function-callingpattern, but also add recursive constraints
logger.info(f'Creating LangGraph agent with {len(self.tools)} tools, llm type: {type(self.llm)}')
logger.info(f'LLM streaming capability: {getattr(self.llm, "streaming", "unknown")}')
self.agent = create_react_agent(self.llm, self.tools, prompt=prompt, checkpointer=False)
logger.info(f'LangGraph agent created: {type(self.agent)}')
# areagentAdd Recursive Limit Configuration
self.agent = self.agent.with_config({'recursion_limit': 100})
logger.info(f'Agent config applied: recursion_limit=100')
async def optimize_assistant_prompt(self):
""" Automatically optimize generationprompt """
chain = ({
'assistant_name': lambda x: x['assistant_name'],
'assistant_description': lambda x: x['assistant_description'],
}
| ASSISTANT_PROMPT_OPT
| self.llm)
async for one in chain.astream({
'assistant_name': self.assistant.name,
'assistant_description': self.assistant.prompt,
}):
yield one
def sync_optimize_assistant_prompt(self):
return optimize_assistant_prompt(self.llm, self.assistant.name, self.assistant.desc)
def generate_guide(self, prompt: str):
""" Generate opening dialogue and opening questions """
return generate_opening_dialog(self.llm, prompt)
def generate_description(self, prompt: str):
""" Generate description dialog """
return generate_breif_description(self.llm, prompt)
def choose_tools(self, tool_list: List[Dict[str, str]], prompt: str) -> List[str]:
"""
Choose A Tool
tool_list: [{name: xxx, description: xxx}]
"""
tool_list = [
ToolInfo(tool_name=one['name'], tool_description=one['description'])
for one in tool_list
]
tool_selector = ToolSelector(llm=self.llm, tools=tool_list)
return tool_selector.select(self.assistant.name, prompt)
async def fake_callback(self, callback: Callbacks):
if not callback:
return
# False callback to call back skills that are offline to the front-end
for one in self.offline_flows:
run_id = uuid.uuid4()
await callback[0].on_tool_start({
'name': one,
},
input_str='flow is offline',
run_id=run_id)
await callback[0].on_tool_end(output='flow is offline', name=one, run_id=run_id)
async def record_chat_history(self, message: List[Any]):
# Record Assistant Chat History
if not os.getenv('BISHENG_RECORD_HISTORY'):
return
try:
os.makedirs('/app/data/history', exist_ok=True)
with open(f'/app/data/history/{self.assistant.id}_{time.time()}.json',
'w',
encoding='utf-8') as f:
json.dump(
{
'system': self.assistant.prompt,
'message': message,
'tools': [format_tool_to_openai_tool(t) for t in self.tools]
},
f,
ensure_ascii=False)
except Exception as e:
logger.error(f'record assistant history error: {str(e)}')
async def trim_messages(self, messages: List[Any]) -> List[Any]:
# Dapatkanencoding
enc = self.cl100k_base()
def get_finally_message(new_messages: List[Any]) -> List[Any]:
# No more processing until only one record has been trimmed
if len(new_messages) == 1:
return new_messages
total_count = 0
for one in new_messages:
if isinstance(one, HumanMessage):
total_count += len(enc.encode(one.content))
elif isinstance(one, AIMessage):
total_count += len(enc.encode(one.content))
if 'tool_calls' in one.additional_kwargs:
total_count += len(
enc.encode(json.dumps(one.additional_kwargs['tool_calls'], ensure_ascii=False))
)
else:
total_count += len(enc.encode(str(one.content)))
if total_count > self.assistant.max_token:
return get_finally_message(new_messages[1:])
return new_messages
return get_finally_message(messages)
async def run(self, query: str, chat_history: List = None, callback: Callbacks = None) -> List[BaseMessage]:
"""
Run Agent Conversation
"""
await self.fake_callback(callback)
if chat_history:
chat_history.append(HumanMessage(content=query))
inputs = chat_history
else:
inputs = [HumanMessage(content=query)]
# trim message
inputs = await self.trim_messages(inputs)
if self.current_agent_executor == 'ReAct':
result = await self.react_run(inputs, callback)
else:
result = await self.agent.ainvoke({'messages': inputs}, config=RunnableConfig(callbacks=callback))
result = result['messages']
# Record Chat History
await self.record_chat_history([one.to_json() for one in result])
return result
async def astream(self, query: str, chat_history: List = None, callback: Callbacks = None):
"""
Run Agent Conversation - Streaming version
"""
await self.fake_callback(callback)
if chat_history:
chat_history.append(HumanMessage(content=query))
inputs = chat_history
else:
inputs = [HumanMessage(content=query)]
# trim message
inputs = await self.trim_messages(inputs)
if self.current_agent_executor == 'ReAct':
# ReActMode temporarily does not support streaming, downgrade to non streaming
result = await self.react_run(inputs, callback)
# Record Chat History
await self.record_chat_history([one.to_json() for one in result])
yield result
else:
# Use Streaming Calls
config = RunnableConfig(callbacks=callback)
final_messages = []
logger.info(f'Using function-calling mode, starting astream...')
chunk_count = 0
try:
# UsemessagesPatternedLangGraph streamingattaintokenLevel of Streaming Output
async for chunk in self.agent.astream({'messages': inputs}, config=config, stream_mode="messages"):
chunk_count += 1
# stream_mode="messages" Return (message, metadata) Meta Group
message = None
if isinstance(chunk, tuple) and len(chunk) >= 2:
message, metadata = chunk[:2]
elif hasattr(chunk, 'content'):
# Directly to the message object
message = chunk
if message:
# stream_mode="messages"Returns Independencechunk, use its content directly
final_messages = [message] # Save message for history
yield [message]
except Exception as astream_error:
logger.exception(f'Error in astream async for loop: {str(astream_error)}')
raise astream_error
logger.info(f'Function calling astream completed, total chunks: {chunk_count}')
if chunk_count == 0:
logger.warning(f'No chunks received from agent.astream()! This indicates a streaming issue.')
# Record Chat History
if final_messages:
await self.record_chat_history([one.to_json() for one in final_messages])
async def react_run(self, inputs: List, callback: Callbacks = None):
""" react Mode input and execution """
result = await self.agent.ainvoke({
'input': inputs[-1].content,
'chat_history': inputs[:-1],
}, config=RunnableConfig(callbacks=callback))
logger.debug(f"react_run result: {result}")
output = result['agent_outcome'].return_values['output']
if isinstance(output, dict):
output = list(output.values())[0]
for one in result['intermediate_steps']:
inputs.append(one[0])
inputs.append(AIMessage(content=output))
return inputs
================================================
FILE: src/backend/bisheng/api/services/assistant_base.py
================================================
import os
from tiktoken.load import load_tiktoken_bpe
from tiktoken.core import Encoding as TikTokenEncoding
class AssistantUtils:
# Ignore assistant configuration has been removed from the system configuration, no such method is required for now
@staticmethod
def cl100k_base() -> TikTokenEncoding:
ENDOFTEXT = "<|endoftext|>"
FIM_PREFIX = "<|fim_prefix|>"
FIM_MIDDLE = "<|fim_middle|>"
FIM_SUFFIX = "<|fim_suffix|>"
ENDOFPROMPT = "<|endofprompt|>"
tiktoken_file = os.path.join(os.path.dirname(__file__), "tiktoken_file/cl100k_base.tiktoken")
mergeable_ranks = load_tiktoken_bpe(
# "https://openaipublic.blob.core.windows.net/encodings/cl100k_base.tiktoken",
tiktoken_file,
expected_hash="223921b76ee99bde995b7ff738513eef100fb51d18c93597a113bcffe865b2a7",
)
special_tokens = {
ENDOFTEXT: 100257,
FIM_PREFIX: 100258,
FIM_MIDDLE: 100259,
FIM_SUFFIX: 100260,
ENDOFPROMPT: 100276,
}
return TikTokenEncoding(**{
"name": "cl100k_base",
"pat_str": r"""'(?i:[sdmt]|ll|ve|re)|[^\r\n\p{L}\p{N}]?+\p{L}++|\p{N}{1,3}+| ?[^\s\p{L}\p{N}]++[\r\n]*+|\s++$|\s*[\r\n]|\s+(?!\S)|\s""",
"mergeable_ranks": mergeable_ranks,
"special_tokens": special_tokens,
})
================================================
FILE: src/backend/bisheng/api/services/audit_log.py
================================================
import asyncio
import csv
from datetime import datetime
from tempfile import NamedTemporaryFile
from typing import Any, List, Optional, Dict, Union, Tuple
from loguru import logger
from sqlalchemy import func
from sqlmodel import col, or_, and_, select
from bisheng.api.v1.schema.chat_schema import AppChatList
from bisheng.api.v1.schema.workflow import WorkflowEventType
from bisheng.api.v1.schemas import resp_200
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.http_error import UnAuthorizedError
from bisheng.common.services.config_service import settings
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync, get_minio_storage
from bisheng.database.models.assistant import AssistantDao, Assistant
from bisheng.database.models.audit_log import AuditLog, SystemId, EventType, ObjectType, AuditLogDao
from bisheng.database.models.flow import FlowDao, Flow, FlowType
from bisheng.database.models.group import Group
from bisheng.database.models.group_resource import GroupResourceDao, ResourceTypeEnum
from bisheng.database.models.message import ChatMessageDao, LikedType
from bisheng.database.models.role import Role
from bisheng.database.models.session import MessageSessionDao, SensitiveStatus, MessageSession
from bisheng.database.models.user_group import UserGroupDao
from bisheng.knowledge.domain.models.knowledge import KnowledgeDao, Knowledge
from bisheng.tool.domain.models.gpts_tools import GptsToolsType
from bisheng.user.domain.models.user import UserDao, User
from bisheng.utils import generate_uuid
# todo change to async or submit thread pool
class AuditLogService:
@classmethod
async def get_audit_log(cls, login_user: UserPayload, group_ids, operator_ids, start_time, end_time,
system_id, event_type, page, limit) -> Any:
groups = group_ids
if not login_user.is_admin():
groups = [str(one.group_id) for one in await UserGroupDao.aget_user_admin_group(login_user.user_id)]
# Not an administrator of any user groups
if not groups:
return UnAuthorizedError.return_resp()
# Filter bygroup_idand administrator permissionsgroupsDoing Intersections
if group_ids:
groups = list(set(groups) & set(group_ids))
if not groups:
return UnAuthorizedError.return_resp()
data, total = await AuditLogDao.get_audit_logs(groups, operator_ids, start_time, end_time,
system_id,
event_type,
page, limit)
return resp_200(data={'data': data, 'total': total})
@classmethod
def get_all_operators(cls, login_user: UserPayload) -> List[Dict]:
groups = []
if not login_user.is_admin():
groups = [one.group_id for one in UserGroupDao.get_user_admin_group(login_user.user_id)]
# not any group admin
if not groups:
raise UnAuthorizedError()
data = AuditLogDao.get_all_operators(groups)
res = {}
for one in data:
if not one[1]:
continue
res[one[0]] = {'user_id': one[0], 'user_name': one[1]}
return list(res.values())
@classmethod
def _chat_log(cls, user: UserPayload, ip_address: str, event_type: EventType, object_type: ObjectType,
object_id: str, object_name: str, resource_type: ResourceTypeEnum):
# Get the group to which the resource belongs
groups = GroupResourceDao.get_resource_group(resource_type, object_id)
group_ids = [one.group_id for one in groups]
audit_log = AuditLog(
operator_id=user.user_id,
operator_name=user.user_name,
group_ids=group_ids,
system_id=SystemId.CHAT.value,
event_type=event_type.value,
object_type=object_type.value,
object_id=object_id,
object_name=object_name,
ip_address=ip_address,
)
AuditLogDao.insert_audit_logs([audit_log])
@classmethod
async def _chat_log_async(cls, user: UserPayload, ip_address: str, event_type: EventType,
object_type: ObjectType,
object_id: str, object_name: str, resource_type: ResourceTypeEnum,
group_ids: List[int] = None):
# Get the group to which the resource belongs
if group_ids is None:
groups = await GroupResourceDao.aget_resource_group(resource_type, object_id)
group_ids = [one.group_id for one in groups]
audit_log = AuditLog(
operator_id=user.user_id,
operator_name=user.user_name,
group_ids=group_ids,
system_id=SystemId.CHAT.value,
event_type=event_type.value,
object_type=object_type.value,
object_id=object_id,
object_name=object_name,
ip_address=ip_address,
)
await AuditLogDao.ainsert_audit_logs([audit_log])
@classmethod
def create_chat_assistant(cls, user: UserPayload, ip_address: str, assistant_id: str):
"""
New Audit Log for Assistant Session
"""
logger.info(f"act=create_chat_assistant user={user.user_name} ip={ip_address} assistant={assistant_id}")
# Getting Assistant Details
assistant_info = AssistantDao.get_one_assistant(assistant_id)
cls._chat_log(user, ip_address, EventType.CREATE_CHAT, ObjectType.ASSISTANT,
assistant_id, assistant_info.name, ResourceTypeEnum.ASSISTANT)
@classmethod
def create_chat_flow(cls, user: UserPayload, ip_address: str, flow_id: str, flow_info=None):
"""
New Skill Session Audit Log
"""
logger.info(f"act=create_chat_flow user={user.user_name} ip={ip_address} flow={flow_id}")
if not flow_info:
flow_info = FlowDao.get_flow_by_id(flow_id)
cls._chat_log(user, ip_address, EventType.CREATE_CHAT, ObjectType.FLOW,
flow_id, flow_info.name, ResourceTypeEnum.FLOW)
@classmethod
def create_chat_workflow(cls, user: UserPayload, ip_address: str, flow_id: str, flow_info=None):
"""
New Workflow Session Audit Log
"""
logger.info(f"act=create_chat_workflow user={user.user_name} ip={ip_address} flow={flow_id}")
if not flow_info:
flow_info = FlowDao.get_flow_by_id(flow_id)
cls._chat_log(user, ip_address, EventType.CREATE_CHAT, ObjectType.WORK_FLOW,
flow_id, flow_info.name, ResourceTypeEnum.WORK_FLOW)
@classmethod
async def delete_chat_flow(cls, user: UserPayload, ip_address: str, flow_info: Flow):
"""
Delete Audit Log for Skill Session
"""
logger.info(f"act=delete_chat_flow user={user.user_name} ip={ip_address} flow={flow_info.id}")
await cls._chat_log_async(user, ip_address, EventType.DELETE_CHAT, ObjectType.FLOW,
flow_info.id, flow_info.name, ResourceTypeEnum.FLOW)
@classmethod
async def delete_chat_workflow(cls, user: UserPayload, ip_address: str, flow_info: Flow):
"""
Delete Audit Log for Skill Session
"""
logger.info(f"act=delete_chat_workflow user={user.user_name} ip={ip_address} flow={flow_info.id}")
await cls._chat_log_async(user, ip_address, EventType.DELETE_CHAT, ObjectType.WORK_FLOW,
flow_info.id, flow_info.name, ResourceTypeEnum.WORK_FLOW)
@classmethod
async def delete_chat_assistant(cls, user: UserPayload, ip_address: str, assistant_info: Assistant):
"""
Delete audit log for assistant session
"""
logger.info(f"act=delete_assistant_flow user={user.user_name} ip={ip_address} assistant={assistant_info.id}")
await cls._chat_log_async(user, ip_address, EventType.DELETE_CHAT, ObjectType.ASSISTANT,
assistant_info.id, assistant_info.name, ResourceTypeEnum.ASSISTANT)
@classmethod
def _build_log(cls, user: UserPayload, ip_address: str, event_type: EventType, object_type: ObjectType,
object_id: str,
object_name: str, resource_type: ResourceTypeEnum):
"""
Build Module Audit Log
"""
# Get which user groups the resource belongs to
groups = GroupResourceDao.get_resource_group(resource_type, object_id)
group_ids = [one.group_id for one in groups]
# Insert Audit Log
audit_log = AuditLog(
operator_id=user.user_id,
operator_name=user.user_name,
group_ids=group_ids,
system_id=SystemId.BUILD.value,
event_type=event_type.value,
object_type=object_type.value,
object_id=object_id,
object_name=object_name,
ip_address=ip_address,
)
AuditLogDao.insert_audit_logs([audit_log])
@classmethod
async def _build_log_async(cls, user: UserPayload, ip_address: str, event_type: EventType, object_type: ObjectType,
object_id: str,
object_name: str, resource_type: ResourceTypeEnum):
"""
Build Module Audit Log
"""
# Get which user groups the resource belongs to
groups = await GroupResourceDao.aget_resource_group(resource_type, object_id)
group_ids = [one.group_id for one in groups]
# Insert Audit Log
audit_log = AuditLog(
operator_id=user.user_id,
operator_name=user.user_name,
group_ids=group_ids,
system_id=SystemId.BUILD.value,
event_type=event_type.value,
object_type=object_type.value,
object_id=object_id,
object_name=object_name,
ip_address=ip_address,
)
await AuditLogDao.ainsert_audit_logs([audit_log])
@classmethod
def create_build_flow(cls, user: UserPayload, ip_address: str, flow_id: str, flow_type: Optional[int] = None):
"""
New Skill Audit Log
"""
obj_type = ObjectType.FLOW
rs_type = ResourceTypeEnum.FLOW
if flow_type == FlowType.WORKFLOW.value:
obj_type = ObjectType.WORK_FLOW
rs_type = ResourceTypeEnum.WORK_FLOW
logger.info(f"act=create_build_flow user={user.user_name} ip={ip_address} flow={flow_id}")
flow_info = FlowDao.get_flow_by_id(flow_id)
cls._build_log(user, ip_address, EventType.CREATE_BUILD, obj_type,
flow_info.id, flow_info.name, rs_type)
@classmethod
async def update_build_flow(cls, user: UserPayload, ip_address: str, flow_id: str, flow_type: Optional[int] = None):
"""
Update Skill Audit Log
"""
obj_type = ObjectType.FLOW
rs_type = ResourceTypeEnum.FLOW
if flow_type == FlowType.WORKFLOW.value:
obj_type = ObjectType.WORK_FLOW
rs_type = ResourceTypeEnum.WORK_FLOW
logger.info(f"act=update_build_flow user={user.user_name} ip={ip_address} flow={flow_id}")
flow_info = await FlowDao.aget_flow_by_id(flow_id)
await cls._build_log_async(user, ip_address, EventType.UPDATE_BUILD, obj_type, flow_info.id, flow_info.name,
rs_type)
@classmethod
def delete_build_flow(cls, user: UserPayload, ip_address: str, flow_info: Flow, flow_type: Optional[int] = None):
"""
Delete Skill Audit Log
"""
obj_type = ObjectType.FLOW
rs_type = ResourceTypeEnum.FLOW
if flow_type == FlowType.WORKFLOW.value:
obj_type = ObjectType.WORK_FLOW
rs_type = ResourceTypeEnum.WORK_FLOW
logger.info(f"act=delete_build_flow user={user.user_name} ip={ip_address} flow={flow_info.id}")
cls._build_log(user, ip_address, EventType.DELETE_BUILD, obj_type,
flow_info.id, flow_info.name, rs_type)
@classmethod
def create_build_assistant(cls, user: UserPayload, ip_address: str, assistant_id: str):
"""
New Assistant Audit Log
"""
logger.info(f"act=create_build_assistant user={user.user_name} ip={ip_address} assistant={assistant_id}")
assistant_info = AssistantDao.get_one_assistant(assistant_id)
cls._build_log(user, ip_address, EventType.CREATE_BUILD, ObjectType.ASSISTANT,
assistant_info.id, assistant_info.name, ResourceTypeEnum.ASSISTANT)
@classmethod
def update_build_assistant(cls, user: UserPayload, ip_address: str, assistant_id: str):
"""
Update the assistant's audit log
"""
logger.info(f"act=update_build_assistant user={user.user_name} ip={ip_address} assistant={assistant_id}")
assistant_info = AssistantDao.get_one_assistant(assistant_id)
cls._build_log(user, ip_address, EventType.UPDATE_BUILD, ObjectType.ASSISTANT,
assistant_info.id, assistant_info.name, ResourceTypeEnum.ASSISTANT)
@classmethod
def delete_build_assistant(cls, user: UserPayload, ip_address: str, assistant_id: str):
"""
Delete Audit Log for Assistant
"""
logger.info(f"act=delete_build_assistant user={user.user_name} ip={ip_address} assistant={assistant_id}")
assistant_info = AssistantDao.get_one_assistant(assistant_id)
cls._build_log(user, ip_address, EventType.DELETE_BUILD, ObjectType.ASSISTANT,
assistant_info.id, assistant_info.name, ResourceTypeEnum.ASSISTANT)
@classmethod
async def create_chat_message(cls, user: UserPayload, ip_address: str, message: Union[str, MessageSession]):
"""
New Chat Message Audit Log for Build Module
"""
if isinstance(message, MessageSession):
message_session = message
else:
message_session = await MessageSessionDao.async_get_one(message)
logger.info(f"act=create_chat_message user={user.user_name} ip={ip_address} session={message_session.chat_id}")
user_groups = await UserGroupDao.aget_user_group(message_session.user_id)
group_ids = [ug.group_id for ug in user_groups]
await cls._chat_log_async(user, ip_address, EventType.CREATE_CHAT, ObjectType.WORKSTATION,
message_session.chat_id, message_session.flow_name, ResourceTypeEnum.WORKSTATION,
group_ids)
@classmethod
async def delete_chat_message(cls, user: UserPayload, ip_address: str, message: Union[str, MessageSession]):
"""
Delete Chat Message Audit Log for Build Module
"""
if isinstance(message, MessageSession):
message_session = message
else:
message_session = await MessageSessionDao.async_get_one(message)
logger.info(f"act=delete_chat_message user={user.user_name} ip={ip_address} session={message_session.chat_id}")
await cls._chat_log_async(user, ip_address, EventType.DELETE_CHAT, ObjectType.WORKSTATION,
message_session.chat_id, message_session.flow_name, ResourceTypeEnum.WORKSTATION)
@classmethod
def _knowledge_log(cls, user: UserPayload, ip_address: str, event_type: EventType, object_type: ObjectType,
object_id: str, object_name: str, resource_type: ResourceTypeEnum, resource_id: str):
"""
Logs of Knowledge Base Modules
"""
# Get which user groups the resource belongs to
groups = GroupResourceDao.get_resource_group(resource_type, resource_id)
group_ids = [one.group_id for one in groups]
# Insert Audit Log
audit_log = AuditLog(
operator_id=user.user_id,
operator_name=user.user_name,
group_ids=group_ids,
system_id=SystemId.KNOWLEDGE.value,
event_type=event_type.value,
object_type=object_type.value,
object_id=object_id,
object_name=object_name,
ip_address=ip_address,
)
AuditLogDao.insert_audit_logs([audit_log])
@classmethod
def create_knowledge(cls, user: UserPayload, ip_address: str, knowledge_id: int):
"""
New Knowledge Base Audit Log
"""
logger.info(f"act=create_knowledge user={user.user_name} ip={ip_address} knowledge={knowledge_id}")
knowledge_info = KnowledgeDao.query_by_id(knowledge_id)
cls._knowledge_log(user, ip_address, EventType.CREATE_KNOWLEDGE, ObjectType.KNOWLEDGE,
str(knowledge_id), knowledge_info.name, ResourceTypeEnum.KNOWLEDGE, str(knowledge_id))
@classmethod
def delete_knowledge(cls, user: UserPayload, ip_address: str, knowledge: Knowledge):
"""
Delete Knowledge Base Audit Log
"""
logger.info(f"act=delete_knowledge user={user.user_name} ip={ip_address} knowledge={knowledge.id}")
cls._knowledge_log(user, ip_address, EventType.DELETE_KNOWLEDGE, ObjectType.KNOWLEDGE,
str(knowledge.id), knowledge.name, ResourceTypeEnum.KNOWLEDGE, str(knowledge.id))
@classmethod
def upload_knowledge_file(cls, user: UserPayload, ip_address: str, knowledge_id: int, file_name: str):
"""
Audit Logs for Knowledge Base Upload Files
"""
logger.info(f"act=upload_knowledge_file user={user.user_name} ip={ip_address}"
f" knowledge={knowledge_id} file={file_name}")
cls._knowledge_log(user, ip_address, EventType.UPLOAD_FILE, ObjectType.FILE,
str(knowledge_id), file_name, ResourceTypeEnum.KNOWLEDGE, str(knowledge_id))
@classmethod
def delete_knowledge_file(cls, user: UserPayload, ip_address: str, knowledge_id: int, file_name: str):
"""
Audit Logs for Knowledge Base Deletion Files
"""
logger.info(f"act=delete_knowledge_file user={user.user_name} ip={ip_address}"
f" knowledge={knowledge_id} file={file_name}")
cls._knowledge_log(user, ip_address, EventType.DELETE_FILE, ObjectType.FILE,
str(knowledge_id), file_name, ResourceTypeEnum.KNOWLEDGE, str(knowledge_id))
@classmethod
def _system_log(cls, user: UserPayload, ip_address: str, group_ids: List[int], event_type: EventType,
object_type: ObjectType, object_id: str, object_name: str, note: str = ''):
audit_log = AuditLog(
operator_id=user.user_id,
operator_name=user.user_name,
group_ids=group_ids,
system_id=SystemId.SYSTEM.value,
event_type=event_type.value,
object_type=object_type.value,
object_id=object_id,
object_name=object_name,
ip_address=ip_address,
note=note,
)
AuditLogDao.insert_audit_logs([audit_log])
@classmethod
def update_user(cls, user: UserPayload, ip_address: str, user_id: int, group_ids: List[int], note: str):
"""
Modify a user's user groups and roles
"""
logger.info(f"act=update_system_user user={user.user_name} ip={ip_address} user_id={user_id} note={note}")
user_info = UserDao.get_user(user_id)
cls._system_log(user, ip_address, group_ids, EventType.UPDATE_USER,
ObjectType.USER_CONF, str(user_id), user_info.user_name, note)
@classmethod
def forbid_user(cls, user: UserPayload, ip_address: str, user_info: User):
"""
user: Action User
user_info: Operated by user
"""
logger.info(f"act=forbid_user user={user.user_name} ip={ip_address} user_id={user.user_id}")
# Get the group to which the user belongs
user_group = UserGroupDao.get_user_group(user_info.user_id)
user_group = [one.group_id for one in user_group]
cls._system_log(user, ip_address, user_group, EventType.FORBID_USER,
ObjectType.USER_CONF, str(user_info.user_id), user_info.user_name)
@classmethod
def recover_user(cls, user: UserPayload, ip_address: str, user_info: User):
logger.info(f"act=recover_user user={user.user_name} ip={ip_address} user_id={user_info.user_id}")
# Get the group to which the user belongs
user_group = UserGroupDao.get_user_group(user_info.user_id)
user_group = [one.group_id for one in user_group]
cls._system_log(user, ip_address, user_group, EventType.RECOVER_USER,
ObjectType.USER_CONF, str(user_info.user_id), user_info.user_name)
@classmethod
def create_user_group(cls, user: UserPayload, ip_address: str, group_info: Group):
logger.info(f"act=create_user_group user={user.user_name} ip={ip_address} group_id={group_info.id}")
cls._system_log(user, ip_address, [group_info.id], EventType.CREATE_USER_GROUP,
ObjectType.USER_GROUP_CONF, str(group_info.id), group_info.group_name)
@classmethod
def update_user_group(cls, user: UserPayload, ip_address: str, group_info: Group):
logger.info(f"act=update_user_group user={user.user_name} ip={ip_address} group_id={group_info.id}")
# Get user group information
cls._system_log(user, ip_address, [group_info.id], EventType.UPDATE_USER_GROUP,
ObjectType.USER_GROUP_CONF, str(group_info.id), group_info.group_name)
@classmethod
def delete_user_group(cls, user: UserPayload, ip_address: str, group_info: Group):
logger.info(f"act=delete_user_group user={user.user_name} ip={ip_address} group_id={group_info.id}")
# Get user group information
cls._system_log(user, ip_address, [group_info.id], EventType.DELETE_USER_GROUP,
ObjectType.USER_GROUP_CONF, str(group_info.id), group_info.group_name)
@classmethod
def create_role(cls, user: UserPayload, ip_address: str, role: Role):
logger.info(f"act=create_role user={user.user_name} ip={ip_address} role_id={role.id}")
cls._system_log(user, ip_address, [role.group_id], EventType.CREATE_ROLE,
ObjectType.ROLE_CONF, str(role.id), role.role_name)
@classmethod
def update_role(cls, user: UserPayload, ip_address: str, role: Role):
logger.info(f"act=update_role user={user.user_name} ip={ip_address} role_id={role.id}")
cls._system_log(user, ip_address, [role.group_id], EventType.UPDATE_ROLE,
ObjectType.ROLE_CONF, str(role.id), role.role_name)
@classmethod
def delete_role(cls, user: UserPayload, ip_address: str, role: Role):
logger.info(f"act=delete_role user={user.user_name} ip={ip_address} role_id={role.id}")
cls._system_log(user, ip_address, [role.group_id], EventType.DELETE_ROLE,
ObjectType.ROLE_CONF, str(role.id), role.role_name)
@classmethod
def create_tool(cls, user: UserPayload, ip_address: str, group_ids: List[int], tool_type: GptsToolsType):
logger.info(f"act=create_tool user={user.user_name} ip={ip_address} tool_type_id={tool_type.id}")
cls._system_log(user, ip_address, group_ids, EventType.ADD_TOOL, ObjectType.TOOL, str(tool_type.id),
tool_type.name)
@classmethod
def update_tool(cls, user: UserPayload, ip_address: str, group_ids: List[int], tool_type: GptsToolsType):
logger.info(f"act=update_tool user={user.user_name} ip={ip_address} tool_type_id={tool_type.id}")
cls._system_log(user, ip_address, group_ids, EventType.UPDATE_TOOL, ObjectType.TOOL, str(tool_type.id),
tool_type.name)
@classmethod
def delete_tool(cls, user: UserPayload, ip_address: str, group_ids: List[int], tool_type: GptsToolsType):
logger.info(f"act=delete_tool user={user.user_name} ip={ip_address} tool_type_id={tool_type.id}")
cls._system_log(user, ip_address, group_ids, EventType.DELETE_TOOL, ObjectType.TOOL, str(tool_type.id),
tool_type.name)
@classmethod
def user_login(cls, user: UserPayload, ip_address: str):
logger.info(f"act=user_login user={user.user_name} ip={ip_address} user_id={user.user_id}")
# Get the group to which the user belongs
user_group = UserGroupDao.get_user_group(user.user_id)
user_group = [one.group_id for one in user_group]
cls._system_log(user, ip_address, user_group, EventType.USER_LOGIN,
ObjectType.NONE, '', '')
@classmethod
async def _dashboard_log(cls, user: UserPayload, ip_address: str, group_ids: List[int], event_type: EventType,
object_id: str, object_name: str):
audit_log = AuditLog(
operator_id=user.user_id,
operator_name=user.user_name,
group_ids=group_ids,
system_id=SystemId.DASHBOARD.value,
event_type=event_type.value,
object_type=ObjectType.DASHBOARD.value,
object_id=object_id,
object_name=object_name,
ip_address=ip_address,
)
await AuditLogDao.ainsert_audit_logs([audit_log])
@classmethod
async def create_dashboard(cls, user: UserPayload, ip_address: str, dashboard_id: str, dashboard_name: str,
group_ids: List[int]):
logger.info(f"act=create_dashboard user={user.user_name} ip={ip_address} dashboard_id={dashboard_id}")
await cls._dashboard_log(user, ip_address, group_ids, EventType.CREATE_DASHBOARD, dashboard_id, dashboard_name)
@classmethod
async def update_dashboard(cls, user: UserPayload, ip_address: str, dashboard_id: str, dashboard_name: str,
group_ids: List[int]):
logger.info(f"act=update_dashboard user={user.user_name} ip={ip_address} dashboard_id={dashboard_id}")
await cls._dashboard_log(user, ip_address, group_ids, EventType.UPDATE_DASHBOARD, dashboard_id, dashboard_name)
@classmethod
async def delete_dashboard(cls, user: UserPayload, ip_address: str, dashboard_id: str, dashboard_name: str,
group_ids: List[int]):
logger.info(f"act=delete_dashboard user={user.user_name} ip={ip_address} dashboard_id={dashboard_id}")
await cls._dashboard_log(user, ip_address, group_ids, EventType.DELETE_DASHBOARD, dashboard_id, dashboard_name)
@classmethod
async def get_filter_flow_ids(cls, user: UserPayload, flow_ids: List[str], group_ids: List[int]) -> (bool, List):
""" Setujuflow_idsAndgroup_idsGet the final SkillidFilters false: Show Back to Empty List"""
flow_ids = [one for one in flow_ids]
group_admins = []
if not user.is_admin():
user_groups = await UserGroupDao.aget_user_admin_group(user.user_id)
# Not a user group administrator, no permissions
if not user_groups:
raise UnAuthorizedError.http_exception()
group_admins = [one.group_id for one in user_groups]
# GroupingidDoing Intersections
if group_ids:
if group_admins:
# Query user group not belonging to user management, return empty
group_admins = list(set(group_admins) & set(group_ids))
if len(group_admins) == 0:
return False, []
else:
group_admins = group_ids
# Get all apps under groupingID
group_flows = []
if group_admins:
group_flows = await GroupResourceDao.get_groups_resource(group_admins,
resource_types=[ResourceTypeEnum.FLOW,
ResourceTypeEnum.WORK_FLOW,
ResourceTypeEnum.ASSISTANT,
ResourceTypeEnum.WORKSTATION])
# User group under user management has no resources
if not group_flows:
return False, []
group_flows = [one.third_id for one in group_flows]
# Acquire the final skillIDRestrict to list
filter_flow_ids = []
if flow_ids and group_flows:
filter_flow_ids = list(set(group_flows) & set(flow_ids))
if not filter_flow_ids:
return False, []
elif flow_ids:
filter_flow_ids = flow_ids
elif group_flows:
filter_flow_ids = group_flows
return True, filter_flow_ids
@classmethod
async def get_session_list(cls, user: UserPayload, flow_ids: List[str], user_ids: List[int], group_ids: List[int],
start_date: datetime, end_date: datetime,
feedback: str, sensitive_status: int, page: int, page_size: int) -> Tuple[
List[AppChatList], int]:
if user.is_admin():
# Administrator: The frontend sends out what it needs to retrieve; if nothing is sent, it retrieves all (an empty list usually means there are no restrictions or the decision is made by the business logic in subsequent logic).
search_group_ids = group_ids or []
else:
# Regular users: Administrative privileges must be verified
user_managed_groups = await UserGroupDao.aget_user_admin_group(user.user_id)
if not user_managed_groups:
raise UnAuthorizedError.http_exception()
managed_group_ids = {one.group_id for one in user_managed_groups}
if group_ids:
# Find the intersection: the intersection of the frontend requests
valid_group_ids = list(set(group_ids) & managed_group_ids)
if not valid_group_ids:
return [], 0
search_group_ids = valid_group_ids
else:
# Default: All managed groups
search_group_ids = list(managed_group_ids)
conditions = []
# Basic equality/range filtering
if sensitive_status:
conditions.append(MessageSession.sensitive_status == sensitive_status)
if user_ids:
conditions.append(col(MessageSession.user_id).in_(user_ids))
if start_date:
conditions.append(col(MessageSession.create_time) >= start_date)
if end_date:
conditions.append(col(MessageSession.create_time) <= end_date)
if flow_ids:
conditions.append(col(MessageSession.flow_id).in_(flow_ids))
# Process type filtering (fixed enumeration)
conditions.append(col(MessageSession.flow_type).in_([
FlowType.FLOW.value,
FlowType.WORKFLOW.value,
FlowType.ASSISTANT.value,
FlowType.WORKSTATION.value
]))
# Feedback status filtering
feedback_map = {
'like': col(MessageSession.like) > 0,
'dislike': col(MessageSession.dislike) > 0,
'copied': col(MessageSession.copied) > 0
}
if feedback in feedback_map:
conditions.append(feedback_map[feedback])
# Group membership filtering
if search_group_ids:
group_filters = [
func.json_contains(MessageSession.group_ids, str(gid))
for gid in search_group_ids
]
conditions.append(or_(*group_filters))
# build query statement
statement = select(MessageSession).where(and_(*conditions)).order_by(col(MessageSession.create_time).desc())
res_task = MessageSessionDao.get_statement_results(statement, page=page, limit=page_size)
total_task = MessageSessionDao.get_statement_count(statement)
res, total = await asyncio.gather(res_task, total_task)
if not res:
return [], total
target_user_ids = set()
target_flow_ids = set() # Flow/Workflow
target_assistant_ids = set() # Assistant
for session in res:
target_user_ids.add(session.user_id)
if session.flow_type in [FlowType.FLOW.value, FlowType.WORKFLOW.value, FlowType.WORKSTATION.value]:
target_flow_ids.add(session.flow_id)
elif session.flow_type == FlowType.ASSISTANT.value:
target_assistant_ids.add(session.flow_id)
target_user_ids_list = list(target_user_ids)
async def get_users_groups_map(u_ids: List[int]):
# get user groups for multiple users
if not u_ids: return {}
tasks = [user.get_user_groups(uid) for uid in u_ids]
results = await asyncio.gather(*tasks)
return dict(zip(u_ids, results))
users_data, flows_data, assistants_data, user_groups_map = await asyncio.gather(
UserDao.aget_user_by_ids(target_user_ids_list),
FlowDao.aget_flow_by_ids(list(target_flow_ids)),
AssistantDao.aget_assistants_by_ids(list(target_assistant_ids)),
get_users_groups_map(target_user_ids_list)
)
user_map = {u.user_id: u.user_name for u in users_data}
flow_map = {f.id: f.name for f in flows_data}
assistant_map = {a.id: a.name for a in assistants_data}
# Construct the return object
result: List[AppChatList] = []
for session in res:
# Determine the current name
current_name = session.flow_name
if session.flow_type in [FlowType.FLOW.value, FlowType.WORKFLOW.value, FlowType.WORKSTATION.value]:
current_name = flow_map.get(session.flow_id, current_name)
elif session.flow_type == FlowType.ASSISTANT.value:
current_name = assistant_map.get(session.flow_id, current_name)
# Append to the result set
result.append(AppChatList(
**session.model_dump(exclude={'flow_name'}),
flow_name=current_name,
like_count=session.like,
dislike_count=session.dislike,
copied_count=session.copied,
user_name=user_map.get(session.user_id, ""), # get user name
user_groups=user_groups_map.get(session.user_id, [])
))
return result, total
@classmethod
async def get_session_messages(cls, user: UserPayload, flow_ids: List[str], user_ids: List[int],
group_ids: List[int],
start_date: datetime, end_date: datetime, feedback: str,
sensitive_status: int) -> List[AppChatList]:
page = 1
page_size = 50
res = []
while True:
result, total = await cls.get_session_list(user, flow_ids, user_ids, group_ids, start_date, end_date,
feedback,
sensitive_status, page, page_size)
if not result:
break
page += 1
res.extend(await cls.get_chat_messages(result))
return res
@classmethod
async def export_session_messages(cls, user: UserPayload, flow_ids: List[str], user_ids: List[int],
group_ids: List[int],
start_date: datetime, end_date: datetime,
feedback: str, sensitive_status: int) -> str:
page = 1
page_size = 30
excel_data = [
['Session ID', 'Application Name', 'Session creation time', 'Username', 'Message Role',
'Message sending time',
'Message text content',
'Like',
'Dislike', 'copy']]
bisheng_pro = await settings.aget_system_login_method().bisheng_pro
if bisheng_pro:
excel_data[0].append('Does it meet the content security review requirements?')
while True:
result, total = await cls.get_session_list(user, flow_ids, user_ids, group_ids, start_date, end_date,
feedback,
sensitive_status, page, page_size)
if not result:
break
page += 1
chat_list = await cls.get_chat_messages(result)
for chat in chat_list:
for message in chat.messages:
message_data = [chat.chat_id, chat.flow_name, chat.create_time.strftime('%Y/%m/%d %H:%M:%S'),
chat.user_name,
'User' if message.category == 'question' else 'AI',
message.create_time.strftime('%Y/%m/%d %H:%M:%S'),
message.message,
'Yes' if message.liked == LikedType.LIKED.value else 'No',
'Yes' if message.liked == LikedType.DISLIKED.value else 'No',
'Yes' if message.copied else 'No']
if bisheng_pro:
message_data.append(
'Yes' if message.sensitive_status == SensitiveStatus.VIOLATIONS.value else 'No')
excel_data.append(message_data)
minio_client = await get_minio_storage()
tmp_object_name = f'tmp/session/export_{generate_uuid()}.csv'
with NamedTemporaryFile(mode='w', newline='') as tmp_file:
csv_writer = csv.writer(tmp_file)
csv_writer.writerows(excel_data)
tmp_file.seek(0)
await minio_client.put_object(object_name=tmp_object_name, file=tmp_file.name,
content_type='application/text',
bucket_name=minio_client.tmp_bucket)
return await minio_client.get_share_link(tmp_object_name, minio_client.tmp_bucket)
@classmethod
async def get_chat_messages(cls, chat_list: List[AppChatList]) -> List[AppChatList]:
chat_ids = [chat.chat_id for chat in chat_list]
chat_messages = await ChatMessageDao.get_all_message_by_chat_ids(chat_ids)
chat_messages_map = {}
for one in chat_messages:
if one.chat_id not in chat_messages_map:
chat_messages_map[one.chat_id] = []
chat_messages_map[one.chat_id].append(one)
for chat in chat_list:
chat_messages = chat_messages_map.get(chat.chat_id, [])
# remove workflow input event, because it's not show in web
chat.messages = [message for message in chat_messages
if message.category != WorkflowEventType.UserInput.value]
return chat_list
================================================
FILE: src/backend/bisheng/api/services/base.py
================================================
from bisheng.core.cache import InMemoryCache
from bisheng.core.cache.redis_manager import get_redis_client_sync, get_redis_client
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync, get_minio_storage
class BaseService:
LogoMemoryCache = InMemoryCache(max_size=200, expiration_time=3600 * 24)
@classmethod
def get_logo_share_link(cls, logo_path: str):
redis_client = get_redis_client_sync()
if not logo_path:
return ''
cache_key = f'logo_cache_new:{logo_path}'
# Fetch from memory first
share_url = cls.LogoMemoryCache.get(cache_key)
if share_url:
return share_url
# Then fromredisFetch in cache
share_url = redis_client.get(cache_key)
if share_url:
cls.LogoMemoryCache.set(cache_key, share_url)
return share_url
minio_client = get_minio_storage_sync()
share_url = minio_client.get_share_link_sync(logo_path)
# Ceacle5Day. Temporary link is valid for7 days
redis_client.set(cache_key, share_url, 3600 * 120)
cls.LogoMemoryCache.set(cache_key, share_url)
return share_url
@classmethod
async def get_logo_share_link_async(cls, logo_path: str):
redis_client = await get_redis_client()
if not logo_path:
return ''
cache_key = f'logo_cache_new:{logo_path}'
# Fetch from memory first
share_url = cls.LogoMemoryCache.get(cache_key)
if share_url:
return share_url
# Then fromredisFetch in cache
share_url = await redis_client.aget(cache_key)
if share_url:
cls.LogoMemoryCache.set(cache_key, share_url)
return share_url
minio_client = await get_minio_storage()
share_url = await minio_client.get_share_link(logo_path)
# Ceacle5Day. Temporary link is valid for7 days
await redis_client.aset(cache_key, share_url, 3600 * 120)
cls.LogoMemoryCache.set(cache_key, share_url)
return share_url
================================================
FILE: src/backend/bisheng/api/services/chat_imp.py
================================================
import asyncio
import json
# Pengaturan websockets Log level is NONE
import logging
from collections import defaultdict
from datetime import datetime, timedelta
from pydantic import BaseModel
from websockets import connect
from bisheng.common.errcode.http_error import ServerError
from bisheng.core.database import get_sync_db_session
from bisheng.database.models.message import ChatMessage
# Maintain a connection pool
connection_pool = defaultdict(asyncio.Queue)
logging.getLogger('websockets').setLevel(logging.ERROR)
expire = 600 # reids 60s Overdue
class TimedQueue:
def __init__(self):
self.queue = asyncio.Queue()
self.last_active = datetime.now()
async def put_nowait(self, item):
self.last_active = datetime.now()
await self.queue.put(item)
async def get_nowait(self):
self.last_active = datetime.now()
return await self.queue.get()
def empty(self):
return self.queue.empty()
def qsize(self):
return self.queue.qsize()
async def clean_inactive_queues(queue: defaultdict, timeout_threshold: timedelta):
while True:
current_time = datetime.now()
for key, timed_queue in list(queue.items()):
# If the queue is not active beyond the set threshold time, clear the queue
if current_time - timed_queue.last_active > timeout_threshold:
while not timed_queue.empty():
timed_queue.get_nowait() # Remove task from queue
del queue[key] # Delete queue
await asyncio.sleep(timeout_threshold.total_seconds())
# Maintain a connection pool
connection_pool = defaultdict(TimedQueue)
# clean_inactive_queues(connection_pool, timedelta(minutes=5))
async def get_connection(uri, identifier):
"""
DapatkanWebSocketConnections. Returns directly if there are connections available in the connection pool;
Otherwise, create a new connection and add it to the connection pool.
"""
if connection_pool[identifier].empty():
# build newWebSocketCONNECT
websocket = await connect(uri)
await connection_pool[identifier].put_nowait(websocket)
# Get Connection from Connection Pool
websocket = await connection_pool[identifier].get_nowait()
return websocket
async def release_connection(identifier, websocket):
"""
releaseWebSocketConnect and put it back into the connection pool.
"""
await connection_pool[identifier].put_nowait(websocket)
def comment_answer(message_id: int, comment: str):
with get_sync_db_session() as session:
message = session.get(ChatMessage, message_id)
if message:
message.remark = comment[:4096]
session.add(message)
session.commit()
class ContentStreamResp(BaseModel):
role: str
content: str
class ChoiceStreamResp(BaseModel):
index: int = 0
delta: ContentStreamResp = 0
session_id: str
def __str__(self) -> str:
jsonData = '{"index": "%s", "delta": %s, "session_id": "%s"}' % (
self.index, json.dumps(self.delta.dict(), ensure_ascii=False), self.session_id)
return '{"choices":[%s]}\n\n' % (jsonData)
async def event_stream(
webosocket: connect,
message: str,
session_id: str,
model: str,
streaming: bool,
):
payload = {'inputs': message, 'flow_id': model, 'chat_id': session_id}
try:
await webosocket.send(json.dumps(payload, ensure_ascii=False))
except Exception as e:
yield ServerError(exception=e).to_sse_event_instance_str()
return
sync = ''
while True:
try:
msg = await webosocket.recv()
except Exception as e:
yield ServerError(exception=e).to_sse_event_instance_str()
break
if msg is None:
continue
# Judgingmsg of income they generate.
res = json.loads(msg)
if streaming:
if res.get('type') != 'end' and res.get('message'):
delta = ContentStreamResp(role='assistant', content=res.get('message'))
yield str(ChoiceStreamResp(index=0, session_id=session_id, delta=delta))
else:
# Control the following via thecloseWhether to send a message
if res.get('type') == 'end':
sync = res.get('message')
if res.get('type') == 'close':
if not streaming and sync:
delta = ContentStreamResp(role='assistant', content=sync)
msg = ChoiceStreamResp(index=0,
session_id=session_id,
delta=delta,
finish_reason='stop')
yield '{"choices":[%s]}' % (json.dumps(msg.dict()))
# Release Connection
elif streaming:
yield 'data: [DONE]'
await release_connection(session_id, webosocket)
break
================================================
FILE: src/backend/bisheng/api/services/component.py
================================================
from typing import List
from bisheng.api.v1.schemas import UnifiedResponseModel, resp_200
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.component import ComponentExistError, ComponentNotExistError
from bisheng.database.models.component import Component, ComponentDao
class ComponentService:
@classmethod
def get_all_component(cls, user: UserPayload) -> UnifiedResponseModel[List[Component]]:
res = ComponentDao.get_user_components(user.user_id)
return resp_200(data=res)
@classmethod
def save_component(cls, component: Component) -> UnifiedResponseModel[Component]:
exist_component = ComponentDao.get_component_by_name(component.user_id, component.name)
if exist_component:
return ComponentExistError.return_resp()
component = ComponentDao.insert_component(component)
return resp_200(data=component)
@classmethod
def update_component(cls, component: Component) -> UnifiedResponseModel[Component]:
exist_component = ComponentDao.get_component_by_name(component.user_id, component.name)
if not exist_component:
return ComponentNotExistError.return_resp()
exist_component.data = component.data
exist_component.description = component.description
exist_component.version = component.version
component = ComponentDao.update_component(exist_component)
return resp_200(data=component)
@classmethod
def delete_component(cls, user_id: int, name: str) -> UnifiedResponseModel[Component]:
exist_component = ComponentDao.get_component_by_name(user_id, name)
if not exist_component:
return ComponentNotExistError.return_resp()
component = ComponentDao.delete_component(exist_component)
return resp_200(data=component)
================================================
FILE: src/backend/bisheng/api/services/dataset_service.py
================================================
from typing import Dict, List, Optional
from fastapi import HTTPException
from bisheng.api.services.base import BaseService
from bisheng.api.v1.schema.dataset_param import CreateDatasetParam
from bisheng.common.errcode.dataset import DatasetNameExistsError
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync
from bisheng.database.models.dataset import Dataset, DatasetCreate, DatasetDao, DatasetRead
from bisheng.user.domain.models.user import UserDao
class DatasetService(BaseService):
@classmethod
def build_dataset_list(cls,
page: int,
limit: int,
keyword: Optional[str] = None) -> (List[Dict], int):
"""completelist DATA"""
dataset_list = DatasetDao.filter_dataset_by_ids(dataset_ids=[],
keyword=keyword,
page=page,
limit=limit)
count_filter = []
if keyword:
count_filter.append(Dataset.name.like('%{}%'.format(keyword)))
total_count = DatasetDao.get_count_by_filter(count_filter)
user_ids = [one.user_id for one in dataset_list]
user_list = UserDao.get_user_by_ids(user_ids)
user_dict = {one.user_id: one for one in user_list}
res = [DatasetRead.model_validate(one) for one in dataset_list]
for one in res:
one.user_name = user_dict[one.user_id].user_name
if one.object_name:
one.url = one.object_name
return res, total_count
@classmethod
def create_dataset(cls, user_id: int, data: CreateDatasetParam):
"""Create Dataset"""
dataset_insert = DatasetCreate.validate(data)
dataset_insert.user_id = user_id
isExist = DatasetDao.get_dataset_by_name(data.name)
if isExist:
raise DatasetNameExistsError()
dataset = DatasetDao.insert(dataset_insert)
# Conditioning Documentation
object_name = f'/dataset/{dataset.id}/{dataset.name}'
if data.file_url:
# MinioClient().upload_minio()
dataset.object_name = object_name
if data.qa_list:
for qa in data.qa_list:
qa.dataset_id = dataset.id
# QADao.insert(qa)
dataset = DatasetDao.update(dataset)
return dataset
@classmethod
def delete_dataset(cls, dataset_id: int):
dataset = DatasetDao.get_dataset_by_id(dataset_id)
if not dataset:
raise HTTPException(status_code=404, detail='Dataset not found')
# Medical Treatment:minio
object_name = dataset.object_name
if object_name:
minio_client = get_minio_storage_sync()
minio_client.remove_object_sync(object_name=object_name)
DatasetDao.delete(dataset)
return True
@classmethod
async def get_one_by_object_name(cls, object_name: str) -> Optional[Dataset]:
dataset = await DatasetDao.aget_dataset_by_object_name(object_name)
================================================
FILE: src/backend/bisheng/api/services/etl4lm_loader.py
================================================
# flake8: noqa
"""Loads PDF with semantic splilter."""
import base64
import logging
import os
from typing import List
from uuid import uuid4
import aiofiles
import cv2
import fitz
import requests
from PIL import Image
from aiohttp import ClientTimeout
from langchain_community.docstore.document import Document
from langchain_community.document_loaders.pdf import BasePDFLoader
from bisheng.core.external.http_client.http_client_manager import get_http_client
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync
logger = logging.getLogger(__name__)
def get_image_tag(results, part):
element_id = part.get("element_id", None)
url = results.get(element_id)
return f""
def get_image_parts(partitions):
page_dict = {}
for part in partitions:
label = part["type"]
if label == "Image":
bboxes = part.get("metadata", {}).get("extra_data", {}).get("bboxes", [])
page = part.get("metadata", {}).get("extra_data", {}).get("pages", -1)
element_id = part.get("element_id", None)
if len(bboxes) == 0 or page == -1 or not element_id:
continue
item = {}
item["bboxes"] = bboxes[0]
item["element_id"] = element_id
page_id = page[0]
if page_id not in page_dict:
page_dict[page_id] = []
page_dict[page_id].append(item)
return page_dict
def crop_image(image_file, item, cropped_imag_base_dir):
element_id = item.get("element_id")
bbox = item.get("bboxes")
img = cv2.imread(image_file)
x1, y1, x2, y2 = bbox
cropped_img = img[y1:y2, x1:x2]
file_name = f"{element_id}.png"
cv2.imwrite(os.path.join(cropped_imag_base_dir, file_name), cropped_img)
return file_name
def extract_pdf_images(file_name, page_dict, doc_id, knowledge_id):
from bisheng.api.services.knowledge_imp import put_images_to_minio
from bisheng.api.services.knowledge_imp import KnowledgeUtils
from bisheng.core.cache.utils import CACHE_DIR
result = {}
base_dir = f"{CACHE_DIR}/{doc_id}"
cropped_image_base_dir = f"{base_dir}/images"
pdf_page_base_dir = f"{base_dir}/images"
if not os.path.exists(pdf_page_base_dir):
os.makedirs(pdf_page_base_dir)
if not os.path.exists(cropped_image_base_dir):
os.makedirs(cropped_image_base_dir)
pdf_document = fitz.open(file_name)
minio_client = get_minio_storage_sync()
for page_number, items in page_dict.items():
page = pdf_document[page_number]
pix = page.get_pixmap()
image = Image.frombytes("RGB", (pix.width, pix.height), pix.samples)
pdf_image_file_name = f"{pdf_page_base_dir}/{page_number}.png"
image.save(pdf_image_file_name)
for item in items:
cropped_image_file = crop_image(
pdf_image_file_name, item, cropped_image_base_dir
)
result[item["element_id"]] = (
f"/{minio_client.bucket}/{KnowledgeUtils.get_knowledge_file_image_dir(doc_id, knowledge_id)}/{cropped_image_file}"
)
put_images_to_minio(cropped_image_base_dir, knowledge_id, doc_id)
return result
def pre_handle(partitions, file_name, knowledge_id):
doc_id = str(uuid4())
image_parts = get_image_parts(partitions=partitions)
if len(image_parts) == 0:
return []
return extract_pdf_images(file_name, image_parts, doc_id, knowledge_id)
def merge_partitions(file_name, partitions, knowledge_id=None):
# Pre-processingpdf, Extracting Images
pre_handle_results = pre_handle(
partitions=partitions, file_name=file_name, knowledge_id=knowledge_id
)
text_elem_sep = "\n"
doc_content = []
is_first_elem = True
last_label = ""
prev_length = 0
metadata = dict(bboxes=[], pages=[], indexes=[], types=[])
for part in partitions:
label, text = part["type"], part["text"]
extra_data = part["metadata"]["extra_data"]
if label == "Image":
part["text"] = get_image_tag(pre_handle_results, part)
text = part["text"]
if is_first_elem:
f_text = text + "\n" if label == "Title" else text
doc_content.append(f_text)
is_first_elem = False
else:
if last_label == "Title" and label == "Title":
doc_content.append("\n" + text)
elif label == "Title":
doc_content.append("\n\n" + text)
elif label == "Table":
doc_content.append("\n\n" + text)
else:
if last_label == "Table":
doc_content.append(text_elem_sep * 2 + text)
else:
doc_content.append(text_elem_sep + text)
last_label = label
metadata["bboxes"].extend(
list(map(lambda x: list(map(int, x)), extra_data["bboxes"]))
)
metadata["pages"].extend(extra_data["pages"])
metadata["types"].extend(extra_data["types"])
indexes = extra_data["indexes"]
up_indexes = [[s + prev_length, e + prev_length] for (s, e) in indexes]
metadata["indexes"].extend(up_indexes)
prev_length += len(doc_content[-1])
content = "".join(doc_content)
return content, metadata
class Etl4lmLoader(BasePDFLoader):
"""Loads a PDF with pypdf and chunks at character level. dummy version
Loader also stores page numbers in metadata.
"""
def __init__(
self,
file_name: str,
file_path: str,
unstructured_api_key: str = None,
unstructured_api_url: str = None,
force_ocr: bool = False,
enable_formular: bool = True,
filter_page_header_footer: bool = False,
ocr_sdk_url: str = None,
timeout: int = 60,
knowledge_id: int = None,
start: int = 0,
n: int = None,
verbose: bool = False,
kwargs: dict = {},
) -> None:
"""Initialize with a file path."""
self.unstructured_api_url = unstructured_api_url
self.unstructured_api_key = unstructured_api_key
self.force_ocr = force_ocr
self.enable_formular = enable_formular
self.filter_page_header_footer = filter_page_header_footer
self.ocr_sdk_url = ocr_sdk_url
self.headers = {"Content-Type": "application/json"}
self.file_name = file_name
self.timemout = timeout
self.start = start
self.n = n
self.extra_kwargs = kwargs
self.partitions = None
self.knowledge_id = knowledge_id
super().__init__(file_path)
def load(self) -> List[Document]:
"""Load given path as pages."""
b64_data = base64.b64encode(open(self.file_path, "rb").read()).decode()
parameters = {"start": self.start, "n": self.n}
parameters.update(self.extra_kwargs)
# TODO: add filter_page_header_footer into payload when elt4llm is ready.
payload = dict(
filename=os.path.basename(self.file_name),
b64_data=[b64_data],
mode="partition",
force_ocr=self.force_ocr,
enable_formula=self.enable_formular,
ocr_sdk_url=self.ocr_sdk_url,
parameters=parameters,
)
try:
resp = requests.post(
self.unstructured_api_url, headers=self.headers, json=payload, timeout=self.timemout
)
except requests.Timeout as e:
logger.error(f"Request to etl4lm API timed out: {e}")
raise Exception("etl4lm server timeout")
except Exception as e:
if str(e).find("Timeout") != -1:
logger.error(f"Request to etl4lm API timed out: {e}")
raise Exception("etl4lm server timeout")
raise e
if resp.status_code != 200:
raise Exception(
f"file partition {os.path.basename(self.file_name)} failed resp={resp.text}"
)
resp = resp.json()
if 200 != resp.get("status_code"):
logger.info(
f"file partition {os.path.basename(self.file_name)} error resp={resp}"
)
raise Exception(
f"file partition error {os.path.basename(self.file_name)} error resp={resp}"
)
partitions = resp["partitions"]
if partitions:
logger.info(f"content_from_partitions")
self.partitions = partitions
content, metadata = merge_partitions(
self.file_path, partitions, self.knowledge_id
)
elif resp.get("text"):
logger.info(f"content_from_text")
content = resp["text"]
metadata = {
"bboxes": [],
"pages": [],
"indexes": [],
"types": [],
}
else:
logger.warning(f"content_is_empty resp={resp}")
content = ""
metadata = {}
logger.info(f'unstruct_return code={resp.get("status_code")}')
if resp.get("b64_pdf"):
with open(self.file_path, "wb") as f:
f.write(base64.b64decode(resp["b64_pdf"]))
metadata["source"] = self.file_name
doc = Document(page_content=content, metadata=metadata)
return [doc]
async def aload(self) -> List[Document]:
"""Asynchronously load given path as pages."""
async with aiofiles.open(self.file_path, "rb") as f:
file_data = await f.read()
b64_data = base64.b64encode(file_data).decode()
parameters = {"start": self.start, "n": self.n}
parameters.update(self.extra_kwargs)
# TODO: add filter_page_header_footer into payload when elt4llm is ready.
payload = dict(
filename=os.path.basename(self.file_name),
b64_data=[b64_data],
mode="partition",
force_ocr=self.force_ocr,
enable_formula=self.enable_formular,
ocr_sdk_url=self.ocr_sdk_url,
parameters=parameters,
)
try:
http_client = await get_http_client()
resp = await http_client.post(
url=self.unstructured_api_url, headers=self.headers, body=payload,
timeout=ClientTimeout(total=self.timemout)
)
except Exception as e:
if str(e).find("Timeout") != -1:
logger.error(f"Request to etl4lm API timed out: {e}")
raise Exception("etl4lm server timeout")
raise e
if (resp.status_code != 200) or (resp.body and resp.body.get("status_code") != 200):
logger.info(
f"file partition {os.path.basename(self.file_name)} error resp={resp}"
)
raise Exception(
f"file partition error {os.path.basename(self.file_name)} error resp={resp}"
)
partitions = resp.body.get("partitions")
if partitions:
logger.info(f"content_from_partitions")
self.partitions = partitions
content, metadata = merge_partitions(
self.file_path, partitions, self.knowledge_id
)
elif resp.body.get("text"):
logger.info(f"content_from_text")
content = resp.body["text"]
metadata = {
"bboxes": [],
"pages": [],
"indexes": [],
"types": [],
}
else:
logger.warning(f"content_is_empty resp={resp.body}")
content = ""
metadata = {}
logger.info(f'unstruct_return code={resp.body.get("status_code")}')
if resp.body.get("b64_pdf"):
with open(self.file_path, "wb") as f:
f.write(base64.b64decode(resp.body["b64_pdf"]))
metadata["source"] = self.file_name
doc = Document(page_content=content, metadata=metadata)
return [doc]
================================================
FILE: src/backend/bisheng/api/services/evaluation.py
================================================
import asyncio
import io
import json
import os
from collections import defaultdict
from copy import deepcopy
from io import BytesIO
from typing import List
import numpy as np
import pandas as pd
from bisheng_ragas import evaluate
from bisheng_ragas.llms.langchain import LangchainLLM
from bisheng_ragas.metrics import AnswerCorrectnessBisheng
from datasets import Dataset
from fastapi import UploadFile, HTTPException
from fastapi.encoders import jsonable_encoder
from loguru import logger
from bisheng.api.services.assistant_agent import AssistantAgent
from bisheng.api.services.flow import FlowService
from bisheng.api.utils import build_flow, build_input_keys_response
from bisheng.api.v1.schema.workflow import WorkflowEventType
from bisheng.api.v1.schemas import (UnifiedResponseModel, resp_200)
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.core.cache import InMemoryCache
from bisheng.core.cache.redis_manager import get_redis_client_sync
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync
from bisheng.database.models.assistant import AssistantDao
from bisheng.database.models.evaluation import (Evaluation, EvaluationDao, ExecType, EvaluationTaskStatus)
from bisheng.database.models.flow import FlowDao
from bisheng.database.models.flow_version import FlowVersionDao, FlowVersion
from bisheng.graph.graph.base import Graph
from bisheng.llm.domain.services import LLMService
from bisheng.user.domain.models.user import UserDao
from bisheng.utils import generate_uuid
from bisheng.worker.workflow.redis_callback import RedisCallback
from bisheng.worker.workflow.tasks import execute_workflow, continue_workflow, workflow_stateful_worker
from bisheng.workflow.common.workflow import WorkflowStatus
expire = 600
class EvaluationService:
UserCache: InMemoryCache = InMemoryCache()
@classmethod
def get_evaluation(cls,
user: UserPayload,
page: int = 1,
limit: int = 20) -> UnifiedResponseModel[List[Evaluation]]:
"""
Get a list of assessment tasks
"""
data = []
res_evaluations, total = EvaluationDao.get_my_evaluations(user.user_id, page, limit)
# SkillIDVertical
flow_ids = []
# assistantIDVertical
assistant_ids = []
# VersionIDVertical
flow_version_ids = []
for one in res_evaluations:
if one.exec_type in [ExecType.FLOW.value, ExecType.WORKFLOW.value]:
flow_ids.append(one.unique_id)
if one.version:
flow_version_ids.append(one.version)
if one.exec_type == ExecType.ASSISTANT.value:
assistant_ids.append(one.unique_id)
flow_names = {}
flow_versions = {}
assistant_names = {}
if flow_ids:
flows = FlowDao.get_flow_by_ids(flow_ids=flow_ids)
flow_names = {str(one.id): one.name for one in flows}
if flow_version_ids:
versions = FlowVersionDao.get_list_by_ids(ids=flow_version_ids)
flow_versions = {one.id: one.name for one in versions}
if assistant_ids:
assistants = AssistantDao.get_assistants_by_ids(assistant_ids=assistant_ids)
assistant_names = {str(one.id): one.name for one in assistants}
redis_client = get_redis_client_sync()
for one in res_evaluations:
evaluation_item = jsonable_encoder(one)
if one.exec_type in [ExecType.FLOW.value, ExecType.WORKFLOW.value]:
evaluation_item['unique_name'] = flow_names.get(one.unique_id)
if one.exec_type == ExecType.ASSISTANT.value:
evaluation_item['unique_name'] = assistant_names.get(one.unique_id)
if one.version:
evaluation_item['version_name'] = flow_versions.get(one.version)
if one.result_score:
evaluation_item['result_score'] = json.loads(one.result_score) if isinstance(one.result_score,
str) else one.result_score
# Processing Task Progress
if one.status != EvaluationTaskStatus.running.value:
evaluation_item['progress'] = f'100%'
elif redis_client.exists(EvaluationService.get_redis_key(one.id)):
evaluation_item['progress'] = f'{redis_client.get(EvaluationService.get_redis_key(one.id))}%'
else:
evaluation_item['progress'] = f'0%'
# Make sure the error description is returned to the front-end
evaluation_item['description'] = one.description or ''
evaluation_item['user_name'] = cls.get_user_name(one.user_id)
data.append(evaluation_item)
return resp_200(data={'data': data, 'total': total})
@classmethod
def delete_evaluation(cls, evaluation_id: int, user_payload: UserPayload) -> UnifiedResponseModel:
evaluation = EvaluationDao.get_user_one_evaluation(user_payload.user_id, evaluation_id)
if not evaluation:
raise HTTPException(status_code=404, detail='Evaluation not found')
EvaluationDao.delete_evaluation(evaluation)
return resp_200()
@classmethod
def get_user_name(cls, user_id: int):
if not user_id:
return 'system'
user = cls.UserCache.get(user_id)
if user:
return user.user_name
user = UserDao.get_user(user_id)
if not user:
return f'{user_id}'
cls.UserCache.set(user_id, user)
return user.user_name
@classmethod
def upload_file(cls, file: UploadFile):
minio_client = get_minio_storage_sync()
file_id = generate_uuid()
file_name = file.filename
file_ext = os.path.basename(file.filename).split('.')[-1]
file_path = f'evaluation/dataset/{file_id}.{file_ext}'
minio_client.put_object_sync(bucket_name=minio_client.bucket, object_name=file_path, file=file.file,
content_type=file.content_type)
return file_name, file_path
@classmethod
def upload_result_file(cls, df: pd.DataFrame):
minio_client = get_minio_storage_sync()
file_id = generate_uuid()
csv_buffer = io.BytesIO()
df.to_csv(csv_buffer, index=False)
csv_buffer.seek(0)
file_path = f'evaluation/result/{file_id}.csv'
minio_client.put_object_sync(
bucket_name=minio_client.bucket,
object_name=file_path,
file=csv_buffer.read(),
content_type='application/csv')
return file_path
@classmethod
def read_csv_file(cls, file_path: str):
minio_client = get_minio_storage_sync()
resp = minio_client.get_object_sync(bucket_name=minio_client.bucket, object_name=file_path)
if resp is None:
return None
return BytesIO(resp)
@classmethod
def parse_csv(cls, file_data: io.BytesIO):
df = pd.read_csv(file_data)
df = df.dropna(axis=0, how='all').dropna(axis=1, how='all')
if df.shape[1] < 2:
raise ValueError("CSV file must have at least two columns")
if df.columns[0] != 'question' or df.columns[1] != 'ground_truth':
raise ValueError(
"CSV file must have 'question' as the first column and 'ground_truth' as the second column")
formatted_data = [{"question": row[0], "ground_truth": row[1]} for row in df.values]
return formatted_data
@classmethod
def get_redis_key(cls, evaluation_id: int):
return f'evaluation_task_progress_{evaluation_id}'
@classmethod
async def get_input_keys(cls, flow_id: str, version_id: int):
artifacts = {}
try:
version_info = FlowVersionDao.get_version_by_id(version_id)
if not version_info:
return {"input": ""}
# L1 Users, usingbuildProcess
try:
async for message in build_flow(graph_data=version_info.data,
artifacts=artifacts,
process_file=False,
flow_id=flow_id,
chat_id=None):
if isinstance(message, Graph):
graph = message
except Exception as e:
logger.error(f'evaluation task get_input_keys {e}')
return {"input": ""}
await graph.abuild()
# Now we need to check the input_keys to send them to the client
input_keys_response = {
'input_keys': []
}
input_nodes = graph.get_input_nodes()
for node in input_nodes:
if hasattr(await node.get_result(), 'input_keys'):
input_keys = build_input_keys_response(await node.get_result(), artifacts)
input_keys['input_keys'].update({'id': node.id})
input_keys_response['input_keys'].append(input_keys.get('input_keys'))
elif 'fileNode' in node.output:
input_keys_response['input_keys'].append({
'file_path': '',
'type': 'file',
'id': node.id
})
if len(input_keys_response.get("input_keys")):
input_item = input_keys_response.get("input_keys")[0]
del input_item["id"]
return input_item
finally:
pass
return {"input": ""}
def execute_workflow_get_answer(workflow_info: FlowVersion, evaluation: Evaluation, question: str) -> str:
# Initialize workflow
unique_id = generate_uuid()
workflow_id = evaluation.unique_id
chat_id = ""
user_id = evaluation.user_id
workflow = RedisCallback(unique_id, workflow_id, chat_id, user_id)
workflow.set_workflow_data(workflow_info.data)
workflow.set_workflow_status(WorkflowStatus.WAITING.value)
hash_key = generate_uuid()
worker_node = workflow_stateful_worker.find_task_node_sync(hash_key)
execute_workflow.apply_async([unique_id, workflow_id, chat_id, user_id], queue=worker_node)
# Listen for execution results of workflows
input_event = None
for event in workflow.sync_get_response_until_break():
input_event = event
status_info = workflow.get_workflow_status()
if status_info["status"] == WorkflowStatus.FAILED.value:
raise Exception(status_info.get("reason", "workflow run failed"))
elif status_info['status'] == WorkflowStatus.SUCCESS.value:
raise Exception("Only Q&A type workflows are currently supported")
elif status_info['status'] == WorkflowStatus.INPUT.value:
if not input_event or input_event.message.get('input_schema', {}).get("tab") == "form_input":
raise Exception("Only Q&A type workflows are currently supported")
# Only workflows entered in dialog boxes are entered by default
workflow.set_user_input({input_event.message.get('node_id'): {"user_input": question}})
workflow.set_workflow_status(WorkflowStatus.INPUT_OVER.value)
worker_node = workflow_stateful_worker.find_task_node_sync(hash_key)
continue_workflow.apply_async([unique_id, workflow_id, chat_id, user_id], queue=worker_node)
events = []
for event in workflow.sync_get_response_until_break():
events.append(event)
status_info = workflow.get_workflow_status()
if status_info['status'] == WorkflowStatus.FAILED.value:
raise Exception(status_info.get("reason", "workflow run failed"))
elif status_info['status'] in [WorkflowStatus.SUCCESS.value, WorkflowStatus.INPUT.value]:
workflow.set_workflow_stop()
# Get the content of the first output event as an answer, if not, report an error
if not events:
raise Exception("Only Q&A type workflows are currently supported")
answer = None
for event in events:
if event.category in [WorkflowEventType.OutputMsg.value, WorkflowEventType.OutputWithInput.value,
WorkflowEventType.OutputWithChoose.value]:
answer = event.message.get('msg', "")
break
elif event.category == WorkflowEventType.StreamMsg.value and event.type != 'stream':
answer = event.message.get('msg', "")
break
if answer is None:
raise Exception("Only Q&A type workflows are currently supported")
return answer
else:
workflow.set_workflow_stop()
raise Exception(f"workflow status is unknown: {status_info}")
else:
raise Exception(f"workflow status is unknown: {status_info}")
async def add_evaluation_task(evaluation_id: int):
evaluation = EvaluationDao.get_one_evaluation(evaluation_id=evaluation_id)
if not evaluation:
return
redis_key = EvaluationService.get_redis_key(evaluation_id)
redis_client = get_redis_client_sync()
try:
file_data = EvaluationService.read_csv_file(evaluation.file_path)
csv_data = EvaluationService.parse_csv(file_data)
progress_increment = 80 / len(csv_data)
current_progress = 0
if evaluation.exec_type == ExecType.FLOW.value:
flow_version = FlowVersionDao.get_version_by_id(version_id=evaluation.version)
if not flow_version:
raise Exception("Flow version not found")
input_keys = await EvaluationService.get_input_keys(flow_id=evaluation.unique_id,
version_id=evaluation.version)
first_key = list(input_keys.keys())[0]
logger.info(f'evaluation task run flow input_keys: {input_keys} first_key: {first_key}')
for index, one in enumerate(csv_data):
input_dict = deepcopy(input_keys)
input_dict[first_key] = one.get('question')
flow_index, flow_result = await FlowService.exec_flow_node(
inputs=input_dict,
tweaks={},
index=0,
versions=[flow_version])
one["answer"] = flow_result.get(flow_version.id)
current_progress += progress_increment
redis_client.set(redis_key, round(current_progress))
elif evaluation.exec_type == ExecType.ASSISTANT.value:
assistant = await AssistantDao.aget_one_assistant(evaluation.unique_id)
if not assistant:
raise Exception("Assistant not found")
gpts_agent = AssistantAgent(assistant_info=assistant, chat_id="", invoke_user_id=evaluation.user_id)
await gpts_agent.init_assistant()
for index, one in enumerate(csv_data):
messages = await gpts_agent.run(one.get('question'))
if len(messages):
one["answer"] = messages[-1].content
current_progress += progress_increment
redis_client.set(redis_key, round(current_progress))
elif evaluation.exec_type == ExecType.WORKFLOW.value:
workflow_info = FlowVersionDao.get_version_by_id(version_id=evaluation.version)
if not workflow_info or workflow_info.flow_id != evaluation.unique_id:
raise Exception("workflow version info not found")
for index, one in enumerate(csv_data):
one["answer"] = await asyncio.to_thread(execute_workflow_get_answer, workflow_info, evaluation,
one.get('question', ""))
_llm = await LLMService.get_evaluation_llm_object(evaluation.user_id)
llm = LangchainLLM(_llm)
data_samples = {
"question": [one.get('question') for one in csv_data],
"answer": [one.get('answer') for one in csv_data],
"ground_truths": [[one.get('ground_truth')] for one in csv_data]
}
dataset = Dataset.from_dict(data_samples)
answer_correctness_bisheng = AnswerCorrectnessBisheng(llm=llm, human_prompt=evaluation.prompt)
score = await asyncio.to_thread(evaluate, dataset, [answer_correctness_bisheng])
df = score.to_pandas()
result = df.to_dict(orient="list")
logger.debug(f'evaluation id = {evaluation_id} result: {result}')
question = result.get('question', [])
columns = [
# Data field:Title:Type(1:Text 2:Numbers 3:%)
("question", "question", 1),
("ground_truths", "ground_truth", 1),
("answer", "answer", 1),
("statements_num_gt_only", "statements_num_gt_only", 2),
("statements_num_answer_only", "statements_num_answer_only", 2),
("statements_num_overlap", "statements_num_overlap", 2),
("answer_recall", "recall", 3),
("answer_precision", "precision", 3),
("answer_f1", "F1", 3)
]
row_list = []
tmp_dict = defaultdict(int)
total_dict = {}
for index, one in enumerate(question):
row_data = {}
for field, title, unit_type in columns:
value = result.get(field)[index]
if unit_type != 1:
tmp_dict[field] += value
if unit_type == 3:
value = f'{value * 100:.2f}%' if value not in ["nan", np.nan] else value
row_data[title] = value
row_list.append(row_data)
total_row_data = {}
for field, title, unit_type in columns:
value = tmp_dict.get(field)
if unit_type == 3:
value = f'{(value / len(row_list)) * 100:.2f}%'
total_dict[field] = value
total_row_data[title] = value
row_list.append(total_row_data)
df = pd.DataFrame(data=row_list, columns=[one[1] for one in columns])
result_file_path = EvaluationService.upload_result_file(df)
evaluation.result_score = total_dict
evaluation.status = EvaluationTaskStatus.success.value
evaluation.result_file_path = result_file_path
EvaluationDao.update_evaluation(evaluation=evaluation)
redis_client.delete(redis_key)
logger.info(f'evaluation task success id={evaluation_id}')
except Exception as e:
logger.exception(f'evaluation task failed id={evaluation_id} {str(e)}')
evaluation.status = EvaluationTaskStatus.failed.value
evaluation.description = str(e)[-500:] # Limit the length of the error description to avoid being too long
EvaluationDao.update_evaluation(evaluation=evaluation)
redis_client.delete(redis_key)
================================================
FILE: src/backend/bisheng/api/services/flow.py
================================================
import asyncio
import copy
from typing import List, Dict, AsyncGenerator, Optional, Union
from fastapi import Request
from fastapi.encoders import jsonable_encoder
from loguru import logger
from bisheng.api.services.audit_log import AuditLogService
from bisheng.api.services.base import BaseService
from bisheng.api.utils import get_L2_param_from_flow
from bisheng.api.v1.schemas import UnifiedResponseModel, resp_200, FlowVersionCreate, FlowCompareReq, resp_500, \
StreamData
from bisheng.chat.utils import process_node_data
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.flow import NotFoundVersionError, CurVersionDelError, VersionNameExistsError, \
NotFoundFlowError, \
FlowOnlineEditError, WorkFlowOnlineEditError
from bisheng.common.errcode.http_error import UnAuthorizedError
from bisheng.common.services import telemetry_service
from bisheng.core.logger import trace_id_var
from bisheng.database.models.flow import FlowDao, FlowStatus, Flow, FlowType
from bisheng.database.models.flow_version import FlowVersionDao, FlowVersionRead, FlowVersion
from bisheng.database.models.group_resource import GroupResourceDao, ResourceTypeEnum, GroupResource
from bisheng.database.models.role_access import RoleAccessDao, AccessType
from bisheng.database.models.session import MessageSessionDao
from bisheng.database.models.tag import TagDao
from bisheng.database.models.user_group import UserGroupDao
from bisheng.database.models.variable_value import VariableDao
from bisheng.processing.process import process_graph_cached, process_tweaks
from bisheng.share_link.domain.models.share_link import ShareLink
from bisheng.user.domain.models.user import UserDao
from bisheng.user.domain.models.user_role import UserRoleDao
from bisheng.utils import get_request_ip
class FlowService(BaseService):
@classmethod
def get_version_list_by_flow(cls, user: UserPayload, flow_id: str) -> UnifiedResponseModel[List[FlowVersionRead]]:
"""
By SkillID Get all versions of a skill
"""
data = FlowVersionDao.get_list_by_flow(flow_id)
# Include Deleted Versions
all_version_num = FlowVersionDao.count_list_by_flow(flow_id, include_delete=True)
return resp_200(data={
'data': data,
'total': all_version_num
})
@classmethod
def get_version_info(cls, user: UserPayload, version_id: int) -> UnifiedResponseModel[FlowVersion]:
"""
According to versionIDGet version details
"""
data = FlowVersionDao.get_version_by_id(version_id)
return resp_200(data=data)
@classmethod
def delete_version(cls, user: UserPayload, version_id: int) -> UnifiedResponseModel[None]:
"""
According to versionIDRemove Version
"""
telemetry_service.log_event_sync(
user_id=user.user_id,
event_type=BaseTelemetryTypeEnum.EDIT_APPLICATION,
trace_id=trace_id_var.get()
)
version_info = FlowVersionDao.get_version_by_id(version_id)
if not version_info:
return NotFoundVersionError.return_resp()
flow_info = FlowDao.get_flow_by_id(version_info.flow_id)
if not flow_info:
return NotFoundFlowError.return_resp()
atype = AccessType.FLOW_WRITE
if flow_info.flow_type == FlowType.WORKFLOW.value:
atype = AccessType.WORKFLOW_WRITE
# Determine permissions
if not user.access_check(flow_info.user_id, flow_info.id, atype):
return UnAuthorizedError.return_resp()
if version_info.is_current == 1:
return CurVersionDelError.return_resp()
FlowVersionDao.delete_flow_version(version_id)
return resp_200()
@classmethod
async def judge_flow_write_permission(cls, user: UserPayload, flow_id: str) -> Flow:
flow_info = await FlowDao.aget_flow_by_id(flow_id)
if not flow_info:
raise NotFoundFlowError.http_exception()
atype = AccessType.FLOW_WRITE
if flow_info.flow_type == FlowType.WORKFLOW.value:
atype = AccessType.WORKFLOW_WRITE
# Determine permissions
if not await user.async_access_check(flow_info.user_id, flow_info.id, atype):
raise UnAuthorizedError.http_exception()
return flow_info
@classmethod
async def change_current_version(cls, request: Request, login_user: UserPayload, flow_id: str, version_id: int) \
-> UnifiedResponseModel[None]:
"""
Modify Current Version
"""
await telemetry_service.log_event(
user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.EDIT_APPLICATION,
trace_id=trace_id_var.get()
)
flow_info = await cls.judge_flow_write_permission(login_user, flow_id)
# Skill go-live status not allowed Switch versions
if flow_info.status == FlowStatus.ONLINE:
return FlowOnlineEditError.return_resp()
# Switch versions
version_info = await FlowVersionDao.aget_version_by_id(version_id)
if not version_info:
return NotFoundVersionError.return_resp()
if version_info.is_current == 1:
return resp_200()
# Modify the version selected by the user for the current version
await FlowVersionDao.change_current_version(flow_id, version_info)
await cls.update_flow_hook(request, login_user, flow_info)
return resp_200()
@classmethod
async def create_new_version(cls, user: UserPayload, flow_id: str, flow_version: FlowVersionCreate) \
-> UnifiedResponseModel[FlowVersion]:
"""
Create New Version
"""
await telemetry_service.log_event(
user_id=user.user_id,
event_type=BaseTelemetryTypeEnum.EDIT_APPLICATION,
trace_id=trace_id_var.get()
)
flow_info = await cls.judge_flow_write_permission(user, flow_id)
exist_version = FlowVersionDao.get_version_by_name(flow_id, flow_version.name)
if exist_version:
return VersionNameExistsError.return_resp()
flow_version = FlowVersion(flow_id=flow_id, name=flow_version.name, description=flow_version.description,
user_id=user.user_id, data=flow_version.data,
original_version_id=flow_version.original_version_id,
flow_type=flow_version.flow_type)
# Create New Version
flow_version = FlowVersionDao.create_version(flow_version)
if flow_info.flow_type == FlowType.FLOW.value:
# Copy the original version of the form data into the new version
VariableDao.copy_variables(flow_version.flow_id, flow_version.original_version_id, flow_version.id)
try:
# Refresh this version of the form data
if not get_L2_param_from_flow(flow_version.data, flow_version.flow_id, flow_version.id):
logger.error(f'flow_id={flow_version.id} version_id={flow_version.id} extract file_node fail')
except:
pass
return resp_200(data=flow_version)
@classmethod
async def update_version_info(cls, request: Request, user: UserPayload, version_id: int,
flow_version: FlowVersionCreate) \
-> UnifiedResponseModel[FlowVersion]:
"""
It updates version information.
"""
await telemetry_service.log_event(
user_id=user.user_id,
event_type=BaseTelemetryTypeEnum.EDIT_APPLICATION,
trace_id=trace_id_var.get()
)
# Contains the deleted version. If the version is deleted, revert to this version
version_info = await FlowVersionDao.aget_version_by_id(version_id, include_delete=True)
if not version_info:
return NotFoundVersionError.return_resp()
flow_info = await cls.judge_flow_write_permission(user, version_info.flow_id)
# Version is the current version, Cannot be edited if the skill is onlinedataData, names and descriptions can be edited
if version_info.is_current == 1 and flow_info.status == FlowStatus.ONLINE.value and flow_version.data:
if flow_info.flow_type == FlowType.WORKFLOW.value:
return WorkFlowOnlineEditError.return_resp()
else:
return FlowOnlineEditError.return_resp()
version_info.name = flow_version.name if flow_version.name else version_info.name
version_info.description = flow_version.description if flow_version.description else version_info.description
version_info.data = flow_version.data if flow_version.data else version_info.data
# Restore this skill version
version_info.is_delete = 0
flow_version = await FlowVersionDao.aupdate_version(version_info)
if flow_info.flow_type == FlowType.FLOW.value:
try:
# Refresh this version of the form data
if not get_L2_param_from_flow(flow_version.data, flow_version.flow_id, flow_version.id):
logger.error(f'flow_id={flow_version.id} version_id={flow_version.id} extract file_node fail')
except:
pass
await cls.update_flow_hook(request, user, flow_info)
return resp_200(data=flow_version)
@classmethod
async def get_one_flow(cls, login_user: UserPayload, flow_id: str, share_link: Union['ShareLink', None] = None) -> \
UnifiedResponseModel[Flow]:
"""
Get details on individual skills
"""
flow_info = await FlowDao.aget_flow_by_id(flow_id)
if not flow_info:
raise NotFoundFlowError()
atype = AccessType.FLOW
if flow_info.flow_type == FlowType.WORKFLOW.value:
atype = AccessType.WORKFLOW
if not await login_user.async_access_check(flow_info.user_id, flow_info.id, atype):
if (share_link is None
or share_link.meta_data is None
or share_link.meta_data.get("flowId") != flow_info.id):
raise UnAuthorizedError()
flow_info.logo = await cls.get_logo_share_link_async(flow_info.logo)
return resp_200(data=flow_info)
@classmethod
def get_all_flows(cls, user: UserPayload, name: str, status: int, tag_id: int = 0, page: int = 1,
page_size: int = 10, flow_type: Optional[int] = FlowType.FLOW.value) -> UnifiedResponseModel[
List[Dict]]:
"""
Get all the skills
"""
flow_ids = []
if tag_id:
ret = TagDao.get_resources_by_tags_batch([tag_id], [ResourceTypeEnum.FLOW, ResourceTypeEnum.WORK_FLOW])
flow_ids = [one.resource_id for one in ret]
assistant_ids = [one.resource_id for one in ret]
if not assistant_ids:
return resp_200(data={
'data': [],
'total': 0
})
# Get a list of skills visible to the user
if user.is_admin():
data = FlowDao.get_flows(user.user_id, "admin", name, status, flow_ids, page, page_size, flow_type)
total = FlowDao.count_flows(user.user_id, "admin", name, status, flow_ids, flow_type)
else:
user_role = UserRoleDao.get_user_roles(user.user_id)
role_ids = [role.role_id for role in user_role]
role_access = RoleAccessDao.get_role_access_batch(role_ids, [AccessType.FLOW, AccessType.WORKFLOW])
flow_id_extra = []
if role_access:
flow_id_extra = [access.third_id for access in role_access]
data = FlowDao.get_flows(user.user_id, flow_id_extra, name, status, flow_ids, page, page_size, flow_type)
total = FlowDao.count_flows(user.user_id, flow_id_extra, name, status, flow_ids, flow_type)
# Get the user information and version information corresponding to the skill list
# SkillIDVertical
flow_ids = []
# Skill Creation User'sIDVertical
user_ids = []
for one in data:
flow_ids.append(one.id)
user_ids.append(one.user_id)
# Get user information in the list
user_infos = UserDao.get_user_by_ids(user_ids)
user_dict = {one.user_id: one.user_name for one in user_infos}
# Get version information in the list
version_infos = FlowVersionDao.get_list_by_flow_ids(flow_ids)
flow_versions = {}
for one in version_infos:
if one.flow_id not in flow_versions:
flow_versions[one.flow_id] = []
flow_versions[one.flow_id].append(jsonable_encoder(one))
# Get the group to which the skill belongs
flow_groups = GroupResourceDao.get_resources_group(ResourceTypeEnum.FLOW, flow_ids)
flow_group_dict = {}
for one in flow_groups:
if one.third_id not in flow_group_dict:
flow_group_dict[one.third_id] = []
flow_group_dict[one.third_id].append(one.group_id)
# Get Skill Associatedtag
flow_tags = TagDao.get_tags_by_resource(ResourceTypeEnum.FLOW, flow_ids)
# Re-stitch Skills ListlistMessage
res = []
for one in data:
one.logo = cls.get_logo_share_link(one.logo)
flow_info = jsonable_encoder(one)
flow_info['user_name'] = user_dict.get(one.user_id, one.user_id)
flow_info['write'] = True if user.is_admin() or user.user_id == one.user_id else False
flow_info['version_list'] = flow_versions.get(one.id, [])
flow_info['group_ids'] = flow_group_dict.get(one.id, [])
flow_info['tags'] = flow_tags.get(one.id, [])
res.append(flow_info)
return resp_200(data={
"data": res,
"total": total
})
@classmethod
async def get_compare_tasks(cls, user: UserPayload, req: FlowCompareReq) -> List:
"""
Get Comparison Tasks
"""
if req.question_list is None or len(req.question_list) == 0:
return []
if req.version_list is None or len(req.version_list) == 0:
return []
if req.node_id is None:
return []
# Get version data
version_infos = FlowVersionDao.get_list_by_ids(req.version_list)
# Start a new event loop
tasks = []
for index, question in enumerate(req.question_list):
question_index = index
tmp_inputs = copy.deepcopy(req.inputs)
tmp_inputs, tmp_tweaks = cls.parse_compare_inputs(tmp_inputs, question)
for version in version_infos:
task = asyncio.create_task(cls.exec_flow_node(
copy.deepcopy(tmp_inputs), tmp_tweaks, question_index, [version]))
tasks.append(task)
return tasks
@classmethod
def parse_compare_inputs(cls, inputs: Dict, question) -> (Dict, Dict):
# Under special treatmentinputs, Hold and PasswebsocketSessions are formatted consistently
if inputs.get('data', None):
for one in inputs['data']:
one['id'] = one['nodeId']
if 'InputFile' in one['id']:
one['file_path'] = one['value']
# Paddingquestion and Generate Replacementtweaks
for key, val in inputs.items():
if key != 'data' and key != 'id':
# Default inputkey, replace the firstkey
logger.info(f"replace_inputs {key} replace to {question}")
inputs[key] = question
break
if 'id' in inputs:
inputs.pop('id')
# Replacement Node Parameters, GantiinputFileNodeAndVariableNodeParameters
tweaks = {}
if 'data' in inputs:
node_data = inputs.pop('data')
if node_data:
tweaks = process_node_data(node_data)
return inputs, tweaks
@classmethod
async def compare_flow_node(cls, user: UserPayload, req: FlowCompareReq) -> UnifiedResponseModel[Dict]:
"""
Compare nodes in two versions Output Results
"""
tasks = await cls.get_compare_tasks(user, req)
if len(tasks) == 0:
return resp_200(data=[])
res = [{} for _ in range(len(req.question_list))]
try:
for one in asyncio.as_completed(tasks):
index, answer = await one
if res[index]:
res[index].update(answer)
else:
res[index] = answer
except Exception as e:
return resp_500(message="Skill comparison error:{}".format(str(e)))
return resp_200(data=res)
@classmethod
async def compare_flow_stream(cls, user: UserPayload, req: FlowCompareReq) -> AsyncGenerator:
"""
Compare nodes in two versions Output Results
"""
tasks = await cls.get_compare_tasks(user, req)
if len(tasks) == 0:
return
for one in asyncio.as_completed(tasks):
index, answer_dict = await one
for version_id, answer in answer_dict.items():
yield str(StreamData(event='message',
data={'question_index': index,
'version_id': version_id,
'answer': answer}))
@classmethod
async def exec_flow_node(cls, inputs: Dict, tweaks: Dict, index: int, versions: List[FlowVersion]):
# Gantianswer
answer_result = {}
# Execute two versions of the node
for one in versions:
graph_data = process_tweaks(one.data, tweaks)
try:
result = await process_graph_cached(graph_data,
inputs,
session_id=None,
history_count=10,
flow_id=one.flow_id)
except Exception as e:
logger.exception(f"exec flow node error version_id: {one.name}")
answer_result[one.id] = f"{one.name}Version skill execution error: {str(e)}"
continue
if isinstance(result, dict) and 'result' in result:
task_result = result['result']
elif hasattr(result, 'result') and hasattr(result, 'session_id'):
task_result = result.result
else:
logger.error(f"exec flow node error version_id: {one.id}, answer: {result}")
task_result = {"answer": "flow exec error"}
answer_result[one.id] = list(task_result.values())[0]
return index, answer_result
@classmethod
def create_flow_hook(cls, request: Request, login_user: UserPayload, flow_info: Flow, version_id,
flow_type: Optional[int] = None) -> bool:
logger.info(f'create_flow_hook flow: {flow_info.id}, user_payload: {login_user.user_id}')
# Write the form required for the skill into the database
try:
if flow_info.data and not get_L2_param_from_flow(flow_info.data, flow_info.id, version_id):
logger.error(f'flow_id={flow_info.id} extract file_node fail')
except Exception:
pass
# Associate the skill to the corresponding user group
user_group = UserGroupDao.get_user_group(login_user.user_id)
if user_group:
batch_resource = []
resource_type = ResourceTypeEnum.FLOW.value
if flow_type and flow_type == FlowType.WORKFLOW.value:
resource_type = ResourceTypeEnum.WORK_FLOW.value
for one in user_group:
batch_resource.append(
GroupResource(group_id=one.group_id,
third_id=flow_info.id,
type=resource_type))
GroupResourceDao.insert_group_batch(batch_resource)
# Write Audit Log
AuditLogService.create_build_flow(login_user, get_request_ip(request), flow_info.id, flow_type)
# WritelogoCeacle
cls.get_logo_share_link(flow_info.logo)
return True
@classmethod
async def update_flow_hook(cls, request: Request, login_user: UserPayload, flow_info: Flow) -> bool:
# Write Audit Log
await AuditLogService.update_build_flow(login_user, get_request_ip(request), flow_info.id,
flow_type=flow_info.flow_type)
# WritelogoCeacle
await cls.get_logo_share_link_async(flow_info.logo)
return True
@classmethod
def delete_flow_hook(cls, request: Request, login_user: UserPayload, flow_info: Flow) -> bool:
logger.info(f'delete_flow_hook flow: {flow_info.id}, user_payload: {login_user.user_id}')
# Write Audit Log
AuditLogService.delete_build_flow(login_user, get_request_ip(request), flow_info, flow_type=flow_info.flow_type)
# Delete Skills Associated Under User Group
GroupResourceDao.delete_group_resource_by_third_id(flow_info.id, ResourceTypeEnum.FLOW)
# Update session information
MessageSessionDao.update_session_info_by_flow(flow_info.name, flow_info.description, flow_info.logo,
flow_info.id, flow_info.flow_type)
return True
================================================
FILE: src/backend/bisheng/api/services/invite_code/__init__.py
================================================
================================================
FILE: src/backend/bisheng/api/services/invite_code/code_validator.py
================================================
import random
import string
class VoucherGenerator:
def __init__(self, length=10):
self.length = length
# Exclude similar letters and numbers: 'I', 'l', 'O', '0', '1'
self.characters = ''.join(set(string.ascii_letters + string.digits) - set('IlOo01'))
self.weights = [7, 9, 10, 5, 8, 4, 2, 1, 3] # Weighting Factor
self.check_digits = ['1', '0', 'X', '9', '8', '7', '6', '5', '4', '3', '2'] # Checksum Correspondence Form
def generate_voucher(self):
voucher_base = ''.join(random.choices(self.characters, k=self.length - 1))
check_digit = self.calculate_check_digit(voucher_base)
return voucher_base + check_digit
def calculate_check_digit(self, voucher_base):
total = sum(self.weights[i] * (ord(char) - ord('A') if char.isalpha() else int(char)) for i, char in
enumerate(voucher_base))
remainder = total % 11
return self.check_digits[remainder]
def validate_voucher(self, voucher):
if len(voucher) != 10:
return False, "Invalid voucher length"
voucher_base = voucher[:-1]
provided_check_digit = voucher[-1]
calculated_check_digit = self.calculate_check_digit(voucher_base)
if provided_check_digit == calculated_check_digit:
return True, "Valid voucher"
else:
return False, "Invalid voucher"
# Example Usage
if __name__ == "__main__":
generator = VoucherGenerator()
voucher_code = generator.generate_voucher() # Generate a unique redemption code
print(f"Generated voucher code: {voucher_code}")
# Verify Redeem Code
is_valid, info = generator.validate_voucher(voucher_code)
print(f"Is valid: {is_valid}, Info: {info}")
# Try to validate an invalid redemption code
invalid_voucher_code = 'ABCDEFGHJK967'
is_valid, info = generator.validate_voucher(invalid_voucher_code)
print(f"Is valid: {is_valid}, Info: {info}")
================================================
FILE: src/backend/bisheng/api/services/invite_code/invite_code.py
================================================
from loguru import logger
from bisheng.api.services.invite_code.code_validator import VoucherGenerator
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.linsight import InviteCodeBindError, InviteCodeInvalidError
from bisheng.database.models.invite_code import InviteCode, InviteCodeDao
from bisheng.utils import generate_uuid
class InviteCodeService:
@classmethod
async def use_invite_code(cls, user_id: int) -> bool:
"""
using referral code
:param user_id: UsersID
:return: Invitation code results
"""
logger.debug(f"use_invite_code {user_id}")
codes = await InviteCodeDao.get_user_bind_code(user_id)
for one in codes:
flag = await InviteCodeDao.use_invite_code(user_id, one.code)
if flag:
logger.debug(f"use_invite_code {user_id}, {one.code} success")
return True
return False
@classmethod
async def revoke_invite_code(cls, user_id: int) -> bool:
"""
Revoke Invitation Code
:param user_id: UsersID
:return: Invitation code revocation result
"""
logger.debug(f"revoke_invite_code {user_id}")
codes = await InviteCodeDao.get_user_all_code(user_id)
for one in codes:
# Description is a brand new invite code and has not been used
if one.used <= 0:
continue
flag = await InviteCodeDao.revoke_invite_code_used(user_id, one.code)
if flag:
logger.debug(f"revoke_invite_code {user_id}, {one.code} success")
return True
return False
@classmethod
async def create_batch_invite_codes(cls, login_user: UserPayload, name: str, num: int, limit: int) -> list[str]:
"""
Bulk create invite codes
:param login_user: Action user information
:param name: Invitation code name
:param num: How many codes
:param limit: Number of uses per invite code
:return: Invitation code list created
"""
generator = VoucherGenerator()
code_list = []
batch_id = generate_uuid()
for i in range(num):
code_list.append(InviteCode(
code=generator.generate_voucher(),
batch_id=batch_id,
batch_name=name,
limit=limit,
created_id=login_user.user_id,
))
# Check if the generated invite code is a duplicate
unique_codes = []
for code in code_list:
if code.code in unique_codes:
raise ValueError(f"Duplicate invite code found: {code.code}")
unique_codes.append(code.code)
# Call the database operation to save the invite code
await InviteCodeDao.insert_invite_code(code_list)
return unique_codes
@classmethod
async def get_invite_code_num(cls, login_user: UserPayload) -> int:
"""
Get the number of times a user can use an invite code
:param login_user: Action user information
:return: Invitation code usage
"""
nums = 0
codes = await InviteCodeDao.get_user_bind_code(login_user.user_id)
for one in codes:
nums += one.limit - one.used
return nums
@classmethod
async def bind_invite_code(cls, login_user: UserPayload, code: str) -> bool:
"""
Binding Invitation Code
:param login_user: Action user information
:param code: Invitation Code
:return: Binding Results
"""
generator = VoucherGenerator()
flag, _ = generator.validate_voucher(code)
if not flag:
raise InviteCodeInvalidError()
codes = await InviteCodeDao.get_user_bind_code(login_user.user_id)
if codes:
raise InviteCodeBindError()
flag = await InviteCodeDao.bind_invite_code(login_user.user_id, code)
if not flag:
raise InviteCodeInvalidError()
return flag
================================================
FILE: src/backend/bisheng/api/services/knowledge.py
================================================
import json
import math
import os
from datetime import datetime
from typing import Any, Dict, List
from fastapi import BackgroundTasks, Request
from loguru import logger
from pymilvus import Collection
from bisheng.api.services.audit_log import AuditLogService
from bisheng.api.services.knowledge_imp import (
KnowledgeUtils,
decide_vectorstores,
delete_knowledge_file_vectors,
process_file_task,
async_read_chunk_text,
)
from bisheng.api.v1.schema.knowledge import KnowledgeFileResp
from bisheng.api.v1.schemas import (
FileChunk,
FileProcessBase,
KnowledgeFileOne,
KnowledgeFileProcess,
UpdatePreviewFileChunk, ExcelRule, KnowledgeFileReProcess,
)
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum
from bisheng.common.constants.vectorstore_metadata import KNOWLEDGE_RAG_METADATA_SCHEMA
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.http_error import NotFoundError, UnAuthorizedError, ServerError
from bisheng.common.errcode.knowledge import (
KnowledgeChunkError,
KnowledgeExistError,
KnowledgeNoEmbeddingError, KnowledgeNotQAError, KnowledgeFileFailedError,
)
from bisheng.common.schemas.telemetry.event_data_schema import NewKnowledgeBaseEventData
from bisheng.common.services import telemetry_service
from bisheng.core.cache.redis_manager import get_redis_client_sync, get_redis_client
from bisheng.core.cache.utils import file_download, async_file_download
from bisheng.core.logger import trace_id_var
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync, get_minio_storage
from bisheng.database.models.group_resource import (
GroupResource,
GroupResourceDao,
ResourceTypeEnum,
)
from bisheng.database.models.role_access import AccessType, RoleAccessDao
from bisheng.database.models.user_group import UserGroupDao
from bisheng.interface.embeddings.custom import FakeEmbedding
from bisheng.knowledge.domain.knowledge_rag import KnowledgeRag
from bisheng.knowledge.domain.models.knowledge import (
Knowledge,
KnowledgeCreate,
KnowledgeDao,
KnowledgeRead,
KnowledgeTypeEnum,
KnowledgeUpdate, KnowledgeState,
)
from bisheng.knowledge.domain.models.knowledge_file import (
KnowledgeFile,
KnowledgeFileDao,
KnowledgeFileStatus, ParseType,
)
from bisheng.llm.domain.const import LLMModelType
from bisheng.llm.domain.models import LLMDao
from bisheng.user.domain.models.user import UserDao
from bisheng.user.domain.models.user_role import UserRoleDao
from bisheng.utils import generate_uuid, generate_knowledge_index_name
from bisheng.utils import get_request_ip
class KnowledgeService(KnowledgeUtils):
@classmethod
async def get_knowledge(
cls,
request: Request,
login_user: UserPayload,
knowledge_type: KnowledgeTypeEnum,
name: str = None,
sort_by: str = "update_time",
page: int = 1,
limit: int = 10,
) -> (List[KnowledgeRead], int):
if not login_user.is_admin():
knowledge_id_extra = []
user_role = await UserRoleDao.aget_user_roles(login_user.user_id)
if user_role:
role_ids = [role.role_id for role in user_role]
role_access = await RoleAccessDao.aget_role_access(role_ids, AccessType.KNOWLEDGE)
if role_access:
knowledge_id_extra = [
int(access.third_id) for access in role_access
]
res = await KnowledgeDao.aget_user_knowledge(
login_user.user_id,
knowledge_id_extra,
knowledge_type,
name,
sort_by,
page,
limit,
)
total = await KnowledgeDao.acount_user_knowledge(
login_user.user_id, knowledge_id_extra, knowledge_type, name
)
else:
res = await KnowledgeDao.aget_all_knowledge(
name, knowledge_type, sort_by, page=page, limit=limit
)
total = await KnowledgeDao.acount_all_knowledge(name, knowledge_type)
result = cls.convert_knowledge_read(login_user, res)
return result, total
@classmethod
def convert_knowledge_read(
cls, login_user: UserPayload, knowledge_list: List[Knowledge]
) -> List[KnowledgeRead]:
db_user_ids = {one.user_id for one in knowledge_list}
db_user_info = UserDao.get_user_by_ids(list(db_user_ids))
db_user_dict = {one.user_id: one.user_name for one in db_user_info}
res = []
for one in knowledge_list:
res.append(
KnowledgeRead(
**one.model_dump(),
user_name=db_user_dict.get(one.user_id, str(one.user_id)),
copiable=login_user.access_check(
one.user_id, str(one.id), AccessType.KNOWLEDGE_WRITE
),
)
)
return res
@classmethod
def get_knowledge_info(
cls, request: Request, login_user: UserPayload, knowledge_id: List[int]
) -> List[KnowledgeRead]:
db_knowledge = KnowledgeDao.get_list_by_ids(knowledge_id)
filter_knowledge = db_knowledge
if not login_user.is_admin():
filter_knowledge = []
for one in db_knowledge:
# Determine if the user has permission
if login_user.access_check(
one.user_id, str(one.id), AccessType.KNOWLEDGE
):
filter_knowledge.append(one)
if not filter_knowledge:
return []
return cls.convert_knowledge_read(login_user, filter_knowledge)
@classmethod
def create_knowledge(
cls, request: Request, login_user: UserPayload, knowledge: KnowledgeCreate
) -> Knowledge:
# Determine if the Knowledge Base is Renamed
repeat_knowledge = KnowledgeDao.get_knowledge_by_name(
knowledge.name, login_user.user_id
)
if repeat_knowledge:
raise KnowledgeExistError.http_exception()
db_knowledge = Knowledge.model_validate(knowledge)
# CorrectionembeddingModels
if not db_knowledge.model:
raise KnowledgeNoEmbeddingError.http_exception()
embed_info = LLMDao.get_model_by_id(int(db_knowledge.model))
if not embed_info:
raise KnowledgeNoEmbeddingError.http_exception()
if embed_info.model_type != LLMModelType.EMBEDDING.value:
raise KnowledgeNoEmbeddingError.http_exception()
# generate index_name and collection_name
db_knowledge.index_name = generate_knowledge_index_name()
db_knowledge.collection_name = db_knowledge.index_name
# Insert into Database
db_knowledge.user_id = login_user.user_id
db_knowledge = KnowledgeDao.insert_one(db_knowledge)
try:
vector_client = KnowledgeRag.init_knowledge_milvus_vectorstore_sync(login_user.user_id,
knowledge=db_knowledge,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
es_client = KnowledgeRag.init_knowledge_es_vectorstore_sync(knowledge=db_knowledge,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
es_client._store._create_index_if_not_exists()
except Exception as e:
logger.exception("create knowledge index name error")
# Handling the next steps in creating a Knowledge Base
cls.create_knowledge_hook(request, login_user, db_knowledge)
return db_knowledge
@classmethod
def create_knowledge_hook(
cls, request: Request, login_user: UserPayload, knowledge: Knowledge
):
# Query the user group the user belongs to under
user_group = UserGroupDao.get_user_group(login_user.user_id)
if user_group:
# Batch Insert Knowledge Base Resources into Associated Tables
batch_resource = []
for one in user_group:
batch_resource.append(
GroupResource(
group_id=one.group_id,
third_id=knowledge.id,
type=ResourceTypeEnum.KNOWLEDGE.value,
)
)
GroupResourceDao.insert_group_batch(batch_resource)
# Log Audit Logs
AuditLogService.create_knowledge(
login_user, get_request_ip(request), knowledge.id
)
telemetry_service.log_event_sync(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.NEW_KNOWLEDGE_BASE,
trace_id=trace_id_var.get(),
event_data=NewKnowledgeBaseEventData(
kb_id=knowledge.id,
kb_name=knowledge.name,
kb_type=knowledge.type
))
return True
@classmethod
def update_knowledge(
cls, request: Request, login_user: UserPayload, knowledge: KnowledgeUpdate
) -> KnowledgeRead:
db_knowledge = KnowledgeDao.query_by_id(knowledge.knowledge_id)
if not db_knowledge:
raise NotFoundError.http_exception()
# judge access
if not login_user.access_check(
db_knowledge.user_id, str(db_knowledge.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError.http_exception()
if knowledge.name and knowledge.name != db_knowledge.name:
repeat_knowledge = KnowledgeDao.get_knowledge_by_name(
knowledge.name, db_knowledge.user_id
)
if repeat_knowledge and repeat_knowledge.id != db_knowledge.id:
raise KnowledgeExistError.http_exception()
db_knowledge.name = knowledge.name
db_knowledge.description = knowledge.description
db_knowledge = KnowledgeDao.update_one(db_knowledge)
user = UserDao.get_user(db_knowledge.user_id)
res = KnowledgeRead(
**db_knowledge.model_dump(),
user_name=user.user_name if user else db_knowledge.user_id,
)
return res
@classmethod
def delete_knowledge(
cls,
request: Request,
login_user: UserPayload,
knowledge_id: int,
only_clear: bool = False,
):
knowledge = KnowledgeDao.query_by_id(knowledge_id)
if not knowledge:
raise NotFoundError.http_exception()
if not login_user.access_check(
knowledge.user_id, str(knowledge_id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError.http_exception()
# Cleaned vectorData in
cls.delete_knowledge_file_in_vector(knowledge)
# CleanedminioData
cls.delete_knowledge_file_in_minio(knowledge_id)
# DeletemysqlDATA
KnowledgeDao.delete_knowledge(knowledge_id, only_clear)
telemetry_service.log_event_sync(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.DELETE_KNOWLEDGE_BASE,
trace_id=trace_id_var.get())
if not only_clear:
cls.delete_knowledge_hook(request, login_user, knowledge)
return True
@classmethod
def delete_knowledge_file_in_vector(cls, knowledge: Knowledge, del_es: bool = True):
# Medical Treatment:vector
embeddings = FakeEmbedding()
vector_client = decide_vectorstores(
knowledge.collection_name, "Milvus", embeddings
)
if isinstance(vector_client.col, Collection):
logger.info(
f"delete_vector col={knowledge.collection_name} knowledge_id={knowledge.id}"
)
if knowledge.collection_name.startswith("col"):
# Singularcollection, simply delete it
vector_client.col.drop()
else:
# partitionMode requires partition key deletion
pk = vector_client.col.query(
expr=f'knowledge_id=="{knowledge.id}"', output_fields=["pk"]
)
vector_client.col.delete(f"pk in {[p['pk'] for p in pk]}")
# Judgingmilvus Are there any moreentity
if vector_client.col.is_empty:
vector_client.col.drop()
if del_es:
# Medical Treatment: es
index_name = knowledge.index_name or knowledge.collection_name # Compatible with older versions
es_client = decide_vectorstores(index_name, "ElasticKeywordsSearch", embeddings)
res = es_client.client.indices.delete(index=index_name, ignore=[400, 404])
logger.info(f"act=delete_es index={index_name} res={res}")
@classmethod
def delete_knowledge_hook(
cls, request: Request, login_user: UserPayload, knowledge: Knowledge
):
logger.info(
f"delete_knowledge_hook id={knowledge.id}, user: {login_user.user_id}"
)
# Delete Knowledge Base Audit Log
AuditLogService.delete_knowledge(login_user, get_request_ip(request), knowledge)
# Purge resources under user groups
GroupResourceDao.delete_group_resource_by_third_id(
str(knowledge.id), ResourceTypeEnum.KNOWLEDGE
)
@classmethod
def delete_knowledge_file_in_minio(cls, knowledge_id: int):
# Qn,1000records to deleteminioDoc.
count = KnowledgeFileDao.count_file_by_knowledge_id(knowledge_id)
if count == 0:
return
page_size = 1000
page_num = math.ceil(count / page_size)
minio_client = get_minio_storage_sync()
for i in range(page_num):
file_list = KnowledgeFileDao.get_file_simple_by_knowledge_id(
knowledge_id, i + 1, page_size
)
for file in file_list:
minio_client.remove_object_sync(object_name=str(file[0]))
if file[1]:
minio_client.remove_object_sync(object_name=file[1])
@classmethod
def get_upload_file_original_name(cls, file_name: str) -> str:
"""
SetujuuuidFile name, get the original name of the uploaded file
"""
if not file_name:
raise ServerError.http_exception("file_name is empty")
# FROMredisGet within
uuid_file_name = file_name.split(".")[0]
original_file_name = get_redis_client_sync().get(f"file_name:{uuid_file_name}") or file_name
return original_file_name
@classmethod
async def save_upload_file_original_name(cls, original_file_name: str) -> str:
"""
Save the original name of the uploaded file toredisGenerate oneuuidFile name
"""
if not original_file_name:
raise ServerError.http_exception("original_file_name is empty")
file_ext = original_file_name.split(".")[-1]
# Generate a uniqueuuidas akey
uuid_file_name = generate_uuid()
redis_client = await get_redis_client()
await redis_client.aset(f"file_name:{uuid_file_name}", original_file_name, expiration=86400)
return f"{uuid_file_name}.{file_ext}"
@classmethod
async def get_preview_file_chunk(
cls, request: Request, login_user: UserPayload, req_data: KnowledgeFileProcess
) -> (str, str, List[FileChunk], Any):
"""
0Parse Mode: uns or local
1: Converted file path
2: After dicingchunkVertical
3: ocrIdentifiedbbox
"""
knowledge = await KnowledgeDao.aquery_by_id(req_data.knowledge_id)
if not await login_user.async_access_check(
knowledge.user_id, str(knowledge.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError.http_exception()
file_path = req_data.file_list[0].file_path
excel_rule = req_data.file_list[0].excel_rule
cache_key = cls.get_preview_cache_key(req_data.knowledge_id, file_path)
redis_client = await get_redis_client()
# Attempt to fetch from cache
if req_data.cache:
if cache_value := await cls.async_get_preview_cache(cache_key):
parse_type = await redis_client.aget(f"{cache_key}_parse_type")
file_share_url = await redis_client.aget(f"{cache_key}_file_path")
partitions = await redis_client.aget(f"{cache_key}_partitions")
res = []
# Sort by segment order
cache_value = dict(sorted(cache_value.items(), key=lambda x: int(x[0])))
for key, val in cache_value.items():
res.append(FileChunk(text=val["text"], metadata=val["metadata"]))
return parse_type, file_share_url, res, partitions
filepath, file_name = await async_file_download(file_path)
file_ext = file_name.split(".")[-1].lower()
file_name = cls.get_upload_file_original_name(file_name)
# Split text
texts, metadatas, parse_type, partitions = await async_read_chunk_text(
login_user.user_id,
filepath,
file_name,
req_data.separator,
req_data.separator_rule,
req_data.chunk_size,
req_data.chunk_overlap,
knowledge_id=req_data.knowledge_id,
force_ocr=req_data.force_ocr,
enable_formula=req_data.enable_formula,
filter_page_header_footer=req_data.filter_page_header_footer,
retain_images=req_data.retain_images,
excel_rule=excel_rule,
no_summary=True,
)
if len(texts) == 0:
raise ValueError("File resolution is empty")
res = []
cache_map = {}
for index, val in enumerate(texts):
metadata_dict = metadatas[index].model_dump()
cache_map[index] = {"text": val, "metadata": metadata_dict}
res.append(FileChunk(text=val, metadata=metadata_dict))
# Default is the address of the source file
file_share_url = file_path
if file_ext in ['doc', 'ppt', 'pptx']:
file_share_url = ''
new_file_name = KnowledgeUtils.get_tmp_preview_file_object_name(filepath)
minio_client = await get_minio_storage()
if await minio_client.object_exists(minio_client.tmp_bucket, new_file_name):
file_share_url = await minio_client.get_share_link(
new_file_name, minio_client.tmp_bucket
)
# Deposit Cache
await cls.async_save_preview_cache(cache_key, mapping=cache_map)
await redis_client.aset(f"{cache_key}_parse_type", parse_type)
await redis_client.aset(f"{cache_key}_file_path", file_share_url)
await redis_client.aset(f"{cache_key}_partitions", partitions)
return parse_type, file_share_url, res, partitions
@classmethod
async def update_preview_file_chunk(
cls, request: Request, login_user: UserPayload, req_data: UpdatePreviewFileChunk
):
knowledge = await KnowledgeDao.aquery_by_id(req_data.knowledge_id)
if not await login_user.async_access_check(
knowledge.user_id, str(knowledge.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError.http_exception()
cache_key = cls.get_preview_cache_key(req_data.knowledge_id, req_data.file_path)
chunk_info = await cls.async_get_preview_cache(cache_key, req_data.chunk_index)
if not chunk_info:
raise NotFoundError.http_exception()
chunk_info["text"] = req_data.text
chunk_info["metadata"]["bbox"] = req_data.bbox
await cls.async_save_preview_cache(
cache_key, chunk_index=req_data.chunk_index, value=chunk_info
)
@classmethod
def delete_preview_file_chunk(
cls, request: Request, login_user: UserPayload, req_data: UpdatePreviewFileChunk
):
knowledge = KnowledgeDao.query_by_id(req_data.knowledge_id)
if not login_user.access_check(
knowledge.user_id, str(knowledge.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError.http_exception()
cache_key = cls.get_preview_cache_key(req_data.knowledge_id, req_data.file_path)
cls.delete_preview_cache(cache_key, chunk_index=req_data.chunk_index)
@classmethod
def save_knowledge_file(
cls, login_user: UserPayload, req_data: KnowledgeFileProcess
):
"""Process uploaded files, Uploaded to onlyminioAndmysql"""
knowledge = KnowledgeDao.query_by_id(req_data.knowledge_id)
if not knowledge:
raise NotFoundError.http_exception()
if not login_user.access_check(
knowledge.user_id, str(knowledge.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError.http_exception()
failed_files = []
# Process each file
process_files = []
preview_cache_keys = []
split_rule_dict = req_data.model_dump(include=set(list(FileProcessBase.model_fields.keys())))
for one in req_data.file_list:
# Upload source files, create data records
db_file = cls.process_one_file(login_user, knowledge, one, split_rule_dict)
# Duplicate file data using asynchronous tasks to execute
if db_file.status != KnowledgeFileStatus.FAILED.value:
# Get a preview cache of this filekey
cache_key = cls.get_preview_cache_key(
req_data.knowledge_id, one.file_path
)
preview_cache_keys.append(cache_key)
process_files.append(db_file)
else:
failed_file_info = db_file.model_dump()
failed_file_info["file_path"] = one.file_path
failed_files.append(failed_file_info)
return knowledge, failed_files, process_files, preview_cache_keys
@classmethod
def process_knowledge_file(
cls,
request: Request,
login_user: UserPayload,
background_tasks: BackgroundTasks,
req_data: KnowledgeFileProcess,
) -> List[KnowledgeFile]:
from bisheng.worker.knowledge import file_worker
"""Process uploaded files"""
knowledge, failed_files, process_files, preview_cache_keys = (
cls.save_knowledge_file(login_user, req_data)
)
# Asynchronous processing of file parsing and warehousing, To voters if approvedcache_keyIf data can be obtained, use thecachefor inbound operations
for index, one in enumerate(process_files):
file_worker.parse_knowledge_file_celery.delay(one.id, preview_cache_keys[index], req_data.callback_url)
cls.upload_knowledge_file_hook(request, login_user, knowledge, process_files)
return failed_files + process_files
@classmethod
def sync_process_knowledge_file(
cls, request: Request, login_user: UserPayload, req_data: KnowledgeFileProcess
) -> List[KnowledgeFile]:
"""Sync uploaded files"""
knowledge, failed_files, process_files, preview_cache_keys = (
cls.save_knowledge_file(login_user, req_data)
)
if process_files:
process_file_task(
knowledge,
process_files,
req_data.separator,
req_data.separator_rule,
req_data.chunk_size,
req_data.chunk_overlap,
req_data.callback_url,
req_data.extra,
preview_cache_keys,
)
process_files = KnowledgeFileDao.select_list([f.id for f in process_files])
cls.upload_knowledge_file_hook(request, login_user, knowledge, process_files)
return failed_files + process_files
@classmethod
async def rebuild_knowledge_file(cls, request: Request,
login_user: UserPayload,
req_data: KnowledgeFileReProcess):
"""
Rebuild Knowledge Base Files
:param request:
:param login_user:
:param req_data:
:return:
"""
from bisheng.worker.knowledge import file_worker
knowledge = await KnowledgeDao.async_query_by_id(req_data.knowledge_id)
if not knowledge:
raise NotFoundError.http_exception()
if not login_user.access_check(
knowledge.user_id, str(knowledge.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError.http_exception()
db_file = await KnowledgeFileDao.query_by_id(req_data.kb_file_id)
if not db_file:
raise NotFoundError.http_exception()
split_rule_dict = req_data.model_dump(include=set(list(FileProcessBase.model_fields.keys())))
if req_data.excel_rule is not None:
split_rule_dict["excel_rule"] = req_data.excel_rule.model_dump()
db_file.split_rule = json.dumps(split_rule_dict)
db_file.status = KnowledgeFileStatus.WAITING.value # Parsing
db_file.updater_id = login_user.user_id
db_file.updater_name = login_user.user_name
db_file = await KnowledgeFileDao.async_update(db_file)
preview_cache_key = cls.get_preview_cache_key(req_data.knowledge_id, file_path=req_data.file_path)
file_worker.retry_knowledge_file_celery.delay(db_file.id, preview_cache_key, req_data.callback_url)
return db_file.model_dump()
@classmethod
def retry_files(
cls,
request: Request,
login_user: UserPayload,
background_tasks: BackgroundTasks,
req_data: dict,
):
from bisheng.worker.knowledge import file_worker
db_file_retry = req_data.get("file_objs")
if not db_file_retry:
return []
id2input = {file.get("id"): file for file in db_file_retry}
file_ids = list(id2input.keys())
db_files: List[KnowledgeFile] = KnowledgeFileDao.select_list(file_ids=file_ids)
if not db_files:
return []
knowledge = KnowledgeDao.query_by_id(db_files[0].knowledge_id)
if not knowledge:
raise NotFoundError.http_exception()
if not login_user.access_check(
knowledge.user_id, str(knowledge.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError.http_exception()
res = []
req_data["knowledge_id"] = knowledge.id
minio_client = get_minio_storage_sync()
for file in db_files:
input_file = id2input.get(file.id)
# file exist
file.object_name = input_file.get("object_name", file.object_name)
file_preview_cache_key = KnowledgeUtils.get_preview_cache_key(
file.knowledge_id, input_file.get("file_path", "")
)
if file.object_name.startswith('tmp'):
# Moving Temporary Files to the Official Directory
new_object_name = KnowledgeUtils.get_knowledge_file_object_name(file.id, file.object_name)
minio_client.copy_object_sync(source_object=file.object_name, dest_object=new_object_name,
source_bucket=minio_client.tmp_bucket,
dest_bucket=minio_client.bucket)
file.object_name = new_object_name
file.file_name = input_file.get("file_name", None) or file.file_name
file.remark = ""
file.split_rule = input_file["split_rule"]
file.status = KnowledgeFileStatus.WAITING.value # Parsing
file.updater_id = login_user.user_id
file.updater_name = login_user.user_name
file = KnowledgeFileDao.update(file)
res.append([file, file_preview_cache_key])
tmp = []
for one_file in res:
file_worker.retry_knowledge_file_celery.delay(one_file[0].id, one_file[1], None)
tmp.append(one_file[0])
cls.upload_knowledge_file_hook(request, login_user, knowledge, tmp)
return []
@classmethod
def upload_knowledge_file_hook(
cls,
request: Request,
login_user: UserPayload,
knowledge: Knowledge,
file_list: List[KnowledgeFile],
):
logger.info(
f"act=upload_knowledge_file_hook user={login_user.user_name} knowledge_id={knowledge.id}"
)
if file_list:
KnowledgeDao.update_knowledge_update_time(knowledge)
# Log Audit Logs
file_name = ""
for one in file_list:
file_name += "\n\n" + one.file_name
AuditLogService.upload_knowledge_file(
login_user, get_request_ip(request), knowledge.id, file_name
)
@classmethod
def process_one_file(
cls,
login_user: UserPayload,
knowledge: Knowledge,
file_info: KnowledgeFileOne,
split_rule: Dict,
) -> KnowledgeFile:
"""Process uploaded files"""
# download original file
filepath, file_name = file_download(file_info.file_path)
md5_ = os.path.splitext(os.path.basename(filepath))[0].split("_")[0]
# Get file size inbytes）
file_size = os.path.getsize(filepath)
file_extension_name = file_name.split(".")[-1]
original_file_name = cls.get_upload_file_original_name(file_name)
# Does it contain duplicate files?
content_repeat = KnowledgeFileDao.get_file_by_condition(
md5_=md5_, knowledge_id=knowledge.id
)
name_repeat = KnowledgeFileDao.get_file_by_condition(
file_name=original_file_name, knowledge_id=knowledge.id
)
if not file_info.excel_rule:
file_info.excel_rule = ExcelRule()
split_rule["excel_rule"] = file_info.excel_rule.model_dump()
str_split_rule = json.dumps(split_rule)
minio_client = get_minio_storage_sync()
if content_repeat or name_repeat:
db_file = content_repeat[0] if content_repeat else name_repeat[0]
old_name = db_file.file_name
file_type = file_name.rsplit(".", 1)[-1]
obj_name = f"tmp/{db_file.id}.{file_type}"
db_file.object_name = obj_name
db_file.remark = json.dumps({
"new_name": original_file_name,
"old_name": old_name}, ensure_ascii=False)
# Uploaded tominio, do not modify the database, it is up to the front-end to decide whether to overwrite or not. If it is overwritten, the retry interface
with open(filepath, "rb") as file:
minio_client.put_object_tmp_sync(db_file.object_name, file.read())
db_file.status = KnowledgeFileStatus.FAILED.value
db_file.split_rule = str_split_rule
# Update file size information
db_file.file_size = file_size
return db_file
# Insert new data, upload the original file tominio
db_file = KnowledgeFile(
knowledge_id=knowledge.id,
file_name=original_file_name,
file_size=file_size,
md5=md5_,
split_rule=str_split_rule,
user_id=login_user.user_id,
user_name=login_user.user_name,
updater_id=login_user.user_id,
updater_name=login_user.user_name,
)
db_file = KnowledgeFileDao.add_file(db_file)
telemetry_service.log_event_sync(
user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.NEW_KNOWLEDGE_FILE,
trace_id=trace_id_var.get(),
)
# Saving original files
db_file.object_name = KnowledgeUtils.get_knowledge_file_object_name(db_file.id, db_file.file_name)
minio_client.put_object_sync(bucket_name=minio_client.bucket, object_name=db_file.object_name,
file=filepath)
logger.info("upload_original_file path={}", db_file.object_name)
KnowledgeFileDao.update(db_file)
return db_file
@classmethod
def get_knowledge_files_title(cls, db_knowledge: Knowledge, files: List[KnowledgeFile]) -> Dict[str, str]:
"""Adoption of documentsidGet file title"""
if not files:
return {}
files = [one for one in files if one.status == KnowledgeFileStatus.SUCCESS.value]
if not files:
return {}
file_title_map: Dict[str, str] = {}
try:
embeddings = FakeEmbedding()
es_client = decide_vectorstores(
db_knowledge.index_name, "ElasticKeywordsSearch", embeddings
)
search_data = {
"size": len(files),
"sort": [
{
"metadata.chunk_index": {
"order": "asc",
"missing": 0,
"unmapped_type": "long",
}
}
],
"post_filter": {
"terms": {"metadata.document_id": [one.id for one in files]}
},
"collapse": {"field": "metadata.document_id"},
}
es_res = es_client.client.search(
index=db_knowledge.index_name, body=search_data
)
for one in es_res["hits"]["hits"]:
file_title_map[str(one["_source"]["metadata"]["document_id"])] = one["_source"]["metadata"]["abstract"]
except Exception as e:
# maybe es index not exist so ignore this error
logger.warning(f"act=get_knowledge_files error={str(e)}")
pass
return file_title_map
@classmethod
def get_knowledge_files(
cls,
request: Request,
login_user: UserPayload,
knowledge_id: int,
file_name: str = None,
status: List[int] = None,
page: int = 1,
page_size: int = 10,
file_ids: List[int] = None,
) -> (List[KnowledgeFileResp], int, bool):
db_knowledge = KnowledgeDao.query_by_id(knowledge_id)
if not db_knowledge:
raise NotFoundError.http_exception()
if not login_user.access_check(
db_knowledge.user_id, str(knowledge_id), AccessType.KNOWLEDGE
):
raise UnAuthorizedError.http_exception()
res = KnowledgeFileDao.get_file_by_filters(
knowledge_id, file_name, status, page, page_size, file_ids
)
total = KnowledgeFileDao.count_file_by_filters(knowledge_id, file_name, status)
# get file title from es
finally_res = []
file_title_map = cls.get_knowledge_files_title(db_knowledge, res)
timeout_files = []
for index, one in enumerate(res):
finally_res.append(KnowledgeFileResp(**one.model_dump()))
# Parsing more than one day, setting status to failed
if one.status in [KnowledgeFileStatus.PROCESSING.value, KnowledgeFileStatus.WAITING.value] and (
datetime.now() - one.update_time).total_seconds() > 86400:
timeout_files.append(one.id)
continue
finally_res[index].title = file_title_map.get(str(one.id), "")
if timeout_files:
KnowledgeFileDao.update_file_status(timeout_files, KnowledgeFileStatus.TIMEOUT,
KnowledgeFileFailedError(
data={"exception": 'Parsing time exceeds 24 hours'}).to_json_str())
return (
finally_res,
total,
login_user.access_check(
db_knowledge.user_id, str(knowledge_id), AccessType.KNOWLEDGE_WRITE
),
)
@classmethod
def delete_knowledge_file(
cls, request: Request, login_user: UserPayload, file_ids: List[int]
):
from bisheng.worker.knowledge import file_worker
knowledge_file = KnowledgeFileDao.select_list(file_ids)
if not knowledge_file:
raise NotFoundError.http_exception()
db_knowledge = KnowledgeDao.query_by_id(knowledge_file[0].knowledge_id)
if not login_user.access_check(
db_knowledge.user_id, str(db_knowledge.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError.http_exception()
# Medical Treatment:vectordb
delete_knowledge_file_vectors(file_ids)
KnowledgeFileDao.delete_batch(file_ids)
telemetry_service.log_event_sync(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.DELETE_KNOWLEDGE_FILE,
trace_id=trace_id_var.get())
# Delete Audit Log for Knowledge Base Files
cls.delete_knowledge_file_hook(
request, login_user, db_knowledge.id, knowledge_file
)
# 5Minutes to check if the file was actually deleted
file_worker.delete_knowledge_file_celery.apply_async(args=(file_ids, knowledge_file[0].knowledge_id, True),
countdown=300)
return True
@classmethod
def delete_knowledge_file_hook(
cls,
request: Request,
login_user: UserPayload,
knowledge_id: int,
file_list: List[KnowledgeFile],
):
logger.info(
f"act=delete_knowledge_file_hook user={login_user.user_name} knowledge_id={knowledge_id}"
)
# Log Audit Logs
# Log Audit Logs
file_name = ""
for one in file_list:
file_name += "\n\n" + one.file_name
AuditLogService.delete_knowledge_file(
login_user, get_request_ip(request), knowledge_id, file_name
)
@classmethod
def judge_knowledge_access(cls, login_user: UserPayload, knowledge_id: int, access_type: AccessType) -> Knowledge:
db_knowledge = KnowledgeDao.query_by_id(knowledge_id)
if not db_knowledge:
raise NotFoundError.http_exception()
if not login_user.access_check(
db_knowledge.user_id, str(knowledge_id), access_type
):
raise UnAuthorizedError.http_exception()
return db_knowledge
@classmethod
def get_knowledge_chunks(
cls,
request: Request,
login_user: UserPayload,
knowledge_id: int,
file_ids: List[int] = None,
keyword: str = None,
page: int = None,
limit: int = None,
) -> (List[FileChunk], int):
db_knowledge = cls.judge_knowledge_access(login_user, knowledge_id, AccessType.KNOWLEDGE)
es_client = KnowledgeRag.init_knowledge_es_vectorstore_sync(db_knowledge)
search_data = {
"from": (page - 1) * limit,
"size": limit,
"sort": [
{
"metadata.document_id": {
"order": "desc",
"missing": 0,
"unmapped_type": "long",
}
},
{
"metadata.chunk_index": {
"order": "asc",
"missing": 0,
"unmapped_type": "long",
}
},
],
}
if file_ids:
search_data["post_filter"] = {"terms": {"metadata.document_id": file_ids}}
if keyword:
search_data["query"] = {"match_phrase": {"text": keyword}}
try:
res = es_client.client.search(index=db_knowledge.index_name, body=search_data)
except Exception as e:
logger.warning(f"act=get_knowledge_chunks error={str(e)}")
raise KnowledgeChunkError.http_exception()
# Query the file information corresponding to the next block
file_ids = set()
result = []
for one in res["hits"]["hits"]:
file_ids.add(one["_source"]["metadata"]["document_id"])
file_map = {}
if file_ids:
file_list = KnowledgeFileDao.get_file_by_ids(list(file_ids))
file_map = {one.id: one for one in file_list}
for one in res["hits"]["hits"]:
file_id = one["_source"]["metadata"]["document_id"]
file_info = file_map.get(file_id, None)
# Filter document summary contents of file names and summaries
result.append(
FileChunk(
text=KnowledgeUtils.split_chunk_metadata(one["_source"]["text"]),
metadata=one["_source"]["metadata"],
parse_type=file_info.parse_type if file_info else None,
)
)
return result, res["hits"]["total"]["value"]
@classmethod
def update_chunk_updater_info(cls, vector_client, es_client, db_knowledge, file_id, login_user):
# Product Requirements！！！！！！！
logger.debug(f"start update_milvus_chunk_updater_info user={login_user.user_name}")
output_fields = [s.name for s in vector_client.col.schema.fields]
iterator = vector_client.col.query_iterator(
expr=f"document_id == {file_id}",
output_fields=output_fields,
timeout=10,
)
update_time = int(datetime.now().timestamp())
while True:
result = iterator.next()
if not result:
iterator.close()
break
for record in result:
if not record.get("pk") or not record.get("vector"):
raise ValueError("milvus chunk pk field or vector field is None")
record["updater"] = login_user.user_name
record["update_time"] = update_time
vector_client.col.upsert(record)
logger.debug(f"update_milvus_chunk_updater_info over")
res = es_client.client.update_by_query(
index=db_knowledge.index_name,
body={
"query": {
"bool": {
"must": {"match": {"metadata.document_id": file_id}},
}
},
"script": {
"source": "ctx._source.metadata.updater=params.updater;ctx._source.metadata.update_time=params.update_time;",
"params": {"updater": login_user.user_name, "update_time": update_time},
},
},
conflicts="proceed",
)
logger.debug(f"update_es_chunk_updater_info: {res}")
@classmethod
def update_knowledge_chunk(
cls,
request: Request,
login_user: UserPayload,
knowledge_id: int,
file_id: int,
chunk_index: int,
text: str,
bbox: str,
):
db_knowledge = cls.judge_knowledge_access(login_user, knowledge_id, AccessType.KNOWLEDGE_WRITE)
logger.info(
f"act=update_vector knowledge_id={knowledge_id} document_id={file_id} chunk_index={chunk_index}"
)
vector_client = KnowledgeRag.init_knowledge_milvus_vectorstore_sync(login_user.user_id, db_knowledge)
# search metadata
output_fields = [s.name for s in vector_client.col.schema.fields if s.name != "vector"]
res = vector_client.col.query(
expr=f"document_id == {file_id} && chunk_index == {chunk_index}",
output_fields=output_fields,
timeout=10,
)
metadata = []
pk = []
for one in res:
pk.append(one.pop("pk"))
metadata.append(one)
if not metadata:
raise ValueError("chunk not found in vector db")
# insert data
logger.info(f"act=add_vector {knowledge_id}")
new_metadata = metadata[0]
new_metadata["bbox"] = bbox
new_text = KnowledgeUtils.aggregate_chunk_metadata(text, new_metadata)
res = vector_client.add_texts([new_text], [new_metadata], timeout=10)
# delete data
logger.info(f"act=delete_vector pk={pk}")
res = vector_client.col.delete(f"pk in {pk}", timeout=10)
logger.info(f"act=update_vector_over {res}")
logger.info(
f"act=update_es knowledge_id={knowledge_id} document_id={file_id} chunk_index={chunk_index}"
)
es_client = KnowledgeRag.init_knowledge_es_vectorstore_sync(db_knowledge)
res = es_client.client.update_by_query(
index=db_knowledge.index_name,
body={
"query": {
"bool": {
"must": {"match": {"metadata.document_id": file_id}},
"filter": {"match": {"metadata.chunk_index": chunk_index}},
}
},
"script": {
"source": "ctx._source.text=params.text;ctx._source.metadata.bbox=params.bbox;",
"params": {"text": new_text, "bbox": bbox},
},
},
)
logger.info(f"act=update_es_chunk_over {res}")
# update metadata updater and update_time
cls.update_chunk_updater_info(vector_client, es_client, db_knowledge, file_id, login_user)
KnowledgeFileDao.update_file_updater(file_id, login_user.user_id, login_user.user_name)
return True
@classmethod
def delete_knowledge_chunk(
cls,
request: Request,
login_user: UserPayload,
knowledge_id: int,
file_id: int,
chunk_index: int,
):
db_knowledge = cls.judge_knowledge_access(login_user, knowledge_id, AccessType.KNOWLEDGE_WRITE)
logger.info(
f"act=delete_vector knowledge_id={knowledge_id} document_id={file_id} chunk_index={chunk_index}"
)
vector_client = KnowledgeRag.init_knowledge_milvus_vectorstore_sync(login_user.user_id, db_knowledge)
res = vector_client.col.delete(
expr=f"document_id == {file_id} && chunk_index == {chunk_index}",
timeout=10,
)
logger.info(f"act=delete_vector_over {res}")
logger.info(
f"act=delete_es knowledge_id={knowledge_id} document_id={file_id} chunk_index={chunk_index} res={res}"
)
es_client = KnowledgeRag.init_knowledge_es_vectorstore_sync(db_knowledge)
res = es_client.client.delete_by_query(
index=db_knowledge.index_name,
query={
"bool": {
"must": {"match": {"metadata.document_id": file_id}},
"filter": {"match": {"metadata.chunk_index": chunk_index}},
}
},
conflicts="proceed",
)
logger.info(f"act=delete_es_over {res}")
cls.update_chunk_updater_info(vector_client, es_client, db_knowledge, file_id, login_user)
KnowledgeFileDao.update_file_updater(file_id, login_user.user_id, login_user.user_name)
return True
@classmethod
def get_file_share_url(cls, file_id: int) -> (str, str):
""" Get the original download address of the file And Corresponding preview file download address """
file = KnowledgeFileDao.get_file_by_ids([file_id])
if not file:
raise NotFoundError()
file = file[0]
minio_client = get_minio_storage_sync()
# 130File parsing prior to version
if file.parse_type in [ParseType.LOCAL.value, ParseType.UNS.value]:
original_url = minio_client.get_share_link_sync(cls.get_knowledge_file_object_name(file.id, file.file_name))
preview_url = minio_client.get_share_link_sync(str(file.id))
else:
original_url = cls.get_file_share_url_with_empty(file.object_name)
preview_url = ""
# 130After the version of the file parsing logic, only the source file and preview file are no longer transferredpdfSettings Updated. what double check raws pls
if file.file_name.endswith(('.doc', '.ppt', '.pptx')):
preview_object_name = KnowledgeUtils.get_knowledge_preview_file_object_name(file.id, file.file_name)
preview_url = cls.get_file_share_url_with_empty(preview_object_name)
return original_url, preview_url
@classmethod
def get_file_share_url_with_empty(cls, object_name: str) -> str:
"""
Get a shared link to a file
:param object_name: Files inminioObject name in
:return: File sharing link
"""
minio_client = get_minio_storage_sync()
if minio_client.object_exists_sync(minio_client.bucket, object_name):
return minio_client.get_share_link_sync(object_name)
return ""
@classmethod
def get_file_bbox(
cls, request: Request, login_user: UserPayload, file_id: int
) -> Any:
file_info = KnowledgeFileDao.select_list([file_id])
file_info = file_info[0]
if not file_info.bbox_object_name:
return None
minio_client = get_minio_storage_sync()
# download bbox file
resp = minio_client.get_object_sync(bucket_name=minio_client.bucket, object_name=file_info.bbox_object_name)
return json.loads(resp.decode("utf-8"))
@classmethod
async def copy_knowledge(
cls,
request,
background_tasks: BackgroundTasks,
login_user: UserPayload,
knowledge: Knowledge,
knowledge_name: str = None,
) -> Any:
from bisheng.worker.knowledge import file_worker
await KnowledgeDao.async_update_state(knowledge.id, KnowledgeState.COPYING, update_time=knowledge.update_time)
knowldge_dict = knowledge.model_dump()
knowldge_dict.pop("id")
knowldge_dict.pop("create_time")
knowldge_dict.pop("update_time", None)
knowldge_dict["user_id"] = login_user.user_id
knowldge_dict["index_name"] = generate_knowledge_index_name()
knowldge_dict["collection_name"] = knowldge_dict["index_name"]
knowldge_dict["name"] = f"{knowledge.name} Copy"[:200] if not knowledge_name else knowledge_name[:200]
knowldge_dict["state"] = KnowledgeState.UNPUBLISHED.value
knowledge_new = Knowledge(**knowldge_dict)
target_knowlege = await KnowledgeDao.async_insert_one(knowledge_new)
# celery not yetok
params = {
"source_knowledge_id": knowledge.id,
"target_id": target_knowlege.id,
"login_user_id": login_user.user_id,
}
cls.create_knowledge_hook(request, login_user, target_knowlege)
file_worker.file_copy_celery.delay(params)
return target_knowlege
@classmethod
async def copy_qa_knowledge(
cls,
request,
login_user: UserPayload,
qa_knowledge: Knowledge,
knowledge_name: str = None,
) -> Any:
await KnowledgeDao.async_update_state(qa_knowledge.id, KnowledgeState.COPYING,
update_time=qa_knowledge.update_time)
qa_knowldge_dict = qa_knowledge.model_dump()
qa_knowldge_dict.pop("id")
qa_knowldge_dict.pop("create_time")
qa_knowldge_dict.pop("update_time", None)
qa_knowldge_dict["user_id"] = login_user.user_id
qa_knowldge_dict["index_name"] = generate_knowledge_index_name()
qa_knowldge_dict["collection_name"] = qa_knowldge_dict["index_name"]
qa_knowldge_dict["name"] = f"{qa_knowledge.name} Copy"[:200] if not knowledge_name else knowledge_name[:200]
qa_knowldge_dict["state"] = KnowledgeState.UNPUBLISHED.value
qa_knowledge_new = Knowledge(**qa_knowldge_dict)
target_qa_knowlege = await KnowledgeDao.async_insert_one(qa_knowledge_new)
cls.create_knowledge_hook(request, login_user, target_qa_knowlege)
from bisheng.worker.knowledge.qa import copy_qa_knowledge_celery
copy_qa_knowledge_celery.delay(source_knowledge_id=qa_knowledge.id, target_knowledge_id=target_qa_knowlege.id,
login_user_id=login_user.user_id)
return target_qa_knowlege
@classmethod
def judge_qa_knowledge_write(
cls, login_user: UserPayload, qa_knowledge_id: int
) -> Knowledge:
db_knowledge = KnowledgeDao.query_by_id(qa_knowledge_id)
# Query the current knowledge base, whether there are write permissions
if not db_knowledge:
raise NotFoundError()
if not login_user.access_check(
db_knowledge.user_id, str(qa_knowledge_id), AccessType.KNOWLEDGE
):
raise UnAuthorizedError.http_exception()
if db_knowledge.type != KnowledgeTypeEnum.QA.value:
raise KnowledgeNotQAError()
return db_knowledge
================================================
FILE: src/backend/bisheng/api/services/knowledge_imp.py
================================================
import asyncio
import json
import os
import re
import time
from pathlib import Path
from typing import Any, Dict, List, Optional, Union
import aiofiles
import requests
from langchain.embeddings.base import Embeddings
from langchain.schema.document import Document
from langchain.text_splitter import CharacterTextSplitter
from langchain.vectorstores.base import VectorStore
from langchain_community.document_loaders import (
BSHTMLLoader,
PyPDFLoader,
TextLoader,
UnstructuredPowerPointLoader,
UnstructuredWordDocumentLoader,
)
from loguru import logger
from pymilvus import Collection
from sqlalchemy import func, or_
from sqlmodel import select
from bisheng.api.services.etl4lm_loader import Etl4lmLoader
from bisheng.api.services.libreoffice_converter import (
convert_doc_to_docx,
convert_ppt_to_pdf, convert_ppt_to_pptx,
)
from bisheng.api.services.md_from_pdf import is_pdf_damaged
from bisheng.api.services.patch_130 import (
convert_file_to_md,
combine_multiple_md_files_to_raw_texts,
)
from bisheng.api.v1.schemas import ExcelRule
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum, ApplicationTypeEnum
from bisheng.common.constants.vectorstore_metadata import KNOWLEDGE_RAG_METADATA_SCHEMA
from bisheng.common.errcode import BaseErrorCode
from bisheng.common.errcode.knowledge import KnowledgeSimilarError, KnowledgeFileDeleteError, KnowledgeFileEmptyError, \
KnowledgeFileChunkMaxError, KnowledgeLLMError, KnowledgeFileDamagedError, KnowledgeFileNotSupportedError, \
KnowledgeEtl4lmTimeoutError, KnowledgeFileFailedError, KnowledgeExcelChunkMaxError, KnowledgeRecommendQuestionError
from bisheng.common.schemas.telemetry.event_data_schema import FileParseEventData
from bisheng.common.services import telemetry_service
from bisheng.common.services.config_service import settings
from bisheng.core.cache.redis_manager import get_redis_client_sync, get_redis_client
from bisheng.core.cache.utils import file_download
from bisheng.core.database import get_sync_db_session
from bisheng.core.logger import trace_id_var
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync, get_minio_storage
from bisheng.interface.embeddings.custom import FakeEmbedding
from bisheng.interface.importing.utils import import_vectorstore
from bisheng.interface.initialize.loading import instantiate_vectorstore
from bisheng.knowledge.domain.knowledge_rag import KnowledgeRag
from bisheng.knowledge.domain.models.knowledge import Knowledge, KnowledgeDao
from bisheng.knowledge.domain.models.knowledge_file import (
KnowledgeFile,
KnowledgeFileDao,
KnowledgeFileStatus,
ParseType,
QAKnoweldgeDao,
QAKnowledge,
QAKnowledgeUpsert,
QAStatus,
)
from bisheng.knowledge.domain.schemas.knowledge_rag_schema import Metadata
from bisheng.llm.domain.services import LLMService
from bisheng.user.domain.models.user import UserDao
from bisheng.utils import md5_hash, util
from bisheng.utils.exceptions import EtlException, FileParseException
from bisheng_langchain.rag.extract_info import extract_title, async_extract_title
from bisheng_langchain.text_splitter import ElemCharacterTextSplitter
filetype_load_map = {
"txt": TextLoader,
"pdf": PyPDFLoader,
"html": BSHTMLLoader,
"md": TextLoader,
"docx": UnstructuredWordDocumentLoader,
"pptx": UnstructuredPowerPointLoader,
}
class KnowledgeUtils:
# Used to distinguishchunkand automated production summary content Format e.g. Filename\nDocument Summary\n--------\n chunkContents
chunk_split = "\n----------\n"
@classmethod
def get_preview_cache_key(cls, knowledge_id: int, file_path: str, md5_value=None) -> str:
if not md5_value:
md5_value = md5_hash(file_path)
return f"preview_file_chunk:{knowledge_id}:{md5_value}"
@classmethod
def aggregate_chunk_metadata(cls, chunk: str, metadata: dict) -> str:
# Video Wall ProcessingchunkAndmetadatadata in to get a newchunk
res = f"{{{metadata.get('document_name', '')}\n"
if metadata.get("abstract", ""):
res += f"{metadata.get('abstract', '')}\n"
res += f"{chunk}}}"
return res
@classmethod
def chunk2promt(cls, chunk: str, metadata: dict) -> str:
# Video Wall ProcessingchunkAndmetadatadata in to get a newchunk
res = f"[file name]:{metadata.get('source', '')}\n[file content begin]\n{chunk}[file content end]\n"
return res
@classmethod
def split_chunk_metadata(cls, chunk: str) -> str:
# After stitching fromchunkis isolated from the originalchunk
# Instructions are old stitching rules
if not chunk.startswith("{"):
return chunk.split(cls.chunk_split)[-1]
chunk = chunk.split("")[-1]
chunk = chunk.split("")[0]
return chunk
@classmethod
async def async_save_preview_cache(
cls, cache_key, mapping: dict = None, chunk_index: int = 0, value: dict = None
):
redis_client = await get_redis_client()
if mapping:
for key, val in mapping.items():
mapping[key] = json.dumps(val, ensure_ascii=False)
await redis_client.ahset(cache_key, mapping=mapping)
else:
await redis_client.ahset(
cache_key, key=chunk_index, value=json.dumps(value, ensure_ascii=False)
)
@classmethod
def delete_preview_cache(cls, cache_key, chunk_index: int = None):
redis_client = get_redis_client_sync()
if chunk_index is None:
redis_client.delete(cache_key)
redis_client.delete(f"{cache_key}_parse_type")
redis_client.delete(f"{cache_key}_file_path")
redis_client.delete(f"{cache_key}_partitions")
else:
redis_client.hdel(cache_key, chunk_index)
@classmethod
def get_preview_cache(cls, cache_key, chunk_index: int = None) -> dict:
redis_client = get_redis_client_sync()
if chunk_index is None:
all_chunk_info = redis_client.hgetall(cache_key)
for key, value in all_chunk_info.items():
all_chunk_info[key] = json.loads(value)
return all_chunk_info
else:
chunk_info = redis_client.hget(cache_key, chunk_index)
if chunk_info:
chunk_info = json.loads(chunk_info)
return chunk_info
@classmethod
async def async_get_preview_cache(cls, cache_key, chunk_index: int = None) -> dict:
redis_client = await get_redis_client()
if chunk_index is None:
all_chunk_info = await redis_client.ahgetall(cache_key)
for key, value in all_chunk_info.items():
all_chunk_info[key] = json.loads(value)
return all_chunk_info
else:
chunk_info = await redis_client.ahget(cache_key, chunk_index)
if chunk_info:
chunk_info = json.loads(chunk_info)
return chunk_info
@classmethod
def get_knowledge_file_image_dir(cls, doc_id: str, knowledge_id: int = None) -> str:
"""Get file image atminioStorage directory for"""
if knowledge_id:
return f"knowledge/images/{knowledge_id}/{doc_id}"
else:
return f"tmp/images/{doc_id}"
@classmethod
def get_knowledge_file_object_name(cls, file_id: int, file_name: str) -> str:
"""Get Knowledge Base Source Files atminioStorage Path for"""
file_ext = file_name.split(".")[-1]
return f"original/{file_id}.{file_ext}"
@classmethod
def get_knowledge_bbox_file_object_name(cls, file_id: int) -> str:
"""Get the corresponding knowledge base filebboxFiles inminioStorage Path for"""
return f"partitions/{file_id}.json"
@classmethod
def get_knowledge_preview_file_object_name(
cls, file_id: int, file_name: str
) -> Optional[str]:
"""Get the preview file corresponding to the knowledge base file atminioStorage Path for This path is stored in the officialbucketand within"""
file_ext = file_name.split(".")[-1]
if file_ext == "doc":
return f"preview/{file_id}.docx"
elif file_ext in ["ppt", "pptx"]:
return f"preview/{file_id}.pdf"
# No preview required for other file types
return None
@classmethod
def get_tmp_preview_file_object_name(cls, file_path: str) -> Optional[str]:
"""Get a temporary preview file atminioStorage Path for This path is stored in a temporarybucket"""
file_name = os.path.basename(file_path)
file_name_no_ext, file_ext = file_name.rsplit(".", 1)
if file_ext == "doc":
return f"preview/{file_name_no_ext}.docx"
elif file_ext in ["ppt", "pptx"]:
return f"preview/{file_name_no_ext}.pdf"
# No preview required for other file types
return None
def put_images_to_minio(local_image_dir, knowledge_id, doc_id):
if not os.path.exists(local_image_dir):
return
minio_client = get_minio_storage_sync()
files = [f for f in os.listdir(local_image_dir)]
for file_name in files:
local_file_name = f"{local_image_dir}/{file_name}"
object_name = f"{KnowledgeUtils.get_knowledge_file_image_dir(doc_id, knowledge_id)}/{file_name}"
with open(local_file_name, "rb") as file_obj:
minio_client.put_object_sync(
object_name=object_name, file=file_obj.read(), bucket_name=minio_client.bucket
)
async def async_images_to_minio(local_image_dir, knowledge_id, doc_id):
if not os.path.exists(local_image_dir):
return
minio_client = await get_minio_storage()
files = [f for f in os.listdir(local_image_dir)]
for file_name in files:
local_file_name = f"{local_image_dir}/{file_name}"
object_name = f"{KnowledgeUtils.get_knowledge_file_image_dir(doc_id, knowledge_id)}/{file_name}"
async with aiofiles.open(local_file_name, "rb") as file_obj:
await minio_client.put_object(
object_name=object_name, file=await file_obj.read(), bucket_name=minio_client.bucket
)
def process_file_task(
knowledge: Knowledge,
db_files: List[KnowledgeFile],
separator: List[str],
separator_rule: List[str],
chunk_size: int,
chunk_overlap: int,
callback_url: str = None,
extra_metadata: Dict = None,
preview_cache_keys: List[str] = None,
retain_images: int = 1,
enable_formula: int = 1,
force_ocr: int = 0,
filter_page_header_footer: int = 0,
):
"""Working with Knowledge Files Tasks"""
try:
index_name = knowledge.index_name or knowledge.collection_name
addEmbedding(
knowledge.collection_name,
index_name,
knowledge.id,
knowledge.model,
separator,
separator_rule,
chunk_size,
chunk_overlap,
db_files,
callback_url,
extra_metadata,
preview_cache_keys=preview_cache_keys,
retain_images=retain_images,
enable_formula=enable_formula,
force_ocr=force_ocr,
filter_page_header_footer=filter_page_header_footer,
)
except Exception as e:
logger.exception("process_file_task error")
new_files = KnowledgeFileDao.select_list([file.id for file in db_files])
new_files_map = {file.id: file for file in new_files}
for file in db_files:
if new_files_map[file.id].status == KnowledgeFileStatus.PROCESSING.value:
file.status = KnowledgeFileStatus.FAILED.value
file.remark = KnowledgeFileFailedError(exception=e).to_json_str()
KnowledgeFileDao.update(file)
logger.info("update files failed status over")
raise e
def delete_vector_files(file_ids: List[int], knowledge: Knowledge) -> bool:
""" Delete vector data andesDATA """
if not file_ids:
return True
logger.info(f"delete_files file_ids={file_ids} knowledge_id={knowledge.id}")
logger.info("start init Milvus")
vector_client = KnowledgeRag.init_knowledge_milvus_vectorstore_sync(0, knowledge=knowledge,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
logger.info("start init ES")
es_client = KnowledgeRag.init_knowledge_es_vectorstore_sync(knowledge=knowledge,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
# Automatically close purchase order aftercollectionIf it does not exist, it will not
if vector_client.col:
vector_client.col.delete(expr=f"document_id in {file_ids}", timeout=10)
logger.info(f"delete_milvus file_ids={file_ids}")
if es_client.client.indices.exists(index=knowledge.index_name):
res = es_client.client.delete_by_query(
index=knowledge.index_name,
query={"terms": {"metadata.document_id": file_ids}},
)
logger.info(f"act=delete_es file_ids={file_ids} res={res}")
return True
def delete_minio_files(file: KnowledgeFile):
"""Delete Knowledge Base Files inminioStorage on"""
minio_client = get_minio_storage_sync()
# Delete source file
if file.object_name:
minio_client.remove_object_sync(bucket_name=minio_client.bucket, object_name=file.object_name)
# DeletebboxDoc.
if file.bbox_object_name:
minio_client.remove_object_sync(bucket_name=minio_client.bucket, object_name=file.bbox_object_name)
# Delete ConvertedpdfDoc.
minio_client.remove_object_sync(bucket_name=minio_client.bucket, object_name=f"{file.id}")
# Delete preview file
preview_object_name = KnowledgeUtils.get_knowledge_preview_file_object_name(
file.id, file.file_name
)
if preview_object_name:
minio_client.remove_object_sync(bucket_name=minio_client.bucket, object_name=preview_object_name)
return True
def delete_knowledge_file_vectors(file_ids: List[int], clear_minio: bool = True):
"""Delete Knowledge File Information"""
knowledge_files = KnowledgeFileDao.select_list(file_ids=file_ids)
knowledge_ids = [file.knowledge_id for file in knowledge_files]
knowledges = KnowledgeDao.get_list_by_ids(knowledge_ids)
if len(knowledges) > 1:
raise KnowledgeFileDeleteError()
knowledge = knowledges[0]
delete_vector_files(file_ids, knowledge)
if clear_minio:
for file in knowledge_files:
delete_minio_files(file)
return True
def decide_vectorstores(
collection_name: str, vector_store: str, embedding: Embeddings, knowledge_id: int = None
) -> Union[VectorStore, Any]:
""" vector db if used by query, must have knowledge_id"""
param: dict = {"embedding": embedding}
if vector_store == "ElasticKeywordsSearch":
vector_config = settings.get_vectors_conf().elasticsearch.model_dump()
if not vector_config:
# No related configurations
raise RuntimeError("vector_stores.elasticsearch not find in config.yaml")
param["index_name"] = collection_name
if isinstance(vector_config["ssl_verify"], str):
vector_config["ssl_verify"] = eval(vector_config["ssl_verify"])
elif vector_store == "Milvus":
if knowledge_id and collection_name.startswith("partition"):
param["partition_key"] = knowledge_id
vector_config = settings.get_vectors_conf().milvus.model_dump()
if not vector_config:
# No related configurations
raise RuntimeError("vector_stores.milvus not find in config.yaml")
param["collection_name"] = collection_name
vector_config.pop("partition_suffix", "")
vector_config.pop("is_partition", "")
else:
raise RuntimeError("unknown vector store type")
param.update(vector_config)
class_obj = import_vectorstore(vector_store)
return instantiate_vectorstore(vector_store, class_object=class_obj, params=param)
def decide_knowledge_llm(invoke_user_id: int) -> Any:
"""Get a summary of the knowledge basechunkright of privacy llmObjects"""
# DapatkanllmConfigure
knowledge_llm = LLMService.get_knowledge_llm()
if not knowledge_llm.extract_title_model_id:
# No related configurations
return None
# DapatkanllmObjects
return LLMService.get_bisheng_llm_sync(
model_id=knowledge_llm.extract_title_model_id,
app_id=ApplicationTypeEnum.KNOWLEDGE_BASE.value,
app_name=ApplicationTypeEnum.KNOWLEDGE_BASE.value,
app_type=ApplicationTypeEnum.KNOWLEDGE_BASE,
user_id=invoke_user_id)
async def async_decide_knowledge_llm(invoke_user_id: int) -> Any:
"""Get a summary of the knowledge basechunkright of privacy llmObjects"""
# DapatkanllmConfigure
knowledge_llm = await LLMService.aget_knowledge_llm()
if not knowledge_llm.extract_title_model_id:
# No related configurations
return None
# DapatkanllmObjects
return await LLMService.get_bisheng_llm(
model_id=knowledge_llm.extract_title_model_id,
app_id=ApplicationTypeEnum.KNOWLEDGE_BASE.value,
app_name=ApplicationTypeEnum.KNOWLEDGE_BASE.value,
app_type=ApplicationTypeEnum.KNOWLEDGE_BASE,
user_id=invoke_user_id)
def addEmbedding(
collection_name: str,
index_name: str,
knowledge_id: int,
model: str,
separator: List[str],
separator_rule: List[str],
chunk_size: int,
chunk_overlap: int,
knowledge_files: List[KnowledgeFile],
callback: str = None,
extra_meta: Dict = None,
preview_cache_keys: List[str] = None,
retain_images: int = 1,
enable_formula: int = 1,
force_ocr: int = 0,
filter_page_header_footer: int = 0,
):
"""Adding Files to Vector SumsesCunene"""
logger.info("start init Milvus")
vector_client = KnowledgeRag.init_knowledge_milvus_vectorstore_sync(knowledge_files[0].updater_id,
knowledge_id=knowledge_id,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
logger.info("start init ES")
es_client = KnowledgeRag.init_knowledge_es_vectorstore_sync(knowledge_id=knowledge_id,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
minio_client = get_minio_storage_sync()
for index, db_file in enumerate(knowledge_files):
# Try to get chunks of a file from the cache
preview_cache_key = None
if preview_cache_keys:
preview_cache_key = (
preview_cache_keys[index] if index < len(preview_cache_keys) else None
)
status = 'failed'
try:
logger.info(
f"process_file_begin file_id={db_file.id} file_name={db_file.file_name}"
)
add_file_embedding(
vector_client,
es_client,
minio_client,
db_file,
separator,
separator_rule,
chunk_size,
chunk_overlap,
extra_meta=extra_meta,
preview_cache_key=preview_cache_key,
# Added parameters
retain_images=retain_images,
knowledge_id=knowledge_id,
enable_formula=enable_formula,
force_ocr=force_ocr,
filter_page_header_footer=filter_page_header_footer,
)
db_file.status = KnowledgeFileStatus.SUCCESS.value
status = 'success'
except FileParseException as e:
logger.exception(
f"process_file_fail file_id={db_file.id} file_name={db_file.file_name}"
)
db_file.status = KnowledgeFileStatus.FAILED.value
if str(e).find("etl4lm server timeout") != -1:
db_file.remark = KnowledgeEtl4lmTimeoutError(exception=e).to_json_str()
else:
db_file.remark = KnowledgeFileFailedError(exception=e).to_json_str()
status = 'parse_failed'
except BaseErrorCode as e:
db_file.status = KnowledgeFileStatus.FAILED.value
db_file.remark = e.to_json_str()
status = 'failed'
except Exception as e:
logger.exception(
f"process_file_fail file_id={db_file.id} file_name={db_file.file_name}"
)
db_file.status = KnowledgeFileStatus.FAILED.value
db_file.remark = KnowledgeFileFailedError(exception=e).to_json_str()
status = 'failed'
finally:
logger.info(
f"process_file_end file_id={db_file.id} file_name={db_file.file_name}"
)
KnowledgeFileDao.update(db_file)
telemetry_service.log_event_sync(user_id=db_file.user_id,
event_type=BaseTelemetryTypeEnum.FILE_PARSE,
trace_id=trace_id_var.get(),
event_data=FileParseEventData(
parse_type=db_file.parse_type,
status=status,
app_type=ApplicationTypeEnum.KNOWLEDGE_BASE
))
if callback:
inp = {
"file_name": db_file.file_name,
"file_status": db_file.status,
"file_id": db_file.id,
"error_msg": db_file.remark,
}
requests.post(url=callback, json=inp, timeout=3)
def add_file_embedding(
vector_client,
es_client,
minio_client,
db_file: KnowledgeFile,
separator: List[str],
separator_rule: List[str],
chunk_size: int,
chunk_overlap: int,
extra_meta: Dict = None,
preview_cache_key: str = None,
knowledge_id: int = None,
retain_images: int = 1,
enable_formula: int = 1,
force_ocr: int = 0,
filter_page_header_footer: int = 0,
):
# download original file
logger.info(
f"start download original file={db_file.id} file_name={db_file.file_name}"
)
file_url = minio_client.get_share_link_sync(db_file.object_name, clear_host=False)
filepath, _ = file_download(file_url)
file_ext = Path(db_file.file_name).suffix.lower()
# Convert split_rule string to dict if needed
excel_rule = ExcelRule()
if db_file.split_rule and isinstance(db_file.split_rule, str):
split_rule = json.loads(db_file.split_rule)
if "excel_rule" in split_rule:
excel_rule = ExcelRule(**split_rule["excel_rule"])
# # extract text from file
try:
texts, metadatas, parse_type, partitions = read_chunk_text(
db_file.user_id,
filepath,
db_file.file_name,
separator,
separator_rule,
chunk_size,
chunk_overlap,
knowledge_id=knowledge_id,
retain_images=retain_images,
enable_formula=enable_formula,
force_ocr=force_ocr,
filter_page_header_footer=filter_page_header_footer,
excel_rule=excel_rule,
)
except EtlException as e:
db_file.parse_type = ParseType.ETL4LM.value
raise FileParseException(str(e)) from e
except BaseErrorCode as e:
raise e
except Exception as e:
raise FileParseException(str(e)) from e
if len(texts) == 0:
raise KnowledgeFileEmptyError()
# If there is data in the cache, the data in the cache is used to go to the warehouse because the user edited it in the interface.
if preview_cache_key:
all_chunk_info = KnowledgeUtils.get_preview_cache(preview_cache_key)
if all_chunk_info:
logger.info(
f"get_preview_cache file={db_file.id} file_name={db_file.file_name}"
)
texts, metadatas = [], []
for key, val in all_chunk_info.items():
texts.append(val["text"])
metadatas.append(Metadata(**val["metadata"]))
for index, one in enumerate(texts):
if len(one) > 10000:
if file_ext in (".xlsx", ".xls", ".csv"):
raise KnowledgeExcelChunkMaxError()
raise KnowledgeFileChunkMaxError()
# On Inbound Stitching file names and document summaries
texts[index] = KnowledgeUtils.aggregate_chunk_metadata(one, metadatas[index].model_dump())
db_file.parse_type = parse_type
# StorageocrIdentifiedpartitionsResult
if partitions:
partition_data = json.dumps(partitions, ensure_ascii=False).encode("utf-8")
db_file.bbox_object_name = KnowledgeUtils.get_knowledge_bbox_file_object_name(
db_file.id
)
minio_client.put_object_sync(
bucket_name=minio_client.bucket,
object_name=db_file.bbox_object_name,
file=partition_data, content_type="application/json",
)
logger.info(
f"chunk_split file={db_file.id} file_name={db_file.file_name} size={len(texts)}"
)
uploader = UserDao.get_user(user_id=db_file.user_id).user_name
if db_file.updater_id:
updater = UserDao.get_user(user_id=db_file.updater_id).user_name
else:
updater = uploader
for metadata in metadatas:
metadata.document_id = db_file.id
metadata.knowledge_id = db_file.knowledge_id
if extra_meta:
metadata.user_metadata = metadata.user_metadata.update(extra_meta)
metadata.upload_time = int(db_file.create_time.timestamp())
metadata.update_time = int(db_file.update_time.timestamp())
metadata.uploader = uploader
metadata.updater = updater
metadatas = [metadata.model_dump() for metadata in metadatas]
logger.info(f"add_vectordb file={db_file.id} file_name={db_file.file_name}")
# Depositmilvus
vector_client.add_texts(texts=texts, metadatas=metadatas)
logger.info(f"add_es file={db_file.id} file_name={db_file.file_name}")
# Deposites
es_client.add_texts(texts=texts, metadatas=metadatas)
logger.info(f"add_complete file={db_file.id} file_name={db_file.file_name}")
if preview_cache_key:
KnowledgeUtils.delete_preview_cache(preview_cache_key)
if file_ext in (".doc", ".ppt", ".pptx"):
tmp_preview_file = KnowledgeUtils.get_tmp_preview_file_object_name(filepath)
preview_object_name = KnowledgeUtils.get_knowledge_preview_file_object_name(
db_file.id, db_file.file_name
)
logger.info(
f"upload_preview_file_to_minio file={db_file.id} tmp_object_name={tmp_preview_file}, preview_object_name={preview_object_name}"
)
if minio_client.object_exists_sync(minio_client.tmp_bucket, tmp_preview_file):
minio_client.copy_object_sync(
source_object=tmp_preview_file,
dest_object=preview_object_name,
source_bucket=minio_client.tmp_bucket,
dest_bucket=minio_client.bucket,
)
logger.info(
f"upload_preview_file_over file={db_file.id} tmp_object_name={tmp_preview_file}, preview_object_name={preview_object_name}"
)
elif file_ext == ".docx":
# special logic for docx replace original file
filepath_dir = os.path.dirname(filepath)
docx_fixed_path = os.path.join(filepath_dir, "tmp")
docx_fixed_path = os.path.join(docx_fixed_path, os.path.basename(filepath))
if os.path.exists(docx_fixed_path):
# replace original docx file
minio_client.put_object_sync(
bucket_name=minio_client.bucket,
object_name=db_file.object_name,
file=docx_fixed_path)
def add_text_into_vector(
vector_client,
es_client,
db_file: KnowledgeFile,
texts: List[str],
metadatas: List[dict],
):
logger.info(f"add_vectordb file={db_file.id} file_name={db_file.file_name}")
# Depositmilvus
vector_client.add_texts(texts=texts, metadatas=metadatas)
logger.info(f"add_es file={db_file.id} file_name={db_file.file_name}")
# Deposites
es_client.add_texts(texts=texts, metadatas=metadatas)
def parse_partitions(partitions: List[Any]) -> Dict:
"""Resolve BuildbboxCorrespondence with text"""
if not partitions:
return {}
res = {}
for part_index, part in enumerate(partitions):
bboxes = part["metadata"]["extra_data"]["bboxes"]
indexes = part["metadata"]["extra_data"]["indexes"]
pages = part["metadata"]["extra_data"]["pages"]
text = part["text"]
for index, bbox in enumerate(bboxes):
key = f"{pages[index]}-" + "-".join([str(int(one)) for one in bbox])
if index == len(bboxes) - 1:
val = text[indexes[index][0]:]
else:
val = text[indexes[index][0]:indexes[index][1]]
res[key] = {"text": val, "type": part["type"], "part_id": part_index}
return res
def upload_preview_file_to_minio(original_file_path: str, preview_file_path: str):
if (
os.path.basename(original_file_path).split(".")[0]
!= os.path.basename(preview_file_path).split(".")[0]
):
logger.error(
f"Original and preview file paths do not match: {original_file_path} vs {preview_file_path}"
)
minio_client = get_minio_storage_sync()
object_name = KnowledgeUtils.get_tmp_preview_file_object_name(original_file_path)
with open(preview_file_path, "rb") as file_obj:
# Upload preview file tominio
minio_client.put_object_tmp_sync(
object_name=object_name, file=file_obj.read()
)
return object_name
async def async_upload_preview_file_to_minio(
original_file_path: str, preview_file_path: str
):
if (
os.path.basename(original_file_path).split(".")[0]
!= os.path.basename(preview_file_path).split(".")[0]
):
logger.error(
f"Original and preview file paths do not match: {original_file_path} vs {preview_file_path}"
)
minio_client = await get_minio_storage()
object_name = KnowledgeUtils.get_tmp_preview_file_object_name(original_file_path)
async with aiofiles.open(preview_file_path, "rb") as file_obj:
# Upload preview file tominio
await minio_client.put_object_tmp(
object_name=object_name, file=await file_obj.read()
)
return object_name
def parse_document_title(title: str) -> str:
"""
Parse document titles, removing special characters and extra spaces
:param title: Original title
:return: Post-processing title
"""
# Removing the Thinking Model'sthinkChange Content
title = re.sub(".*", "", title, flags=re.S).strip()
# If there is amd The code fast marker removes the code block marker
if final_title := extract_code_blocks(title):
title = "\n".join(final_title)
return title
def read_chunk_text(
invoke_user_id: int,
input_file: str,
file_name: str,
separator: Optional[List[str]],
separator_rule: Optional[List[str]],
chunk_size: int,
chunk_overlap: int,
knowledge_id: Optional[int] = None,
retain_images: int = 1,
enable_formula: int = 1,
force_ocr: int = 1,
filter_page_header_footer: int = 0,
excel_rule: ExcelRule = None,
no_summary: bool = False,
) -> (List[str], List[dict], str, Any): # type: ignore
"""
0：chunks text
1：chunks metadata
2：parse_type: etl4lm or un_etl4lm
3: ocr bbox data: maybe None
"""
# Gets the title of the document summaryllm
llm = None
if not no_summary:
try:
llm = decide_knowledge_llm(invoke_user_id)
knowledge_llm = LLMService.get_knowledge_llm()
except Exception as e:
logger.exception("knowledge_llm_error:")
raise KnowledgeLLMError()
text_splitter = ElemCharacterTextSplitter(
separators=separator,
separator_rule=separator_rule,
chunk_size=chunk_size,
chunk_overlap=chunk_overlap,
is_separator_regex=True,
)
# Load document content
logger.info(f"start_file_loader file_name={file_name}")
parse_type = ParseType.UN_ETL4LM.value
# excel File processing comes out separately
partitions = []
texts = []
etl_for_lm_url = settings.get_knowledge().etl4lm.url
file_extension_name = file_name.split(".")[-1].lower()
if file_extension_name in ["xls", "xlsx", "csv"]:
# set default values.
if not excel_rule:
excel_rule = ExcelRule()
# convert excel contents to markdown
md_files_path, local_image_dir, doc_id = convert_file_to_md(
file_name=file_name,
input_file_name=input_file,
header_rows=[
excel_rule.header_start_row - 1, # convert to 0-based index
excel_rule.header_end_row - 1,
],
data_rows=excel_rule.slice_length,
append_header=excel_rule.append_header,
retain_images=bool(retain_images),
)
# skip following processes and return splited values.
texts, documents = combine_multiple_md_files_to_raw_texts(path=md_files_path)
elif file_extension_name in ["doc", "docx", "html", "mhtml", "ppt", "pptx"]:
if file_extension_name == "doc":
# convert doc to docx
input_file = convert_doc_to_docx(input_doc_path=input_file)
if not input_file:
raise Exception(
f"failed to convert {file_name} to docx, please check backend log"
)
elif file_extension_name == "ppt":
input_file = convert_ppt_to_pptx(input_path=input_file)
if not input_file:
raise Exception("failed convert ppt to pptx, please check backend log")
md_file_name, local_image_dir, doc_id = convert_file_to_md(
file_name=file_name,
input_file_name=input_file,
knowledge_id=knowledge_id,
retain_images=bool(retain_images),
)
if not md_file_name:
raise Exception(f"failed to parse {file_name}, please check backend log")
# save images to minio
if local_image_dir and retain_images == 1:
put_images_to_minio(
local_image_dir=local_image_dir,
knowledge_id=knowledge_id,
doc_id=doc_id,
)
# will bepptxSave as preview file to
if file_extension_name in ["ppt", "pptx"]:
ppt_pdf_path = convert_ppt_to_pdf(input_path=input_file)
if ppt_pdf_path:
upload_preview_file_to_minio(input_file, ppt_pdf_path)
elif file_extension_name == "doc":
upload_preview_file_to_minio(
input_file.replace(".docx", ".doc"), input_file
)
# Handle it the same way you didmdDoc.
loader = filetype_load_map["md"](file_path=md_file_name, autodetect_encoding=True)
documents = loader.load()
elif file_extension_name in ["txt", "md"]:
loader = filetype_load_map[file_extension_name](file_path=input_file, autodetect_encoding=True)
documents = loader.load()
else:
if etl_for_lm_url:
if file_extension_name in ["pdf"]:
# Determine if the document is damaged
if is_pdf_damaged(input_file):
raise KnowledgeFileDamagedError()
etl4lm_settings = settings.get_knowledge().etl4lm
parse_type = ParseType.ETL4LM.value
try:
loader = Etl4lmLoader(
file_name,
input_file,
unstructured_api_url=etl4lm_settings.url,
ocr_sdk_url=etl4lm_settings.ocr_sdk_url,
force_ocr=bool(force_ocr),
enable_formular=bool(enable_formula),
timeout=etl4lm_settings.timeout,
filter_page_header_footer=bool(filter_page_header_footer),
knowledge_id=knowledge_id,
)
documents = loader.load()
except Exception as e:
raise EtlException(str(e)) from e
partitions = loader.partitions
partitions = parse_partitions(partitions)
else:
if file_extension_name in ['pdf']:
md_file_name, local_image_dir, doc_id = convert_file_to_md(
file_name=file_name,
input_file_name=input_file,
knowledge_id=knowledge_id,
retain_images=bool(retain_images),
)
if not md_file_name: raise Exception(f"failed to parse {file_name}, please check backend log")
# save images to minio
if local_image_dir and retain_images == 1:
put_images_to_minio(
local_image_dir=local_image_dir,
knowledge_id=knowledge_id,
doc_id=doc_id,
)
# Handle it the same way you didmdDoc.
loader = filetype_load_map["md"](file_path=md_file_name)
documents = loader.load()
else:
if file_extension_name not in filetype_load_map:
raise KnowledgeFileNotSupportedError()
loader = filetype_load_map[file_extension_name](file_path=input_file)
documents = loader.load()
logger.info(f"start_extract_title file_name={file_name}")
if llm:
t = time.time()
for one in documents:
# Configured correlationllmIf so, summarize the document
title = extract_title(
llm=llm,
text=one.page_content,
abstract_prompt=knowledge_llm.abstract_prompt,
)
# remove .* tag content
one.metadata["title"] = parse_document_title(title)
logger.info("file_extract_title=success timecost={}", time.time() - t)
if file_extension_name in ["xls", "xlsx", "csv"]:
for one in texts:
one.metadata["title"] = documents[0].metadata.get("title", "")
else:
logger.info(f"start_split_text file_name={file_name}")
texts = text_splitter.split_documents(documents)
raw_texts = [t.page_content for t in texts]
logger.info(f"start_process_metadata file_name={file_name}")
metadatas = [
Metadata(
bbox=json.dumps({"chunk_bboxes": t.metadata.get("chunk_bboxes", "")}),
page=(
t.metadata["chunk_bboxes"][0].get("page")
if t.metadata.get("chunk_bboxes", None)
else t.metadata.get("page", 0)
),
document_name=file_name,
abstract=t.metadata.get("title", ""),
chunk_index=t_index,
user_metadata={}
)
for t_index, t in enumerate(texts)
]
logger.info(f"file_chunk_over file_name=={file_name}")
return raw_texts, metadatas, parse_type, partitions
async def async_read_chunk_text(
invoke_user_id: int,
input_file: str,
file_name: str,
separator: Optional[List[str]],
separator_rule: Optional[List[str]],
chunk_size: int,
chunk_overlap: int,
knowledge_id: Optional[int] = None,
retain_images: int = 1,
enable_formula: int = 1,
force_ocr: int = 1,
filter_page_header_footer: int = 0,
excel_rule: ExcelRule = None,
no_summary: bool = False,
) -> (List[str], List[Metadata], str, Any): # type: ignore
"""Asynchronous version of read_chunk_text"""
llm = None
if not no_summary:
try:
llm = await async_decide_knowledge_llm(invoke_user_id)
knowledge_llm = await LLMService.aget_knowledge_llm()
except Exception as e:
logger.exception("knowledge_llm_error:")
raise Exception(
f"Documentation Knowledge Base Summary Model is no longer valid, please go to Model Management-Configure in System Model Settings.{str(e)}"
)
text_splitter = ElemCharacterTextSplitter(
separators=separator,
separator_rule=separator_rule,
chunk_size=chunk_size,
chunk_overlap=chunk_overlap,
is_separator_regex=True,
)
# Load document content
logger.info(f"start_file_loader file_name={file_name}")
parse_type = ParseType.UN_ETL4LM.value
# excel File processing comes out separately
partitions = []
texts = []
etl_for_lm_url = (await settings.async_get_knowledge()).etl4lm.url
file_extension_name = file_name.split(".")[-1].lower()
if file_extension_name in ["xls", "xlsx", "csv"]:
# set default values.
if not excel_rule:
excel_rule = ExcelRule()
# convert excel contents to markdown
md_files_path, local_image_dir, doc_id = await util.sync_func_to_async(convert_file_to_md)(
file_name=file_name,
input_file_name=input_file,
header_rows=[
excel_rule.header_start_row - 1, # convert to 0-based index
excel_rule.header_end_row - 1,
],
data_rows=excel_rule.slice_length,
append_header=excel_rule.append_header,
retain_images=bool(retain_images),
)
# skip following processes and return splited values.
texts, documents = await util.sync_func_to_async(combine_multiple_md_files_to_raw_texts)(path=md_files_path)
elif file_extension_name in ["doc", "docx", "html", "mhtml", "ppt", "pptx"]:
if file_extension_name == "doc":
# convert doc to docx
input_file = await util.sync_func_to_async(convert_doc_to_docx)(input_doc_path=input_file)
if not input_file:
raise Exception(
f"failed to convert {file_name} to docx, please check backend log"
)
elif file_extension_name == "ppt":
input_file = await asyncio.to_thread(convert_ppt_to_pptx, input_path=input_file)
if not input_file:
raise Exception("failed convert ppt to pptx, please check backend log")
md_file_name, local_image_dir, doc_id = await util.sync_func_to_async(convert_file_to_md)(
file_name=file_name,
input_file_name=input_file,
knowledge_id=knowledge_id,
retain_images=bool(retain_images),
)
if not md_file_name:
raise Exception(f"failed to parse {file_name}, please check backend log")
# save images to minio
if local_image_dir and retain_images == 1:
await async_images_to_minio(
local_image_dir=local_image_dir,
knowledge_id=knowledge_id,
doc_id=doc_id,
)
# will bepptxSave as preview file to
if file_extension_name in ["ppt", "pptx"]:
ppt_pdf_path = await util.sync_func_to_async(convert_ppt_to_pdf)(input_path=input_file)
if ppt_pdf_path:
await async_upload_preview_file_to_minio(input_file, ppt_pdf_path)
elif file_extension_name == "doc":
await async_upload_preview_file_to_minio(
input_file.replace(".docx", ".doc"), input_file
)
# Handle it the same way you didmdDoc.
loader = filetype_load_map["md"](file_path=md_file_name, autodetect_encoding=True)
documents = await loader.aload()
elif file_extension_name in ["txt", "md"]:
loader = filetype_load_map[file_extension_name](file_path=input_file, autodetect_encoding=True)
documents = await loader.aload()
else:
if etl_for_lm_url:
if file_extension_name in ["pdf"]:
# Determine if the document is damaged
if is_pdf_damaged(input_file):
raise Exception('The file is damaged.')
etl4lm_settings = (await settings.async_get_knowledge()).etl4lm
loader = Etl4lmLoader(
file_name,
input_file,
unstructured_api_url=etl4lm_settings.url,
ocr_sdk_url=etl4lm_settings.ocr_sdk_url,
force_ocr=bool(force_ocr),
enable_formular=bool(enable_formula),
timeout=etl4lm_settings.timeout,
filter_page_header_footer=bool(filter_page_header_footer),
knowledge_id=knowledge_id,
)
documents = await loader.aload()
parse_type = ParseType.ETL4LM.value
partitions = loader.partitions
partitions = parse_partitions(partitions)
else:
if file_extension_name in ['pdf']:
md_file_name, local_image_dir, doc_id = await util.sync_func_to_async(convert_file_to_md)(
file_name=file_name,
input_file_name=input_file,
knowledge_id=knowledge_id,
retain_images=bool(retain_images),
)
if not md_file_name: raise Exception(f"failed to parse {file_name}, please check backend log")
# save images to minio
if local_image_dir and retain_images == 1:
await async_images_to_minio(
local_image_dir=local_image_dir,
knowledge_id=knowledge_id,
doc_id=doc_id,
)
# Handle it the same way you didmdDoc.
loader = filetype_load_map["md"](file_path=md_file_name)
documents = await loader.aload()
else:
if file_extension_name not in filetype_load_map:
raise Exception("Type not supported")
loader = filetype_load_map[file_extension_name](file_path=input_file)
documents = await loader.aload()
logger.info(f"start_extract_title file_name={file_name}")
if llm:
t = time.time()
for one in documents:
# Configured correlationllmIf so, summarize the document
title = await async_extract_title(
llm=llm,
text=one.page_content,
abstract_prompt=knowledge_llm.abstract_prompt,
)
# remove .* tag content
one.metadata["title"] = parse_document_title(title)
logger.info("file_extract_title=success timecost={}", time.time() - t)
if file_extension_name in ["xls", "xlsx", "csv"]:
for one in texts:
one.metadata["title"] = documents[0].metadata.get("title", "")
else:
logger.info(f"start_split_text file_name={file_name}")
texts = text_splitter.split_documents(documents)
raw_texts = [t.page_content for t in texts]
logger.info(f"start_process_metadata file_name={file_name}")
metadatas = [
Metadata(
bbox=json.dumps({"chunk_bboxes": t.metadata.get("chunk_bboxes", "")}),
page=(
t.metadata["chunk_bboxes"][0].get("page")
if t.metadata.get("chunk_bboxes", None)
else t.metadata.get("page", 0)
),
document_name=file_name,
abstract=t.metadata.get("title", ""),
chunk_index=t_index,
user_metadata={}
)
for t_index, t in enumerate(texts)
]
logger.info(f"file_chunk_over file_name=={file_name}")
return raw_texts, metadatas, parse_type, partitions
def text_knowledge(
db_knowledge: Knowledge, db_file: KnowledgeFile, documents: List[Document]
):
"""Usetext Importknowledge"""
embeddings = LLMService.get_bisheng_knowledge_embedding_sync(model_id=int(db_knowledge.model),
invoke_user_id=db_file.user_id)
vectore_client = decide_vectorstores(
db_knowledge.collection_name, "Milvus", embeddings
)
logger.info("vector_init_conn_done milvus={}", db_knowledge.collection_name)
index_name = db_knowledge.index_name or db_knowledge.collection_name
es_client = decide_vectorstores(index_name, "ElasticKeywordsSearch", embeddings)
separator = "\n\n"
chunk_size = 1000
chunk_overlap = 100
text_splitter = CharacterTextSplitter(
separator=separator,
chunk_size=chunk_size,
chunk_overlap=chunk_overlap,
add_start_index=True,
)
texts = text_splitter.split_documents(documents)
logger.info(f"chunk_split knowledge_id={db_knowledge.id} size={len(texts)}")
# Storage mysql
file_name = documents[0].metadata.get("source")
db_file.file_name = file_name
with get_sync_db_session() as session:
session.add(db_file)
session.commit()
session.refresh(db_file)
result = db_file.model_dump()
try:
metadata = [
{
"file_id": db_file.id,
"knowledge_id": f"{db_knowledge.id}",
"page": doc.metadata.pop("page", 1),
"source": doc.metadata.pop("source", ""),
"bbox": doc.metadata.pop("bbox", ""),
"extra": json.dumps(doc.metadata, ensure_ascii=False),
"title": "",
"chunk_index": index,
}
for index, doc in enumerate(documents)
]
vectore_client.add_texts(
texts=[t.page_content for t in texts], metadatas=metadata
)
# Storagees
if es_client:
es_client.add_texts(
texts=[t.page_content for t in texts], metadatas=metadata
)
db_file.status = 2
result["status"] = 2
with get_sync_db_session() as session:
session.add(db_file)
session.commit()
except Exception as e:
logger.error(e)
setattr(db_file, "status", 3)
setattr(db_file, "remark", str(e)[:500])
with get_sync_db_session() as session:
session.add(db_file)
session.commit()
result["status"] = 3
result["remark"] = str(e)[:500]
return result
def delete_vector(collection_name: str, partition_key: str):
try:
embeddings = FakeEmbedding()
vectore_client = decide_vectorstores(collection_name, 'Milvus', embeddings)
if isinstance(vectore_client.col, Collection):
if partition_key:
pass
else:
res = vectore_client.col.drop(timeout=1)
logger.info('act=delete_milvus col={} res={}', collection_name, res)
except Exception as e:
# Handle situations where a collection does not exist or where there are other errors
logger.warning(f'act=delete_milvus_failed col={collection_name} error={str(e)}')
# Even an error is considered a successful deletion as the goal is to ensure that there is no dirty data
def delete_es(index_name: str):
try:
embeddings = FakeEmbedding()
esvectore_client = decide_vectorstores(index_name, 'ElasticKeywordsSearch', embeddings)
if esvectore_client:
res = esvectore_client.client.indices.delete(index=index_name, ignore=[400, 404])
logger.info(f'act=delete_es index={index_name} res={res}')
except Exception as e:
# Dealing with non-existent indexes or other errors
logger.warning(f'act=delete_es_failed index={index_name} error={str(e)}')
# Even an error is considered a successful deletion as the goal is to ensure that there is no dirty data
def QA_save_knowledge(db_knowledge: Knowledge, QA: QAKnowledge):
"""Usetext Importknowledge"""
questions = QA.questions
answer = json.loads(QA.answers)[0]
extra = {}
if QA.extra_meta:
extra = json.loads(QA.extra_meta) or {}
extra.update({"answer": answer, "main_question": questions[0]})
docs = [Document(page_content=question, metadata=extra) for question in questions]
try:
embeddings = LLMService.get_bisheng_knowledge_embedding_sync(invoke_user_id=QA.user_id,
model_id=int(db_knowledge.model))
vector_client = decide_vectorstores(
db_knowledge.collection_name, "Milvus", embeddings
)
es_client = decide_vectorstores(
db_knowledge.index_name, "ElasticKeywordsSearch", embeddings
)
logger.info(
f"vector_init_conn_done col={db_knowledge.collection_name} index={db_knowledge.index_name}"
)
# Unificationdocument
metadata = [
{
"file_id": QA.id,
"knowledge_id": f"{db_knowledge.id}",
"page": doc.metadata.pop("page", 1),
"source": doc.metadata.pop("source", ""),
"bbox": doc.metadata.pop("bbox", ""),
"title": doc.metadata.pop("title", ""),
"chunk_index": index,
"extra": json.dumps(doc.metadata, ensure_ascii=False),
}
for index, doc in enumerate(docs)
]
vector_client.add_texts(
texts=[t.page_content for t in docs], metadatas=metadata
)
logger.info(f"qa_save_knowledge add vector over")
es_client.add_texts(texts=[t.page_content for t in docs], metadatas=metadata)
logger.info(f"qa_save_knowledge add es over")
QA.status = QAStatus.ENABLED.value
KnowledgeFileDao.update(QA)
except Exception as e:
logger.error(e)
setattr(QA, "status", QAStatus.FAILED.value)
setattr(QA, "remark", KnowledgeFileFailedError(exception=e).to_json_str())
KnowledgeFileDao.update(QA)
return QA
def add_qa(db_knowledge: Knowledge, data: QAKnowledgeUpsert) -> QAKnowledge:
"""Usetext ImportQAknowledge"""
if db_knowledge.type != 1:
raise Exception("knowledge type error")
try:
# Similar question unified insertion
questions = data.questions
if questions:
if data.id:
qa_db = QAKnoweldgeDao.get_qa_knowledge_by_primary_id(data.id)
qa_db.questions = questions
qa_db.answers = data.answers
qa = QAKnoweldgeDao.update(qa_db)
# Needs to be deleted before insertion
delete_vector_data(db_knowledge, [data.id])
else:
qa = QAKnoweldgeDao.insert_qa(data)
telemetry_service.log_event_sync(
user_id=qa.user_id,
event_type=BaseTelemetryTypeEnum.NEW_KNOWLEDGE_FILE,
trace_id=trace_id_var.get()
)
# Right.questionTo be performedembedding, and then enter the Knowledge Base
qa = QA_save_knowledge(db_knowledge, qa)
return qa
except Exception as e:
logger.exception(e)
raise e
def qa_status_change(qa_db: QAKnowledge, target_status: int, db_knowledge: Knowledge):
"""QA State toggle"""
if qa_db.status == target_status:
logger.info("qa status is same, skip")
return
if target_status == QAStatus.DISABLED.value:
delete_vector_data(db_knowledge, [qa_db.id])
qa_db.status = target_status
QAKnoweldgeDao.update(qa_db)
else:
qa_db.status = QAStatus.PROCESSING.value
QAKnoweldgeDao.update(qa_db)
QA_save_knowledge(db_knowledge, qa_db)
return qa_db
async def list_qa_by_knowledge_id(
knowledge_id: int,
page_size: int = 10,
page_num: int = 1,
question: Optional[str] = None,
answer: Optional[str] = None,
keyword: Optional[str] = None,
status: Optional[int] = None,
) -> list[Any] | tuple[Any, Any]:
"""Get all under knowledge baseqa"""
if not knowledge_id:
return []
count_sql = select(func.count(QAKnowledge.id)).where(
QAKnowledge.knowledge_id == knowledge_id
)
list_sql = select(QAKnowledge).where(QAKnowledge.knowledge_id == knowledge_id)
if status:
count_sql = count_sql.where(QAKnowledge.status == status)
list_sql = list_sql.where(QAKnowledge.status == status)
if question:
count_sql = count_sql.where(QAKnowledge.questions.like(f"%{question}%"))
list_sql = list_sql.where(QAKnowledge.questions.like(f"%{question}%"))
if answer:
count_sql = count_sql.where(QAKnowledge.answers.like(f"%{answer}%"))
list_sql = list_sql.where(QAKnowledge.answers.like(f"%{answer}%"))
if keyword:
count_sql = count_sql.where(
or_(
QAKnowledge.questions.like(f"%{keyword}%"),
QAKnowledge.answers.like(f"%{keyword}%"),
)
)
list_sql = list_sql.where(
or_(
QAKnowledge.answers.like(f"%{keyword}%"),
QAKnowledge.questions.like(f"%{keyword}%"),
)
)
list_sql = (
list_sql.order_by(QAKnowledge.update_time.desc())
.limit(page_size)
.offset((page_num - 1) * page_size)
)
count = await QAKnoweldgeDao.total_count(count_sql)
list_qa = await QAKnoweldgeDao.query_by_condition(list_sql)
return list_qa, count
def delete_vector_data(knowledge: Knowledge, file_ids: List[int]):
"""Delete vector data, Want to make a general purpose that can be dockedlangchainright of privacyvectorDB"""
# embeddings = FakeEmbedding()
# vectore_config_dict: dict = settings.get_knowledge().get('vectorstores')
# if not vectore_config_dict:
# raise Exception('Vector database must be configured')
# elastic_index = knowledge.index_name or knowledge.collection_name
# vectore_client_list = [
# decide_vectorstores(elastic_index, db, embeddings) if db == 'ElasticKeywordsSearch' else
# decide_vectorstores(knowledge.collection_name, db, embeddings)
# for db in vectore_config_dict.keys()
# ]
# logger.info('vector_init_conn_done col={} dbs={}', knowledge.collection_name,
# vectore_config_dict.keys())
# for vectore_client in vectore_client_list:
# Inquiryvector primary key
embeddings = FakeEmbedding()
collection_name = knowledge.collection_name
# Medical Treatment:vectordb
vectore_client = decide_vectorstores(collection_name, "Milvus", embeddings)
try:
if isinstance(vectore_client.col, Collection):
pk = vectore_client.col.query(
expr=f"file_id in {file_ids}", output_fields=["pk"], timeout=10
)
else:
pk = []
except Exception:
# Want to try that again?
logger.error("timeout_except")
vectore_client.close_connection(vectore_client.alias)
vectore_client = decide_vectorstores(collection_name, "Milvus", embeddings)
pk = vectore_client.col.query(
expr=f"file_id in {file_ids}", output_fields=["pk"], timeout=10
)
logger.info("query_milvus pk={}", pk)
if pk:
res = vectore_client.col.delete(f"pk in {[p['pk'] for p in pk]}", timeout=10)
logger.info(f"act=delete_vector file_id={file_ids} res={res}")
vectore_client.close_connection(vectore_client.alias)
# elastic
index_name = knowledge.index_name or collection_name
esvectore_client = decide_vectorstores(
index_name, "ElasticKeywordsSearch", embeddings
)
if esvectore_client:
res = esvectore_client.client.delete_by_query(
index=index_name, body={"query": {"terms": {"metadata.file_id": file_ids}}}
)
logger.info(f"act=delete_es res={res}")
return True
def recommend_question(invoke_user_id: int, question: str, answer: str, number: int = 3) -> List[str]:
from langchain.chains.llm import LLMChain
from langchain_core.prompts.prompt import PromptTemplate
prompt = """- Role: Problem Generation Specialist
- Background: Users want to generate similar questions based on given questions and answers through artificial intelligence models in order to expand the knowledge base or for educational and testing purposes.
- Profile: You are a professional data analyst and language modeler who specializes in extracting patterns from existing data and generating new relevant questions.
- Constrains: Ensure that the generated questions are semantically similar to the original questions, while maintaining diversity and avoiding duplication.
- Workflow:
1. Analyze questions and answers entered by users and extract keywords and topics.
2. Create similar questions based on extracted keywords and topics.
3. Verify that the generated questions are semantically similar to the original questions and ensure diversity.
- Examples:
Question:"What is the capital of France?"
Answers:"Paris"
Buat3similar questions:
- "What is the name of the capital of France?"
- "Which city is the capital of France?"
- "What country's capital is Paris?"
Please usejson Return
{{"questions": Generated Question List}}
Here are the questions and answers provided by the user:
Question:{question}
Answers:{answer}
You generated{number}similar questions:
"""
llm = LLMService.get_knowledge_similar_llm(invoke_user_id)
if not llm:
raise KnowledgeSimilarError.http_exception()
llm_chain = LLMChain(llm=llm, prompt=PromptTemplate.from_template(prompt))
gen_question = llm_chain.predict(question=question, answer=answer, number=number)
try:
code_ret = extract_code_blocks(gen_question)
if code_ret:
question_dict = json.loads(code_ret[0])
return question_dict["questions"]
elif gen_question:
question_dict = json.loads(gen_question)
return question_dict.get("questions", [])
else:
logger.info("md_code_extract_error {}", gen_question)
return []
except Exception as exc:
logger.error("recommend_question json.loads error:{}", gen_question)
raise KnowledgeRecommendQuestionError(exception=exc, message=gen_question)
def extract_code_blocks(markdown_code_block: str):
# Define regular expression patterns
pattern = r"```\w*\s*(.*?)```"
# Use re.DOTALL letting . Ability to match line breaks
matches = re.findall(pattern, markdown_code_block, re.DOTALL)
# Remove whitespace at both ends of each code block
return [match.strip() for match in matches]
================================================
FILE: src/backend/bisheng/api/services/libreoffice_converter.py
================================================
import os
import shutil # For checking if the executable is in PATH
import subprocess
import tempfile
from loguru import logger
def get_libreoffice_path():
"""
Tries to find the LibreOffice executable.
prerequisites:
1. install libreoffice
2. linux:
sudo apt-get install libreoffice
sudo yum install libreoffice-headless
3. macos:
brew install libreoffice
"""
if shutil.which("soffice"):
return "soffice"
if shutil.which("libreoffice"):
return "libreoffice"
# Common Windows paths
windows_paths = [
r"C:\Program Files\LibreOffice\program\soffice.exe",
r"C:\Program Files (x86)\LibreOffice\program\soffice.exe",
]
for path in windows_paths:
if os.path.exists(path):
return path
return None
def _convert_file_extension(input_path, convert_extension, output_dir=None, except_file_ext=None):
if not os.path.isabs(input_path):
input_path = os.path.abspath(input_path)
if not os.path.exists(input_path):
logger.debug(f"Error: Input file not found at '{input_path}'")
return None
if output_dir is None:
output_dir = os.path.dirname(input_path)
else:
if not os.path.isabs(output_dir):
output_dir = os.path.abspath(output_dir)
if not os.path.exists(output_dir):
try:
os.makedirs(output_dir, exist_ok=True)
logger.debug(f"Created output directory: '{output_dir}'")
except OSError as e:
logger.debug(f"Error creating output directory '{output_dir}': {e}")
return None
# Check if libreoffice_exec is in PATH if it's not a full path
soffice_path = get_libreoffice_path()
if not soffice_path:
logger.debug(
"Error: LibreOffice (soffice) command not found. Please install LibreOffice and ensure it's in your PATH, or adjust 'get_libreoffice_path()'."
)
return False
base_name = os.path.basename(input_path)
file_name_no_ext = os.path.splitext(base_name)[0]
output_path = os.path.join(output_dir, f"{file_name_no_ext}.{except_file_ext}")
try:
with tempfile.TemporaryDirectory() as tmp_dir:
command = [
soffice_path,
"--headless", # Run in headless mode (no GUI)
"-env:SingleAppInstance=false",
f"-env:UserInstallation=file://{tmp_dir}",
"--convert-to",
convert_extension,
"--outdir",
output_dir, # Specify the output directory
input_path, # The input file
]
logger.debug(f"Executing command: {' '.join(command)}")
process = subprocess.run(
command, check=True, capture_output=True, text=True, timeout=180
) # 120 seconds timeout
logger.debug(f"LibreOffice STDOUT: {process.stdout}")
if process.stderr: # LibreOffice sometimes logger.debugs info to stderr even on success
logger.debug(f"LibreOffice STDERR: {process.stderr}")
# Check if the file was actually created
# LibreOffice creates the file with the correct name in the output_dir
if os.path.exists(output_path):
# If output_docx_path is different (it shouldn't be with this logic, but for safety)
logger.debug(f"Successfully converted '{input_path}' to '{output_path}'")
return output_path
else:
# This case should ideally not happen if subprocess.run didn't raise an error
# and LibreOffice worked as expected.
logger.debug(
f"Error: Conversion command seemed to succeed, but output file '{output_path}' not found."
)
return None
except FileNotFoundError:
logger.debug(
f"Error: The LibreOffice executable '{soffice_path}' was not found."
)
logger.debug(
"Ensure LibreOffice is installed and the command is in your PATH or provide the full path."
)
return None
except subprocess.CalledProcessError as e:
logger.debug(f"Error during LibreOffice conversion for '{input_path}':")
logger.debug(f"Command: {' '.join(e.cmd)}")
logger.debug(f"Return code: {e.returncode}")
logger.debug(f"STDOUT: {e.stdout}")
logger.debug(f"STDERR: {e.stderr}")
return None
except subprocess.TimeoutExpired:
logger.debug(f"Error: LibreOffice conversion for '{input_path}' timed out.")
return None
except Exception as e:
logger.debug(
f"An unexpected error occurred during conversion of '{input_path}': {e}"
)
return None
def convert_doc_to_docx(input_doc_path, output_dir=None):
"""
Converts a .doc file to .docx using LibreOffice/soffice command line.
Args:
input_doc_path (str): The absolute path to the input .doc file.
output_dir (str, optional): The directory to save the converted .docx file.
If None, saves in the same directory as the input file.
libreoffice_exec (str, optional): The command name or full path of the
LibreOffice executable (e.g., 'libreoffice',
'soffice', or '/opt/libreoffice7.x/program/soffice').
Returns:
str: The path to the converted .docx file if successful, None otherwise.
"""
if not input_doc_path.lower().endswith((".doc", ".docx")):
logger.debug(f"Error: Input file '{input_doc_path}' is not a .doc file.")
return None
return _convert_file_extension(input_doc_path, "docx:Office Open XML Text", output_dir, except_file_ext="docx")
def convert_ppt_to_pdf(input_path, output_dir=None):
"""
Converts .ppt or .pptx to PDF using LibreOffice soffice command.
Args:
input_path (str): Path to the .ppt or .pptx file.
output_dir (str, optional): Directory to save the PDF.
Defaults to the same directory as the input file.
"""
if not input_path.lower().endswith((".ppt", ".pptx")):
logger.debug(f"Error: {input_path} is not a .ppt or .pptx file.")
return False
return _convert_file_extension(input_path, "pdf", output_dir, except_file_ext="pdf")
def convert_ppt_to_pptx(input_path, output_dir=None):
"""
Converts .ppt to .pptx using LibreOffice soffice command.
Args:
input_path (str): Path to the .ppt file.
output_dir (str, optional): Directory to save the .pptx.
Defaults to the same directory as the input file.
"""
if not input_path.lower().endswith(".ppt"):
logger.debug(f"Error: {input_path} is not a .ppt file.")
return False
return _convert_file_extension(input_path, "pptx", output_dir, except_file_ext="pptx")
if __name__ == "__main__":
file_name = "/Users/tju/Resources/docs/docx/resume.doc"
convert_doc_to_docx(file_name, output_dir="/Users/tju/Resources/docs/docx")
logger.debug(f"{os.path.basename(file_name)}/x")
================================================
FILE: src/backend/bisheng/api/services/md_from_docx.py
================================================
import os
from pathlib import Path
from uuid import uuid4
import pypandoc
from loguru import logger
from bisheng.api.services.libreoffice_converter import convert_doc_to_docx
try:
# Try checking pandoc version, try to download if it fails
pandoc_path = pypandoc.get_pandoc_path()
logger.debug(f"Pandoc found at: {pandoc_path}")
except OSError: # OSError Yes get_pandoc_path Thrown when not found
logger.debug("Pandoc not found. Attempting to download pandoc...")
try:
pypandoc.download_pandoc() # This will download to pypandoc in the package directory of
logger.debug("Pandoc downloaded successfully by pypandoc.")
# You may need to re-fetch the path or pypandoc After that, it will be automatically found
except Exception as e_download:
logger.debug(f"Failed to download pandoc using pypandoc: {e_download}")
exit() # Exit if unable to download
def convert_doc_to_md_pandoc_high_quality(
doc_path_str: str, output_md_str: str, image_dir_name: str = "media", retry_convert_docx: bool = True,
):
"""
Use Pandoc will be .doc OR .docx Convert files with high quality to Markdown, and extract the image.
Parameters:
doc_path_str (str): Entered Word Document path
output_md_str (str): Output Markdown FilePath
image_dir_name (str): The subdirectory name used to store the extracted image. This directory will be created in Markdown Next to the file.
"""
doc_path = Path(doc_path_str)
output_md_path = Path(output_md_str)
if not doc_path.exists():
logger.debug(f"Error: Input file {doc_path} %s does not exist.")
return
# Ensure Output Markdown The file's parent directory exists
output_md_path.parent.mkdir(parents=True, exist_ok=True)
# Pandoc Output Format Options (gfm Often a good choice)
pandoc_format_to = "gfm"
# Pandoc Extra arguments
# --extract-media=Directory name: told Pandoc Extracts all media files, such as pictures, to the specified subdirectory.
# Pandoc will automatically create this directory and Markdown The image link in points to this directory.
# --atx-headers: If your Pandoc version support, this option will use '#' The title of the style.
# If you previously reported an error due to a version issue and you did not upgrade Pandoc, you can comment out this line.
extra_args = [
"--wrap=none",
# '--atx-headers', # Automatically close purchase order after Pandoc Older version causes this option to report an error, please comment it out or upgrade Pandoc
f"--extract-media={image_dir_name}", # Key: Extract images to specified subdirectories
]
# Images will be extracted to output_md_path Under the sibling directory image_dir_name In subdirectories
# For example: if output_md_path Yes "output/document.md" Dan image_dir_name Yes "images",
# Images will be stored in "output/images/" Under the directory, the link would be "images/image1.png"
try:
pypandoc.convert_file(
source_file=str(doc_path),
to=pandoc_format_to,
outputfile=str(output_md_path),
extra_args=extra_args,
)
logger.debug(f"Pandoc Conversion Complete: {output_md_path}")
except RuntimeError as e: # Pandoc Often thrown when an error is not found or executed RuntimeError
if "Unknown option --atx-headers" in str(e):
logger.debug(
" error message '--atx-headers' The option is unknown, which usually means your Pandoc Older version."
)
except Exception as e: # Other potential errors
logger.debug(f"Convert File {doc_path} An unknown error occurred while: {e}")
# If the conversion fails, try todocxConvert to StandarddocxTry Again
if not os.path.exists(output_md_path):
if retry_convert_docx:
output_dir = os.path.dirname(doc_path_str)
output_dir = os.path.join(output_dir, "tmp")
output_docx_path = convert_doc_to_docx(doc_path_str, output_dir)
if not output_docx_path:
raise RuntimeError("convert to docx failed")
convert_doc_to_md_pandoc_high_quality(doc_path_str=output_docx_path, output_md_str=output_md_str,
image_dir_name=image_dir_name, retry_convert_docx=False)
return
raise RuntimeError("convert to md failed")
def handler(cache_dir, file_name):
"""
The main function that handles file conversions.
Parameters:
file_name (str): Entered Word Document path
knowledge_id (str): Knowledge ID, which is used to generate the output file name.
"""
doc_id = str(uuid4())
md_file_name = f"{cache_dir}/{doc_id}.md"
local_image_dir = f"{cache_dir}/{doc_id}"
convert_doc_to_md_pandoc_high_quality(
doc_path_str=file_name,
output_md_str=md_file_name,
image_dir_name=local_image_dir,
)
return md_file_name, f"{local_image_dir}/media", doc_id
if __name__ == "__main__":
# Define test parameters
test_cache_dir = "/Users/tju/Desktop"
test_file_name = "/Users/tju/Resources/docs/docx/resume.docx"
# test_file_name = "/Users/tju/Resources/docs/docx/2307.09288.docx"
# Recall handler Function is tested
md_file_name, image_dir, doc_id = handler(
cache_dir=test_cache_dir,
file_name=test_file_name,
)
# Output Results
print(f"Markdown FilePath: {md_file_name}")
print(f"Picture directory path: {image_dir}")
print(f"Documentation ID: {doc_id}")
================================================
FILE: src/backend/bisheng/api/services/md_from_excel.py
================================================
import math
import os
from typing import List
from uuid import uuid4
import openpyxl
import pandas as pd
from loguru import logger
def xls_to_xlsx(xls_path):
if not xls_path.lower().endswith(".xls"):
return None
if not os.path.exists(xls_path):
return None
try:
xls_file = pd.ExcelFile(xls_path)
sheets_to_write = {}
# 2. Iterate through all worksheets, check if empty, and save non-empty content to the dictionary
for sheet_name in xls_file.sheet_names:
df = xls_file.parse(sheet_name)
# df.empty will judge DataFrame No data (the number of rows is0）
if not df.empty:
sheets_to_write[sheet_name] = df
else:
# Discard Blank Sheet
pass
# 3. Do not create a new file if there are any non-empty worksheets
if not sheets_to_write:
return None
# 4. Write a new file if a non-empty worksheet exists
xlsx_path = os.path.splitext(xls_path)[0] + ".xlsx"
with pd.ExcelWriter(xlsx_path, engine="openpyxl") as writer:
for sheet_name, df in sheets_to_write.items():
df.to_excel(writer, sheet_name=sheet_name, index=False)
return xlsx_path
except Exception as e:
logger.exception(f'xls_to_xlsx error: ')
return None
def remove_characters(s, chars_to_remove=["\n", "\r"]):
"""
Removes the specified character from the string.
"""
if not isinstance(s, str):
return s
for char in chars_to_remove:
s = s.replace(char, "")
return s.strip()
def unmerge_and_read_sheet(sheet_obj):
"""
read out openpyxl Sheet object, unmerge cells by populating the top-left value of the merge range into all cells in the range.
and returns the data as a list of lists.
"""
if sheet_obj.max_row == 0 or sheet_obj.max_column == 0:
return []
max_row = sheet_obj.max_row
max_column = sheet_obj.max_column
data_grid = [
[None for _ in range(max_column)] for _ in range(max_row)
]
# Berturut-turut50Row Blank Row Stop Reading Content
empty_row_num = 0
max_empty_rows = 50
empty_row_end = 0
for r_idx, row in enumerate(sheet_obj.iter_rows()):
if empty_row_num > max_empty_rows:
break
row_empty = True
for c_idx, cell in enumerate(row):
data_grid[r_idx][c_idx] = cell.value
if cell.value:
row_empty = False
if row_empty:
empty_row_num += 1
else:
empty_row_num = 0
if empty_row_num > 0:
data_grid = data_grid[:-empty_row_num]
merged_cell_ranges = list(sheet_obj.merged_cells.ranges)
for merged_range in merged_cell_ranges:
min_col, min_row, max_col, max_row = merged_range.bounds
top_left_cell_value = sheet_obj.cell(row=min_row, column=min_col).value
# ignore empty rows
if min_row > len(data_grid):
continue
for r in range(min_row, max_row + 1):
for c in range(min_col, max_col + 1):
data_grid[r - 1][c - 1] = top_left_cell_value
return data_grid
def generate_markdown_table_string(
header_rows_list_of_lists,
data_rows_list_of_lists,
num_columns,
separator_placement_index=1,
):
"""
Generate from new rulesMarkdownTable String
Automatically close purchase order afterheader_rows_list_of_listsIf empty, no headers and delimiters are generated.
"""
md_lines = []
# Handle headers and delimiters only if a header row is provided
if header_rows_list_of_lists:
pre_separator_header = header_rows_list_of_lists[:separator_placement_index]
for row_values in pre_separator_header:
md_lines.append(
"| "
+ " | ".join(
remove_characters(str(v)) if v is not None else ""
for v in row_values
)
+ " |"
)
# Insert below the header in the first rowMarkdownSeparator
if num_columns > 0:
md_lines.append("|" + "---|" * num_columns)
post_separator_header = header_rows_list_of_lists[separator_placement_index:]
for row_values in post_separator_header:
md_lines.append(
"| "
+ " | ".join(
remove_characters(str(v)) if v is not None else ""
for v in row_values
)
+ " |"
)
# Always process data rows
for row_values in data_rows_list_of_lists:
md_lines.append(
"| "
+ " | ".join(
remove_characters(str(v)) if v is not None else "" for v in row_values
)
+ " |"
)
return "\n".join(md_lines)
def process_dataframe_to_markdown_files(
df,
sheet_index: str,
num_header_rows,
rows_per_markdown,
output_dir,
append_header=True,
):
"""
- append_header=True: Tekan num_header_rows Separate the header and data.
- append_header=False: All content is treated as data, table header is empty, ignored num_header_rows。
"""
if df.empty:
logger.warning(f" feed '{sheet_index}' DataDataFrameEmpty, skippingMarkdownBuat")
return
num_columns = df.shape[1]
rows = df.shape[0]
if rows == 0 or num_columns == 0:
return
header_block_df = pd.DataFrame()
start_header_idx, end_header_idx = num_header_rows[0], num_header_rows[1]
if start_header_idx >= rows:
append_header = False
# --- Core Logic Modified: According to append_header Decide how to split the data ---
if append_header:
# Handle header index outliers based on user rules
if start_header_idx >= rows:
logger.warning(
f"Table Header Start Row {start_header_idx} Total lines exceeded {rows}. The first row will be used as the table header.")
start_header_idx, end_header_idx = 0, 0
elif end_header_idx >= rows:
logger.warning(
f"Table Header End Row {end_header_idx} Total lines exceeded {rows}. will be truncated to the last line.")
end_header_idx = rows - 1
# Make sure the index is legitimate
if start_header_idx < 0: start_header_idx = 0
if end_header_idx < start_header_idx: end_header_idx = start_header_idx
try:
header_slice = slice(start_header_idx, end_header_idx + 1)
header_block_df = df.iloc[header_slice]
data_block_df = df.drop(df.index[header_slice]).reset_index(drop=True)
header_rows_as_lists = header_block_df.values.tolist()
except Exception as e:
logger.error(
f" At Source '{sheet_index}' Index by header in [{start_header_idx}, {end_header_idx}] Error Splitting Data: {e}Skip")
return
else:
# when append_header are False , everything is treated as data and the header list is empty
header_rows_as_lists = []
data_block_df = df.reset_index(drop=True)
# --- Subsequent pagination logic ---
if data_block_df.empty:
if append_header and not header_block_df.empty:
markdown_content = generate_markdown_table_string(
header_rows_as_lists, [], num_columns
)
# BUG FIX: Use zfill for proper padding. This is file '000' for the sheet.
file_name = f"{str(sheet_index).zfill(2)}000.md"
file_path = os.path.join(output_dir, file_name)
try:
with open(file_path, "w", encoding="utf-8") as f:
f.write(markdown_content)
logger.debug(f" Header-only files saved:'{file_path}'")
except Exception as e:
logger.debug(f" Save file '{file_path}' Error during: {e}")
return
num_data_rows_total = len(data_block_df)
num_files_to_create = math.ceil(num_data_rows_total / rows_per_markdown) if rows_per_markdown > 0 else (
1 if num_data_rows_total > 0 else 0)
for i in range(num_files_to_create):
start_idx = i * rows_per_markdown
end_idx = min(start_idx + rows_per_markdown, num_data_rows_total)
current_data_chunk_as_lists = data_block_df.iloc[start_idx:end_idx].values.tolist()
final_header_for_chunk = header_rows_as_lists
final_data_for_chunk = current_data_chunk_as_lists
# If no real header is attached and the current data block is not empty, the first row of data is used as the “pseudo header” to generate the delimiter
if not append_header and current_data_chunk_as_lists:
final_header_for_chunk = [current_data_chunk_as_lists[0]]
final_data_for_chunk = current_data_chunk_as_lists[1:]
markdown_content = generate_markdown_table_string(
final_header_for_chunk, final_data_for_chunk, num_columns
)
# BUG FIX: Use zfill for proper 2-digit sheet and 3-digit file padding.
file_name = f"{str(sheet_index).zfill(2)}{str(i).zfill(3)}.md"
file_path = os.path.join(output_dir, file_name)
try:
with open(file_path, "w", encoding="utf-8") as f:
f.write(markdown_content)
logger.debug(
f" Sudah disimpan'{file_path}' (incl. {len(current_data_chunk_as_lists)} Row Raw Data)"
)
except Exception as e:
logger.debug(f" Save file '{file_path}' Error during: {e}")
def is_list_of_lists_empty(data_list):
"""
Determine if a 2D list is empty or contains only empty values (None, '')。
"""
if not data_list:
return True
# Use any() and generator expressions for efficient judgment
# any(row) Check for non-empty lines
# any(cell is not None and cell != '' for cell in row) Check if there are non-empty cells in the row
return not any(any(cell is not None and str(cell).strip() != '' for cell in row) for row in data_list)
def excel_file_to_markdown(
excel_path, num_header_rows, rows_per_markdown, output_dir, append_header=True
):
logger.debug(f"\nStart ProcessingExcelDocumentation:'{excel_path}'")
try:
workbook = openpyxl.load_workbook(excel_path, data_only=True, read_only=False)
except Exception as e:
logger.debug(f"Error: Unable to loadExcelDoc. '{excel_path}'Reason: {e}")
return
sheet_index = 0
for sheet_name in workbook.sheetnames:
logger.debug(f"\n (In work)ExcelWorksheet'{sheet_name}'...")
sheet_obj = workbook[sheet_name]
unmerged_data_list_of_lists = unmerge_and_read_sheet(sheet_obj)
logger.debug(f"\n Excel'{sheet_name}'...{len(unmerged_data_list_of_lists)}")
# Using the new decision function
if is_list_of_lists_empty(unmerged_data_list_of_lists):
logger.debug(f" Worksheet '{sheet_name}' Empty or no valid data, skipping.")
continue
df = pd.DataFrame(unmerged_data_list_of_lists)
df.fillna("", inplace=True)
if df.empty:
logger.debug(f" Worksheet '{sheet_name}' Empty after processingDataFrameSkip")
continue
process_dataframe_to_markdown_files(
df,
str(sheet_index),
num_header_rows,
rows_per_markdown,
output_dir,
append_header=append_header,
)
sheet_index += 1
if workbook:
workbook.close()
logger.debug(f"\nExcelDoc. '{excel_path}' Process Completed.")
def csv_file_to_markdown(
csv_path,
num_header_rows,
rows_per_markdown,
output_dir,
csv_encoding="utf-8",
csv_delimiter=",",
append_header=True,
):
logger.debug(f"\nStart ProcessingCSVDocumentation:'{csv_path}'")
try:
df = pd.read_csv(
csv_path,
header=None,
dtype=str,
encoding=csv_encoding,
sep=csv_delimiter,
keep_default_na=False,
)
df.fillna("", inplace=True)
except pd.errors.EmptyDataError:
logger.debug(f"Error: CSVDoc. '{csv_path}' Empty")
return
except FileNotFoundError:
logger.debug(f"Error: CSVDoc. '{csv_path}' Nothing found.")
return
except Exception as e:
logger.debug(f"Error: UnreadableCSVDoc. '{csv_path}'Reason: {e}")
return
if df.empty:
logger.debug(f"CSVDoc. '{csv_path}' Empty or empty after processing, skipping.")
return
process_dataframe_to_markdown_files(
df,
"0",
num_header_rows,
rows_per_markdown,
output_dir,
append_header,
)
logger.debug(f"\nCSVDoc. '{csv_path}' Process Completed.")
def convert_file_to_markdown(
input_file_path,
num_header_rows,
rows_per_markdown,
base_output_dir="output_markdown_files",
csv_encoding="utf-8",
csv_delimiter=",",
append_header=True,
):
"""
will be Excel OR CSV Convert files to multiple Markdown files.
"""
if not os.path.exists(input_file_path):
logger.debug(f"Error: Input file '{input_file_path}' Nothing found.")
return
if not os.path.exists(base_output_dir):
os.makedirs(base_output_dir)
logger.debug(f"To create an output directory:'{base_output_dir}'")
_, file_extension = os.path.splitext(input_file_path)
file_extension = file_extension.lower()
if file_extension == ".xls":
input_file_path = xls_to_xlsx(input_file_path)
if file_extension in [".xlsx", ".xls"]:
excel_file_to_markdown(
input_file_path,
num_header_rows,
rows_per_markdown,
base_output_dir,
append_header,
)
elif file_extension == ".csv":
csv_file_to_markdown(
input_file_path,
num_header_rows,
rows_per_markdown,
base_output_dir,
csv_encoding,
csv_delimiter,
append_header,
)
else:
logger.debug(
f"Error: Unsupported file type '{file_extension}'Please provide user. Excel (.xlsx, .xls) OR CSV (.csv) files."
)
def handler(
cache_dir,
file_name: str,
header_rows: List[int] = [0, 1],
data_rows: int = 12,
append_header=True,
):
"""
The main function that handles file conversions.
"""
doc_id = uuid4()
md_file_name = f"{cache_dir}/{doc_id}"
convert_file_to_markdown(
input_file_path=file_name,
base_output_dir=md_file_name,
num_header_rows=header_rows,
rows_per_markdown=data_rows,
append_header=append_header,
)
return md_file_name, None, doc_id
if __name__ == "__main__":
# Define test parameters
test_cache_dir = "/Users/zhangguoqing/Downloads/tmp"
test_file_name = "/Users/zhangguoqing/Downloads/124327.xlsx"
# Test append_header=True and the index is out of bounds
test_header_rows = [0, 0] # start_header_index Out of Scope
test_data_rows = 2
test_append_header = True
# Recall handler Function
print("--- Test Scenarios: append_header=True, Table header index out of bounds ---")
handler(
cache_dir=test_cache_dir,
file_name=test_file_name,
header_rows=test_header_rows,
data_rows=test_data_rows,
append_header=test_append_header,
)
================================================
FILE: src/backend/bisheng/api/services/md_from_html.py
================================================
import requests
from bs4 import BeautifulSoup, Comment
from markdownify import markdownify as md
import os
import re
import base64
from urllib.parse import urljoin, urlparse
from uuid import uuid4
from loguru import logger
import shutil
from pathlib import Path
# Configure logger
USER_AGENT = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36"
class HTML2MarkdownConverter:
def __init__(
self,
output_dir="output",
media_download_timeout=60,
):
self.output_dir = output_dir
self.MEDIA_DOWNLOAD_TIMEOUT = media_download_timeout
os.makedirs(self.output_dir, exist_ok=True)
self.current_image_absolute_path = None
self.current_video_absolute_path = None
self.base_url = None
# mhtml_resources is kept for cid processing in case it's used by other parts, but parsing is removed.
self.mhtml_resources = {}
self.source_html_filepath = None
def _clean_html(self, html_content):
logger.debug("Starting HTML cleaning (refined logic).")
soup = BeautifulSoup(html_content, "html.parser")
for D_tag in soup.find_all(["script", "style", "link", "meta"]):
D_tag.decompose()
for comment in soup.find_all(string=lambda text: isinstance(text, Comment)):
comment.extract()
potentially_problematic_container_tags = ["header", "footer", "nav", "aside"]
non_content_patterns = re.compile(
r"adsbygoogle|ad-slot|advertisement|promo(tion)?|banner-ad|popup-ad|cookie-notice|gdpr-banner|newsletter-signup|social-share-buttons|flyout-menu",
re.IGNORECASE,
)
non_content_roles = [
"banner",
"navigation",
"search",
"complementary",
"contentinfo",
"dialog",
"menubar",
"toolbar",
"directory",
"log",
"status",
"timer",
]
media_tags_to_check = ["img", "video", "picture", "figure", "svg", "audio"]
for tag in list(soup.find_all(True)):
if not tag.parent:
continue
decomposed_this_iteration = False
if tag.name in potentially_problematic_container_tags:
if not tag.find_all(media_tags_to_check):
tag.decompose()
decomposed_this_iteration = True
if decomposed_this_iteration:
continue
if tag.name not in media_tags_to_check:
class_match = any(
non_content_patterns.search(cls) for cls in tag.get("class", [])
)
id_match = (
non_content_patterns.search(tag.get("id", ""))
if tag.get("id")
else False
)
role_match = tag.get("role", "") in non_content_roles
if class_match or id_match or role_match:
if not tag.find_all(media_tags_to_check):
tag.decompose()
decomposed_this_iteration = True
if decomposed_this_iteration:
continue
form_elements_to_remove = [
"form",
"button",
"input",
"select",
"textarea",
"fieldset",
"legend",
]
for tag_name_to_remove in form_elements_to_remove:
for form_tag in list(soup.find_all(tag_name_to_remove)):
if not form_tag.parent:
continue
if not form_tag.find_all(media_tags_to_check):
form_tag.decompose()
for tag in soup.find_all(True):
if not tag.parent and tag.name not in ["html", "head", "body"]:
continue
attrs_to_remove = [
attr for attr in tag.attrs if attr.startswith("on") or attr == "style"
]
for attr in attrs_to_remove:
del tag.attrs[attr]
logger.debug("HTML cleaning (refined logic) finished.")
return str(soup)
def _download_media_file(
self,
media_url,
base_url_for_relative,
media_absolute_save_dir,
markdown_relative_media_folder,
media_type_prefixes=("image/", "video/", "audio/"),
):
if not media_absolute_save_dir:
logger.error(
f"Absolute path for saving media (media_absolute_save_dir) is not set for URL: {media_url}"
)
return None, media_url
original_media_url_for_error_logger = media_url
try:
parsed_media_url = urlparse(media_url)
if media_url.startswith("data:"):
if not any(
prefix in media_url
for prefix in media_type_prefixes
if prefix == "image/"
):
return None, media_url
try:
header, encoded = media_url.split(",", 1)
media_data = base64.b64decode(encoded)
ext_match = re.search(
r"data:(?Pimage|video|audio)/(?P[a-zA-Z0-9+]+);",
header,
)
ext = ext_match.group("ext").lower() if ext_match else "png"
if ext == "svg+xml":
ext = "svg"
elif ext == "jpeg":
ext = "jpg"
if not ext or len(ext) > 5 or not ext.isalnum():
ext = "png"
unique_filename = f"media_{uuid4().hex}.{ext}"
absolute_filepath = os.path.join(
media_absolute_save_dir, unique_filename
)
markdown_path = os.path.join(
markdown_relative_media_folder, unique_filename
)
with open(absolute_filepath, "wb") as f:
f.write(media_data)
logger.info(f"Data URI media saved to {absolute_filepath}")
return markdown_path, media_url
except Exception as e:
logger.error(f"Failed to decode/save data URI media: {e}")
return None, media_url
actual_media_url_str = media_url
if not parsed_media_url.scheme or not parsed_media_url.netloc:
if not base_url_for_relative:
logger.warning(
f"Cannot resolve relative media URL {actual_media_url_str} without a base URL."
)
return None, actual_media_url_str
actual_media_url_str = urljoin(
base_url_for_relative, actual_media_url_str
)
parsed_actual_url = urlparse(actual_media_url_str)
ext = None
path_part_for_ext = parsed_actual_url.path
filename_from_url_for_ext = os.path.basename(path_part_for_ext)
if "." in filename_from_url_for_ext:
candidate_ext = filename_from_url_for_ext.split(".")[-1].lower()
if (
len(candidate_ext) <= 5
and candidate_ext.isalnum()
and candidate_ext
in [
"jpg",
"jpeg",
"png",
"gif",
"svg",
"webp",
"bmp",
"tiff",
"mp4",
"webm",
"ogg",
"mov",
"avi",
"mkv",
"mp3",
"wav",
"aac",
]
):
ext = candidate_ext
if parsed_actual_url.scheme == "file":
local_file_path_str = parsed_actual_url.path
if (
os.name == "nt"
): # Windows: remove leading '/' if path starts like /C:/...
if (
len(local_file_path_str) > 2
and local_file_path_str[0] == "/"
and local_file_path_str[2] == ":"
):
local_file_path_str = local_file_path_str[1:]
local_file_to_copy = Path(local_file_path_str)
if local_file_to_copy.exists() and local_file_to_copy.is_file():
if not ext:
ext = (
local_file_to_copy.suffix[1:].lower() or "dat"
) # Get ext from local file if not from URL
unique_filename = f"media_{uuid4().hex}.{ext}"
absolute_filepath_dest = os.path.join(
media_absolute_save_dir, unique_filename
)
markdown_path = os.path.join(
markdown_relative_media_folder, unique_filename
)
shutil.copy(str(local_file_to_copy), absolute_filepath_dest)
logger.info(
f"Local media file '{local_file_to_copy}' copied to '{absolute_filepath_dest}'"
)
return (
markdown_path,
media_url,
) # Return original media_url for consistency
else:
logger.warning(
f"Local file '{local_file_to_copy}' referenced by '{actual_media_url_str}' not found or not a file."
)
return None, media_url
elif parsed_actual_url.scheme in ["http", "https"]:
response = requests.get(
actual_media_url_str,
headers={"User-Agent": USER_AGENT},
timeout=self.MEDIA_DOWNLOAD_TIMEOUT,
stream=True,
)
response.raise_for_status()
content_type = response.headers.get("Content-Type", "").lower()
if not ext: # Try to get extension from Content-Type if not from URL
if any(
content_type.startswith(prefix)
for prefix in media_type_prefixes
):
type_part = content_type.split(";")[0]
candidate_ext_ct = type_part.split("/")[-1]
if candidate_ext_ct == "svg+xml":
ext = "svg"
elif candidate_ext_ct == "jpeg":
ext = "jpg"
elif candidate_ext_ct in [
"png",
"gif",
"webp",
"bmp",
"tiff",
"mp4",
"webm",
"ogg",
"mov",
"avi",
"mkv",
"mp3",
"wav",
"aac",
]:
ext = candidate_ext_ct
ext = ext if ext else "dat" # Final fallback extension
unique_filename = f"media_{uuid4().hex}.{ext}"
absolute_filepath = os.path.join(
media_absolute_save_dir, unique_filename
)
markdown_path = os.path.join(
markdown_relative_media_folder, unique_filename
)
with open(absolute_filepath, "wb") as f:
for chunk in response.iter_content(chunk_size=81920):
f.write(chunk)
logger.info(
f"HTTP/S media {actual_media_url_str} downloaded to {absolute_filepath}"
)
return (
markdown_path,
actual_media_url_str,
) # Return resolved URL for HTTP/S
else:
logger.warning(
f"Skipping download for unsupported scheme: {actual_media_url_str}"
)
return None, actual_media_url_str
except requests.exceptions.Timeout:
logger.error(
f"Timeout processing media {original_media_url_for_error_logger}"
)
except requests.exceptions.HTTPError as e:
logger.error(
f"HTTP error {e.response.status_code} processing media {original_media_url_for_error_logger}: {e.response.reason}"
)
except requests.exceptions.RequestException as e:
logger.error(
f"RequestException processing media {original_media_url_for_error_logger}: {e}"
)
except IOError as e:
logger.error(
f"IOError processing media {original_media_url_for_error_logger}: {e}"
)
except Exception as e:
logger.error(
f"Unexpected error processing media {original_media_url_for_error_logger}: {e}"
)
return None, original_media_url_for_error_logger
def _process_images_in_html(
self, html_content, base_url_for_relative, markdown_relative_image_folder
):
logger.debug(
f"Starting image processing. MD relative image folder: {markdown_relative_image_folder}"
)
soup = BeautifulSoup(html_content, "html.parser")
for img_tag in soup.find_all("img"):
original_src = img_tag.get("src")
alt_text = img_tag.get("alt", "").strip()
if not original_src:
img_tag.decompose()
continue
original_src = original_src.strip()
if not original_src:
img_tag.decompose()
continue
if original_src.startswith("cid:"):
cid = original_src[4:]
if hasattr(self, "mhtml_resources") and cid in self.mhtml_resources:
media_data, resource_filename_ext = self.mhtml_resources[cid]
ext_from_mhtml = "png"
if "." in resource_filename_ext:
candidate_ext = resource_filename_ext.split(".")[-1].lower()
if len(candidate_ext) <= 5 and candidate_ext.isalnum():
ext_from_mhtml = candidate_ext
unique_filename = f"image_{uuid4().hex}.{ext_from_mhtml}"
absolute_filepath = os.path.join(
self.current_image_absolute_path, unique_filename
)
markdown_path = os.path.join(
markdown_relative_image_folder, unique_filename
)
try:
with open(absolute_filepath, "wb") as f:
f.write(media_data)
img_tag["src"] = markdown_path
if not alt_text:
alt_text = f"Embedded image {unique_filename}"
img_tag["alt"] = alt_text
except IOError as e:
img_tag.decompose()
else:
img_tag.decompose()
continue
markdown_image_path, _ = self._download_media_file(
original_src,
base_url_for_relative,
self.current_image_absolute_path,
markdown_relative_image_folder,
media_type_prefixes=("image/",),
)
if markdown_image_path:
img_tag["src"] = markdown_image_path
if not alt_text:
alt_text = (
f"Downloaded image {os.path.basename(markdown_image_path)}"
)
img_tag["alt"] = alt_text
else:
img_tag.decompose()
logger.debug("Image processing finished.")
return str(soup)
def _process_videos_in_html(
self,
html_content,
base_url_for_relative,
markdown_relative_video_folder,
markdown_relative_image_folder_for_poster,
):
logger.debug(
f"Starting video processing. MD video folder: {markdown_relative_video_folder}, MD poster folder: {markdown_relative_image_folder_for_poster}"
)
soup = BeautifulSoup(html_content, "html.parser")
for video_tag in soup.find_all("video"):
original_poster_src = video_tag.get("poster")
if original_poster_src:
original_poster_src = original_poster_src.strip()
if original_poster_src:
logger.info(f"Processing poster for video: {original_poster_src}")
if (
self.current_image_absolute_path
): # Ensure image path is set for saving posters
poster_md_path, _ = self._download_media_file(
original_poster_src,
base_url_for_relative,
self.current_image_absolute_path, # Save posters in the image asset directory
markdown_relative_image_folder_for_poster, # Use the image folder's relative path for MD link
media_type_prefixes=("image/",),
)
if poster_md_path:
video_tag["poster"] = poster_md_path
else:
if "poster" in video_tag.attrs:
del video_tag["poster"]
else:
logger.warning(
f"Cannot process poster {original_poster_src} as image asset path is not initialized."
)
source_tags = video_tag.find_all("source")
processed_source_successfully = False
if source_tags:
for source_tag in source_tags:
original_src = source_tag.get("src")
if original_src:
original_src = original_src.strip()
if not original_src:
continue
if original_src.startswith("cid:"):
cid = original_src[4:]
if (
hasattr(self, "mhtml_resources")
and cid in self.mhtml_resources
):
media_data, resource_filename_ext = (
self.mhtml_resources[cid]
)
ext_from_mhtml = "mp4"
if "." in resource_filename_ext:
candidate_ext = resource_filename_ext.split(".")[
-1
].lower()
if (
len(candidate_ext) <= 5
and candidate_ext.isalnum()
):
ext_from_mhtml = candidate_ext
unique_filename = (
f"video_{uuid4().hex}.{ext_from_mhtml}"
)
absolute_filepath = os.path.join(
self.current_video_absolute_path, unique_filename
)
markdown_path = os.path.join(
markdown_relative_video_folder, unique_filename
)
try:
with open(absolute_filepath, "wb") as f:
f.write(media_data)
source_tag["src"] = markdown_path
processed_source_successfully = True
except IOError as e:
source_tag.decompose()
else:
source_tag.decompose()
continue
markdown_video_path, _ = self._download_media_file(
original_src,
base_url_for_relative,
self.current_video_absolute_path,
markdown_relative_video_folder,
media_type_prefixes=("video/", "application/octet-stream"),
)
if markdown_video_path:
source_tag["src"] = markdown_video_path
processed_source_successfully = True
else:
source_tag.decompose()
original_video_src_attr = video_tag.get("src")
if original_video_src_attr and not processed_source_successfully:
original_video_src_attr = original_video_src_attr.strip()
if original_video_src_attr:
if original_video_src_attr.startswith("cid:"):
cid = original_video_src_attr[4:]
if (
hasattr(self, "mhtml_resources")
and cid in self.mhtml_resources
):
media_data, resource_filename_ext = self.mhtml_resources[
cid
]
ext_from_mhtml = "mp4"
if "." in resource_filename_ext:
candidate_ext = resource_filename_ext.split(".")[
-1
].lower()
if len(candidate_ext) <= 5 and candidate_ext.isalnum():
ext_from_mhtml = candidate_ext
unique_filename = f"video_{uuid4().hex}.{ext_from_mhtml}"
absolute_filepath = os.path.join(
self.current_video_absolute_path, unique_filename
)
markdown_path = os.path.join(
markdown_relative_video_folder, unique_filename
)
try:
with open(absolute_filepath, "wb") as f:
f.write(media_data)
video_tag["src"] = markdown_path
processed_source_successfully = True
except IOError as e:
if "src" in video_tag.attrs:
del video_tag["src"]
else:
if "src" in video_tag.attrs:
del video_tag["src"]
else:
markdown_video_path, _ = self._download_media_file(
original_video_src_attr,
base_url_for_relative,
self.current_video_absolute_path,
markdown_relative_video_folder,
media_type_prefixes=("video/", "application/octet-stream"),
)
if markdown_video_path:
video_tag["src"] = markdown_video_path
processed_source_successfully = True
else:
if "src" in video_tag.attrs:
del video_tag["src"]
# If no video source was successfully processed, remove the entire video tag.
if not processed_source_successfully:
logger.warning(
f"Decomposing video tag as no downloadable sources were found."
)
video_tag.decompose()
logger.debug("Video processing finished.")
return str(soup)
def _cleanup_markdown(self, markdown_text):
logger.debug("Starting Markdown cleanup.")
markdown_text = re.sub(r"\[\s*\]\(\s*\)", "", markdown_text)
markdown_text = re.sub(r"!\[(.*?)\]\s+\((.*?)\)", r"", markdown_text)
markdown_text = re.sub(r"\[(.*?)\]\s+\((.*?)\)", r"[\1](\2)", markdown_text)
markdown_text = re.sub(r"\n([*-+])(\S)", r"\n\1 \2", markdown_text)
markdown_text = re.sub(r"\n(\d+\.)(\S)", r"\n\1 \2", markdown_text)
markdown_text = re.sub(r"!\[\s*\]\((.*?)\)", r"", markdown_text)
markdown_text = re.sub(r"\n{3,}", "\n\n", markdown_text)
lines = markdown_text.splitlines()
stripped_lines = [line.strip() for line in lines]
markdown_text = "\n".join(stripped_lines)
logger.debug("Markdown cleanup finished.")
return markdown_text.strip()
def convert(self, source, output_filename_stem=None):
html_content = None
self.base_url = None
self.mhtml_resources = {}
self.source_html_filepath = None
if not output_filename_stem:
stem = os.path.splitext(os.path.basename(source))[0]
output_filename_stem = (
stem if stem else f"file_conversion_{uuid4().hex[:8]}"
)
logger.info(
f"Starting conversion. Source: {source}, Type: html_file, Output stem: {output_filename_stem}"
)
self.source_html_filepath = os.path.abspath(
source
) # Store absolute path of source HTML
try:
with open(
self.source_html_filepath, "r", encoding="utf-8", errors="replace"
) as f:
html_content = f.read()
self.base_url = (
Path(self.source_html_filepath).parent.as_uri() + "/"
) # file:///path/to/containing_directory/
except FileNotFoundError:
logger.error(f"HTML file not found: {source}")
return None
except IOError as e:
logger.error(f"Could not read HTML file {source}: {e}")
return None
except Exception as e:
logger.error(f"Unexpected error reading HTML file {source}: {e}")
return None
if not html_content:
logger.error(f"No HTML content to process from {source}.")
return None
md_img_rel_folder = f"{output_filename_stem}"
self.current_image_absolute_path = os.path.join(
self.output_dir, md_img_rel_folder
)
md_vid_rel_folder = f"{output_filename_stem}"
self.current_video_absolute_path = os.path.join(
self.output_dir, md_vid_rel_folder
)
try:
os.makedirs(self.current_image_absolute_path, exist_ok=True)
os.makedirs(self.current_video_absolute_path, exist_ok=True)
except OSError as e:
logger.error(f"Could not create asset directories: {e}")
return None
logger.info("Cleaning HTML...")
cleaned_html = self._clean_html(html_content)
logger.info("Processing and downloading images...")
html_after_images = self._process_images_in_html(
cleaned_html, self.base_url, md_img_rel_folder
)
logger.info("Processing and downloading videos...")
# Pass md_img_rel_folder for posters
html_after_videos = self._process_videos_in_html(
html_after_images, self.base_url, md_vid_rel_folder, md_img_rel_folder
)
logger.info("Converting HTML to Markdown...")
try:
markdown_output = md(
html_after_videos,
heading_style="atx",
bullets="-",
default_title=False,
strip=[],
)
except Exception as e:
logger.error(f"Error during Markdown conversion for {source}: {e}.")
try:
markdown_output = md(
html_after_images,
heading_style="atx",
bullets="-",
default_title=False,
strip=[],
) # Fallback
except Exception as e2:
debug_html_path = os.path.join(
self.output_dir, f"{output_filename_stem}_debug_processed.html"
)
try:
with open(debug_html_path, "w", encoding="utf-8") as f_debug:
f_debug.write(html_after_videos)
except IOError:
pass
return None
logger.info("Cleaning Markdown...")
final_markdown = self._cleanup_markdown(markdown_output)
output_md_path = os.path.join(self.output_dir, f"{output_filename_stem}.md")
try:
with open(output_md_path, "w", encoding="utf-8") as f:
f.write(final_markdown)
logger.info(f"Markdown file saved to {output_md_path}")
for asset_path in [
self.current_image_absolute_path,
self.current_video_absolute_path,
]:
if os.path.exists(asset_path) and not os.listdir(asset_path):
try:
os.rmdir(asset_path)
except OSError as e_rmdir:
logger.warning(
f"Could not remove empty asset folder {asset_path}: {e_rmdir}"
)
return output_md_path
except IOError as e:
logger.error(f"Could not write Markdown file {output_md_path}: {e}")
return None
except Exception as e:
logger.error(
f"Unexpected error writing Markdown file {output_md_path}: {e}"
)
return None
def html_handler(input_file_name, doc_id, converter):
if os.path.exists(input_file_name):
md_path_local_html = converter.convert(
input_file_name,
output_filename_stem=doc_id,
)
if not md_path_local_html:
logger.warning(f"Failed to convert local HTML: {input_file_name}")
else:
logger.debug(f"\nLocal HTML test file not found at '{input_file_name}'.")
def handler(cache_dir, file_or_url: str):
output_dir = f"{cache_dir}"
converter = HTML2MarkdownConverter(
output_dir=output_dir,
media_download_timeout=60,
)
doc_id = str(uuid4())
if file_or_url.endswith((".html", ".htm")):
html_handler(file_or_url, doc_id, converter)
else:
logger.error(
f"Unsupported file type: {file_or_url}. Only .html and .htm files are supported."
)
return None, None, None
return f"{cache_dir}/{doc_id}.md", f"{cache_dir}/{doc_id}", doc_id
if __name__ == "__main__":
output_dir = "/Users/tju/Library/Caches/bisheng"
input_file = "/Users/tju/Resources/docs/html/f.html"
output_md, asset_dir, doc_id = handler(output_dir, input_file)
if output_md and os.path.exists(output_md):
print(f"Markdown saved to: {output_md}")
else:
print("Conversion failed.")
================================================
FILE: src/backend/bisheng/api/services/md_from_pdf.py
================================================
import os
import threading
from uuid import uuid4
import fitz
from loguru import logger
pymu_lock = threading.Lock()
def convert_pdf_to_md(output_dir, pdf_path, doc_id):
"""
will specify the PDF Convert file to Markdown files and keep the contents in their original order.
This function extracts PDF Text, tables and pictures in and based on what they are on the page
Vertical position to sort and then consolidate to one Markdown in the file.
The image is saved as a separate file in the specified output directory.
Args:
pdf_path (str): Entered PDF FilePath
output_dir (str): SAVING Markdown A directory of files and pictures.
"""
# Make sure the output directory exists
if not os.path.exists(output_dir):
os.makedirs(output_dir)
md_filename = f"{doc_id}.md"
md_filepath = os.path.join(output_dir, md_filename)
img_dir = os.path.join(output_dir, f"images")
if not os.path.exists(img_dir):
os.makedirs(img_dir)
doc = None
try:
doc = fitz.open(pdf_path)
except Exception as e:
raise Exception('The file is damaged.')
try:
md_content = ""
image_counter = 1
for page_num in range(len(doc)):
with pymu_lock:
page = doc.load_page(page_num)
page_elements = []
tables = page.find_tables()
if tables.tables:
for tab in tables.tables:
if not tab.to_pandas().empty:
md_table = tab.to_pandas().to_markdown(index=False)
table_bbox = fitz.Rect(tab.bbox)
page_elements.append(
{
"type": "table",
"bbox": table_bbox,
"content": md_table,
}
)
image_info_list = page.get_image_info(xrefs=True)
if image_info_list:
for img_info in image_info_list:
xref = img_info["xref"]
if xref == 0:
continue
base_image = doc.extract_image(xref)
if not base_image:
continue
image_bytes = base_image["image"]
image_ext = base_image["ext"]
img_filename = f"image_{page_num + 1}_{image_counter}.{image_ext}"
img_path = os.path.join(img_dir, img_filename)
with open(img_path, "wb") as img_file:
img_file.write(image_bytes)
md_image = f""
image_bbox = fitz.Rect(img_info["bbox"])
page_elements.append(
{"type": "image", "bbox": image_bbox, "content": md_image}
)
image_counter += 1
table_bboxes = (
[fitz.Rect(tab.bbox) for tab in tables.tables]
if tables.tables
else []
)
text_blocks = page.get_text("blocks")
for b in text_blocks:
block_rect = fitz.Rect(b[:4])
block_text = b[4].strip()
is_in_table = False
for table_bbox in table_bboxes:
if block_rect.intersects(table_bbox):
is_in_table = True
break
if block_text and not is_in_table:
page_elements.append(
{"type": "text", "bbox": block_rect, "content": block_text}
)
page_elements.sort(key=lambda el: el["bbox"].y0)
for elem in page_elements:
md_content += elem["content"] + "\n\n"
with open(md_filepath, "w", encoding="utf-8") as md_file:
md_file.write(md_content)
except Exception as e:
logger.exception(f"Error processing pdf: {e}")
raise Exception(f"Document parsing failed: {str(e)[-100:]}") # Capture last100characters to avoid overly long error messages
finally:
with pymu_lock:
if doc:
doc.close()
def is_pdf_damaged(pdf_path: str) -> bool:
"""
Others PDF Whether the file is corrupt.
Args:
pdf_path (str): PDF Path of file
Returns:
bool: If the file is damaged, go back True; otherwise go back to False。
"""
try:
doc = fitz.open(pdf_path)
doc.close()
return False
except Exception as e:
logger.error(f"PDF file is damaged: {e}")
return True
def handler(cache_dir, file_or_url: str):
doc_id = uuid4()
ouput_dir = f"{cache_dir}/{doc_id}"
convert_pdf_to_md(ouput_dir, file_or_url, doc_id)
return f"{ouput_dir}/{doc_id}.md", f"{ouput_dir}/images", doc_id
def exec_thread_safe():
pdf_path = "/Users/tju/Documents/Resources/pdf/bisheng/chen4.pdf"
output_directory = "/Users/tju/Desktop/output"
md_file, local_image, doc_id = handler(output_directory, pdf_path)
if __name__ == "__main__":
import multiprocessing
processes = []
for _ in range(10):
process = multiprocessing.Process(target=exec_thread_safe)
processes.append(process)
process.start()
for process in processes:
process.join()
threads = []
for i in range(4):
thread = threading.Thread(target=exec_thread_safe, name=f"Thread-{i}")
threads.append(thread)
thread.start()
for thread in threads:
thread.join()
================================================
FILE: src/backend/bisheng/api/services/md_from_pptx.py
================================================
from bisheng.pptx2md import convert, ConversionConfig
from pathlib import Path
from uuid import uuid4
def parser_pptx2md(
pptx_file: str,
md_file: str,
image_dir: str = None,
):
"""
Convert a PowerPoint file to Markdown format.
Args:
pptx_file (str): Path to the PowerPoint file.
md_file (str): Path to the output Markdown file.
image_dir (str, optional): Directory to save images. Defaults to None
"""
# Basic usage
convert(
ConversionConfig(
pptx_path=Path(pptx_file),
output_path=Path(md_file),
image_dir=Path(image_dir),
disable_notes=True,
)
)
def handler(
cache_dir,
file_name,
):
doc_id = str(uuid4())
md_file_name = f"{cache_dir}/{doc_id}.md"
image_dir = f"{cache_dir}/{doc_id}"
parser_pptx2md(
pptx_file=file_name,
md_file=md_file_name,
image_dir=image_dir,
)
# Image,Can be used asynchronously
# GantimdPicture path in file
return md_file_name, image_dir, doc_id
if __name__ == "__main__":
pptx_file = "/Users/tju/Resources/docs/ppt/you-lian.pptx"
cache_dir = "/Users/tju/Desktop"
md_file_name, image_dir, doc_id = handler(cache_dir, pptx_file)
print(f"Markdown file: {md_file_name}")
print(f"Image directory: {image_dir}")
print(f"Document ID: {doc_id}")
================================================
FILE: src/backend/bisheng/api/services/md_post_processing.py
================================================
import html
import os
import re
from bs4 import BeautifulSoup
def post_processing(file_path, retain_images=True):
"""
(Final full version)
Comprehensively combine oneMarkdownin the fileHTMLConvert tags to standardMarkdownFormat and process the picture correctly according to the parameters.
"""
try:
with open(file_path, "r", encoding="utf-8") as file:
content = file.read()
# Step 1: Image Handling (Top Priority Execution)
if not retain_images:
# If the image is not preserved, delete the image in all formats globally before any conversion
content = re.sub(r"]*>", "", content, flags=re.IGNORECASE)
content = re.sub(
r"\[!\[.*?\]\(.*?\)\]\(.*?\)", "", content, flags=re.DOTALL
)
content = re.sub(r"!\[.*?\]\(.*?\)", "", content, flags=re.DOTALL)
else:
# Convert only if image is preservedHTMLright of privacyimgTaggedMarkdownFormat
# Use a helper function to extractsrcAndalt
def _img_to_md(match):
img_tag = match.group(0)
src_match = re.search(r'src="([^"]+)"', img_tag, re.IGNORECASE)
alt_match = re.search(r'alt="([^"]*)"', img_tag, re.IGNORECASE)
src = src_match.group(1) if src_match else ""
alt = alt_match.group(1) if alt_match else ""
return f""
content = re.sub(r"]*>", _img_to_md, content, flags=re.IGNORECASE)
# Step 2: ComplexityHTMLBlock Level Element Conversion (UseBeautifulSoupSupport)
def _table_to_md(match):
soup = BeautifulSoup(match.group(0), "html.parser")
headers = [
th.get_text(strip=True).replace("|", r"\|")
for th in soup.find_all("th")
]
if not headers: # If no evidence of microbial
label
elif has_paragraph_tag:
para_count = 0
for child in item.children:
if hasattr(child, 'name') and child.name == 'p':
if para_count == 0:
# first
Use list style
self.add_paragraph(child, p_style=MDX_STYLE.LIST_NUMBER) \
.style.paragraph_format.space_after = Pt(1)
else:
# Additional
Use continuation style
self.add_paragraph(child, p_style=MDX_STYLE.LIST_CONTINUE) \
.style.paragraph_format.space_after = Pt(1)
para_count += 1
# Working with Nested Ordered Lists
if hasattr(item, "ol") and item.ol is not None:
sub_num: int = 1 # Sub-Serial Number
for item2 in item.ol.children:
if item2.name != 'li':
continue
self.add_paragraph(item2, prefix="(%d). " % sub_num, p_style=MDX_STYLE.LIST_CONTINUE) \
.style.paragraph_format.first_line_indent = 0 # TODO Number List Style
sub_num += 1
# Working with Nested Unordered Lists
if hasattr(item, "ul") and item.ul is not None:
for item2 in item.ul.children:
if item2.name != 'li':
continue
self.add_paragraph(item2, prefix="• ", p_style=MDX_STYLE.LIST_CONTINUE) \
.style.paragraph_format.space_after = Pt(1)
def add_bullet_list(self, bullet_list):
# It may be. list
text = str(bullet_list.contents[1].string).strip()
if text.startswith("[ ]") or text.startswith("[x]"):
self.add_todo_list(bullet_list)
return
for item in bullet_list.children:
if item.name != 'li': # Skip non- li label
continue
# Get direct text content(Exclude subtags)
direct_text = ''.join([str(s) for s in item.find_all(text=True, recursive=False)]).strip()
# Check if there are paragraph tags
has_paragraph_tag = item.find('p') is not None
# If there is direct text,Add your paragraph
if direct_text:
self.add_paragraph(item, p_style=MDX_STYLE.LIST_BULLET) \
.style.paragraph_format.space_after = Pt(1)
# If there is no direct text but there is
label,Medical Treatment:
label
elif has_paragraph_tag:
para_count = 0
for child in item.children:
if hasattr(child, 'name') and child.name == 'p':
if para_count == 0:
# first
Use list style
self.add_paragraph(child, p_style=MDX_STYLE.LIST_BULLET) \
.style.paragraph_format.space_after = Pt(1)
else:
# Additional
Use continuation style
self.add_paragraph(child, p_style=MDX_STYLE.LIST_CONTINUE) \
.style.paragraph_format.space_after = Pt(1)
para_count += 1
# Medical Treatment: ul Above listings
if hasattr(item, "ul") and item.ul is not None:
for item2 in item.ul.children:
if item2.string == "\n" or (not item2.string):
continue
self.add_paragraph(item2, prefix="• ", p_style=MDX_STYLE.LIST_CONTINUE) \
.style.paragraph_format.space_after = Pt(1)
# Medical Treatment: ol Above listings
if hasattr(item, "ol") and item.ol is not None:
# read out start Property,If not present, defaults to1
sub_num = int(item.ol.get('start', 1))
for item2 in item.ol.children:
if item2.name != 'li':
continue
self.add_paragraph(item2, prefix=f"{sub_num}. ", p_style=MDX_STYLE.LIST_CONTINUE) \
.style.paragraph_format.space_after = Pt(1)
sub_num += 1
# falseTODO list
def add_todo_list(self, todo_list):
# list_para.style.font.name = "Consolas"
for item in todo_list.children:
if item.string == "\n" or (not item.string):
continue
text: str = item.string
list_para = self.document.add_paragraph(style=MDX_STYLE.PLAIN_LIST)
if text.startswith("[x]"):
list_para.add_run("[ √ ]").font.name = "Consolas"
list_para.add_run(text.replace("[x]", " ", 1))
elif text.startswith("[ ]"):
list_para.add_run("[ ]").font.name = "Consolas"
list_para.add_run(text.replace("[ ]", " ", 1))
# Split Lines
def add_split_line(self):
p = self.document.add_paragraph()
p.paragraph_format.space_before = Pt(6)
p.paragraph_format.space_after = Pt(6)
border_elm = parse_xml(
r''.format(nsdecls('w')))
p._p.get_or_add_pPr().append(border_elm)
# Hyperlinks
def add_link(self, p: Paragraph, text: str, href: str):
self.debug("[link]:", text, "[href]:", href)
add_hyperlink(p, href, text)
# run = p.add_run(text)
def add_paragraph(self, children, p_style: str = None, prefix: str = ""):
"""
children: list|str
An element (including an image) within a paragraph. Divided according to whether there is a style, forming a list.
There are style text such as bold, italics, images, etc.
As`I am plain _while_ he is **bold**`will be converted to:
["I am plain", "while", "he is", "bold"]
"""
p = self.document.add_paragraph(prefix, style=p_style)
if type(children) == str:
p.add_run(children)
return p
for elem in children.contents: # Traverse all elements in a paragraph
if elem.name == "a":
self.add_link(p, elem.string, elem["href"])
elif elem.name == "img":
self.add_picture(elem, parent_paragraph=p) # Pass the current paragraph
elif elem.name is not None: # Substring with character style
self.add_run(p, elem.string, elem.name)
elif not elem.string == "\n": # Substring without character style
self.add_run(p, elem)
return p
# from docx.enum.style import WD_STYLE
def add_blockquote(self, children):
# TODO Place Blockquote on1x1In the table, optimize the display effect of the reference block
# Set left indent, up and down spacing
table: Table = self.document.add_table(0, 1)
row_cells = table.add_row().cells
# Supports multiple paragraphs
para_count = 0
for child in children.contents:
# Skip line breaks
if isinstance(child, str) and child.strip() == "":
continue
# Working with paragraph tags
if hasattr(child, 'name') and child.name == 'p':
if para_count > 0:
# Add a new passage
row_cells[0].add_paragraph()
p = row_cells[0].paragraphs[para_count]
for elem in child.contents: # Traverse all elements in a paragraph
if elem.name == "a":
self.add_link(p, elem.string, elem["href"])
elif elem.name == "img":
self.add_picture(elem, parent_paragraph=p) # Pass the current paragraph
elif elem.name is not None: # Substring with character style
self.add_run(p, elem.string, elem.name)
elif elem.string and elem.string != "\n": # Substring without character style
self.add_run(p, elem.string)
para_count += 1
# Working with Direct Text Nodes
elif isinstance(child, str) and child.strip():
if para_count == 0:
p = row_cells[0].paragraphs[0]
para_count = 1
else:
p = row_cells[0].add_paragraph()
para_count += 1
p.add_run(child.strip())
shading_elm_1 = parse_xml(r''.format(nsdecls('w')))
table.rows[0].cells[0]._tc.get_or_add_tcPr().append(shading_elm_1)
def html2docx(self, html_str: str):
# OpenHTML
soup = BeautifulSoup(html_str, 'html.parser')
# Find it securely body label
body_tag = soup.find('body')
if body_tag is None:
# If no evidence of microbial body,Use entire soup
body_tag = soup
# Title
title_text = ""
# parse the labels one by one and writewordII
for root in body_tag.children:
# Skip plain text nodes and line breaks
if isinstance(root, str):
if root.strip():
# Working with Direct Text Nodes
self.document.add_paragraph(root.strip(), style=MDX_STYLE.PLAIN_TEXT)
continue
# debug("<%s>" % root.name)
if root.name == "p": # Normal paragraph
self.add_paragraph(root, p_style=MDX_STYLE.PLAIN_TEXT)
elif root.name == "blockquote": # Blockquote
self.add_blockquote(root)
elif root.name == "ol": # Numbered
self.add_number_list(root)
elif root.name == "ul": # Unordered List OR List
self.add_bullet_list(root)
elif root.name == "table": # Table Filter
self.add_table(root)
elif root.name == "hr":
self.add_split_line()
elif root.name == "pre":
self.add_code_block(root)
elif root.name in ["h1", "h2", "h3", "h4", "h5", "h6"]:
if title_text == "":
title_text = root.string or ""
self.add_heading(root.string or "", root.name)
docx_bytes = io.BytesIO()
self.document.save(docx_bytes)
docx_bytes.seek(0)
return docx_bytes.getbuffer(), title_text
================================================
FILE: src/backend/bisheng/common/utils/markdown_cmpnt/md_to_docx/provider/simple_style.py
================================================
import docx
from docx.enum.style import WD_STYLE_TYPE
# Chinese Font Size Conversion pt THE METHOD
def _zihao_to_pt(chn_name: str):
chn_name = chn_name.strip()
pt_mapping = {
"No. 1": 42,
"Little Primary": 36,
"No. 1": 26,
"Little Yi": 24,
"No. 2": 22,
"Second": 18,
"Number three": 16,
"Little San": 15,
"No. 4": 14,
"Shikama": 12,
"No. 5": 10.5,
"Xiao Wu": 9,
"No. 6": 7.5,
"Little Six": 6.5,
"No. 7": 5.5,
"No. 8": 5,
}
if pt_mapping.get(chn_name):
return pt_mapping[chn_name]
else:
print("[YAML ERROR]:", chn_name,
"Not a canonical title.(In the Chinese context, the largest font size is'No. 1', the minimum is'No. 8')。")
class SimpleStyle:
style_name: str
base_style_name: str
style_type: WD_STYLE_TYPE = WD_STYLE_TYPE.PARAGRAPH
font_default: str = "Times New Roman"
font_east_asia: str = "Song Ti"
font_size: float = 14
font_color: str = "000000"
font_bold: bool = False
font_italic: bool = False
font_underline: bool = False
font_strike: bool = False
first_line_indent: int = 0
line_spacing: int = 1
space_before: int = 0
space_after: int = 0
def __init__(self,
style_name: str, # Style name
base_style_name: str, # Styles based on
conf: dict, # Specific style data fromyamlDeserialized
style_type: docx.enum.style = WD_STYLE_TYPE.PARAGRAPH # Is it a paragraph or a list or some other type
):
self.style_name = style_name
self.base_style_name = base_style_name
self.style_type = style_type
try:
self.font_default = conf["font"]["default"]
self.font_east_asia = conf["font"]["east-asia"]
if str(conf["font"]["size"]).isdigit():
self.font_size = conf["font"]["size"]
else: # If the font size is given in Chinese, convert it
self.font_size = _zihao_to_pt(conf["font"]["size"])
except KeyError:
print("[YAML ERROR]:", style_name,
"| Error occurred in setting font style. Set to:",
self.font_default, self.font_east_asia, str(self.font_size) + "pt")
# Check when color is specified, do not specify default black
if conf["font"].get("color") is not None:
try:
if type(conf["font"]["color"]) != str \
or len(conf["font"]["color"]) != 6:
raise TypeError
# Try to convert to16binary number and whether it matchesRGBsize
hex_num = int(conf["font"]["color"], 16)
if 0 <= hex_num <= 0xFFFFFF:
self.font_color = str(conf["font"]["color"])
else:
raise ValueError
except ValueError:
print("[YAML ERROR]:", style_name,
"| Value of color isn't a hex or out of [000000, FFFFFF].",
"Default to black(000000).")
except TypeError:
print("[YAML ERROR]:", style_name, "| Value of color must be string with 6 characters.",
"Default to black(000000).")
# Bold, Italic, Underline, Strikeout
if conf.get("font").get("extra"):
# print(conf["font"]["extra"])
self.font_bold = "bold" in (conf["font"]["extra"])
self.font_italic = "italic" in list(conf["font"]["extra"])
self.font_underline = "underline" in list(conf["font"]["extra"])
self.font_strike = "strike" in list(conf["font"]["extra"])
if conf.get("first-line-indent"):
self.first_line_indent = conf["first-line-indent"]
if conf.get("line-spacing"):
self.line_spacing = conf["line-spacing"]
if conf.get("space"):
if conf.get("space").get("before"):
self.space_before = conf["space"]["before"]
if conf.get("space").get("after"):
self.space_after = conf["space"]["after"]
def __str__(self) -> str:
return "".join(str(i) for i in (
self.style_name, " ",
self.font_default, " ",
self.font_east_asia, " ",
self.font_size, " ",
self.font_color, " ",
"space:(", self.space_before, " ", self.space_after, ") ",
"bold-", self.font_bold, " ",
"italic-", self.font_italic, " ",
"underline-", self.font_underline, " ",
"strike-", self.font_strike
))
================================================
FILE: src/backend/bisheng/common/utils/markdown_cmpnt/md_to_docx/provider/style_manager.py
================================================
import docx
from docx import Document
from docx.oxml.ns import qn
from docx.shared import Inches, RGBColor, Pt
from docx.styles.style import _ParagraphStyle, BaseStyle
from docx.enum.style import WD_STYLE_TYPE
from docx.styles.styles import Styles
from bisheng.common.utils.markdown_cmpnt.md_to_docx.provider.simple_style import SimpleStyle
from bisheng.common.utils.markdown_cmpnt.md_to_docx.utils.style_enum import MDX_STYLE
class StyleManager:
def __init__(self, doc: Document, yaml_conf: dict):
self.styles: Styles = doc.styles
self.style_conf = yaml_conf
def init_styles(self):
# Pengaturanheading 1~4
for i in range(1, 5):
s = SimpleStyle("Heading%d" % i, "Heading %d" % i,
self.style_conf["h%d" % i])
self.set_style(s)
# TODO What other styling settings are there?
# CaptionsCaptionTable Style(?)
# s = SimpleStyle("Normal", "Normal", self.style_conf["normal"])
s = SimpleStyle(MDX_STYLE.PLAIN_TEXT, "Normal", self.style_conf["normal"])
self.set_style(s)
# General style settings
def set_style(self, _style: SimpleStyle):
new_style: _ParagraphStyle
if _style.style_name not in self.styles:
new_style: _ParagraphStyle = self.styles.add_style(_style.style_name, _style.style_type)
new_style.base_style = self.styles[_style.base_style_name]
else:
new_style = self.styles[_style.style_name]
new_style.quick_style = True
# ##### Font Related #####
# Set font, color, size
new_style.font.name = _style.font_default # Set onlynameIs Set Western Font
new_style.font.size = Pt(_style.font_size)
new_style._element.rPr.rFonts.set(qn('w:eastAsia'), _style.font_east_asia) # To set additional Chinese fonts
new_style.font.color.rgb = RGBColor.from_string(_style.font_color)
# Bold, Italic, Underline, Strikeout
new_style.font.bold = _style.font_bold
new_style.font.italic = _style.font_italic
new_style.font.underline = _style.font_underline
new_style.font.strike = _style.font_strike
# ##### Paragraph Related #####
# Set indent, before paragraph/Space after paragraph, line spacing between paragraphs
new_style.paragraph_format.first_line_indent = (Pt(_style.font_size) * int(_style.first_line_indent))
new_style.paragraph_format.space_before = Pt(_style.space_before)
new_style.paragraph_format.space_after = Pt(_style.space_after)
new_style.paragraph_format.line_spacing = _style.line_spacing
# ##### Other _________ #####
# Remove the black dot in the top left corner before the paragraph
new_style.paragraph_format.keep_together = False
# Header Style Retention keep_with_next are True, avoid separate pagination of titles
if not _style.style_name.startswith("Heading"):
new_style.paragraph_format.keep_with_next = False
# Title does not precede pagination
if _style.style_name.startswith("Heading"):
new_style.paragraph_format.page_break_before = False
# Show on shortcut style window
new_style.quick_style = True
return
================================================
FILE: src/backend/bisheng/common/utils/markdown_cmpnt/md_to_docx/utils/__init__.py
================================================
================================================
FILE: src/backend/bisheng/common/utils/markdown_cmpnt/md_to_docx/utils/style_enum.py
================================================
class MDX_STYLE:
PLAIN_TEXT = "Normal"
PLAIN_LIST = "List"
LIST_NUMBER = "List Number"
LIST_CONTINUE = "List Continue"
LIST_BULLET = "List Bullet"
BLOCKQUOTE = "Quote"
HEADING = "Heading"
TABLE = "Table Grid"
CAPTION = "Caption"
================================================
FILE: src/backend/bisheng/common/utils/markdown_cmpnt/md_to_docx/utils/yaml_utils.py
================================================
import yaml
from yaml import FullLoader
def read_style_yaml(path: str) -> dict:
with open(path, "r", encoding="utf-8") as file:
return yaml.load(file, FullLoader)
================================================
FILE: src/backend/bisheng/common/utils/markdown_cmpnt/md_to_pdf.py
================================================
"""
Typora Style Markdown Transfer PDF Converter
Use Playwright (Chromium) will be HTML Render As PDF Professional Markdown Transfer PDF Exporter
Support Typora of style styles and mathematical expressions MathJax rendered
Feature:
- Pass Headless Chromium will be Markdown Convert file or string to stylized PDF
- Contains the default Typora Style CSS Styles
- Supports customization CSS Override
- MathJax Integration for LaTeX Mathematical rendering
- Configurable page formatting and margins
- Robust error handling and resource management
dependent:
pip install playwright markdown
playwright install chromium
Usage Sample:
from to_pdf import MarkdownToPdfConverter
converter = MarkdownToPdfConverter()
converter.convert_file("README.md.md", "output.pdf")
"""
import logging
import os
import tempfile
from pathlib import Path
from typing import Optional, Union
from bisheng.common.utils.markdown_cmpnt.md_to_docx.parser.ext_md_syntax import ExtMdSyntax
logger = logging.getLogger(__name__)
# Dependent Import and Error Handling
try:
import markdown
except ImportError as e:
logger.error("Required dependency is missing: %s", e)
raise ImportError("Please Install 'markdown' Package: pip install markdown") from e
try:
from playwright.sync_api import sync_playwright
PLAYWRIGHT_AVAILABLE = True
except ImportError:
PLAYWRIGHT_AVAILABLE = False
logger.warning("Playwright Not available. Please install: pip install playwright && playwright install chromium")
DEFAULT_CSS = r"""
top: 6px;
font-size: 9px;
color: var(--muted-color);
background: transparent;
}
/* Tables */
table {
width: 100%;
border-collapse: collapse;
margin: 0 0 0.9em 0;
font-size: 11pt;
}
th, td {
border: 1px solid #e6ecf0;
padding: 8px 10px;
text-align: left;
}
th { background: #fafbfc; font-weight: 600; }
/* Images */
img { max-width: 100%; height: auto; display: block; margin: 0.5em 0; }
/* Caption support (if markdown-to-html generator provides figure/figcaption) */
figure { margin: 0 0 0.9em 0; }
figcaption { font-size: 10pt; color: var(--muted-color); text-align: center; margin-top: 0.35em; }
/* Table of contents (toc) */
.toc {
margin: 0 0 1em 0;
padding-left: 0;
list-style: none;
}
.toc li { margin: 0.25em 0; }
.toc a { color: var(--accent-color); }
/* Footnotes */
.footnotes { font-size: 10pt; border-top: 1px dashed var(--hr-color); margin-top: 1em; padding-top: 0.6em; }
.footnotes li { margin: 0.4em 0; }
/* Page break helpers */
.page-break { page-break-after: always; }
.page-break-before { page-break-before: always; }
/* Fixed header/footer for printing — Chromium will include them in the page content when using print() */
.header, .footer {
display: block;
position: fixed;
left: 0; right: 0;
color: var(--muted-color);
font-size: 9pt;
padding: 6px 12mm;
}
.header { top: 0; }
.footer { bottom: 0; }
.footer .page-number::after { content: counter(page); }
/* Print-specific rules */
@media print {
body { background: white; }
/* Hide interactive elements if present */
a[href]:after { content: ""; }
/* Avoid background colors bleeding if printer doesn't support it */
* { -webkit-print-color-adjust: exact; }
}
/* Small-screen fallback (not usually needed for print) */
@media screen and (max-width: 800px) {
#content { padding: 16px; }
}
/* MathJax tweaks */
.mjx-svg-href { vertical-align: middle; }
/* Accessibility tweaks */
:focus { outline: 2px dashed rgba(3,102,214,0.35); outline-offset: 2px; }
/* Utility classes */
.center { text-align: center; }
.right { text-align: right; }
.small { font-size: 10pt; color: var(--muted-color); }
/* End of CSS */
"""
MATHJAX_SNIPPET = r"""
"""
HTML_TEMPLATE = r"""
{mathjax}
{body}
"""
class MarkdownToPdfError(Exception):
"""Markdown Transfer PDF Conversion error custom exception."""
pass
class MarkdownToPdfConverter:
"""
Has Typora The power of style rendering Markdown Transfer PDF Converter
Such provision will Markdown Convert file or string to PDF short circuit exist.
Use Playwright Render and include comprehensive error handling.
"""
SUPPORTED_PAGE_FORMATS = {'A4', 'A3', 'A5', 'Letter', 'Legal', 'Tabloid'}
DEFAULT_MARKDOWN_EXTENSIONS = [ExtMdSyntax(), 'extra', 'codehilite', 'toc', 'tables', 'sane_lists', 'fenced_code']
DEFAULT_TIMEOUT = 60000 # ms
def __init__(self,
default_css: Optional[str] = None,
enable_math: bool = False,
page_format: str = 'A4',
margin_mm: int = 20):
"""
Initializes the converter with default settings.
Args:
default_css: Used instead of built-in Typora Custom Default for Styles CSS
enable_math: enabled MathJax Math expression support
page_format: Default page format (A4, A3, A5, Letter, Legal, Tabloid)
margin_mm: Default margin (mm)
Raises:
MarkdownToPdfError: Automatically close purchase order after Playwright Unavailable or invalid parameters
"""
if not PLAYWRIGHT_AVAILABLE:
raise MarkdownToPdfError(
'Playwright Not installed. Please install: pip install playwright && playwright install chromium'
)
if page_format not in self.SUPPORTED_PAGE_FORMATS:
raise MarkdownToPdfError(
f'Unsupported page format: {page_format}Supported Formats: {self.SUPPORTED_PAGE_FORMATS}'
)
if not isinstance(margin_mm, (int, float)) or margin_mm < 0:
raise MarkdownToPdfError('Margins must be non-negative')
self.default_css = default_css or DEFAULT_CSS
self.enable_math = enable_math
self.page_format = page_format
self.margin_mm = margin_mm
logger.info("MarkdownToPdfConverter Initialized, Format=%s Margin=%dmm",
page_format, margin_mm)
def _validate_input_path(self, file_path: Union[str, Path]) -> Path:
"""Validate input file path and go back Path of research."""
path = Path(file_path)
if not path.exists():
raise MarkdownToPdfError(f'This input file does not exist.: {path}')
if not path.is_file():
raise MarkdownToPdfError(f'Input path is not a file: {path}')
if not path.suffix.lower() in {'.md', '.markdown', '.txt'}:
logger.warning("Input file not available markdown extension: %s", path.suffix)
return path
def _validate_output_path(self, file_path: Union[str, Path]) -> Path:
"""Verify the output file path and make sure the directory exists."""
path = Path(file_path)
# Make sure the parent directory exists
path.parent.mkdir(parents=True, exist_ok=True)
if path.suffix.lower() != '.pdf':
logger.warning("Output file not available .pdf extension: %s", path.suffix)
return path
def _load_css_file(self, css_path: Union[str, Path]) -> str:
"""Load from file CSS Content with error handling."""
try:
css_file = Path(css_path)
if not css_file.exists():
raise MarkdownToPdfError(f'CSS File don\'t exists: {css_file}')
with open(css_file, 'r', encoding='utf-8') as f:
css_content = f.read().strip()
if not css_content:
logger.warning("CSS File is empty: %s", css_file)
return css_content
except (OSError, UnicodeDecodeError) as e:
raise MarkdownToPdfError(f'read out CSS Doc. {css_path} Kalah: {e}') from e
def render_markdown_to_html(self,
markdown_text: str,
custom_css: Optional[str] = None,
enable_math: Optional[bool] = None) -> str:
"""
will be Markdown Convert text to with style and MathJax Supportive HTML。
Args:
markdown_text: To be converted Markdown Contents
custom_css: Optional customizations CSSfor overriding default styles
enable_math: Mathematical support for overriding instance settings
Returns:
Available for PDF Conversion Complete HTML Documentation
Raises:
MarkdownToPdfError: Automatically close purchase order after Markdown Failed to Process
"""
if not isinstance(markdown_text, str):
raise MarkdownToPdfError('Markdown Text must be a string')
if not markdown_text.strip():
logger.warning("SERVICES Markdown Konten kosong")
try:
# Using the extension will Markdown Convert To HTML
html_body = markdown.markdown(
markdown_text,
extensions=self.DEFAULT_MARKDOWN_EXTENSIONS
)
# OK CSS And MathJax Pengaturan
css_content = custom_css or self.default_css
use_math = enable_math if enable_math is not None else self.enable_math
math_snippet = MATHJAX_SNIPPET if use_math else '\n'
# Generate full HTML Documentation
html_document = HTML_TEMPLATE.format(
css=css_content,
mathjax=math_snippet,
body=html_body
)
logger.debug("Success will %d character (s) of Markdown Convert To HTML", len(markdown_text))
return html_document
except Exception as e:
raise MarkdownToPdfError(f'will be Markdown Render As HTML Kalah: {e}') from e
def convert_html_to_pdf(self,
html_content: str,
output_path: Union[str, Path],
page_format: Optional[str] = None,
margin_mm: Optional[int] = None) -> None:
"""
Use Playwright will be HTML Convert content to PDF。
Args:
html_content: To be converted HTML Contents
output_path: PDF SAVE PATH
page_format: Override default page formatting
margin_mm: Override default margin
Raises:
MarkdownToPdfError: Automatically close purchase order after PDF failed to transform
"""
if not isinstance(html_content, str) or not html_content.strip():
raise MarkdownToPdfError('HTML Content cannot be empty')
output_file = self._validate_output_path(output_path)
format_to_use = page_format or self.page_format
margin_to_use = margin_mm if margin_mm is not None else self.margin_mm
if format_to_use not in self.SUPPORTED_PAGE_FORMATS:
raise MarkdownToPdfError(f'Unsupported page format: {format_to_use}')
# Create Temporary HTML Doc.
temp_html_path = None
try:
with tempfile.NamedTemporaryFile(
mode='w',
suffix='.html',
delete=False,
encoding='utf-8'
) as temp_file:
temp_html_path = temp_file.name
temp_file.write(html_content)
logger.debug("Temporary Created HTML Doc.: %s", temp_html_path)
# Use Playwright Convert To PDF
self._render_pdf_with_playwright(temp_html_path, output_file, format_to_use, margin_to_use)
logger.info("Slider Created Successfully. PDF: %s", output_file)
except Exception as e:
raise MarkdownToPdfError(f'will be HTML Convert To PDF Kalah: {e}') from e
finally:
# Clean Up Temp Files
if temp_html_path and os.path.exists(temp_html_path):
try:
os.unlink(temp_html_path)
logger.debug("Temporary files cleaned: %s", temp_html_path)
except OSError as e:
logger.warning("Clean Up Temp Files %s Kalah: %s", temp_html_path, e)
def convert_html_to_pdf_bytes(self,
html_content: str,
page_format: Optional[str] = None,
margin_mm: Optional[int] = None) -> bytes:
"""
Use Playwright will be HTML Convert content to PDF Bytes of data.
Args:
html_content: To be converted HTML Contents
page_format: Override default page formatting
margin_mm: Override default margin
Returns:
PDF Bytes data of the file
Raises:
MarkdownToPdfError: Automatically close purchase order after PDF failed to transform
"""
if not isinstance(html_content, str) or not html_content.strip():
raise MarkdownToPdfError('HTML Content cannot be empty')
format_to_use = page_format or self.page_format
margin_to_use = margin_mm if margin_mm is not None else self.margin_mm
if format_to_use not in self.SUPPORTED_PAGE_FORMATS:
raise MarkdownToPdfError(f'Unsupported page format: {format_to_use}')
# Create Temporary HTML Doc.
temp_html_path = None
try:
with tempfile.NamedTemporaryFile(
mode='w',
suffix='.html',
delete=False,
encoding='utf-8'
) as temp_file:
temp_html_path = temp_file.name
temp_file.write(html_content)
logger.debug("Temporary Created HTML Doc.: %s", temp_html_path)
# Use Playwright Convert To PDF Bytes of data
pdf_bytes = self._render_pdf_bytes_with_playwright(temp_html_path, format_to_use, margin_to_use)
logger.debug("Successfully Generated! PDF bytes data, size: %d bytes", len(pdf_bytes))
return pdf_bytes
except Exception as e:
raise MarkdownToPdfError(f'will be HTML Convert To PDF Byte Data Failure: {e}') from e
finally:
# Clean Up Temp Files
if temp_html_path and os.path.exists(temp_html_path):
try:
os.unlink(temp_html_path)
logger.debug("Temporary files cleaned: %s", temp_html_path)
except OSError as e:
logger.warning("Clean Up Temp Files %s Kalah: %s", temp_html_path, e)
def _render_pdf_with_playwright(self,
html_file_path: str,
output_path: Path,
page_format: str,
margin_mm: int) -> None:
"""Medical Treatment: Playwright PDF The internal method of the build."""
try:
with sync_playwright() as playwright:
browser = playwright.chromium.launch()
try:
page = browser.new_page()
file_url = f'file:///{html_file_path.replace(os.sep, "/")}'
logger.debug("Load in browser HTML: %s", file_url)
page.goto(file_url, timeout=self.DEFAULT_TIMEOUT)
# If enabled MathJax, waiting for typesetting to complete
if self.enable_math:
self._wait_for_mathjax(page)
# Build with specified settings PDF
margin_config = {
'top': f'{margin_mm}mm',
'bottom': f'{margin_mm}mm',
'left': f'{margin_mm}mm',
'right': f'{margin_mm}mm'
}
page.pdf(
path=str(output_path),
format=page_format,
margin=margin_config,
print_background=True
)
finally:
browser.close()
except Exception as e:
raise MarkdownToPdfError(f'Playwright PDF Generation Failed: {e}') from e
def _render_pdf_bytes_with_playwright(self,
html_file_path: str,
page_format: str,
margin_mm: int) -> bytes:
"""Medical Treatment: Playwright PDF Internal method for byte data generation."""
try:
with sync_playwright() as playwright:
browser = playwright.chromium.launch()
try:
page = browser.new_page()
file_url = f'file:///{html_file_path.replace(os.sep, "/")}'
logger.debug("Load in browser HTML: %s", file_url)
page.goto(file_url, timeout=self.DEFAULT_TIMEOUT)
# If enabled MathJax, waiting for typesetting to complete
if self.enable_math:
self._wait_for_mathjax(page)
# Build with specified settings PDF Bytes of data
margin_config = {
'top': f'{margin_mm}mm',
'bottom': f'{margin_mm}mm',
'left': f'{margin_mm}mm',
'right': f'{margin_mm}mm'
}
pdf_bytes = page.pdf(
format=page_format,
margin=margin_config,
print_background=True
)
return pdf_bytes
finally:
browser.close()
except Exception as e:
raise MarkdownToPdfError(f'Playwright PDF Byte data generation failed: {e}') from e
def _wait_for_mathjax(self, page) -> None:
"""Menunggu MathJax Finish typography with timeout."""
try:
logger.debug("Menunggu MathJax Format...")
page.wait_for_function(
"() => window.MathJax && window.MathJax.typesetPromise",
timeout=self.DEFAULT_TIMEOUT
)
page.evaluate("() => window.MathJax && window.MathJax.typesetPromise()")
logger.debug("MathJax Typography complete")
except Exception as e:
logger.debug("MathJax Timeout or non-existent (this is normal): %s", e)
def convert_file(self,
input_path: Union[str, Path],
output_path: Union[str, Path],
css_file: Optional[Union[str, Path]] = None,
**kwargs) -> None:
"""
will be Markdown Convert file to PDF。
Args:
input_path: Masukkan Markdown FilePath
output_path: Output PDF FilePath
css_file: Optional customizations CSS FilePath
**kwargs: Add Ons (page_format, margin_mm, enable_math)
Raises:
MarkdownToPdfError: If the conversion fails
"""
input_file = self._validate_input_path(input_path)
try:
# read out Markdown Contents
with open(input_file, 'r', encoding='utf-8') as f:
markdown_content = f.read()
logger.info("FROM %s read %d characters", input_file, len(markdown_content))
# If customization is provided CSS, then load
custom_css = None
if css_file:
custom_css = self._load_css_file(css_file)
logger.info("FROM %s Customization loaded CSS", css_file)
# Convert To PDF
self.convert_string(markdown_content, output_path, custom_css, **kwargs)
except (OSError, UnicodeDecodeError) as e:
raise MarkdownToPdfError(f'Read input file {input_file} Kalah: {e}') from e
def convert_string(self,
markdown_text: str,
output_path: Union[str, Path],
custom_css: Optional[str] = None,
**kwargs) -> None:
"""
will be Markdown Convert string to PDF。
Args:
markdown_text: Markdown String content
output_path: Output PDF FilePath
custom_css: Optional customizations CSS Contents
**kwargs: Add Ons (page_format, margin_mm, enable_math)
Raises:
MarkdownToPdfError: If the conversion fails
"""
try:
# will be Markdown Render As HTML
html_content = self.render_markdown_to_html(
markdown_text,
custom_css=custom_css,
enable_math=kwargs.get('enable_math')
)
# will be HTML Convert To PDF
self.convert_html_to_pdf(
html_content,
output_path,
page_format=kwargs.get('page_format'),
margin_mm=kwargs.get('margin_mm')
)
except MarkdownToPdfError:
raise
except Exception as e:
raise MarkdownToPdfError(f'failed to transform: {e}') from e
def convert_file_to_bytes(self,
input_path: Union[str, Path],
css_file: Optional[Union[str, Path]] = None,
**kwargs) -> bytes:
"""
will be Markdown Convert file to PDF Bytes of data.
Args:
input_path: Masukkan Markdown FilePath
css_file: Optional customizations CSS FilePath
**kwargs: Add Ons (page_format, margin_mm, enable_math)
Returns:
PDF Bytes data of the file
Raises:
MarkdownToPdfError: If the conversion fails
"""
input_file = self._validate_input_path(input_path)
try:
# read out Markdown Contents
with open(input_file, 'r', encoding='utf-8') as f:
markdown_content = f.read()
logger.info("FROM %s read %d characters", input_file, len(markdown_content))
# If customization is provided CSS, then load
custom_css = None
if css_file:
custom_css = self._load_css_file(css_file)
logger.info("FROM %s Customization loaded CSS", css_file)
# Convert To PDF Bytes of data
return self.convert_string_to_bytes(markdown_content, custom_css, **kwargs)
except (OSError, UnicodeDecodeError) as e:
raise MarkdownToPdfError(f'Read input file {input_file} Kalah: {e}') from e
def convert_string_to_bytes(self,
markdown_text: str,
custom_css: Optional[str] = None,
**kwargs) -> bytes:
"""
will be Markdown Convert string to PDF Bytes of data.
Args:
markdown_text: Markdown String content
custom_css: Optional customizations CSS Contents
**kwargs: Add Ons (page_format, margin_mm, enable_math)
Returns:
PDF Bytes data of the file
Raises:
MarkdownToPdfError: If the conversion fails
"""
try:
# will be Markdown Render As HTML
html_content = self.render_markdown_to_html(
markdown_text,
custom_css=custom_css,
enable_math=kwargs.get('enable_math')
)
# will be HTML Convert To PDF Bytes of data
return self.convert_html_to_pdf_bytes(
html_content,
page_format=kwargs.get('page_format'),
margin_mm=kwargs.get('margin_mm')
)
except MarkdownToPdfError:
raise
except Exception as e:
raise MarkdownToPdfError(f'failed to transform: {e}') from e
# Backward compatibility function
def render_markdown_to_html(md_text: str, css: str = None, enable_math: bool = False) -> str:
"""Legacy functions for backward compatibility."""
converter = MarkdownToPdfConverter(enable_math=enable_math)
return converter.render_markdown_to_html(md_text, custom_css=css, enable_math=enable_math)
def html_to_pdf_with_playwright(html: str, output_path: str, format: str = 'A4', margin_mm: int = 20):
"""Legacy functions for backward compatibility."""
converter = MarkdownToPdfConverter(page_format=format, margin_mm=margin_mm)
converter.convert_html_to_pdf(html, output_path, page_format=format, margin_mm=margin_mm)
def md_to_pdf(input_md: str, output_pdf: str, css_file: str = None, is_path: bool = False,
enable_math: bool = False, page_format: str = 'A4', margin_mm: int = 20):
"""Legacy functions for backward compatibility."""
converter = MarkdownToPdfConverter(
enable_math=enable_math,
page_format=page_format,
margin_mm=margin_mm
)
if is_path:
converter.convert_file(input_md, output_pdf, css_file=css_file)
else:
custom_css = None
if css_file:
custom_css = converter._load_css_file(css_file)
converter.convert_string(input_md, output_pdf, custom_css=custom_css)
def md_to_pdf_bytes(input_md: str, css_file: str = None, is_path: bool = False,
enable_math: bool = False, page_format: str = 'A4', margin_mm: int = 20) -> bytes:
"""
will be Markdown Convert To PDF Handy function for byte data.
Args:
input_md: Markdown File path (if is_path=True or: Markdown String
css_file: Optional customizations CSS FilePath
is_path: input_md Is file path
enable_math: enabled MathJax Math expression support
page_format: Post Format (A4, A3, A5, Letter, Legal, Tabloid)
margin_mm: Margins (in mm)
Returns:
PDF Bytes data of the file
Raises:
MarkdownToPdfError: If the conversion fails
"""
converter = MarkdownToPdfConverter(
enable_math=enable_math,
page_format=page_format,
margin_mm=margin_mm
)
if is_path:
return converter.convert_file_to_bytes(input_md, css_file=css_file)
else:
custom_css = None
if css_file:
custom_css = converter._load_css_file(css_file)
return converter.convert_string_to_bytes(input_md, custom_css=custom_css)
================================================
FILE: src/backend/bisheng/common/utils/util.py
================================================
import base64
import contextvars
import functools
import hashlib
import hmac
import os
import uuid
import orjson
def orjson_dumps(v, *, default=None, sort_keys=False, indent_2=True):
option = orjson.OPT_SORT_KEYS if sort_keys else None
if indent_2:
# orjson.dumps returns bytes, to match standard json.dumps we need to decode
# option
# To modify how data is serialized, specify option. Each option is an integer constant in orjson.
# To specify multiple options, mask them together, e.g., option=orjson.OPT_STRICT_INTEGER | orjson.OPT_NAIVE_UTC
if option is None:
option = orjson.OPT_INDENT_2
else:
option |= orjson.OPT_INDENT_2
if default is None:
return orjson.dumps(v, option=option).decode()
return orjson.dumps(v, default=default, option=option).decode()
def generate_short_high_entropy_string(length=32):
"""
Generate a short string of high entropy of the specified length (default32characters.
Parameters:
length: Target Length
"""
# Calculate the required number of raw bytes based on the target length (Base64Corresponds to each character6bit
byte_length = (length * 6 + 7) // 8 # Round Up
# Generate Encryption-Level Random Keys
key = os.urandom(16) # 128Bit Key
# Generate basic random data (UUIDright of privacy128bits are sufficient as a source)
base_random = uuid.uuid4().bytes
# HMACCrypto Hybrid Enhanced Randomness
h = hmac.new(key, base_random, hashlib.sha256)
hmac_bytes = h.digest()
# Truncate bytes of desired length
combined = hmac_bytes[:byte_length]
# Convert ToURLSAFETYBase64and truncate to target length (without padding)
short_str = base64.urlsafe_b64encode(combined).decode().rstrip('=')[:length]
return short_str
# --- Define decorator ---
def transfer_trace_id(func):
"""
Decorator: Automatically convert the current context (including trace_id) into the execution environment of the decorated function.
Available for Thread OR Executor。
"""
@functools.wraps(func)
def wrapper(*args, **kwargs):
# 1. The code here is still executing in the parent thread, capturing the context
ctx = contextvars.copy_context()
# 2. Use ctx.run Run the original function in the captured context
return ctx.run(func, *args, **kwargs)
return wrapper
================================================
FILE: src/backend/bisheng/components/__init__.py
================================================
from bisheng.interface.custom.custom_component import CustomComponent
__all__ = ["CustomComponent"]
================================================
FILE: src/backend/bisheng/components/custom_components/CustomComponent.py
================================================
from bisheng import CustomComponent
from bisheng.field_typing import Data
class Component(CustomComponent):
documentation: str = 'http://docs.bisheng.org/components/custom'
def build_config(self):
return {'param': {'display_name': 'Parameter'}}
def build(self, param: Data) -> Data:
return param
================================================
FILE: src/backend/bisheng/components/custom_components/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/ai/__init__.py
================================================
from langchain_anthropic import ChatAnthropic
from langchain_community.chat_models import ChatTongyi, ChatZhipuAI, MiniMaxChat, MoonshotChat
from langchain_community.document_compressors import DashScopeRerank
from langchain_community.embeddings import DashScopeEmbeddings
from langchain_deepseek import ChatDeepSeek
from langchain_ollama import OllamaEmbeddings, ChatOllama
from langchain_openai import ChatOpenAI, AzureChatOpenAI, OpenAIEmbeddings, AzureOpenAIEmbeddings
from .asr import OpenAIASRClient, AliyunASRClient, AzureOpenAIASRClient
from .base import BaseASRClient, BaseTTSClient
from .llm.chat_openai_compatible import ChatOpenAICompatible
from .rerank.common_rerank import CommonRerank
from .rerank.xinference_rerank import XinferenceRerank
from .tts import OpenAITTSClient, AliyunTTSClient, AzureOpenAITTSClient
__all__ = [
'BaseASRClient',
'BaseTTSClient',
'OpenAIASRClient',
'AliyunASRClient',
'AzureOpenAIASRClient',
'OpenAITTSClient',
'AliyunTTSClient',
'AzureOpenAITTSClient',
'ChatOllama',
'ChatOpenAI',
'AzureChatOpenAI',
'ChatTongyi',
'ChatZhipuAI',
'MiniMaxChat',
'ChatAnthropic',
'ChatDeepSeek',
'MoonshotChat',
'ChatOpenAICompatible',
'OllamaEmbeddings',
'OpenAIEmbeddings',
'AzureOpenAIEmbeddings',
'DashScopeEmbeddings',
'DashScopeRerank',
'CommonRerank',
'XinferenceRerank'
]
================================================
FILE: src/backend/bisheng/core/ai/asr/__init__.py
================================================
from .aliyun_asr_client import AliyunASRClient
from .azure_openai_asr_client import AzureOpenAIASRClient
from .openai_asr_client import OpenAIASRClient
__all__ = [
'OpenAIASRClient',
'AliyunASRClient',
'AzureOpenAIASRClient'
]
================================================
FILE: src/backend/bisheng/core/ai/asr/aliyun_asr_client.py
================================================
import asyncio
from typing import Optional
from dashscope.audio.asr import Recognition, RecognitionResult
from ..base import BaseASRClient
class AliyunASRClient(BaseASRClient):
"""Alibaba CloudASRClient"""
def __init__(self, api_key: str, model: str, **kwargs):
"""
Initialize Alibaba CloudASRClient
"""
self.api_key = api_key
self.recognition = Recognition(
model=model,
format="wav",
sample_rate=16000,
callback=None,
**kwargs
)
# Time-consuming operation, asynchronous execution
def sync_func(self, temp_file, language=None, model=None):
result: RecognitionResult = self.recognition.call(temp_file, api_key=self.api_key, language=language,
model=model)
return result
async def _transcribe(
self,
audio: str,
language: Optional[str] = None,
model: Optional[str] = None,
**kwargs
) -> str:
result: RecognitionResult = await asyncio.to_thread(self.sync_func, audio, language, model)
if result.status_code != 200:
raise RuntimeError(
f"ASR request failed with status code {result.code} and message {result.message}"
)
sentence = result.get_sentence()
if sentence and sentence[0]:
return sentence[0]["text"]
return ""
================================================
FILE: src/backend/bisheng/core/ai/asr/azure_openai_asr_client.py
================================================
from typing import Optional
import openai
from ..base import BaseASRClient
class AzureOpenAIASRClient(BaseASRClient):
"""MicrosoftAzure OpenAI ASRClient"""
def __init__(self, api_key: str, **kwargs):
self.model = kwargs.pop("model", "whisper-1")
self.client = openai.AsyncAzureOpenAI(api_key=api_key, **kwargs)
async def _transcribe(
self,
audio: str,
language: str = "auto",
model: Optional[str] = None,
**kwargs
) -> str:
"""
UseAzure OpenAI Whisper APISpeech Recognition
:param audio:
:param language:
:param model:
:return:
"""
with open(audio, "rb") as f:
response = await self.client.audio.transcriptions.create(
file=f,
model=model or self.model,
language=language if language != "auto" else None,
**kwargs
)
return response.text
================================================
FILE: src/backend/bisheng/core/ai/asr/openai_asr_client.py
================================================
from typing import Optional, Union, BinaryIO
import openai
from ..base import BaseASRClient
class OpenAIASRClient(BaseASRClient):
"""OpenAI ASRClient"""
def __init__(self, api_key: str, **kwargs):
self.model = kwargs.pop("model", "whisper-1")
self.client = openai.AsyncOpenAI(api_key=api_key, **kwargs)
async def _transcribe(
self,
audio: Union[str, bytes, BinaryIO],
language: str = "auto",
model: Optional[str] = None,
**kwargs
) -> str:
"""
UseOpenAI Whisper APISpeech Recognition
Args:
audio: Audio File Host
language: Language code, e.g. 'zh', 'en'
model: Model name, defaults to 'whisper-1'
Returns:
Recognized text content
"""
with open(audio, "rb") as f:
transcript = await self.client.audio.transcriptions.create(
model=model or self.model,
file=f,
language=language,
**kwargs
)
return transcript.text
================================================
FILE: src/backend/bisheng/core/ai/base.py
================================================
import asyncio
import os
import tempfile
import uuid
from abc import ABC, abstractmethod
from typing import Optional, Union, BinaryIO, Sequence
import aiofiles
from langchain_core.callbacks import Callbacks
from langchain_core.documents import BaseDocumentCompressor, Document
from pydantic import ConfigDict
class BaseASRClient(ABC):
"""ASR (Automatic Speech Recognition) Base Interface Class"""
async def transcribe(
self,
audio: Union[str, bytes, BinaryIO],
language: Optional[str] = None,
model: Optional[str] = None,
**kwargs
) -> str:
"""
Convert Audio to Text
Args:
audio: Audio file path, audio byte data, or file object
language: Language code, e.g. 'zh', 'en'
model: Used Model Name
Returns:
Transcribed text content
"""
if not audio:
raise ValueError("Audio input is required")
if isinstance(audio, str):
with open(audio, 'rb') as audio_file:
audio_bytes = audio_file.read()
elif isinstance(audio, bytes):
audio_bytes = audio
elif hasattr(audio, 'read'):
audio_bytes = audio.read()
else:
raise ValueError("Invalid audio input type")
tmp_dir = tempfile.gettempdir()
tmp_file_path = os.path.join(tmp_dir, uuid.uuid4().hex + '.wav')
# ffmpeg Convert To16kSampling Rate MonowavDoc.
converted_file_path = os.path.join(tmp_dir, uuid.uuid4().hex + '_16k_mono.wav')
try:
async with aiofiles.open(tmp_file_path, 'wb') as f:
await f.write(audio_bytes)
command = f'ffmpeg -y -i "{tmp_file_path}" -ar 16000 -ac 1 "{converted_file_path}"'
process = await asyncio.create_subprocess_shell(
command,
stdout=asyncio.subprocess.PIPE,
stderr=asyncio.subprocess.PIPE
)
await process.communicate()
return await self._transcribe(converted_file_path, language=language, model=model, **kwargs)
finally:
if os.path.exists(tmp_file_path):
os.remove(tmp_file_path)
if os.path.exists(converted_file_path):
os.remove(converted_file_path)
@abstractmethod
async def _transcribe(self, audio: str, language: Optional[str] = None, model: Optional[str] = None,
**kwargs) -> str:
"""
Internal Method: Convert Audio to Text
Args:
audio: Audio File Host, Files are automatically deleted when processing is complete
language: Language code, e.g. 'zh', 'en'
model: Used Model Name
Returns:
Transcribed text content
"""
raise NotImplementedError
class BaseTTSClient(ABC):
"""TTS (Text To Speech) Base Interface Class"""
@abstractmethod
async def synthesize(
self,
text: str,
voice: Optional[str] = None,
language: Optional[str] = None,
format: str = "mp3"
) -> bytes:
"""
Synthesize text into speech
Args:
text: Text to compose
voice: Audio Options
language: Language code
format: Audio formats, such as 'mp3', 'wav'
Returns:
Audio Bytes Data
"""
pass
class BaseRerank(BaseDocumentCompressor):
"""Rerank base interface class"""
model_config = ConfigDict(arbitrary_types_allowed=True)
@abstractmethod
def compress_documents(
self,
documents: Sequence[Document],
query: str,
callbacks: Optional[Callbacks] = None,
) -> Sequence[Document]:
"""Compress retrieved documents given the query context.
Args:
documents: The retrieved documents.
query: The query context.
callbacks: Optional callbacks to run during compression.
Returns:
The compressed documents.
"""
@staticmethod
def sort_rerank_result(documents: Sequence[Document], results: list[dict]) -> Sequence[Document]:
"""Sort and annotate original documents based on rerank results."""
sorted_docs = []
for res in results:
index = res.get("index")
if index is not None and 0 <= index < len(documents):
doc = documents[index]
doc.metadata["relevance_score"] = res.get("relevance_score")
sorted_docs.append(doc)
return sorted_docs
================================================
FILE: src/backend/bisheng/core/ai/embeddings/__init__.py
================================================
from langchain_community.embeddings import DashScopeEmbeddings
from langchain_ollama import OllamaEmbeddings
from langchain_openai import OpenAIEmbeddings, AzureOpenAIEmbeddings
__all__ = [
'OllamaEmbeddings',
'OpenAIEmbeddings',
'AzureOpenAIEmbeddings',
'DashScopeEmbeddings',
]
================================================
FILE: src/backend/bisheng/core/ai/llm/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/ai/llm/chat_openai_compatible.py
================================================
from typing import Optional, Any
from langchain_core.language_models import LanguageModelInput
from langchain_openai import ChatOpenAI
class ChatOpenAICompatible(ChatOpenAI):
"""
A ChatOpenAI subclass that ensures compatibility with older parameter names.
use max_tokens instead of max_completion_tokens.
"""
def _get_request_payload(
self,
input_: LanguageModelInput,
*,
stop: Optional[list[str]] = None,
**kwargs: Any,
) -> dict:
payload = super()._get_request_payload(input_, stop=stop, **kwargs)
# max_tokens was deprecated in favor of max_completion_tokens
# in September 2024 release
if "max_completion_tokens" in payload:
payload["max_tokens"] = payload.pop("max_completion_tokens")
return payload
================================================
FILE: src/backend/bisheng/core/ai/llm/custom_chat_deepseek.py
================================================
from typing import Optional, Any
from langchain_core.language_models import LanguageModelInput
from langchain_core.messages import AIMessage
from langchain_deepseek import ChatDeepSeek
class CustomChatDeepSeek(ChatDeepSeek):
def _get_request_payload(
self,
input_: LanguageModelInput,
*,
stop: Optional[list[str]] = None,
**kwargs: Any,
) -> dict:
payload = super()._get_request_payload(input_, stop=stop, **kwargs)
messages = self._convert_input(input_).to_messages()
for payload_msg, msg in zip(payload['messages'], messages):
if isinstance(msg, AIMessage):
payload_msg['reasoning_content'] = msg.additional_kwargs.get('reasoning_content', '')
return payload
================================================
FILE: src/backend/bisheng/core/ai/llm/custom_chat_tongyi.py
================================================
from typing import Any, List, Dict
from langchain_community.chat_models import ChatTongyi
from langchain_core.messages import BaseMessage
class CustomChatTongYi(ChatTongyi):
"""
Custom ChatTongYi Model to handle multi-modal input conversion.
"""
def _invocation_params(
self, messages: List[BaseMessage], stop: Any, **kwargs: Any
) -> Dict[str, Any]:
# Get basic call parameters
params = super()._invocation_params(messages, stop, **kwargs)
# Iterate through and clean the messages data
if "messages" in params:
for msg in params["messages"]:
# Only when the content is a list (multimodal) is processing required.
if "content" in msg and isinstance(msg["content"], list):
new_content = []
for item in msg["content"]:
# Check if the image_url is in OpenAI format.
if isinstance(item, dict) and item.get("type") == "image_url":
# Extract the URL string
image_url_data = item.get("image_url")
url_str = ""
if isinstance(image_url_data, dict):
url_str = image_url_data.get("url", "")
elif isinstance(image_url_data, str):
url_str = image_url_data
# Append the cleaned image data
if url_str:
new_content.append({
"type": "image",
"image": url_str
})
else:
# Non-image data is directly appended
new_content.append(item)
# Update the message content with cleaned data
msg["content"] = new_content
return params
================================================
FILE: src/backend/bisheng/core/ai/rerank/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/ai/rerank/common_rerank.py
================================================
from typing import Optional, Sequence
from httpx import Client, AsyncClient
from langchain_core.callbacks import Callbacks
from langchain_core.documents import Document
from pydantic import Field, model_validator
from typing_extensions import Self
from ..base import BaseRerank
class CommonRerank(BaseRerank):
"""Document compressor that support `/v1/rerank` router and results simial vllm."""
base_url: str = Field(..., description="server base url， example: http://localhost:9997")
rerank_endpoint: str = Field(default="/v1/rerank", description="rerank endpoint for base url")
api_key: Optional[str] = Field(default="", description="api key for rerank server")
model: str = Field(..., description="model name for rerank model")
client: Optional[Client] = Field(default=None, description="client instance")
async_client: Optional[AsyncClient] = Field(default=None, description="async client instance")
@model_validator(mode="after")
def validate_client(self) -> Self:
"""Validate that client exists in environment."""
self.base_url = self.base_url.rstrip("/") if self.base_url else self.base_url
if not self.client:
self.client = Client(base_url=self.base_url)
if not self.async_client:
self.async_client = AsyncClient(base_url=self.base_url)
return self
def get_default_headers(self):
headers = {}
if self.api_key:
headers["Authorization"] = f"Bearer {self.api_key}"
return headers
def get_default_params(self):
params = {}
if self.model:
params["model"] = self.model
return params
def get_req_data(
self,
documents: Sequence[Document],
query: str
) -> dict:
if len(documents) == 0: # to avoid empty api call
return {}
docs = [
doc.page_content if isinstance(doc, Document) else doc for doc in documents
]
req_data = self.get_default_params()
req_data["query"] = query
req_data["documents"] = docs
return req_data
def compress_documents(
self,
documents: Sequence[Document],
query: str,
callbacks: Optional[Callbacks] = None,
) -> Sequence[Document]:
"""Rerank retrieved documents given the query context.
Args:
documents: The retrieved documents.
query: The query context.
callbacks: Optional callbacks to run during compression.
Returns:
The compressed documents.
"""
req_data = self.get_req_data(documents, query)
if not req_data:
return []
resp = self.client.post(self.rerank_endpoint, json=req_data, headers=self.get_default_headers())
if resp.status_code != 200:
raise ValueError(f"Rerank request failed with status code {resp.status_code}: {resp.text}")
results = resp.json().get("results", [])
return self.sort_rerank_result(documents, results)
async def acompress_documents(
self,
documents: Sequence[Document],
query: str,
callbacks: Optional[Callbacks] = None,
) -> Sequence[Document]:
"""Async rerank retrieved documents given the query context.
Args:
documents: The retrieved documents.
query: The query context.
callbacks: Optional callbacks to run during compression.
Returns:
The compressed documents.
"""
req_data = self.get_req_data(documents, query)
if not req_data:
return []
resp = await self.async_client.post(self.rerank_endpoint, json=req_data, headers=self.get_default_headers())
if resp.status_code != 200:
raise ValueError(f"Rerank request failed with status code {resp.status_code}: {resp.text}")
results = resp.json().get("results", [])
return self.sort_rerank_result(documents, results)
================================================
FILE: src/backend/bisheng/core/ai/rerank/rrf_rerank.py
================================================
from typing import Sequence, Optional, List, Any, Dict
from langchain_core.callbacks import Callbacks
from langchain_core.documents import Document
from pydantic import model_validator
from bisheng.core.ai.base import BaseRerank
class RRFRerank(BaseRerank):
"""
Perform weighted Reciprocal Rank Fusion on multiple rank lists.
You can find more details about RRF here:
https://plg.uwaterloo.ca/~gvcormac/cormacksigir09-rrf.pdf
It uses a rank fusion.
Args:
retrievers: A list of retrievers to ensemble.
weights: A list of weights corresponding to the retrievers. Defaults to equal
weighting for all retrievers.
c: A constant added to the rank, controlling the balance between the importance
of high-ranked items and the consideration given to lower-ranked items.
Default is 60.
"""
retrievers: List[Any]
weights: List[float] = None
c: int = 60
remove_zero_score: bool = False
@model_validator(mode='before')
@classmethod
def set_weights(cls, values: Dict[str, Any]) -> Dict[str, Any]:
if not values.get("weights"):
n_retrievers = len(values["retrievers"])
values["weights"] = [1 / n_retrievers] * n_retrievers
return values
def compress_documents(
self,
documents: List[List[Document]],
query: str,
callbacks: Optional[Callbacks] = None,
) -> Sequence[Document]:
"""
Args:
documents: A list of document lists for each retriever search list[document].
query: The query to use for computing weighted reciprocal rank.
callbacks: Optional callbacks to use for computing weighted reciprocal rank.
Returns:
list: The final aggregated list of items sorted by their weighted RRF
scores in descending order. and remove duplicates document.
"""
if len(documents) != len(self.weights):
raise ValueError("Number of rank lists must be equal to the number of weights.")
# Create a union of all unique documents in the input doc_lists
all_documents = set()
for doc_list in documents:
for doc in doc_list:
all_documents.add(doc.page_content)
# Initialize the RRF score dictionary for each document
rrf_score_dic = {doc: 0.0 for doc in all_documents}
# Calculate RRF scores for each document
for doc_list, weight in zip(documents, self.weights):
for rank, doc in enumerate(doc_list, start=1):
rrf_score = weight * (1 / (rank + self.c))
rrf_score_dic[doc.page_content] += rrf_score
# Sort documents by their RRF scores in descending order
sorted_documents = sorted(rrf_score_dic.keys(), key=lambda x: rrf_score_dic[x], reverse=True)
# Map the sorted page_content back to the original document objects
page_content_to_doc_map = {doc.page_content: doc for doc_list in documents for doc in doc_list}
if self.remove_zero_score:
sorted_docs = [page_content_to_doc_map[page_content] for page_content in sorted_documents if
rrf_score_dic[page_content] > 0]
else:
sorted_docs = [page_content_to_doc_map[page_content] for page_content in sorted_documents]
return sorted_docs
================================================
FILE: src/backend/bisheng/core/ai/rerank/xinference/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/ai/rerank/xinference/_compat.py
================================================
# Copyright 2022-2024 XProbe Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from typing import Dict, Iterable, List, Literal, Optional, Union
from pydantic.version import VERSION as PYDANTIC_VERSION
PYDANTIC_V2 = PYDANTIC_VERSION.startswith("2.")
if PYDANTIC_V2:
from pydantic.v1 import ( # noqa: F401
BaseModel,
Field,
Protocol,
ValidationError,
create_model,
create_model_from_namedtuple,
create_model_from_typeddict,
parse_file_as,
validate_arguments,
validator,
)
from pydantic.v1.error_wrappers import ErrorWrapper # noqa: F401
from pydantic.v1.parse import load_str_bytes # noqa: F401
from pydantic.v1.types import StrBytes # noqa: F401
from pydantic.v1.utils import ROOT_KEY # noqa: F401
else:
from pydantic import ( # noqa: F401
BaseModel,
Field,
Protocol,
ValidationError,
create_model,
create_model_from_namedtuple,
create_model_from_typeddict,
parse_file_as,
validate_arguments,
validator,
)
from pydantic.error_wrappers import ErrorWrapper # noqa: F401
from pydantic.parse import load_str_bytes # noqa: F401
from pydantic.types import StrBytes # noqa: F401
from pydantic.utils import ROOT_KEY # noqa: F401
from openai.types.chat.chat_completion_named_tool_choice_param import (
ChatCompletionNamedToolChoiceParam,
)
from openai.types.chat.chat_completion_stream_options_param import (
ChatCompletionStreamOptionsParam,
)
from openai.types.chat.chat_completion_tool_param import ChatCompletionToolParam
from openai.types.shared_params.response_format_json_object import (
ResponseFormatJSONObject,
)
from openai.types.shared_params.response_format_text import ResponseFormatText
OpenAIChatCompletionStreamOptionsParam = create_model_from_typeddict(
ChatCompletionStreamOptionsParam
)
OpenAIChatCompletionToolParam = create_model_from_typeddict(ChatCompletionToolParam)
OpenAIChatCompletionNamedToolChoiceParam = create_model_from_typeddict(
ChatCompletionNamedToolChoiceParam
)
from openai._types import Body
class JSONSchema(BaseModel):
name: str
description: Optional[str] = None
schema_: Optional[Dict[str, object]] = Field(alias="schema", default=None)
strict: Optional[bool] = None
class ResponseFormatJSONSchema(BaseModel):
json_schema: JSONSchema
type: Literal["json_schema"]
ResponseFormat = Union[
ResponseFormatText, ResponseFormatJSONObject, ResponseFormatJSONSchema
]
class CreateChatCompletionOpenAI(BaseModel):
"""
Comes from source code: https://github.com/openai/openai-python/blob/main/src/openai/types/chat/completion_create_params.py
"""
messages: List[Dict]
model: str
frequency_penalty: Optional[float]
logit_bias: Optional[Dict[str, int]]
logprobs: Optional[bool]
max_completion_tokens: Optional[int]
max_tokens: Optional[int]
n: Optional[int]
parallel_tool_calls: Optional[bool]
presence_penalty: Optional[float]
response_format: Optional[ResponseFormat]
seed: Optional[int]
service_tier: Optional[Literal["auto", "default"]]
stop: Union[Optional[str], List[str]]
stream_options: Optional[OpenAIChatCompletionStreamOptionsParam] # type: ignore
temperature: Optional[float]
tool_choice: Optional[ # type: ignore
Union[
Literal["none", "auto", "required"],
OpenAIChatCompletionNamedToolChoiceParam,
]
]
tools: Optional[Iterable[OpenAIChatCompletionToolParam]] # type: ignore
top_logprobs: Optional[int]
top_p: Optional[float]
extra_body: Optional[Body]
user: Optional[str]
================================================
FILE: src/backend/bisheng/core/ai/rerank/xinference/client/__init__.py
================================================
# Copyright 2022-2023 XProbe Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from .restful.async_restful_client import AsyncClient
from .restful.restful_client import Client
# For compatibility
RESTfulClient = Client
AsyncRESTfulClient = AsyncClient
================================================
FILE: src/backend/bisheng/core/ai/rerank/xinference/client/common.py
================================================
# Copyright 2022-2023 XProbe Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import json
from typing import Any, AsyncIterator, Iterator, Union
def convert_float_to_int_or_str(model_size: float) -> Union[int, str]:
"""convert float to int or string
if float can be presented as int, convert it to int, otherwise convert it to string
"""
if int(model_size) == model_size:
return int(model_size)
else:
return str(model_size)
def streaming_response_iterator(
response_lines: Iterator[bytes],
) -> Iterator[Any]:
"""
Create an Iterator to handle the streaming type of generation.
Note
----------
This method is for compatible with openai. Please refer to:
https://github.com/openai/openai-python/blob/v0.28.1/openai/api_requestor.py#L99
Parameters
----------
response_lines: Iterator[bytes]
Generated lines by the Model Generator.
Returns
-------
Iterator["CompletionChunk"]
Iterator of CompletionChunks generated by models.
"""
for line in response_lines:
line = line.strip()
if line.startswith(b"data:"):
json_str = line[len(b"data:") :].strip()
if json_str == b"[DONE]":
continue
data = json.loads(json_str.decode("utf-8"))
error = data.get("error", None)
if error is not None:
raise Exception(str(error))
yield data
async def async_streaming_response_iterator(
response_lines: AsyncIterator[bytes],
) -> AsyncIterator[Any]:
"""
Create an AsyncIterator to handle the streaming type of generation.
Note
----------
This method is for compatible with openai. Please refer to:
https://github.com/openai/openai-python/blob/v0.28.1/openai/api_requestor.py#L99
Parameters
----------
response_lines: AsyncIterator[bytes]
Generated lines by the Model Generator.
Returns
-------
AsyncIterator["CompletionChunk"]
AsyncIterator of CompletionChunks generated by models.
"""
async for line in response_lines:
line = line.strip()
if line.startswith(b"data:"):
json_str = line[len(b"data:") :].strip()
if json_str == b"[DONE]":
continue
data = json.loads(json_str.decode("utf-8"))
error = data.get("error", None)
if error is not None:
raise Exception(str(error))
yield data
================================================
FILE: src/backend/bisheng/core/ai/rerank/xinference/client/handlers.py
================================================
from .restful.async_restful_client import ( # noqa: F401
AsyncRESTfulAudioModelHandle as AsyncAudioModelHandle,
)
from .restful.async_restful_client import ( # noqa: F401
AsyncRESTfulChatModelHandle as AsyncChatModelHandle,
)
from .restful.async_restful_client import ( # noqa: F401
AsyncRESTfulEmbeddingModelHandle as AsyncEmbeddingModelHandle,
)
from .restful.async_restful_client import ( # noqa: F401
AsyncRESTfulGenerateModelHandle as AsyncGenerateModelHandle,
)
from .restful.async_restful_client import ( # noqa: F401
AsyncRESTfulImageModelHandle as AsyncImageModelHandle,
)
from .restful.async_restful_client import ( # noqa: F401
AsyncRESTfulVideoModelHandle as AsyncVideoModelHandle,
)
from .restful.restful_client import ( # noqa: F401
RESTfulAudioModelHandle as AudioModelHandle,
)
from .restful.restful_client import ( # noqa: F401
RESTfulChatModelHandle as ChatModelHandle,
)
from .restful.restful_client import ( # noqa: F401
RESTfulEmbeddingModelHandle as EmbeddingModelHandle,
)
from .restful.restful_client import ( # noqa: F401
RESTfulGenerateModelHandle as GenerateModelHandle,
)
from .restful.restful_client import ( # noqa: F401
RESTfulImageModelHandle as ImageModelHandle,
)
from .restful.restful_client import ( # noqa: F401
RESTfulVideoModelHandle as VideoModelHandle,
)
================================================
FILE: src/backend/bisheng/core/ai/rerank/xinference/client/restful/__init__.py
================================================
# Copyright 2022-2023 XProbe Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
================================================
FILE: src/backend/bisheng/core/ai/rerank/xinference/client/restful/async_restful_client.py
================================================
# Copyright 2022-2023 XProbe Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import asyncio
import json
from typing import TYPE_CHECKING, Any, AsyncIterator, Dict, List, Optional, Union
import aiohttp
from ..common import async_streaming_response_iterator, convert_float_to_int_or_str
if TYPE_CHECKING:
from ...types import (
ChatCompletion,
ChatCompletionChunk,
Completion,
CompletionChunk,
Embedding,
ImageList,
PytorchGenerateConfig,
VideoList,
)
def _filter_params(params: Dict[Any, Any]) -> Dict[Any, Any]:
filtered = {}
for key, value in params.items():
if value is not None:
filtered[key] = value
return filtered
async def _get_error_string(response: aiohttp.ClientResponse) -> str:
result = None
try:
if response.content:
json_data = await response.json()
result = json_data["detail"]
except Exception:
pass
try:
response.raise_for_status()
except aiohttp.ClientError as e:
result = str(e)
if result is None:
result = "Unknown error"
await _release_response(response)
return result
async def _release_response(response: aiohttp.ClientResponse):
"""Release the aiohttp response."""
response.release()
await response.wait_for_close()
class AsyncRESTfulModelHandle:
"""
A sync model interface (for RESTful client) which provides type hints that makes it much easier to use xinference
programmatically.
"""
def __init__(self, model_uid: str, base_url: str, auth_headers: Dict):
self._model_uid = model_uid
self._base_url = base_url
self.auth_headers = auth_headers
self.timeout = aiohttp.ClientTimeout(total=1800)
self.session = aiohttp.ClientSession(
connector=aiohttp.TCPConnector(force_close=True)
)
async def close(self):
"""Close the AsyncRESTfulModelHandle session."""
if self.session:
await self.session.close()
self.session = None
def __del__(self):
if self.session:
loop = asyncio.get_event_loop()
loop.create_task(self.close())
class AsyncRESTfulEmbeddingModelHandle(AsyncRESTfulModelHandle):
async def create_embedding(
self, input: Union[str, List[str]], **kwargs
) -> "Embedding":
"""
Create an Embedding from user input via RESTful APIs.
Parameters
----------
input: Union[str, List[str]]
Input text to embed, encoded as a string or array of tokens.
To embed multiple inputs in a single request, pass an array of strings or array of token arrays.
Returns
-------
Embedding
The resulted Embedding vector that can be easily consumed by machine learning models and algorithms.
Raises
------
RuntimeError
Report the failure of embeddings and provide the error message.
"""
url = f"{self._base_url}/v1/embeddings"
request_body = {
"model": self._model_uid,
"input": input,
}
request_body.update(kwargs)
response = await self.session.post(
url, json=request_body, headers=self.auth_headers
)
if response.status != 200:
raise RuntimeError(
f"Failed to create the embeddings, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def convert_ids_to_tokens(
self, input: Union[List, List[List]], **kwargs
) -> List[str]:
"""
Convert token IDs to human readable tokens via RESTful APIs.
Parameters
----------
input: Union[List, List[List]]
Input token IDs to convert, can be a single list of token IDs or a list of token ID lists.
To convert multiple sequences in a single request, pass a list of token ID lists.
Returns
-------
list
A list of decoded tokens in human readable format.
Raises
------
RuntimeError
Report the failure of token conversion and provide the error message.
"""
url = f"{self._base_url}/v1/convert_ids_to_tokens"
request_body = {
"model": self._model_uid,
"input": input,
}
request_body.update(kwargs)
response = await self.session.post(
url, json=request_body, headers=self.auth_headers
)
if response.status != 200:
raise RuntimeError(
f"Failed to decode token ids, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
class AsyncRESTfulRerankModelHandle(AsyncRESTfulModelHandle):
async def rerank(
self,
documents: List[str],
query: str,
top_n: Optional[int] = None,
max_chunks_per_doc: Optional[int] = None,
return_documents: Optional[bool] = None,
return_len: Optional[bool] = None,
**kwargs,
):
"""
Returns an ordered list of documents ordered by their relevance to the provided query.
Parameters
----------
query: str
The search query
documents: List[str]
The documents to rerank
top_n: int
The number of results to return, defaults to returning all results
max_chunks_per_doc: int
The maximum number of chunks derived from a document
return_documents: bool
if return documents
return_len: bool
if return tokens len
Returns
-------
Scores
The scores of documents ordered by their relevance to the provided query
Raises
------
RuntimeError
Report the failure of rerank and provide the error message.
"""
url = f"{self._base_url}/v1/rerank"
request_body = {
"model": self._model_uid,
"documents": documents,
"query": query,
"top_n": top_n,
"max_chunks_per_doc": max_chunks_per_doc,
"return_documents": return_documents,
"return_len": return_len,
"kwargs": json.dumps(kwargs),
}
request_body.update(kwargs)
response = await self.session.post(
url, json=request_body, headers=self.auth_headers
)
if response.status != 200:
raise RuntimeError(
f"Failed to rerank documents, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
class AsyncRESTfulImageModelHandle(AsyncRESTfulModelHandle):
async def text_to_image(
self,
prompt: str,
n: int = 1,
size: str = "1024*1024",
response_format: str = "url",
**kwargs,
) -> "ImageList":
"""
Creates an image by the input text.
Parameters
----------
prompt: `str` or `List[str]`
The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
n: `int`, defaults to 1
The number of images to generate per prompt. Must be between 1 and 10.
size: `str`, defaults to `1024*1024`
The width*height in pixels of the generated image. Must be one of 256x256, 512x512, or 1024x1024.
response_format: `str`, defaults to `url`
The format in which the generated images are returned. Must be one of url or b64_json.
Returns
-------
ImageList
A list of image objects.
"""
url = f"{self._base_url}/v1/images/generations"
request_body = {
"model": self._model_uid,
"prompt": prompt,
"n": n,
"size": size,
"response_format": response_format,
"kwargs": json.dumps(kwargs),
}
response = await self.session.post(
url, json=request_body, headers=self.auth_headers
)
if response.status != 200:
raise RuntimeError(
f"Failed to create the images, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def image_to_image(
self,
image: Union[str, bytes, List[Union[str, bytes]]],
prompt: str,
negative_prompt: Optional[str] = None,
n: int = 1,
size: Optional[str] = None,
response_format: str = "url",
**kwargs,
) -> "ImageList":
"""
Creates an image by the input text.
Parameters
----------
image: `Union[str, bytes, List[Union[str, bytes]]]`
The ControlNet input condition to provide guidance to the `unet` for generation. If the type is
specified as `torch.FloatTensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be
accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height
and/or width are passed, `image` is resized accordingly. If multiple ControlNets are specified in
`init`, images must be passed as a list such that each element of the list can be correctly batched for
input to a single ControlNet.
prompt: `str` or `List[str]`
The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
negative_prompt (`str` or `List[str]`, *optional*):
The prompt or prompts not to guide the image generation. If not defined, one has to pass
`negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
less than `1`).
n: `int`, defaults to 1
The number of images to generate per prompt. Must be between 1 and 10.
size: `str`, defaults to `1024*1024`
The width*height in pixels of the generated image. Must be one of 256x256, 512x512, or 1024x1024.
response_format: `str`, defaults to `url`
The format in which the generated images are returned. Must be one of url or b64_json.
Returns
-------
ImageList
A list of image objects.
:param prompt:
:param image:
"""
url = f"{self._base_url}/v1/images/variations"
params = {
"model": self._model_uid,
"prompt": prompt,
"negative_prompt": negative_prompt,
"n": n,
"size": size,
"response_format": response_format,
"kwargs": json.dumps(kwargs),
}
params = _filter_params(params)
files: List[Any] = []
for key, value in params.items():
files.append((key, (None, value)))
# Handle both single image and list of images
if isinstance(image, list):
if len(image) == 0:
raise ValueError("Image list cannot be empty")
elif len(image) == 1:
# Single image in list, use it directly
files.append(("image", ("image", image[0], "application/octet-stream")))
else:
# Multiple images - send all images with same field name
# FastAPI will collect them into a list
for img_data in image:
files.append(
("image", ("image", img_data, "application/octet-stream"))
)
else:
# Single image
files.append(("image", ("image", image, "application/octet-stream")))
response = await self.session.post(url, data=files, headers=self.auth_headers)
if response.status != 200:
raise RuntimeError(
f"Failed to variants the images, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def image_edit(
self,
image: Union[Union[str, bytes], List[Union[str, bytes]]],
prompt: str,
mask: Optional[Union[str, bytes]] = None,
n: int = 1,
size: Optional[str] = None,
response_format: str = "url",
**kwargs,
) -> "ImageList":
"""
Edit image(s) by the input text and optional mask.
Parameters
----------
image: `Union[Union[str, bytes], List[Union[str, bytes]]]`
The input image(s) to edit. Can be:
- Single image: file path, URL, or binary image data
- Multiple images: list of file paths, URLs, or binary image data
When multiple images are provided, the first image is used as the primary image
and subsequent images are used as reference images for better editing results.
prompt: `str`
The prompt or prompts to guide image editing. If not defined, you need to pass `prompt_embeds`.
mask: `Optional[Union[str, bytes]]`, optional
An optional mask image. White pixels in the mask are repainted while black pixels are preserved.
If provided, this will trigger inpainting mode. If not provided, this will trigger image-to-image mode.
n: `int`, defaults to 1
The number of images to generate per prompt. Must be between 1 and 10.
size: `Optional[str]`, optional
The width*height in pixels of the generated image. If not specified, uses the original image size.
response_format: `str`, defaults to `url`
The format in which the generated images are returned. Must be one of url or b64_json.
**kwargs
Additional parameters to pass to the model.
Returns
-------
ImageList
A list of edited image objects.
Raises
------
RuntimeError
If the image editing request fails.
Examples
--------
# Single image editing
result = await model.image_edit(
image="path/to/image.png",
prompt="make this image look like a painting"
)
# Multiple image editing with reference images
result = await model.image_edit(
image=["primary_image.png", "reference1.jpg", "reference2.png"],
prompt="edit the main image using the style from reference images"
)
"""
url = f"{self._base_url}/v1/images/edits"
params = {
"model": self._model_uid,
"prompt": prompt,
"n": n,
"size": size,
"response_format": response_format,
"kwargs": json.dumps(kwargs),
}
params = _filter_params(params)
files: List[Any] = []
for key, value in params.items():
files.append((key, (None, value)))
# Handle single image or multiple images
import aiohttp
data = aiohttp.FormData()
# Add all parameters as form fields
for key, value in params.items():
if value is not None:
data.add_field(key, str(value))
# Handle single image or multiple images
if isinstance(image, list):
# Validate image list is not empty
if len(image) == 0:
raise ValueError("Image list cannot be empty")
# Multiple images - send as image[] array
for i, img in enumerate(image):
if isinstance(img, str):
# File path - read file content
with open(img, "rb") as f:
content = f.read()
data.add_field(
f"image[]",
content,
filename=f"image_{i}.png",
content_type="image/png",
)
else:
# Binary data
data.add_field(
f"image[]",
img,
filename=f"image_{i}.png",
content_type="image/png",
)
else:
# Single image
if isinstance(image, str):
# File path - read file content
with open(image, "rb") as f:
content = f.read()
data.add_field(
"image", content, filename="image.png", content_type="image/png"
)
else:
# Binary data
data.add_field(
"image", image, filename="image.png", content_type="image/png"
)
if mask is not None:
if isinstance(mask, str):
# File path - read file content
with open(mask, "rb") as f:
content = f.read()
data.add_field(
"mask", content, filename="mask.png", content_type="image/png"
)
else:
# Binary data
data.add_field(
"mask", mask, filename="mask.png", content_type="image/png"
)
try:
response = await self.session.post(
url, data=data, headers=self.auth_headers
)
if response.status != 200:
raise RuntimeError(
f"Failed to edit the images, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
return response_data
finally:
await _release_response(response) if "response" in locals() else None
async def inpainting(
self,
image: Union[str, bytes],
mask_image: Union[str, bytes],
prompt: str,
negative_prompt: Optional[str] = None,
n: int = 1,
size: Optional[str] = None,
response_format: str = "url",
**kwargs,
) -> "ImageList":
"""
Inpaint an image by the input text.
Parameters
----------
image: `Union[str, bytes]`
an image batch to be inpainted (which parts of the image to
be masked out with `mask_image` and repainted according to `prompt`). For both numpy array and pytorch
tensor, the expected value range is between `[0, 1]` If it's a tensor or a list or tensors, the
expected shape should be `(B, C, H, W)` or `(C, H, W)`. If it is a numpy array or a list of arrays, the
expected shape should be `(B, H, W, C)` or `(H, W, C)` It can also accept image latents as `image`, but
if passing latents directly it is not encoded again.
mask_image: `Union[str, bytes]`
representing an image batch to mask `image`. White pixels in the mask
are repainted while black pixels are preserved. If `mask_image` is a PIL image, it is converted to a
single channel (luminance) before use. If it's a numpy array or pytorch tensor, it should contain one
color channel (L) instead of 3, so the expected shape for pytorch tensor would be `(B, 1, H, W)`, `(B,
H, W)`, `(1, H, W)`, `(H, W)`. And for numpy array would be for `(B, H, W, 1)`, `(B, H, W)`, `(H, W,
1)`, or `(H, W)`.
prompt: `str` or `List[str]`
The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
negative_prompt (`str` or `List[str]`, *optional*):
The prompt or prompts not to guide the image generation. If not defined, one has to pass
`negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
less than `1`).
n: `int`, defaults to 1
The number of images to generate per prompt. Must be between 1 and 10.
size: `str`, defaults to None
The width*height in pixels of the generated image.
response_format: `str`, defaults to `url`
The format in which the generated images are returned. Must be one of url or b64_json.
Returns
-------
ImageList
A list of image objects.
:param prompt:
:param image:
"""
url = f"{self._base_url}/v1/images/inpainting"
params = {
"model": self._model_uid,
"prompt": prompt,
"negative_prompt": negative_prompt,
"n": n,
"size": size,
"response_format": response_format,
"kwargs": json.dumps(kwargs),
}
params = _filter_params(params)
files: List[Any] = []
for key, value in params.items():
files.append((key, (None, value)))
files.append(("image", ("image", image, "application/octet-stream")))
files.append(
(
"mask_image",
("mask_image", mask_image, "application/octet-stream"),
)
)
response = await self.session.post(url, data=files, headers=self.auth_headers)
if response.status != 200:
raise RuntimeError(
f"Failed to inpaint the images, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def ocr(self, image: Union[str, bytes], **kwargs):
url = f"{self._base_url}/v1/images/ocr"
params = {
"model": self._model_uid,
"kwargs": json.dumps(kwargs),
}
params = _filter_params(params)
files: List[Any] = []
for key, value in params.items():
files.append((key, (None, value)))
files.append(("image", ("image", image, "application/octet-stream")))
response = await self.session.post(url, data=files, headers=self.auth_headers)
if response.status != 200:
raise RuntimeError(
f"Failed to ocr the images, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
class AsyncRESTfulVideoModelHandle(AsyncRESTfulModelHandle):
async def text_to_video(
self,
prompt: str,
n: int = 1,
**kwargs,
) -> "VideoList":
"""
Creates a video by the input text.
Parameters
----------
prompt: `str` or `List[str]`
The prompt or prompts to guide video generation. If not defined, you need to pass `prompt_embeds`.
n: `int`, defaults to 1
The number of videos to generate per prompt. Must be between 1 and 10.
Returns
-------
VideoList
A list of video objects.
"""
url = f"{self._base_url}/v1/video/generations"
request_body = {
"model": self._model_uid,
"prompt": prompt,
"n": n,
"kwargs": json.dumps(kwargs),
}
response = await self.session.post(
url, json=request_body, headers=self.auth_headers
)
if response.status != 200:
raise RuntimeError(
f"Failed to create the video, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def image_to_video(
self,
image: Union[str, bytes],
prompt: str,
negative_prompt: Optional[str] = None,
n: int = 1,
**kwargs,
) -> "VideoList":
"""
Creates a video by the input image and text.
Parameters
----------
image: `Union[str, bytes]`
The input image to condition the generation on.
prompt: `str` or `List[str]`
The prompt or prompts to guide video generation. If not defined, you need to pass `prompt_embeds`.
negative_prompt (`str` or `List[str]`, *optional*):
The prompt or prompts not to guide the image generation.
n: `int`, defaults to 1
The number of videos to generate per prompt. Must be between 1 and 10.
Returns
-------
VideoList
A list of video objects.
"""
url = f"{self._base_url}/v1/video/generations/image"
params = {
"model": self._model_uid,
"prompt": prompt,
"negative_prompt": negative_prompt,
"n": n,
"kwargs": json.dumps(kwargs),
}
params = _filter_params(params)
files: List[Any] = []
for key, value in params.items():
files.append((key, (None, value)))
files.append(("image", ("image", image, "application/octet-stream")))
response = await self.session.post(url, data=files, headers=self.auth_headers)
if response.status != 200:
raise RuntimeError(
f"Failed to create the video from image, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def flf_to_video(
self,
first_frame: Union[str, bytes],
last_frame: Union[str, bytes],
prompt: str,
negative_prompt: Optional[str] = None,
n: int = 1,
**kwargs,
) -> "VideoList":
"""
Creates a video by the first frame, last frame and text.
Parameters
----------
first_frame: `Union[str, bytes]`
The first frame to condition the generation on.
last_frame: `Union[str, bytes]`
The last frame to condition the generation on.
prompt: `str` or `List[str]`
The prompt or prompts to guide video generation. If not defined, you need to pass `prompt_embeds`.
negative_prompt (`str` or `List[str]`, *optional*):
The prompt or prompts not to guide the image generation.
n: `int`, defaults to 1
The number of videos to generate per prompt. Must be between 1 and 10.
Returns
-------
VideoList
A list of video objects.
"""
url = f"{self._base_url}/v1/video/generations/flf"
params = {
"model": self._model_uid,
"prompt": prompt,
"negative_prompt": negative_prompt,
"n": n,
"kwargs": json.dumps(kwargs),
}
params = _filter_params(params)
files: List[Any] = []
for key, value in params.items():
files.append((key, (None, value)))
files.append(
("first_frame", ("image", first_frame, "application/octet-stream"))
)
files.append(("last_frame", ("image", last_frame, "application/octet-stream")))
response = await self.session.post(url, files=files, headers=self.auth_headers)
if response.status != 200:
raise RuntimeError(
f"Failed to create the video from image, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
class AsyncRESTfulGenerateModelHandle(AsyncRESTfulModelHandle):
async def generate(
self,
prompt: str,
generate_config: Optional["PytorchGenerateConfig"] = None,
) -> Union["Completion", AsyncIterator["CompletionChunk"]]:
"""
Creates a completion for the provided prompt and parameters via RESTful APIs.
Parameters
----------
prompt: str
The user's message or user's input.
generate_config: Optional["PytorchGenerateConfig"]
Additional configuration for the chat generation.
"PytorchGenerateConfig" -> Configuration for pytorch model
Returns
-------
Union["Completion", AsyncIterator["CompletionChunk"]]
Stream is a parameter in generate_config.
When stream is set to True, the function will return AsyncIterator["CompletionChunk"].
When stream is set to False, the function will return "Completion".
Raises
------
RuntimeError
Fail to generate the completion from the server. Detailed information provided in error message.
"""
url = f"{self._base_url}/v1/completions"
request_body: Dict[str, Any] = {"model": self._model_uid, "prompt": prompt}
if generate_config is not None:
for key, value in generate_config.items():
request_body[key] = value
stream = bool(generate_config and generate_config.get("stream"))
response = await self.session.post(
url, json=request_body, headers=self.auth_headers
)
if response.status != 200:
raise RuntimeError(
f"Failed to generate completion, detail: {await _get_error_string(response)}"
)
if stream:
return async_streaming_response_iterator(response.content)
response_data = await response.json()
await _release_response(response)
return response_data
class AsyncRESTfulChatModelHandle(AsyncRESTfulGenerateModelHandle):
async def chat(
self,
messages: List[Dict],
tools: Optional[List[Dict]] = None,
generate_config: Optional["PytorchGenerateConfig"] = None,
) -> Union["ChatCompletion", AsyncIterator["ChatCompletionChunk"]]:
"""
Given a list of messages comprising a conversation, the model will return a response via RESTful APIs.
Parameters
----------
messages: List[Dict]
A list of messages comprising the conversation so far.
tools: Optional[List[Dict]]
A tool list.
generate_config: Optional["PytorchGenerateConfig"]
Additional configuration for the chat generation.
"PytorchGenerateConfig" -> configuration for pytorch model
Returns
-------
Union["ChatCompletion", AsyncIterator["ChatCompletionChunk"]]
Stream is a parameter in generate_config.
When stream is set to True, the function will return AsyncIterator["ChatCompletionChunk"].
When stream is set to False, the function will return "ChatCompletion".
Raises
------
RuntimeError
Report the failure to generate the chat from the server. Detailed information provided in error message.
"""
url = f"{self._base_url}/v1/chat/completions"
request_body: Dict[str, Any] = {
"model": self._model_uid,
"messages": messages,
}
if tools is not None:
request_body["tools"] = tools
if generate_config is not None:
for key, value in generate_config.items():
request_body[key] = value
stream = bool(generate_config and generate_config.get("stream"))
response = await self.session.post(
url, json=request_body, headers=self.auth_headers
)
if response.status != 200:
raise RuntimeError(
f"Failed to generate chat completion, detail: {await _get_error_string(response)}"
)
if stream:
return async_streaming_response_iterator(response.content)
response_data = await response.json()
await _release_response(response)
return response_data
class AsyncRESTfulAudioModelHandle(AsyncRESTfulModelHandle):
async def transcriptions(
self,
audio: bytes,
language: Optional[str] = None,
prompt: Optional[str] = None,
response_format: Optional[str] = "json",
temperature: Optional[float] = 0,
timestamp_granularities: Optional[List[str]] = None,
**kwargs,
):
"""
Transcribes audio into the input language.
Parameters
----------
audio: bytes
The audio file object (not file name) to transcribe, in one of these formats: flac, mp3, mp4, mpeg,
mpga, m4a, ogg, wav, or webm.
language: Optional[str]
The language of the input audio. Supplying the input language in ISO-639-1
(https://en.wikipedia.org/wiki/List_of_ISO_639_language_codes) format will improve accuracy and latency.
prompt: Optional[str]
An optional text to guide the model's style or continue a previous audio segment.
The prompt should match the audio language.
response_format: Optional[str], defaults to json
The format of the transcript output, in one of these options: json, text, srt, verbose_json, or vtt.
temperature: Optional[float], defaults to 0
The sampling temperature, between 0 and 1. Higher values like 0.8 will make the output more random,
while lower values like 0.2 will make it more focused and deterministic.
If set to 0, the model will use log probability to automatically increase the temperature
until certain thresholds are hit.
timestamp_granularities: Optional[List[str]], default is None.
The timestamp granularities to populate for this transcription. response_format must be set verbose_json
to use timestamp granularities. Either or both of these options are supported: word, or segment.
Note: There is no additional latency for segment timestamps, but generating word timestamps incurs
additional latency.
Returns
-------
The transcribed text.
"""
url = f"{self._base_url}/v1/audio/transcriptions"
params = {
"model": self._model_uid,
"language": language,
"prompt": prompt,
"response_format": response_format,
"temperature": temperature,
"timestamp_granularities[]": timestamp_granularities,
"kwargs": json.dumps(kwargs),
}
params = _filter_params(params)
files: List[Any] = []
files.append(("file", ("file", audio, "application/octet-stream")))
response = await self.session.post(
url, data=params, files=files, headers=self.auth_headers
)
if response.status != 200:
raise RuntimeError(
f"Failed to transcribe the audio, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def translations(
self,
audio: bytes,
language: Optional[str] = None,
prompt: Optional[str] = None,
response_format: Optional[str] = "json",
temperature: Optional[float] = 0,
timestamp_granularities: Optional[List[str]] = None,
):
"""
Translates audio into English.
Parameters
----------
audio: bytes
The audio file object (not file name) to transcribe, in one of these formats: flac, mp3, mp4, mpeg,
mpga, m4a, ogg, wav, or webm.
language: Optional[str]
The language of the input audio. Supplying the input language in ISO-639-1
(https://en.wikipedia.org/wiki/List_of_ISO_639_language_codes) format will improve accuracy and latency.
prompt: Optional[str]
An optional text to guide the model's style or continue a previous audio segment.
The prompt should match the audio language.
response_format: Optional[str], defaults to json
The format of the transcript output, in one of these options: json, text, srt, verbose_json, or vtt.
temperature: Optional[float], defaults to 0
The sampling temperature, between 0 and 1. Higher values like 0.8 will make the output more random,
while lower values like 0.2 will make it more focused and deterministic.
If set to 0, the model will use log probability to automatically increase the temperature
until certain thresholds are hit.
timestamp_granularities: Optional[List[str]], default is None.
The timestamp granularities to populate for this transcription. response_format must be set verbose_json
to use timestamp granularities. Either or both of these options are supported: word, or segment.
Note: There is no additional latency for segment timestamps, but generating word timestamps incurs
additional latency.
Returns
-------
The translated text.
"""
url = f"{self._base_url}/v1/audio/translations"
params = {
"model": self._model_uid,
"language": language,
"prompt": prompt,
"response_format": response_format,
"temperature": temperature,
"timestamp_granularities[]": timestamp_granularities,
}
params = _filter_params(params)
files: List[Any] = []
files.append(("file", ("file", audio, "application/octet-stream")))
response = await self.session.post(
url, data=params, files=files, headers=self.auth_headers
)
if response.status != 200:
raise RuntimeError(
f"Failed to translate the audio, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def speech(
self,
input: str,
voice: str = "",
response_format: str = "mp3",
speed: float = 1.0,
stream: bool = False,
prompt_speech: Optional[bytes] = None,
prompt_latent: Optional[bytes] = None,
**kwargs,
):
"""
Generates audio from the input text.
Parameters
----------
input: str
The text to generate audio for. The maximum length is 4096 characters.
voice: str
The voice to use when generating the audio.
response_format: str
The format to audio in.
speed: str
The speed of the generated audio.
stream: bool
Use stream or not.
prompt_speech: bytes
The audio bytes to be provided to the model.
prompt_latent: bytes
The latent bytes to be provided to the model.
Returns
-------
bytes
The generated audio binary.
"""
url = f"{self._base_url}/v1/audio/speech"
params = {
"model": self._model_uid,
"input": input,
"voice": voice,
"response_format": response_format,
"speed": speed,
"stream": stream,
"kwargs": json.dumps(kwargs),
}
params = _filter_params(params)
files: List[Any] = []
if prompt_speech:
files.append(
(
"prompt_speech",
("prompt_speech", prompt_speech, "application/octet-stream"),
)
)
if prompt_latent:
files.append(
(
"prompt_latent",
("prompt_latent", prompt_latent, "application/octet-stream"),
)
)
if files:
response = await self.session.post(
url, data=params, files=files, headers=self.auth_headers
)
else:
response = await self.session.post(
url, json=params, headers=self.auth_headers
)
if response.status != 200:
raise RuntimeError(
f"Failed to speech the text, detail: {await _get_error_string(response)}"
)
if stream:
await _release_response(response)
return response.content.iter_chunked(1024)
await _release_response(response)
return response.content
class AsyncRESTfulFlexibleModelHandle(AsyncRESTfulModelHandle):
async def infer(
self,
**kwargs,
):
"""
Call flexible model.
Parameters
----------
kwargs: dict
The inference arguments.
Returns
-------
bytes
The inference result.
"""
url = f"{self._base_url}/v1/flexible/infers"
params: Dict = { # type: ignore
"model": self._model_uid,
}
params.update(kwargs)
response = await self.session.post(url, json=params, headers=self.auth_headers)
if response.status != 200:
raise RuntimeError(
f"Failed to predict, detail: {await _get_error_string(response)}"
)
await _release_response(response)
return response.content
class AsyncClient:
def __init__(self, base_url, api_key: Optional[str] = None):
self.base_url = base_url
self._headers: Dict[str, str] = {}
self._cluster_authed = False
self.timeout = aiohttp.ClientTimeout(total=1800)
self.session = aiohttp.ClientSession(
connector=aiohttp.TCPConnector(force_close=True), timeout=self.timeout
)
self._check_cluster_authenticated()
if api_key is not None and self._cluster_authed:
self._headers["Authorization"] = f"Bearer {api_key}"
async def close(self):
"""Close the AsyncClient session."""
if self.session:
await self.session.close()
self.session = None
def __del__(self):
if self.session:
loop = asyncio.get_event_loop()
loop.create_task(self.close())
def _set_token(self, token: Optional[str]):
if not self._cluster_authed or token is None:
return
self._headers["Authorization"] = f"Bearer {token}"
def _get_token(self) -> Optional[str]:
return (
str(self._headers["Authorization"]).replace("Bearer ", "")
if "Authorization" in self._headers
else None
)
def _check_cluster_authenticated(self):
import requests
session = requests.Session()
url = f"{self.base_url}/v1/cluster/auth"
response = session.get(url)
# compatible with old version of xinference
if response.status_code == 404:
self._cluster_authed = False
else:
if response.status_code != 200:
response_data = response.json()
raise RuntimeError(
f"Failed to get cluster information, detail: {response_data['detail']}"
)
response_data = response.json()
self._cluster_authed = bool(response_data["auth"])
async def vllm_models(self) -> Dict[str, Any]:
url = f"{self.base_url}/v1/models/vllm-supported"
response = await self.session.get(url, headers=self._headers)
if response.status != 200:
response_data = await response.json()
await _release_response(response)
raise RuntimeError(
f"Failed to get cluster information, detail: {response_data['detail']}"
)
try:
response_data = await response.json()
await _release_response(response)
return response_data
except Exception as e:
raise RuntimeError(f"Error parsing JSON response: {e}")
async def login(self, username: str, password: str):
if not self._cluster_authed:
return
url = f"{self.base_url}/token"
payload = {"username": username, "password": password}
response = await self.session.post(url, json=payload)
if response.status != 200:
response_data = await response.json()
await _release_response(response)
raise RuntimeError(
f"Failed to get cluster information, detail: {response_data['detail']}"
)
response_data = await response.json()
await _release_response(response)
# Only bearer token for now
access_token = response_data["access_token"]
self._headers["Authorization"] = f"Bearer {access_token}"
async def list_models(self) -> Dict[str, Dict[str, Any]]:
"""
Retrieve the model specifications from the Server.
Returns
-------
Dict[str, Dict[str, Any]]
The collection of model specifications with their names on the server.
"""
url = f"{self.base_url}/v1/models"
response = await self.session.get(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Failed to list model, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
model_list = response_data["data"]
return {item["id"]: item for item in model_list}
async def launch_model(
self,
model_name: str,
model_type: str = "LLM",
model_engine: Optional[str] = None,
model_uid: Optional[str] = None,
model_size_in_billions: Optional[Union[int, str, float]] = None,
model_format: Optional[str] = None,
quantization: Optional[str] = None,
replica: int = 1,
n_worker: int = 1,
n_gpu: Optional[Union[int, str]] = "auto",
peft_model_config: Optional[Dict] = None,
request_limits: Optional[int] = None,
worker_ip: Optional[str] = None,
gpu_idx: Optional[Union[int, List[int]]] = None,
model_path: Optional[str] = None,
**kwargs,
) -> str:
"""
Launch the model based on the parameters on the server via RESTful APIs.
Parameters
----------
model_name: str
The name of model.
model_type: str
type of model.
model_engine: Optional[str]
Specify the inference engine of the model when launching LLM.
model_uid: str
UID of model, auto generate a UUID if is None.
model_size_in_billions: Optional[Union[int, str, float]]
The size (in billions) of the model.
model_format: Optional[str]
The format of the model.
quantization: Optional[str]
The quantization of model.
replica: Optional[int]
The replica of model, default is 1.
n_worker: int
Number of workers to run.
n_gpu: Optional[Union[int, str]],
The number of GPUs used by the model, default is "auto". If n_worker>1, means number of GPUs per worker.
``n_gpu=None`` means cpu only, ``n_gpu=auto`` lets the system automatically determine the best number of GPUs to use.
peft_model_config: Optional[Dict]
- "lora_list": A List of PEFT (Parameter-Efficient Fine-Tuning) model and path.
- "image_lora_load_kwargs": A Dict of lora load parameters for image model
- "image_lora_fuse_kwargs": A Dict of lora fuse parameters for image model
request_limits: Optional[int]
The number of request limits for this model, default is None.
``request_limits=None`` means no limits for this model.
worker_ip: Optional[str]
Specify the worker ip where the model is located in a distributed scenario.
gpu_idx: Optional[Union[int, List[int]]]
Specify the GPU index where the model is located.
model_path: Optional[str]
Model path, if gguf format, should be the file path, otherwise, should be directory of the model.
**kwargs:
Any other parameters been specified. e.g. multimodal_projector for multimodal inference with the llama.cpp backend.
Returns
-------
str
The unique model_uid for the launched model.
"""
url = f"{self.base_url}/v1/models"
# convert float to int or string since the RESTful API does not accept float.
if isinstance(model_size_in_billions, float):
model_size_in_billions = convert_float_to_int_or_str(model_size_in_billions)
payload = {
"model_uid": model_uid,
"model_name": model_name,
"model_engine": model_engine,
"peft_model_config": peft_model_config,
"model_type": model_type,
"model_size_in_billions": model_size_in_billions,
"model_format": model_format,
"quantization": quantization,
"replica": replica,
"n_worker": n_worker,
"n_gpu": n_gpu,
"request_limits": request_limits,
"worker_ip": worker_ip,
"gpu_idx": gpu_idx,
"model_path": model_path,
}
wait_ready = kwargs.pop("wait_ready", True)
for key, value in kwargs.items():
payload[str(key)] = value
if wait_ready:
response = await self.session.post(url, json=payload, headers=self._headers)
else:
response = await self.session.post(
url, json=payload, headers=self._headers, params={"wait_ready": False}
)
if response.status != 200:
raise RuntimeError(
f"Failed to launch model, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data["model_uid"]
async def terminate_model(self, model_uid: str):
"""
Terminate the specific model running on the server.
Parameters
----------
model_uid: str
The unique id that identify the model we want.
Raises
------
RuntimeError
Report failure to get the wanted model with given model_uid. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/models/{model_uid}"
response = await self.session.delete(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Failed to terminate model, detail: {await _get_error_string(response)}"
)
await _release_response(response)
async def get_launch_model_progress(self, model_uid: str) -> dict:
"""
Get progress of the specific model.
Parameters
----------
model_uid: str
The unique id that identify the model we want.
Returns
-------
result: dict
Result that contains progress.
Raises
------
RuntimeError
Report failure to get the wanted model with given model_uid. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/models/{model_uid}/progress"
response = await self.session.get(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Fail to get model launching progress, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def cancel_launch_model(self, model_uid: str):
"""
Cancel launching model.
Parameters
----------
model_uid: str
The unique id that identify the model we want.
Raises
------
RuntimeError
Report failure to get the wanted model with given model_uid. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/models/{model_uid}/cancel"
response = await self.session.post(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Fail to cancel launching model, detail: {await _get_error_string(response)}"
)
await _release_response(response)
async def get_instance_info(self, model_name: str, model_uid: str):
url = f"{self.base_url}/v1/models/instances"
response = await self.session.get(
url,
headers=self._headers,
params={"model_name": model_name, "model_uid": model_uid},
)
if response.status != 200:
raise RuntimeError("Failed to get instance info")
response_data = await response.json()
await _release_response(response)
return response_data
async def _get_supervisor_internal_address(self):
url = f"{self.base_url}/v1/address"
response = await self.session.get(url, headers=self._headers)
if response.status != 200:
raise RuntimeError("Failed to get supervisor internal address")
response_data = await response.json()
await _release_response(response)
return response_data
async def get_model(self, model_uid: str) -> AsyncRESTfulModelHandle:
"""
Launch the model based on the parameters on the server via RESTful APIs.
Parameters
----------
model_uid: str
The unique id that identify the model.
Returns
-------
ModelHandle
The corresponding Model Handler based on the Model specified in the uid:
- :obj:`xinference.client.handlers.GenerateModelHandle` -> provide handle to basic generate Model. e.g. Baichuan.
- :obj:`xinference.client.handlers.ChatModelHandle` -> provide handle to chat Model. e.g. Baichuan-chat.
Raises
------
RuntimeError
Report failure to get the wanted model with given model_uid. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/models/{model_uid}"
response = await self.session.get(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Failed to get the model description, detail: {await _get_error_string(response)}"
)
desc = await response.json()
await _release_response(response)
if desc["model_type"] == "LLM":
if "chat" in desc["model_ability"]:
return AsyncRESTfulChatModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
elif "generate" in desc["model_ability"]:
return AsyncRESTfulGenerateModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
else:
raise ValueError(f"Unrecognized model ability: {desc['model_ability']}")
elif desc["model_type"] == "embedding":
return AsyncRESTfulEmbeddingModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
elif desc["model_type"] == "image":
return AsyncRESTfulImageModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
elif desc["model_type"] == "rerank":
return AsyncRESTfulRerankModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
elif desc["model_type"] == "audio":
return AsyncRESTfulAudioModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
elif desc["model_type"] == "video":
return AsyncRESTfulVideoModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
elif desc["model_type"] == "flexible":
return AsyncRESTfulFlexibleModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
else:
raise ValueError(f"Unknown model type:{desc['model_type']}")
async def describe_model(self, model_uid: str):
"""
Get model information via RESTful APIs.
Parameters
----------
model_uid: str
The unique id that identify the model.
Returns
-------
dict
A dictionary containing the following keys:
- "model_type": str
the type of the model determined by its function, e.g. "LLM" (Large Language Model)
- "model_name": str
the name of the specific LLM model family
- "model_lang": List[str]
the languages supported by the LLM model
- "model_ability": List[str]
the ability or capabilities of the LLM model
- "model_description": str
a detailed description of the LLM model
- "model_format": str
the format specification of the LLM model
- "model_size_in_billions": int
the size of the LLM model in billions
- "quantization": str
the quantization applied to the model
- "revision": str
the revision number of the LLM model specification
- "context_length": int
the maximum text length the LLM model can accommodate (include all input & output)
Raises
------
RuntimeError
Report failure to get the wanted model with given model_uid. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/models/{model_uid}"
response = await self.session.get(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Failed to get the model description, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def register_model(
self,
model_type: str,
model: str,
persist: bool,
worker_ip: Optional[str] = None,
):
"""
Register a custom model.
Parameters
----------
model_type: str
The type of model.
model: str
The model definition. (refer to: https://inference.readthedocs.io/en/latest/models/custom.html)
worker_ip: Optional[str]
The IP address of the worker on which the model is running.
persist: bool
Raises
------
RuntimeError
Report failure to register the custom model. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/model_registrations/{model_type}"
request_body = {"model": model, "worker_ip": worker_ip, "persist": persist}
response = await self.session.post(
url, json=request_body, headers=self._headers
)
if response.status != 200:
raise RuntimeError(
f"Failed to register model, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def unregister_model(self, model_type: str, model_name: str):
"""
Unregister a custom model.
Parameters
----------
model_type: str
The type of model.
model_name: str
The name of the model
Raises
------
RuntimeError
Report failure to unregister the custom model. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/model_registrations/{model_type}/{model_name}"
response = await self.session.delete(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Failed to register model, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def list_model_registrations(self, model_type: str) -> List[Dict[str, Any]]:
"""
List models registered on the server.
Parameters
----------
model_type: str
The type of the model.
Returns
-------
List[Dict[str, Any]]
The collection of registered models on the server.
Raises
------
RuntimeError
Report failure to list model registration. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/model_registrations/{model_type}"
response = await self.session.get(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Failed to list model registration, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def list_cached_models(
self, model_name: Optional[str] = None, worker_ip: Optional[str] = None
) -> List[Dict[Any, Any]]:
"""
Get a list of cached models.
Parameters
----------
model_name: Optional[str]
The name of model.
worker_ip: Optional[str]
Specify the worker ip where the model is located in a distributed scenario.
Returns
-------
List[Dict[Any, Any]]
The collection of cached models on the server.
Raises
------
RuntimeError
Raised when the request fails, including the reason for the failure.
"""
url = f"{self.base_url}/v1/cache/models"
params = {
"model_name": model_name,
"worker_ip": worker_ip,
}
params = _filter_params(params)
response = await self.session.get(url, headers=self._headers, params=params)
if response.status != 200:
raise RuntimeError(
f"Failed to list cached model, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data.get("list")
async def list_deletable_models(
self, model_version: str, worker_ip: Optional[str] = None
) -> Dict[str, Any]:
"""
Get the cached models with the model path cached on the server.
Parameters
----------
model_version: str
The version of the model.
worker_ip: Optional[str]
Specify the worker ip where the model is located in a distributed scenario.
Returns
-------
Dict[str, Dict[str,str]]]
Dictionary with keys "model_name" and values model_file_location.
"""
url = f"{self.base_url}/v1/cache/models/files"
params = {
"model_version": model_version,
"worker_ip": worker_ip,
}
params = _filter_params(params)
response = await self.session.get(url, headers=self._headers, params=params)
if response.status != 200:
raise RuntimeError(
f"Failed to get paths by model name, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def confirm_and_remove_model(
self, model_version: str, worker_ip: Optional[str] = None
) -> bool:
"""
Remove the cached models with the model name cached on the server.
Parameters
----------
model_version: str
The version of the model.
worker_ip: Optional[str]
Specify the worker ip where the model is located in a distributed scenario.
Returns
-------
str
The response of the server.
"""
url = f"{self.base_url}/v1/cache/models"
params = {
"model_version": model_version,
"worker_ip": worker_ip,
}
params = _filter_params(params)
response = await self.session.delete(url, headers=self._headers, params=params)
if response.status != 200:
raise RuntimeError(
f"Failed to remove cached models, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data.get("result", False)
async def get_model_registration(
self, model_type: str, model_name: str
) -> Dict[str, Any]:
"""
Get the model with the model type and model name registered on the server.
Parameters
----------
model_type: str
The type of the model.
model_name: str
The name of the model.
Returns
-------
List[Dict[str, Any]]
The collection of registered models on the server.
"""
url = f"{self.base_url}/v1/model_registrations/{model_type}/{model_name}"
response = await self.session.get(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Failed to list model registration, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def query_engine_by_model_name(
self, model_name: str, model_type: Optional[str] = "LLM"
):
"""
Get the engine parameters with the model name registered on the server.
Parameters
----------
model_name: str
The name of the model.
model_type: str
Model type, LLM by default.
Returns
-------
Dict[str, List[Dict[str, Any]]]
The supported engine parameters of registered models on the server.
"""
if not model_type:
url = f"{self.base_url}/v1/engines/{model_name}"
else:
url = f"{self.base_url}/v1/engines/{model_type}/{model_name}"
response = await self.session.get(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Failed to query engine parameters by model name, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def abort_request(
self, model_uid: str, request_id: str, block_duration: int = 30
):
"""
Abort a request.
Abort a submitted request. If the request is finished or not found, this method will be a no-op.
Currently, this interface is only supported when batching is enabled for models on transformers backend.
Parameters
----------
model_uid: str
Model uid.
request_id: str
Request id.
block_duration: int
The duration to make the request id abort. If set to 0, the abort_request will be immediate, which may
prevent it from taking effect if it arrives before the request operation.
Returns
-------
Dict
Return empty dict.
"""
url = f"{self.base_url}/v1/models/{model_uid}/requests/{request_id}/abort"
response = await self.session.post(
url, headers=self._headers, json={"block_duration": block_duration}
)
if response.status != 200:
raise RuntimeError(
f"Failed to abort request, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def get_workers_info(self):
url = f"{self.base_url}/v1/workers"
response = await self.session.get(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Failed to get workers info, detail: {await _get_error_string(response)}"
)
response_data = await response.json()
await _release_response(response)
return response_data
async def get_supervisor_info(self):
url = f"{self.base_url}/v1/supervisor"
response = await self.session.get(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Failed to get supervisor info, detail: {await _get_error_string(response)}"
)
response_json = await response.json()
await _release_response(response)
return response_json
async def get_progress(self, request_id: str):
url = f"{self.base_url}/v1/requests/{request_id}/progress"
response = await self.session.get(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Failed to get progress, detail: {await _get_error_string(response)}"
)
response_json = await response.json()
await _release_response(response)
return response_json
async def abort_cluster(self):
url = f"{self.base_url}/v1/clusters"
response = await self.session.delete(url, headers=self._headers)
if response.status != 200:
raise RuntimeError(
f"Failed to abort cluster, detail: {await _get_error_string(response)}"
)
response_json = await response.json()
await _release_response(response)
return response_json
================================================
FILE: src/backend/bisheng/core/ai/rerank/xinference/client/restful/restful_client.py
================================================
# Copyright 2022-2023 XProbe Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import json
from typing import TYPE_CHECKING, Any, Dict, Iterator, List, Optional, Union
import requests
from ..common import convert_float_to_int_or_str, streaming_response_iterator
if TYPE_CHECKING:
from ...types import (
ChatCompletion,
ChatCompletionChunk,
Completion,
CompletionChunk,
Embedding,
ImageList,
PytorchGenerateConfig,
VideoList,
)
def _get_error_string(response: requests.Response) -> str:
try:
if response.content:
return response.json()["detail"]
except Exception:
pass
try:
response.raise_for_status()
except requests.HTTPError as e:
return str(e)
return "Unknown error"
class RESTfulModelHandle:
"""
A sync model interface (for RESTful client) which provides type hints that makes it much easier to use xinference
programmatically.
"""
def __init__(self, model_uid: str, base_url: str, auth_headers: Dict):
self._model_uid = model_uid
self._base_url = base_url
self.auth_headers = auth_headers
self.session = requests.Session()
def close(self):
"""
Close the session.
"""
if self.session:
self.session.close()
self.session = None
def __del__(self):
if self.session:
self.close()
class RESTfulEmbeddingModelHandle(RESTfulModelHandle):
def create_embedding(self, input: Union[str, List[str]], **kwargs) -> "Embedding":
"""
Create an Embedding from user input via RESTful APIs.
Parameters
----------
input: Union[str, List[str]]
Input text to embed, encoded as a string or array of tokens.
To embed multiple inputs in a single request, pass an array of strings or array of token arrays.
Returns
-------
Embedding
The resulted Embedding vector that can be easily consumed by machine learning models and algorithms.
Raises
------
RuntimeError
Report the failure of embeddings and provide the error message.
"""
url = f"{self._base_url}/v1/embeddings"
request_body = {
"model": self._model_uid,
"input": input,
}
request_body.update(kwargs)
response = self.session.post(url, json=request_body, headers=self.auth_headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to create the embeddings, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def convert_ids_to_tokens(
self, input: Union[List, List[List]], **kwargs
) -> List[str]:
"""
Convert token IDs to human readable tokens via RESTful APIs.
Parameters
----------
input: Union[List, List[List]]
Input token IDs to convert, can be a single list of token IDs or a list of token ID lists.
To convert multiple sequences in a single request, pass a list of token ID lists.
Returns
-------
list
A list of decoded tokens in human readable format.
Raises
------
RuntimeError
Report the failure of token conversion and provide the error message.
"""
url = f"{self._base_url}/v1/convert_ids_to_tokens"
request_body = {
"model": self._model_uid,
"input": input,
}
request_body.update(kwargs)
response = self.session.post(url, json=request_body, headers=self.auth_headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to decode token ids, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
class RESTfulRerankModelHandle(RESTfulModelHandle):
def rerank(
self,
documents: List[str],
query: str,
top_n: Optional[int] = None,
max_chunks_per_doc: Optional[int] = None,
return_documents: Optional[bool] = None,
return_len: Optional[bool] = None,
**kwargs,
):
"""
Returns an ordered list of documents ordered by their relevance to the provided query.
Parameters
----------
query: str
The search query
documents: List[str]
The documents to rerank
top_n: int
The number of results to return, defaults to returning all results
max_chunks_per_doc: int
The maximum number of chunks derived from a document
return_documents: bool
if return documents
return_len: bool
if return tokens len
Returns
-------
Scores
The scores of documents ordered by their relevance to the provided query
Raises
------
RuntimeError
Report the failure of rerank and provide the error message.
"""
url = f"{self._base_url}/v1/rerank"
request_body = {
"model": self._model_uid,
"documents": documents,
"query": query,
"top_n": top_n,
"max_chunks_per_doc": max_chunks_per_doc,
"return_documents": return_documents,
"return_len": return_len,
"kwargs": json.dumps(kwargs),
}
request_body.update(kwargs)
response = self.session.post(url, json=request_body, headers=self.auth_headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to rerank documents, detail: {response.json()['detail']}"
)
response_data = response.json()
return response_data
class RESTfulImageModelHandle(RESTfulModelHandle):
def text_to_image(
self,
prompt: str,
n: int = 1,
size: str = "1024*1024",
response_format: str = "url",
**kwargs,
) -> "ImageList":
"""
Creates an image by the input text.
Parameters
----------
prompt: `str` or `List[str]`
The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
n: `int`, defaults to 1
The number of images to generate per prompt. Must be between 1 and 10.
size: `str`, defaults to `1024*1024`
The width*height in pixels of the generated image. Must be one of 256x256, 512x512, or 1024x1024.
response_format: `str`, defaults to `url`
The format in which the generated images are returned. Must be one of url or b64_json.
Returns
-------
ImageList
A list of image objects.
"""
url = f"{self._base_url}/v1/images/generations"
request_body = {
"model": self._model_uid,
"prompt": prompt,
"n": n,
"size": size,
"response_format": response_format,
"kwargs": json.dumps(kwargs),
}
response = self.session.post(url, json=request_body, headers=self.auth_headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to create the images, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def image_to_image(
self,
image: Union[str, bytes, List[Union[str, bytes]]],
prompt: str,
negative_prompt: Optional[str] = None,
n: int = 1,
size: Optional[str] = None,
response_format: str = "url",
**kwargs,
) -> "ImageList":
"""
Creates an image by the input text.
Parameters
----------
image: `Union[str, bytes, List[Union[str, bytes]]]`
The ControlNet input condition to provide guidance to the `unet` for generation. If the type is
specified as `torch.FloatTensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be
accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height
and/or width are passed, `image` is resized accordingly. If multiple ControlNets are specified in
`init`, images must be passed as a list such that each element of the list can be correctly batched for
input to a single ControlNet.
prompt: `str` or `List[str]`
The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
negative_prompt (`str` or `List[str]`, *optional*):
The prompt or prompts not to guide the image generation. If not defined, one has to pass
`negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
less than `1`).
n: `int`, defaults to 1
The number of images to generate per prompt. Must be between 1 and 10.
size: `str`, defaults to `1024*1024`
The width*height in pixels of the generated image. Must be one of 256x256, 512x512, or 1024x1024.
response_format: `str`, defaults to `url`
The format in which the generated images are returned. Must be one of url or b64_json.
Returns
-------
ImageList
A list of image objects.
:param prompt:
:param image:
"""
url = f"{self._base_url}/v1/images/variations"
params = {
"model": self._model_uid,
"prompt": prompt,
"negative_prompt": negative_prompt,
"n": n,
"size": size,
"response_format": response_format,
"kwargs": json.dumps(kwargs),
}
files: List[Any] = []
for key, value in params.items():
files.append((key, (None, value)))
# Handle both single image and list of images
if isinstance(image, list):
if len(image) == 0:
raise ValueError("Image list cannot be empty")
elif len(image) == 1:
# Single image in list, use it directly
files.append(("image", ("image", image[0], "application/octet-stream")))
else:
# Multiple images - send all images with same field name
# FastAPI will collect them into a list
for img_data in image:
files.append(
("image", ("image", img_data, "application/octet-stream"))
)
else:
# Single image
files.append(("image", ("image", image, "application/octet-stream")))
response = self.session.post(url, files=files, headers=self.auth_headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to variants the images, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def image_edit(
self,
image: Union[Union[str, bytes], List[Union[str, bytes]]],
prompt: str,
mask: Optional[Union[str, bytes]] = None,
n: int = 1,
size: Optional[str] = None,
response_format: str = "url",
**kwargs,
) -> "ImageList":
"""
Edit image(s) by the input text and optional mask.
Parameters
----------
image: `Union[Union[str, bytes], List[Union[str, bytes]]]`
The input image(s) to edit. Can be:
- Single image: file path, URL, or binary image data
- Multiple images: list of file paths, URLs, or binary image data
When multiple images are provided, the first image is used as the primary image
and subsequent images are used as reference images for better editing results.
prompt: `str`
The prompt or prompts to guide image editing. If not defined, you need to pass `prompt_embeds`.
mask: `Optional[Union[str, bytes]]`, optional
An optional mask image. White pixels in the mask are repainted while black pixels are preserved.
If provided, this will trigger inpainting mode. If not provided, this will trigger image-to-image mode.
n: `int`, defaults to 1
The number of images to generate per prompt. Must be between 1 and 10.
size: `Optional[str]`, optional
The width*height in pixels of the generated image. If not specified, uses the original image size.
response_format: `str`, defaults to `url`
The format in which the generated images are returned. Must be one of url or b64_json.
**kwargs
Additional parameters to pass to the model.
Returns
-------
ImageList
A list of edited image objects.
Raises
------
RuntimeError
If the image editing request fails.
Examples
--------
# Single image editing
result = model.image_edit(
image="path/to/image.png",
prompt="make this image look like a painting"
)
# Multiple image editing with reference images
result = model.image_edit(
image=["primary_image.png", "reference1.jpg", "reference2.png"],
prompt="edit the main image using the style from reference images"
)
"""
url = f"{self._base_url}/v1/images/edits"
params = {
"model": self._model_uid,
"prompt": prompt,
"n": n,
"size": size,
"response_format": response_format,
"kwargs": json.dumps(kwargs),
}
files: List[Any] = []
for key, value in params.items():
if value is not None:
files.append((key, (None, value)))
# Handle single image or multiple images using requests format
if isinstance(image, list):
# Validate image list is not empty
if len(image) == 0:
raise ValueError("Image list cannot be empty")
# Multiple images - send as image[] array
for i, img in enumerate(image):
if isinstance(img, str):
# File path - open file
f = open(img, "rb")
files.append(
(f"image[]", (f"image_{i}", f, "application/octet-stream"))
)
else:
# Binary data
files.append(
(f"image[]", (f"image_{i}", img, "application/octet-stream"))
)
else:
# Single image
if isinstance(image, str):
# File path - open file
f = open(image, "rb")
files.append(("image", ("image", f, "application/octet-stream")))
else:
# Binary data
files.append(("image", ("image", image, "application/octet-stream")))
if mask is not None:
if isinstance(mask, str):
# File path - open file
f = open(mask, "rb")
files.append(("mask", ("mask", f, "application/octet-stream")))
else:
# Binary data
files.append(("mask", ("mask", mask, "application/octet-stream")))
try:
response = self.session.post(url, files=files, headers=self.auth_headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to edit the images, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
finally:
# Close all opened files
for file_item in files:
if (
len(file_item) >= 2
and hasattr(file_item[1], "__len__")
and len(file_item[1]) >= 2
):
file_obj = file_item[1][1]
if hasattr(file_obj, "close"):
file_obj.close()
def inpainting(
self,
image: Union[str, bytes],
mask_image: Union[str, bytes],
prompt: str,
negative_prompt: Optional[str] = None,
n: int = 1,
size: Optional[str] = None,
response_format: str = "url",
**kwargs,
) -> "ImageList":
"""
Inpaint an image by the input text.
Parameters
----------
image: `Union[str, bytes]`
an image batch to be inpainted (which parts of the image to
be masked out with `mask_image` and repainted according to `prompt`). For both numpy array and pytorch
tensor, the expected value range is between `[0, 1]` If it's a tensor or a list or tensors, the
expected shape should be `(B, C, H, W)` or `(C, H, W)`. If it is a numpy array or a list of arrays, the
expected shape should be `(B, H, W, C)` or `(H, W, C)` It can also accept image latents as `image`, but
if passing latents directly it is not encoded again.
mask_image: `Union[str, bytes]`
representing an image batch to mask `image`. White pixels in the mask
are repainted while black pixels are preserved. If `mask_image` is a PIL image, it is converted to a
single channel (luminance) before use. If it's a numpy array or pytorch tensor, it should contain one
color channel (L) instead of 3, so the expected shape for pytorch tensor would be `(B, 1, H, W)`, `(B,
H, W)`, `(1, H, W)`, `(H, W)`. And for numpy array would be for `(B, H, W, 1)`, `(B, H, W)`, `(H, W,
1)`, or `(H, W)`.
prompt: `str` or `List[str]`
The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
negative_prompt (`str` or `List[str]`, *optional*):
The prompt or prompts not to guide the image generation. If not defined, one has to pass
`negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
less than `1`).
n: `int`, defaults to 1
The number of images to generate per prompt. Must be between 1 and 10.
size: `str`, defaults to None
The width*height in pixels of the generated image.
response_format: `str`, defaults to `url`
The format in which the generated images are returned. Must be one of url or b64_json.
Returns
-------
ImageList
A list of image objects.
:param prompt:
:param image:
"""
url = f"{self._base_url}/v1/images/inpainting"
params = {
"model": self._model_uid,
"prompt": prompt,
"negative_prompt": negative_prompt,
"n": n,
"size": size,
"response_format": response_format,
"kwargs": json.dumps(kwargs),
}
files: List[Any] = []
for key, value in params.items():
files.append((key, (None, value)))
files.append(("image", ("image", image, "application/octet-stream")))
files.append(
("mask_image", ("mask_image", mask_image, "application/octet-stream"))
)
response = self.session.post(url, files=files, headers=self.auth_headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to inpaint the images, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def ocr(self, image: Union[str, bytes], **kwargs):
url = f"{self._base_url}/v1/images/ocr"
params = {
"model": self._model_uid,
"kwargs": json.dumps(kwargs),
}
files: List[Any] = []
for key, value in params.items():
files.append((key, (None, value)))
files.append(("image", ("image", image, "application/octet-stream")))
response = self.session.post(url, files=files, headers=self.auth_headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to ocr the images, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
class RESTfulVideoModelHandle(RESTfulModelHandle):
def text_to_video(
self,
prompt: str,
n: int = 1,
**kwargs,
) -> "VideoList":
"""
Creates a video by the input text.
Parameters
----------
prompt: `str` or `List[str]`
The prompt or prompts to guide video generation. If not defined, you need to pass `prompt_embeds`.
n: `int`, defaults to 1
The number of videos to generate per prompt. Must be between 1 and 10.
Returns
-------
VideoList
A list of video objects.
"""
url = f"{self._base_url}/v1/video/generations"
request_body = {
"model": self._model_uid,
"prompt": prompt,
"n": n,
"kwargs": json.dumps(kwargs),
}
response = self.session.post(url, json=request_body, headers=self.auth_headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to create the video, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def image_to_video(
self,
image: Union[str, bytes],
prompt: str,
negative_prompt: Optional[str] = None,
n: int = 1,
**kwargs,
) -> "VideoList":
"""
Creates a video by the input image and text.
Parameters
----------
image: `Union[str, bytes]`
The input image to condition the generation on.
prompt: `str` or `List[str]`
The prompt or prompts to guide video generation. If not defined, you need to pass `prompt_embeds`.
negative_prompt (`str` or `List[str]`, *optional*):
The prompt or prompts not to guide the image generation.
n: `int`, defaults to 1
The number of videos to generate per prompt. Must be between 1 and 10.
Returns
-------
VideoList
A list of video objects.
"""
url = f"{self._base_url}/v1/video/generations/image"
params = {
"model": self._model_uid,
"prompt": prompt,
"negative_prompt": negative_prompt,
"n": n,
"kwargs": json.dumps(kwargs),
}
files: List[Any] = []
for key, value in params.items():
files.append((key, (None, value)))
files.append(("image", ("image", image, "application/octet-stream")))
response = self.session.post(url, files=files, headers=self.auth_headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to create the video from image, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def flf_to_video(
self,
first_frame: Union[str, bytes],
last_frame: Union[str, bytes],
prompt: str,
negative_prompt: Optional[str] = None,
n: int = 1,
**kwargs,
) -> "VideoList":
"""
Creates a video by the first frame, last frame and text.
Parameters
----------
first_frame: `Union[str, bytes]`
The first frame to condition the generation on.
last_frame: `Union[str, bytes]`
The last frame to condition the generation on.
prompt: `str` or `List[str]`
The prompt or prompts to guide video generation. If not defined, you need to pass `prompt_embeds`.
negative_prompt (`str` or `List[str]`, *optional*):
The prompt or prompts not to guide the image generation.
n: `int`, defaults to 1
The number of videos to generate per prompt. Must be between 1 and 10.
Returns
-------
VideoList
A list of video objects.
"""
url = f"{self._base_url}/v1/video/generations/flf"
params = {
"model": self._model_uid,
"prompt": prompt,
"negative_prompt": negative_prompt,
"n": n,
"kwargs": json.dumps(kwargs),
}
files: List[Any] = []
for key, value in params.items():
files.append((key, (None, value)))
files.append(
("first_frame", ("image", first_frame, "application/octet-stream"))
)
files.append(("last_frame", ("image", last_frame, "application/octet-stream")))
response = self.session.post(url, files=files, headers=self.auth_headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to create the video from image, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
class RESTfulGenerateModelHandle(RESTfulModelHandle):
def generate(
self,
prompt: str,
generate_config: Optional["PytorchGenerateConfig"] = None,
) -> Union["Completion", Iterator["CompletionChunk"]]:
"""
Creates a completion for the provided prompt and parameters via RESTful APIs.
Parameters
----------
prompt: str
The user's message or user's input.
generate_config: Optional["PytorchGenerateConfig"]
Additional configuration for the chat generation.
"PytorchGenerateConfig" -> Configuration for pytorch model
Returns
-------
Union["Completion", Iterator["CompletionChunk"]]
Stream is a parameter in generate_config.
When stream is set to True, the function will return Iterator["CompletionChunk"].
When stream is set to False, the function will return "Completion".
Raises
------
RuntimeError
Fail to generate the completion from the server. Detailed information provided in error message.
"""
url = f"{self._base_url}/v1/completions"
request_body: Dict[str, Any] = {"model": self._model_uid, "prompt": prompt}
if generate_config is not None:
for key, value in generate_config.items():
request_body[key] = value
stream = bool(generate_config and generate_config.get("stream"))
response = self.session.post(
url, json=request_body, stream=stream, headers=self.auth_headers
)
if response.status_code != 200:
raise RuntimeError(
f"Failed to generate completion, detail: {_get_error_string(response)}"
)
if stream:
return streaming_response_iterator(response.iter_lines())
response_data = response.json()
return response_data
class RESTfulChatModelHandle(RESTfulGenerateModelHandle):
def chat(
self,
messages: List[Dict],
tools: Optional[List[Dict]] = None,
generate_config: Optional["PytorchGenerateConfig"] = None,
) -> Union["ChatCompletion", Iterator["ChatCompletionChunk"]]:
"""
Given a list of messages comprising a conversation, the model will return a response via RESTful APIs.
Parameters
----------
messages: List[Dict]
A list of messages comprising the conversation so far.
tools: Optional[List[Dict]]
A tool list.
generate_config: Optional["PytorchGenerateConfig"]
Additional configuration for the chat generation.
"PytorchGenerateConfig" -> configuration for pytorch model
Returns
-------
Union["ChatCompletion", Iterator["ChatCompletionChunk"]]
Stream is a parameter in generate_config.
When stream is set to True, the function will return Iterator["ChatCompletionChunk"].
When stream is set to False, the function will return "ChatCompletion".
Raises
------
RuntimeError
Report the failure to generate the chat from the server. Detailed information provided in error message.
"""
url = f"{self._base_url}/v1/chat/completions"
request_body: Dict[str, Any] = {
"model": self._model_uid,
"messages": messages,
}
if tools is not None:
request_body["tools"] = tools
if generate_config is not None:
for key, value in generate_config.items():
request_body[key] = value
stream = bool(generate_config and generate_config.get("stream"))
response = self.session.post(
url, json=request_body, stream=stream, headers=self.auth_headers
)
if response.status_code != 200:
raise RuntimeError(
f"Failed to generate chat completion, detail: {_get_error_string(response)}"
)
if stream:
return streaming_response_iterator(response.iter_lines())
response_data = response.json()
return response_data
class RESTfulAudioModelHandle(RESTfulModelHandle):
def transcriptions(
self,
audio: bytes,
language: Optional[str] = None,
prompt: Optional[str] = None,
response_format: Optional[str] = "json",
temperature: Optional[float] = 0,
timestamp_granularities: Optional[List[str]] = None,
**kwargs,
):
"""
Transcribes audio into the input language.
Parameters
----------
audio: bytes
The audio file object (not file name) to transcribe, in one of these formats: flac, mp3, mp4, mpeg,
mpga, m4a, ogg, wav, or webm.
language: Optional[str]
The language of the input audio. Supplying the input language in ISO-639-1
(https://en.wikipedia.org/wiki/List_of_ISO_639_language_codes) format will improve accuracy and latency.
prompt: Optional[str]
An optional text to guide the model's style or continue a previous audio segment.
The prompt should match the audio language.
response_format: Optional[str], defaults to json
The format of the transcript output, in one of these options: json, text, srt, verbose_json, or vtt.
temperature: Optional[float], defaults to 0
The sampling temperature, between 0 and 1. Higher values like 0.8 will make the output more random,
while lower values like 0.2 will make it more focused and deterministic.
If set to 0, the model will use log probability to automatically increase the temperature
until certain thresholds are hit.
timestamp_granularities: Optional[List[str]], default is None.
The timestamp granularities to populate for this transcription. response_format must be set verbose_json
to use timestamp granularities. Either or both of these options are supported: word, or segment.
Note: There is no additional latency for segment timestamps, but generating word timestamps incurs
additional latency.
Returns
-------
The transcribed text.
"""
url = f"{self._base_url}/v1/audio/transcriptions"
params = {
"model": self._model_uid,
"language": language,
"prompt": prompt,
"response_format": response_format,
"temperature": temperature,
"timestamp_granularities[]": timestamp_granularities,
"kwargs": json.dumps(kwargs),
}
files: List[Any] = []
files.append(("file", ("file", audio, "application/octet-stream")))
response = self.session.post(
url, data=params, files=files, headers=self.auth_headers
)
if response.status_code != 200:
raise RuntimeError(
f"Failed to transcribe the audio, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def translations(
self,
audio: bytes,
language: Optional[str] = None,
prompt: Optional[str] = None,
response_format: Optional[str] = "json",
temperature: Optional[float] = 0,
timestamp_granularities: Optional[List[str]] = None,
):
"""
Translates audio into English.
Parameters
----------
audio: bytes
The audio file object (not file name) to transcribe, in one of these formats: flac, mp3, mp4, mpeg,
mpga, m4a, ogg, wav, or webm.
language: Optional[str]
The language of the input audio. Supplying the input language in ISO-639-1
(https://en.wikipedia.org/wiki/List_of_ISO_639_language_codes) format will improve accuracy and latency.
prompt: Optional[str]
An optional text to guide the model's style or continue a previous audio segment.
The prompt should match the audio language.
response_format: Optional[str], defaults to json
The format of the transcript output, in one of these options: json, text, srt, verbose_json, or vtt.
temperature: Optional[float], defaults to 0
The sampling temperature, between 0 and 1. Higher values like 0.8 will make the output more random,
while lower values like 0.2 will make it more focused and deterministic.
If set to 0, the model will use log probability to automatically increase the temperature
until certain thresholds are hit.
timestamp_granularities: Optional[List[str]], default is None.
The timestamp granularities to populate for this transcription. response_format must be set verbose_json
to use timestamp granularities. Either or both of these options are supported: word, or segment.
Note: There is no additional latency for segment timestamps, but generating word timestamps incurs
additional latency.
Returns
-------
The translated text.
"""
url = f"{self._base_url}/v1/audio/translations"
params = {
"model": self._model_uid,
"language": language,
"prompt": prompt,
"response_format": response_format,
"temperature": temperature,
"timestamp_granularities[]": timestamp_granularities,
}
files: List[Any] = []
files.append(("file", ("file", audio, "application/octet-stream")))
response = self.session.post(
url, data=params, files=files, headers=self.auth_headers
)
if response.status_code != 200:
raise RuntimeError(
f"Failed to translate the audio, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def speech(
self,
input: str,
voice: str = "",
response_format: str = "mp3",
speed: float = 1.0,
stream: bool = False,
prompt_speech: Optional[bytes] = None,
prompt_latent: Optional[bytes] = None,
**kwargs,
):
"""
Generates audio from the input text.
Parameters
----------
input: str
The text to generate audio for. The maximum length is 4096 characters.
voice: str
The voice to use when generating the audio.
response_format: str
The format to audio in.
speed: str
The speed of the generated audio.
stream: bool
Use stream or not.
prompt_speech: bytes
The audio bytes to be provided to the model.
prompt_latent: bytes
The latent bytes to be provided to the model.
Returns
-------
bytes
The generated audio binary.
"""
url = f"{self._base_url}/v1/audio/speech"
params = {
"model": self._model_uid,
"input": input,
"voice": voice,
"response_format": response_format,
"speed": speed,
"stream": stream,
"kwargs": json.dumps(kwargs),
}
files: List[Any] = []
if prompt_speech:
files.append(
(
"prompt_speech",
("prompt_speech", prompt_speech, "application/octet-stream"),
)
)
if prompt_latent:
files.append(
(
"prompt_latent",
("prompt_latent", prompt_latent, "application/octet-stream"),
)
)
if files:
response = self.session.post(
url, data=params, files=files, headers=self.auth_headers, stream=stream
)
else:
response = self.session.post(
url, json=params, headers=self.auth_headers, stream=stream
)
if response.status_code != 200:
raise RuntimeError(
f"Failed to speech the text, detail: {_get_error_string(response)}"
)
if stream:
return response.iter_content(chunk_size=1024)
return response.content
class RESTfulFlexibleModelHandle(RESTfulModelHandle):
def infer(
self,
*args,
**kwargs,
):
"""
Call flexible model.
Parameters
----------
kwargs: dict
The inference arguments.
Returns
-------
bytes
The inference result.
"""
url = f"{self._base_url}/v1/flexible/infers"
params = {
"model": self._model_uid,
"args": args,
}
params.update(kwargs)
response = self.session.post(url, json=params, headers=self.auth_headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to predict, detail: {_get_error_string(response)}"
)
return response.json()
class Client:
def __init__(self, base_url, api_key: Optional[str] = None):
self.base_url = base_url
self._headers: Dict[str, str] = {}
self._cluster_authed = False
self.session = requests.Session()
self._check_cluster_authenticated()
if api_key is not None and self._cluster_authed:
self._headers["Authorization"] = f"Bearer {api_key}"
def close(self):
"""
Close the session.
"""
if self.session:
self.session.close()
self.session = None
def __del__(self):
if self.session:
self.close()
def _set_token(self, token: Optional[str]):
if not self._cluster_authed or token is None:
return
self._headers["Authorization"] = f"Bearer {token}"
def _get_token(self) -> Optional[str]:
return (
str(self._headers["Authorization"]).replace("Bearer ", "")
if "Authorization" in self._headers
else None
)
def _check_cluster_authenticated(self):
url = f"{self.base_url}/v1/cluster/auth"
response = self.session.get(url)
# compatible with old version of xinference
if response.status_code == 404:
self._cluster_authed = False
else:
if response.status_code != 200:
raise RuntimeError(
f"Failed to get cluster information, detail: {response.json()['detail']}"
)
response_data = response.json()
self._cluster_authed = bool(response_data["auth"])
def vllm_models(self) -> Dict[str, Any]:
url = f"{self.base_url}/v1/models/vllm-supported"
response = self.session.get(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to fetch VLLM models. detail: {response.json()['detail']}"
)
try:
return response.json()
except Exception as e:
raise RuntimeError(f"Error parsing JSON response: {e}")
def login(self, username: str, password: str):
if not self._cluster_authed:
return
url = f"{self.base_url}/token"
payload = {"username": username, "password": password}
response = self.session.post(url, json=payload)
if response.status_code != 200:
raise RuntimeError(f"Failed to login, detail: {response.json()['detail']}")
response_data = response.json()
# Only bearer token for now
access_token = response_data["access_token"]
self._headers["Authorization"] = f"Bearer {access_token}"
def list_models(self) -> Dict[str, Dict[str, Any]]:
"""
Retrieve the model specifications from the Server.
Returns
-------
Dict[str, Dict[str, Any]]
The collection of model specifications with their names on the server.
"""
url = f"{self.base_url}/v1/models"
response = self.session.get(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to list model, detail: {_get_error_string(response)}"
)
response_data = response.json()
model_list = response_data["data"]
return {item["id"]: item for item in model_list}
def launch_model(
self,
model_name: str,
model_type: str = "LLM",
model_engine: Optional[str] = None,
model_uid: Optional[str] = None,
model_size_in_billions: Optional[Union[int, str, float]] = None,
model_format: Optional[str] = None,
quantization: Optional[str] = None,
replica: int = 1,
n_worker: int = 1,
n_gpu: Optional[Union[int, str]] = "auto",
peft_model_config: Optional[Dict] = None,
request_limits: Optional[int] = None,
worker_ip: Optional[str] = None,
gpu_idx: Optional[Union[int, List[int]]] = None,
model_path: Optional[str] = None,
enable_virtual_env: Optional[bool] = None,
virtual_env_packages: Optional[List[str]] = None,
envs: Optional[Dict[str, str]] = None,
**kwargs,
) -> str:
"""
Launch the model based on the parameters on the server via RESTful APIs.
Parameters
----------
model_name: str
The name of model.
model_type: str
type of model.
model_engine: Optional[str]
Specify the inference engine of the model when launching LLM.
model_uid: str
UID of model, auto generate a UUID if is None.
model_size_in_billions: Optional[Union[int, str, float]]
The size (in billions) of the model.
model_format: Optional[str]
The format of the model.
quantization: Optional[str]
The quantization of model.
replica: Optional[int]
The replica of model, default is 1.
n_worker: int
Number of workers to run.
n_gpu: Optional[Union[int, str]],
The number of GPUs used by the model, default is "auto". If n_worker>1, means number of GPUs per worker.
``n_gpu=None`` means cpu only, ``n_gpu=auto`` lets the system automatically determine the best number of GPUs to use.
peft_model_config: Optional[Dict]
- "lora_list": A List of PEFT (Parameter-Efficient Fine-Tuning) model and path.
- "image_lora_load_kwargs": A Dict of lora load parameters for image model
- "image_lora_fuse_kwargs": A Dict of lora fuse parameters for image model
request_limits: Optional[int]
The number of request limits for this model, default is None.
``request_limits=None`` means no limits for this model.
worker_ip: Optional[str]
Specify the worker ip where the model is located in a distributed scenario.
gpu_idx: Optional[Union[int, List[int]]]
Specify the GPU index where the model is located.
model_path: Optional[str]
Model path, if gguf format, should be the file path, otherwise, should be directory of the model.
enable_virtual_env: Optional[bool]
If enable virtual env.
virtual_env_packages: Optional[List[str]]
Packages to specify in virtual env, can be used to override builtin packages in virtual env.
envs: Optional[Dict[str, str]]
Environment variables to pass when launching model.
**kwargs:
Any other parameters been specified. e.g. multimodal_projector for multimodal inference with the llama.cpp backend.
Returns
-------
str
The unique model_uid for the launched model.
"""
url = f"{self.base_url}/v1/models"
# convert float to int or string since the RESTful API does not accept float.
if isinstance(model_size_in_billions, float):
model_size_in_billions = convert_float_to_int_or_str(model_size_in_billions)
payload = {
"model_uid": model_uid,
"model_name": model_name,
"model_engine": model_engine,
"peft_model_config": peft_model_config,
"model_type": model_type,
"model_size_in_billions": model_size_in_billions,
"model_format": model_format,
"quantization": quantization,
"replica": replica,
"n_worker": n_worker,
"n_gpu": n_gpu,
"request_limits": request_limits,
"worker_ip": worker_ip,
"gpu_idx": gpu_idx,
"model_path": model_path,
"enable_virtual_env": enable_virtual_env,
"virtual_env_packages": virtual_env_packages,
"envs": envs,
}
wait_ready = kwargs.pop("wait_ready", True)
for key, value in kwargs.items():
payload[str(key)] = value
if wait_ready:
response = self.session.post(url, json=payload, headers=self._headers)
else:
response = self.session.post(
url, json=payload, headers=self._headers, params={"wait_ready": False}
)
if response.status_code != 200:
raise RuntimeError(
f"Failed to launch model, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data["model_uid"]
def terminate_model(self, model_uid: str):
"""
Terminate the specific model running on the server.
Parameters
----------
model_uid: str
The unique id that identify the model we want.
Raises
------
RuntimeError
Report failure to get the wanted model with given model_uid. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/models/{model_uid}"
response = self.session.delete(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to terminate model, detail: {_get_error_string(response)}"
)
def get_launch_model_progress(self, model_uid: str) -> dict:
"""
Get progress of the specific model.
Parameters
----------
model_uid: str
The unique id that identify the model we want.
Returns
-------
result: dict
Result that contains progress.
Raises
------
RuntimeError
Report failure to get the wanted model with given model_uid. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/models/{model_uid}/progress"
response = self.session.get(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Fail to get model launching progress, detail: {_get_error_string(response)}"
)
return response.json()
def cancel_launch_model(self, model_uid: str):
"""
Cancel launching model.
Parameters
----------
model_uid: str
The unique id that identify the model we want.
Raises
------
RuntimeError
Report failure to get the wanted model with given model_uid. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/models/{model_uid}/cancel"
response = self.session.post(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Fail to cancel launching model, detail: {_get_error_string(response)}"
)
def get_instance_info(self, model_name: str, model_uid: str):
url = f"{self.base_url}/v1/models/instances"
response = self.session.get(
url,
headers=self._headers,
params={"model_name": model_name, "model_uid": model_uid},
)
if response.status_code != 200:
raise RuntimeError("Failed to get instance info")
response_data = response.json()
return response_data
def _get_supervisor_internal_address(self):
url = f"{self.base_url}/v1/address"
response = self.session.get(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError("Failed to get supervisor internal address")
response_data = response.json()
return response_data
def get_model(self, model_uid: str) -> RESTfulModelHandle:
"""
Launch the model based on the parameters on the server via RESTful APIs.
Parameters
----------
model_uid: str
The unique id that identify the model.
Returns
-------
ModelHandle
The corresponding Model Handler based on the Model specified in the uid:
- :obj:`xinference.client.handlers.GenerateModelHandle` -> provide handle to basic generate Model. e.g. Baichuan.
- :obj:`xinference.client.handlers.ChatModelHandle` -> provide handle to chat Model. e.g. Baichuan-chat.
Raises
------
RuntimeError
Report failure to get the wanted model with given model_uid. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/models/{model_uid}"
response = self.session.get(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to get the model description, detail: {_get_error_string(response)}"
)
desc = response.json()
if desc["model_type"] == "LLM":
if "chat" in desc["model_ability"]:
return RESTfulChatModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
elif "generate" in desc["model_ability"]:
return RESTfulGenerateModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
else:
raise ValueError(f"Unrecognized model ability: {desc['model_ability']}")
elif desc["model_type"] == "embedding":
return RESTfulEmbeddingModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
elif desc["model_type"] == "image":
return RESTfulImageModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
elif desc["model_type"] == "rerank":
return RESTfulRerankModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
elif desc["model_type"] == "audio":
return RESTfulAudioModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
elif desc["model_type"] == "video":
return RESTfulVideoModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
elif desc["model_type"] == "flexible":
return RESTfulFlexibleModelHandle(
model_uid, self.base_url, auth_headers=self._headers
)
else:
raise ValueError(f"Unknown model type:{desc['model_type']}")
def describe_model(self, model_uid: str):
"""
Get model information via RESTful APIs.
Parameters
----------
model_uid: str
The unique id that identify the model.
Returns
-------
dict
A dictionary containing the following keys:
- "model_type": str
the type of the model determined by its function, e.g. "LLM" (Large Language Model)
- "model_name": str
the name of the specific LLM model family
- "model_lang": List[str]
the languages supported by the LLM model
- "model_ability": List[str]
the ability or capabilities of the LLM model
- "model_description": str
a detailed description of the LLM model
- "model_format": str
the format specification of the LLM model
- "model_size_in_billions": int
the size of the LLM model in billions
- "quantization": str
the quantization applied to the model
- "revision": str
the revision number of the LLM model specification
- "context_length": int
the maximum text length the LLM model can accommodate (include all input & output)
Raises
------
RuntimeError
Report failure to get the wanted model with given model_uid. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/models/{model_uid}"
response = self.session.get(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to get the model description, detail: {_get_error_string(response)}"
)
return response.json()
def register_model(
self,
model_type: str,
model: str,
persist: bool,
worker_ip: Optional[str] = None,
):
"""
Register a custom model.
Parameters
----------
model_type: str
The type of model.
model: str
The model definition. (refer to: https://inference.readthedocs.io/en/latest/models/custom.html)
worker_ip: Optional[str]
The IP address of the worker on which the model is running.
persist: bool
Raises
------
RuntimeError
Report failure to register the custom model. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/model_registrations/{model_type}"
request_body = {"model": model, "worker_ip": worker_ip, "persist": persist}
response = self.session.post(url, json=request_body, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to register model, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def unregister_model(self, model_type: str, model_name: str):
"""
Unregister a custom model.
Parameters
----------
model_type: str
The type of model.
model_name: str
The name of the model
Raises
------
RuntimeError
Report failure to unregister the custom model. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/model_registrations/{model_type}/{model_name}"
response = self.session.delete(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to register model, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def list_model_registrations(
self, model_type: str, detailed: bool = False
) -> List[Dict[str, Any]]:
"""
List models registered on the server.
Parameters
----------
model_type: str
The type of the model.
detailed: bool
Whether to display detailed information.
Returns
-------
List[Dict[str, Any]]
The collection of registered models on the server.
Raises
------
RuntimeError
Report failure to list model registration. Provide details of failure through error message.
"""
url = f"{self.base_url}/v1/model_registrations/{model_type}?detailed={'true' if detailed else 'false'}"
response = self.session.get(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to list model registration, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def list_cached_models(
self, model_name: Optional[str] = None, worker_ip: Optional[str] = None
) -> List[Dict[Any, Any]]:
"""
Get a list of cached models.
Parameters
----------
model_name: Optional[str]
The name of model.
worker_ip: Optional[str]
Specify the worker ip where the model is located in a distributed scenario.
Returns
-------
List[Dict[Any, Any]]
The collection of cached models on the server.
Raises
------
RuntimeError
Raised when the request fails, including the reason for the failure.
"""
url = f"{self.base_url}/v1/cache/models"
params = {
"model_name": model_name,
"worker_ip": worker_ip,
}
response = self.session.get(url, headers=self._headers, params=params)
if response.status_code != 200:
raise RuntimeError(
f"Failed to list cached model, detail: {_get_error_string(response)}"
)
response_data = response.json()
response_data = response_data.get("list")
return response_data
def list_deletable_models(
self, model_version: str, worker_ip: Optional[str] = None
) -> Dict[str, Any]:
"""
Get the cached models with the model path cached on the server.
Parameters
----------
model_version: str
The version of the model.
worker_ip: Optional[str]
Specify the worker ip where the model is located in a distributed scenario.
Returns
-------
Dict[str, Dict[str,str]]]
Dictionary with keys "model_name" and values model_file_location.
"""
url = f"{self.base_url}/v1/cache/models/files"
params = {
"model_version": model_version,
"worker_ip": worker_ip,
}
response = self.session.get(url, headers=self._headers, params=params)
if response.status_code != 200:
raise RuntimeError(
f"Failed to get paths by model name, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def confirm_and_remove_model(
self, model_version: str, worker_ip: Optional[str] = None
) -> bool:
"""
Remove the cached models with the model name cached on the server.
Parameters
----------
model_version: str
The version of the model.
worker_ip: Optional[str]
Specify the worker ip where the model is located in a distributed scenario.
Returns
-------
str
The response of the server.
"""
url = f"{self.base_url}/v1/cache/models"
params = {
"model_version": model_version,
"worker_ip": worker_ip,
}
response = self.session.delete(url, headers=self._headers, params=params)
if response.status_code != 200:
raise RuntimeError(
f"Failed to remove cached models, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data.get("result", False)
def get_model_registration(
self, model_type: str, model_name: str
) -> Dict[str, Any]:
"""
Get the model with the model type and model name registered on the server.
Parameters
----------
model_type: str
The type of the model.
model_name: str
The name of the model.
Returns
-------
List[Dict[str, Any]]
The collection of registered models on the server.
"""
url = f"{self.base_url}/v1/model_registrations/{model_type}/{model_name}"
response = self.session.get(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to list model registration, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def query_engine_by_model_name(
self, model_name: str, model_type: Optional[str] = "LLM"
):
"""
Get the engine parameters with the model name registered on the server.
Parameters
----------
model_name: str
The name of the model.
model_type: str
Model type, LLM by default.
Returns
-------
Dict[str, List[Dict[str, Any]]]
The supported engine parameters of registered models on the server.
"""
if not model_type:
url = f"{self.base_url}/v1/engines/{model_name}"
else:
url = f"{self.base_url}/v1/engines/{model_type}/{model_name}"
response = self.session.get(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to query engine parameters by model name, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def abort_request(self, model_uid: str, request_id: str, block_duration: int = 30):
"""
Abort a request.
Abort a submitted request. If the request is finished or not found, this method will be a no-op.
Currently, this interface is only supported when batching is enabled for models on transformers backend.
Parameters
----------
model_uid: str
Model uid.
request_id: str
Request id.
block_duration: int
The duration to make the request id abort. If set to 0, the abort_request will be immediate, which may
prevent it from taking effect if it arrives before the request operation.
Returns
-------
Dict
Return empty dict.
"""
url = f"{self.base_url}/v1/models/{model_uid}/requests/{request_id}/abort"
response = self.session.post(
url, headers=self._headers, json={"block_duration": block_duration}
)
if response.status_code != 200:
raise RuntimeError(
f"Failed to abort request, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def get_workers_info(self):
url = f"{self.base_url}/v1/workers"
response = self.session.get(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to get workers info, detail: {_get_error_string(response)}"
)
response_data = response.json()
return response_data
def get_supervisor_info(self):
url = f"{self.base_url}/v1/supervisor"
response = self.session.get(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to get supervisor info, detail: {_get_error_string(response)}"
)
response_json = response.json()
return response_json
def get_progress(self, request_id: str):
url = f"{self.base_url}/v1/requests/{request_id}/progress"
response = self.session.get(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to get progress, detail: {_get_error_string(response)}"
)
response_json = response.json()
return response_json
def abort_cluster(self):
url = f"{self.base_url}/v1/clusters"
response = self.session.delete(url, headers=self._headers)
if response.status_code != 200:
raise RuntimeError(
f"Failed to abort cluster, detail: {_get_error_string(response)}"
)
response_json = response.json()
return response_json
================================================
FILE: src/backend/bisheng/core/ai/rerank/xinference/fields.py
================================================
# Copyright 2022-2023 XProbe Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from ._compat import Field
none_field = Field(None)
stream_interval_field = Field(default=2)
echo_field = Field(
default=False,
description="Whether to echo the prompt in the generated text. Useful for chatbots.",
)
logprobs_field = Field(
default=None,
ge=0,
description="The number of logprobs to generate. If None, no logprobs are generated.",
)
# Note: changed from 1024 to None to let the model output maximum content which has better user experience
max_tokens_field = Field(
default=None,
ge=1,
description="The maximum number of tokens to generate.",
)
temperature_field = Field(
default=0.8,
ge=0.0,
le=2.0,
description="Adjust the randomness of the generated text.\n\n"
"Temperature is a hyperparameter that controls the randomness of the generated text. "
"It affects the probability distribution of the model's output tokens. "
"A higher temperature (e.g., 1.5) makes the output more random and creative, "
"while a lower temperature (e.g., 0.5) makes the output more focused, deterministic, and conservative. "
"The default value is 0.8, which provides a balance between randomness and determinism. "
"At the extreme, a temperature of 0 will always pick the most likely next token, "
"leading to identical outputs in each run.",
)
top_p_field = Field(
default=0.95,
ge=0.0,
le=1.0,
description="Limit the next token selection to a subset of tokens with "
"a cumulative probability above a threshold P.\n\n"
"Top-p sampling, also known as nucleus sampling, "
"is another text generation method that selects the next token from a subset of tokens "
"that together have a cumulative probability of at least p. "
"This method provides a balance between diversity and quality by considering "
"both the probabilities of tokens and the number of tokens to sample from. "
"A higher value for top_p (e.g., 0.95) will lead to more diverse text, "
"while a lower value (e.g., 0.5) will generate more focused and conservative text.",
)
stop_field = Field(
default=[],
description="A list of tokens at which to stop generation. If None, no stop tokens are used.",
)
stream_field = Field(
default=False,
description="Whether to stream the results as they are generated. Useful for chatbots.",
)
stream_option_field = Field(
default={
"include_usage": False,
},
description="If set, an additional chunk will be streamed before the `data: [DONE]` message.",
)
top_k_field = Field(
default=40,
ge=0,
description="Limit the next token selection to the K most probable tokens.\n\n"
"Top-k sampling is a text generation method that selects the next token "
"only from the top k most likely tokens predicted by the model. "
"It helps reduce the risk of generating low-probability or nonsensical tokens, "
"but it may also limit the diversity of the output. "
"A higher value for top_k (e.g., 100) will consider more tokens and lead to more diverse text, "
"while a lower value (e.g., 10) will focus on the most probable tokens and "
"generate more conservative text.",
)
repeat_penalty_field = Field(
default=1.1,
ge=0.0,
description="A penalty applied to each token that is already generated. "
"This helps prevent the model from repeating itself.\n\n"
"Repeat penalty is a hyperparameter used to penalize the repetition of token sequences "
"during text generation. "
"It helps prevent the model from generating repetitive or monotonous text. "
"A higher value (e.g., 1.5) will penalize repetitions more strongly, "
"while a lower value (e.g., 0.9) will be more lenient.",
)
presence_penalty_field = Field(
default=0.0,
ge=-2.0,
le=2.0,
description="Positive values penalize new tokens based on whether they appear in the text so far, "
"increasing the model's likelihood to talk about new topics.",
)
frequency_penalty_field = Field(
default=0.0,
ge=-2.0,
le=2.0,
description="Positive values penalize new tokens based on their existing frequency in the text so far, "
"decreasing the model's likelihood to repeat the same line verbatim.",
)
mirostat_mode_field = Field(
default=0,
ge=0,
le=2,
description="Enable Mirostat constant-perplexity algorithm of the specified version (1 or 2; 0 = disabled)",
)
mirostat_tau_field = Field(
default=5.0,
ge=0.0,
le=10.0,
description="Mirostat target entropy, i.e. the target perplexity - lower values produce focused and coherent text, "
"larger values produce more diverse and less coherent text",
)
mirostat_eta_field = Field(
default=0.1, ge=0.001, le=1.0, description="Mirostat learning rate"
)
================================================
FILE: src/backend/bisheng/core/ai/rerank/xinference/types.py
================================================
# Copyright 2022-2023 XProbe Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from typing import Any, Callable, Dict, ForwardRef, Iterable, List, Optional, Union
from typing_extensions import Literal, NotRequired, TypedDict
from ._compat import (
BaseModel,
create_model,
create_model_from_typeddict,
validate_arguments,
)
from .fields import (
echo_field,
max_tokens_field,
none_field,
repeat_penalty_field,
stop_field,
stream_field,
stream_interval_field,
stream_option_field,
temperature_field,
top_k_field,
top_p_field,
)
class Image(TypedDict):
url: Optional[str]
b64_json: Optional[str]
class ImageList(TypedDict):
created: int
data: List[Image]
class ImageEditRequest(TypedDict, total=False):
image: Union[Union[str, bytes], List[Union[str, bytes]]]
mask: Optional[Union[str, bytes]]
prompt: str
n: int
size: Optional[str]
response_format: str
class SDAPIResult(TypedDict):
images: List[str]
parameters: dict
info: dict
class Video(TypedDict):
url: Optional[str]
b64_json: Optional[str]
class VideoList(TypedDict):
created: int
data: List[Video]
class EmbeddingUsage(TypedDict):
prompt_tokens: int
total_tokens: int
class EmbeddingData(TypedDict):
index: int
object: str
# support sparse embedding
embedding: Union[List[float], Dict[str, float]]
class Embedding(TypedDict):
object: Literal["list"]
model: str
model_replica: str
data: List[EmbeddingData]
usage: EmbeddingUsage
class Document(TypedDict):
text: str
class DocumentObj(TypedDict):
index: int
relevance_score: float
document: Optional[Document]
# Cohere API compatibility
class ApiVersion(TypedDict):
version: str
is_deprecated: bool
is_experimental: bool
# Cohere API compatibility
class BilledUnit(TypedDict):
input_tokens: int
output_tokens: int
search_units: int
classifications: int
class RerankTokens(TypedDict):
input_tokens: int
output_tokens: int
class Meta(TypedDict):
api_version: Optional[ApiVersion]
billed_units: Optional[BilledUnit]
tokens: RerankTokens
warnings: Optional[List[str]]
class Rerank(TypedDict):
id: str
results: List[DocumentObj]
meta: Meta
class CompletionLogprobs(TypedDict):
text_offset: List[int]
token_logprobs: List[Optional[float]]
tokens: List[str]
top_logprobs: List[Optional[Dict[str, float]]]
class ToolCallFunction(TypedDict):
name: str
arguments: str
class ToolCalls(TypedDict):
id: str
type: Literal["function"]
function: ToolCallFunction
class CompletionChoice(TypedDict):
text: NotRequired[str]
index: int
logprobs: Optional[CompletionLogprobs]
finish_reason: Optional[str]
tool_calls: NotRequired[List[ToolCalls]]
class CompletionUsage(TypedDict):
prompt_tokens: int
completion_tokens: int
total_tokens: int
class CompletionChunk(TypedDict):
id: str
object: Literal["text_completion"]
created: int
model: str
choices: List[CompletionChoice]
usage: NotRequired[CompletionUsage]
class Completion(TypedDict):
id: str
object: Literal["text_completion"]
created: int
model: str
choices: List[CompletionChoice]
usage: CompletionUsage
class ChatCompletionAudio(TypedDict):
id: str
data: str
expires_at: int
transcript: str
class ChatCompletionMessage(TypedDict):
role: str
reasoning_content: NotRequired[str]
content: Optional[str]
audio: NotRequired[ChatCompletionAudio]
user: NotRequired[str]
tool_calls: NotRequired[List]
class ChatCompletionChoice(TypedDict):
index: int
message: ChatCompletionMessage
finish_reason: Optional[str]
class ChatCompletion(TypedDict):
id: str
object: Literal["chat.completion"]
created: int
model: str
choices: List[ChatCompletionChoice]
usage: CompletionUsage
class ChatCompletionChunkDelta(TypedDict):
role: NotRequired[str]
reasoning_content: NotRequired[Union[str, None]]
content: NotRequired[Union[str, None]]
tool_calls: NotRequired[List[ToolCalls]]
class ChatCompletionChunkChoice(TypedDict):
index: int
delta: ChatCompletionChunkDelta
finish_reason: Optional[str]
class ChatCompletionChunk(TypedDict):
id: str
model: str
object: Literal["chat.completion.chunk"]
created: int
choices: List[ChatCompletionChunkChoice]
usage: NotRequired[CompletionUsage]
StoppingCriteria = Callable[[List[int], List[float]], bool]
class StoppingCriteriaList(List[StoppingCriteria]):
def __call__(self, input_ids: List[int], logits: List[float]) -> bool:
return any([stopping_criteria(input_ids, logits) for stopping_criteria in self])
LogitsProcessor = Callable[[List[int], List[float]], List[float]]
class LogitsProcessorList(List[LogitsProcessor]):
def __call__(self, input_ids: List[int], scores: List[float]) -> List[float]:
for processor in self:
scores = processor(input_ids, scores)
return scores
class PytorchGenerateConfig(TypedDict, total=False):
temperature: float
repetition_penalty: float
top_p: float
top_k: int
stream: bool
max_tokens: int
echo: bool
stop: Optional[Union[str, List[str]]]
stop_token_ids: Optional[Union[int, List[int]]]
stream_interval: int
model: Optional[str]
tools: Optional[List[Dict]]
lora_name: Optional[str]
stream_options: Optional[Union[dict, None]]
request_id: Optional[str]
class CogagentGenerateConfig(PytorchGenerateConfig, total=False):
platform: Optional[Literal["Mac", "WIN", "Mobile"]]
format: Optional[
Literal[
"(Answer in Action-Operation-Sensitive format.)",
"(Answer in Status-Plan-Action-Operation format.)",
"(Answer in Status-Action-Operation-Sensitive format.)",
"(Answer in Status-Action-Operation format.)",
"(Answer in Action-Operation format.)",
]
]
class PytorchModelConfig(TypedDict, total=False):
revision: Optional[str]
device: str
gpus: Optional[str]
num_gpus: int
max_gpu_memory: str
gptq_ckpt: Optional[str]
gptq_wbits: int
gptq_groupsize: int
gptq_act_order: bool
trust_remote_code: bool
max_num_seqs: int
enable_tensorizer: Optional[bool]
reasoning_content: bool
min_pixels: NotRequired[int]
max_pixels: NotRequired[int]
quantization_config: NotRequired[Dict]
context_length: NotRequired[int]
torch_dtype: NotRequired[str]
enable_flash_attn: NotRequired[bool]
def get_pydantic_model_from_method(
meth,
exclude_fields: Optional[Iterable[str]] = None,
include_fields: Optional[Dict[str, Any]] = None,
) -> BaseModel:
# The validate_arguments set Config.extra = "forbid" by default.
f = validate_arguments(meth, config={"arbitrary_types_allowed": True})
model = f.model
model.Config.extra = "ignore"
model.__fields__.pop("self", None)
model.__fields__.pop("args", None)
model.__fields__.pop("kwargs", None)
pydantic_private_keys = [
key for key in model.__fields__.keys() if key.startswith("v__")
]
for key in pydantic_private_keys:
model.__fields__.pop(key)
if exclude_fields is not None:
for key in exclude_fields:
model.__fields__.pop(key, None)
if include_fields is not None:
dummy_model = create_model("DummyModel", **include_fields)
model.__fields__.update(dummy_model.__fields__)
return model
def fix_forward_ref(model):
"""
pydantic in Python 3.8 generates ForwardRef field, we replace them
by the Optional[Any]
"""
exclude_fields = []
include_fields = {}
for key, field in model.__fields__.items():
if isinstance(field.annotation, ForwardRef):
exclude_fields.append(key)
include_fields[key] = (Optional[Any], None)
if exclude_fields:
for key in exclude_fields:
model.__fields__.pop(key, None)
if include_fields:
dummy_model = create_model("DummyModel", **include_fields)
model.__fields__.update(dummy_model.__fields__)
return model
class ModelAndPrompt(BaseModel):
model: str
prompt: str
class ModelAndMessages(BaseModel):
model: str
messages: List[Dict[str, Any]]
class CreateCompletionTorch(BaseModel):
echo: bool = echo_field
max_tokens: Optional[int] = max_tokens_field
repetition_penalty: float = repeat_penalty_field
stop: Optional[Union[str, List[str]]] = stop_field
stop_token_ids: Optional[Union[int, List[int]]] = none_field
stream: bool = stream_field
stream_options: Optional[Union[dict, None]] = stream_option_field
stream_interval: int = stream_interval_field
temperature: float = temperature_field
top_p: float = top_p_field
top_k: int = top_k_field
lora_name: Optional[str]
request_id: Optional[str]
chat_template_kwargs: Optional[Union[str, Dict[str, Any]]]
# This type is for openai API compatibility
CreateCompletionOpenAI: BaseModel
from openai.types.completion_create_params import CompletionCreateParamsNonStreaming
CreateCompletionOpenAI = create_model_from_typeddict(
CompletionCreateParamsNonStreaming,
)
CreateCompletionOpenAI = fix_forward_ref(CreateCompletionOpenAI)
class CreateCompletion(
ModelAndPrompt,
CreateCompletionTorch,
CreateCompletionOpenAI,
):
pass
class CreateChatModel(BaseModel):
model: str
# Currently, chat calls generates, so the params share the same one.
CreateChatCompletionTorch = CreateCompletionTorch
from ._compat import CreateChatCompletionOpenAI
class CreateChatCompletion( # type: ignore
CreateChatModel,
CreateChatCompletionTorch,
CreateChatCompletionOpenAI,
):
pass
class LoRA:
def __init__(self, lora_name: str, local_path: str):
self.lora_name = lora_name
self.local_path = local_path
def to_dict(self):
return {
"lora_name": self.lora_name,
"local_path": self.local_path,
}
@classmethod
def from_dict(cls, data: Dict):
return cls(
lora_name=data["lora_name"],
local_path=data["local_path"],
)
class PeftModelConfig:
def __init__(
self,
peft_model: Optional[List[LoRA]] = None,
image_lora_load_kwargs: Optional[Dict] = None,
image_lora_fuse_kwargs: Optional[Dict] = None,
):
self.peft_model = peft_model
self.image_lora_load_kwargs = image_lora_load_kwargs
self.image_lora_fuse_kwargs = image_lora_fuse_kwargs
def to_dict(self):
return {
"lora_list": (
[lora.to_dict() for lora in self.peft_model]
if self.peft_model
else None
),
"image_lora_load_kwargs": self.image_lora_load_kwargs,
"image_lora_fuse_kwargs": self.image_lora_fuse_kwargs,
}
@classmethod
def from_dict(cls, data: Dict):
peft_model_list = data.get("lora_list", None)
peft_model = (
[LoRA.from_dict(lora_dict) for lora_dict in peft_model_list]
if peft_model_list is not None
else None
)
return cls(
peft_model=peft_model,
image_lora_load_kwargs=data.get("image_lora_load_kwargs"),
image_lora_fuse_kwargs=data.get("image_lora_fuse_kwargs"),
)
# This type is for Anthropic API compatibility
ANTHROPIC_AVAILABLE = False
try:
from anthropic.types import ContentBlock, Usage
ANTHROPIC_AVAILABLE = True
except ImportError:
ContentBlock = None
Usage = None
# Use TYPE_CHECKING to avoid runtime issues with mypy
from typing import TYPE_CHECKING
if TYPE_CHECKING:
# For type checking, define the types as if Anthropic is available
from anthropic.types import ContentBlock as ContentBlock_
from anthropic.types import Usage as Usage_
class AnthropicMessage(TypedDict):
id: str
type: str
role: str
content: List[ContentBlock_]
model: str
stop_reason: str
stop_sequence: str
usage: Usage_
container: Dict[str, Any]
class MessageCreateParams(TypedDict):
model: str
messages: List[Dict[str, Any]]
max_tokens: int
stream: NotRequired[bool]
temperature: NotRequired[float]
top_p: NotRequired[float]
top_k: NotRequired[int]
stop_sequences: NotRequired[List[str]]
metadata: NotRequired[Dict[str, Any]]
tools: NotRequired[List[Dict[str, Any]]]
tool_choice: NotRequired[Union[str, Dict[str, Any]]]
CreateMessageAnthropic: BaseModel
class CreateMessage(
ModelAndMessages,
):
pass
else:
# Runtime definitions
if ANTHROPIC_AVAILABLE:
class AnthropicMessage(TypedDict):
id: str
type: str
role: str
content: List[ContentBlock]
model: str
stop_reason: str
stop_sequence: str
usage: Usage
container: Dict[str, Any]
class MessageCreateParams(TypedDict):
model: str
messages: List[Dict[str, Any]]
max_tokens: int
stream: NotRequired[bool]
temperature: NotRequired[float]
top_p: NotRequired[float]
top_k: NotRequired[int]
stop_sequences: NotRequired[List[str]]
metadata: NotRequired[Dict[str, Any]]
tools: NotRequired[List[Dict[str, Any]]]
tool_choice: NotRequired[Union[str, Dict[str, Any]]]
CreateMessageAnthropic: BaseModel = create_model_from_typeddict(
MessageCreateParams,
)
CreateMessageAnthropic = fix_forward_ref(CreateMessageAnthropic)
class CreateMessage(CreateMessageAnthropic):
pass
else:
# Define dummy types when Anthropic is not available
class AnthropicMessage:
pass
class MessageCreateParams:
pass
CreateMessageAnthropic = None
class CreateMessage:
pass
================================================
FILE: src/backend/bisheng/core/ai/rerank/xinference_rerank.py
================================================
from typing import Sequence, Optional
from langchain_core.callbacks import Callbacks
from langchain_core.documents import Document
from pydantic import Field, model_validator
from typing_extensions import Self
from .xinference.client import Client
from ..base import BaseRerank
class XinferenceRerank(BaseRerank):
"""Document compressor that uses `Xinference Rerank API`."""
base_url: str = Field(..., description="xinference server base url， example: http://localhost:9997")
api_key: Optional[str] = Field(default="", description="api key for xinference server")
model_uid: str = Field(..., description="model uid for xinference rerank model")
client: Optional[Client] = Field(default=None, description="xinference client instance")
@model_validator(mode="after")
def validate_client(self) -> Self:
"""Validate that client exists in environment."""
if self.base_url.endswith(("/v1", "/v1/")):
self.base_url = self.base_url.rsplit("/v1", 1)[0]
if not self.client:
self.client = Client(base_url=self.base_url, api_key=self.api_key)
return self
def compress_documents(
self,
documents: Sequence[Document],
query: str,
callbacks: Optional[Callbacks] = None,
) -> Sequence[Document]:
if len(documents) == 0: # to avoid empty api call
return []
docs = [
doc.page_content if isinstance(doc, Document) else doc for doc in documents
]
model = self.client.get_model(self.model_uid) # ensure model exists
result = model.rerank(docs, query)
results = result.get("results", [])
return self.sort_rerank_result(documents, results)
================================================
FILE: src/backend/bisheng/core/ai/test/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/ai/test/test.py
================================================
import asyncio
import os
import pytest
from bisheng.core.ai import AzureOpenAIASRClient
@pytest.mark.asyncio
async def test_aliyun_asr():
from ..asr import AliyunASRClient
# api_key = os.environ.get('ALIYUN_API_KEY')
api_key = "sk-b4c85ea486a04aab8041dc9f36d20617"
client = AliyunASRClient(api_key=api_key, model="paraformer-realtime-v2")
with open("../asr/tmpm103lam_.wav", "rb") as f:
audio = f.read()
text = await client.transcribe(audio)
print(text)
# assert text == "Hello word, This is Alibaba Voice Lab."
@pytest.mark.asyncio
async def test_aliyun_tts():
from ..tts import AliyunTTSClient
api_key = os.environ.get('ALIYUN_API_KEY')
client = AliyunTTSClient(api_key=api_key)
audio_bytes = await client.synthesize("Hello World!")
with open("./data/aliyun_result.mp3", "wb") as f:
f.write(audio_bytes)
assert os.path.exists("./data/aliyun_result.mp3")
@pytest.mark.asyncio
async def test_azure_openai_asr():
api_key = os.environ.get('AZURE_OPENAI_API_KEY')
api_version = os.environ.get('AZURE_OPENAI_API_VERSION')
azure_endpoint = os.environ.get('AZURE_OPENAI_ENDPOINT')
client = AzureOpenAIASRClient(api_key=api_key, model="gpt-4o-transcribe", azure_endpoint=azure_endpoint,
api_version=api_version)
text = await client.transcribe("./data/asr_example.wav")
assert text == "Hello word, This is Alibaba Voice Lab."
@pytest.mark.asyncio
async def test_azure_openai_tts():
from ..tts import AzureOpenAITTSClient
api_key = os.environ.get('AZURE_OPENAI_API_KEY')
api_version = os.environ.get('AZURE_OPENAI_API_VERSION')
azure_endpoint = os.environ.get('AZURE_OPENAI_ENDPOINT')
client = AzureOpenAITTSClient(api_key=api_key, model="gpt-4o-transcribe", azure_endpoint=azure_endpoint,
api_version=api_version)
audio_bytes = await client.synthesize("Hello, world!")
with open("./data/azure_openai_result.mp3", "wb") as f:
f.write(audio_bytes)
assert os.path.exists("./data/azure_openai_result.mp3")
def test_xinference_rerank():
from ..rerank.xinference_rerank import XinferenceRerank
api_key = os.environ.get('XINFERENCE_API_KEY')
base_url = os.environ.get('XINFERENCE_BASE_URL')
model_uid = os.environ.get('XINFERENCE_RERANK_MODEL_UID')
reranker = XinferenceRerank(base_url=base_url, api_key=api_key, model_uid=model_uid)
async def main():
await test_aliyun_asr()
# await test_aliyun_tts()
if __name__ == "__main__":
asyncio.run(main())
================================================
FILE: src/backend/bisheng/core/ai/tts/__init__.py
================================================
from .aliyun_tts_client import AliyunTTSClient
from .azure_openai_tts_client import AzureOpenAITTSClient
from .openai_tts_client import OpenAITTSClient
__all__ = [
'OpenAITTSClient',
'AliyunTTSClient',
'AzureOpenAITTSClient'
]
================================================
FILE: src/backend/bisheng/core/ai/tts/aliyun_tts_client.py
================================================
import asyncio
from typing import Optional
from dashscope.audio.tts_v2 import SpeechSynthesizer
from ..base import BaseTTSClient
class AliyunTTSClient(BaseTTSClient):
"""Alibaba CloudTTSClient"""
def __init__(self, api_key: str, **kwargs):
"""
Initialize Alibaba CloudTTSClient
"""
self.model = kwargs.get("model", "cosyvoice-v2")
self.voice = kwargs.get("voice", "longxiaochun_v2")
self.app_key = api_key
self.synthesizer = SpeechSynthesizer(model=self.model, voice=self.voice)
self.synthesizer.request.apikey = self.app_key
def sync_func(self, text: str):
audio = self.synthesizer.call(text=text)
return audio
async def synthesize(
self,
text: str,
voice: Optional[str] = None,
language: Optional[str] = None,
format: str = "mp3"
) -> bytes:
"""
Convert text to audio
:param text:
:param voice:
:param language:
:param format:
:return:
"""
audio = await asyncio.to_thread(self.sync_func, text=text)
if audio is None:
raise ValueError("TTS synthesis failed, no audio data returned.")
return audio
================================================
FILE: src/backend/bisheng/core/ai/tts/azure_openai_tts_client.py
================================================
import openai
from typing import Optional, Literal
from ..base import BaseTTSClient
class AzureOpenAITTSClient(BaseTTSClient):
"""OpenAI TTSClient"""
def __init__(self, api_key: str, **kwargs):
"""
InisialisasiOpenAI TTSClient
:param api_key:
:param kwargs:
"""
self.model = kwargs.pop("model", "tts-1")
self.voice = kwargs.pop("voice", "alloy")
self.client = openai.AsyncAzureOpenAI(api_key=api_key, **kwargs)
async def synthesize(
self,
text: str,
voice: Optional[str] = None,
language: Optional[str] = None,
format: Literal["mp3", "opus", "aac", "flac", "wav", "pcm"] = "mp3"
) -> bytes:
"""
UseOpenAI TTS APIText-to-speech synthesis
:param text:
:param voice:
:param language:
:param format:
:return:
"""
response = await self.client.audio.speech.create(
model=self.model,
voice=voice or self.voice,
input=text,
response_format=format
)
if response.response.status_code != 200:
raise Exception(
f"OpenAI TTS APIRequest failed, status code:{response.response.status_code}Error Msg{response.response.text}")
return response.content
================================================
FILE: src/backend/bisheng/core/ai/tts/openai_tts_client.py
================================================
from typing import Optional, Literal
import openai
from ..base import BaseTTSClient
class OpenAITTSClient(BaseTTSClient):
"""OpenAI TTSClient"""
def __init__(self, api_key: str, **kwargs):
"""
InisialisasiOpenAI TTSClient
:param api_key:
:param kwargs:
"""
self.model = kwargs.pop("model", "tts-1")
self.voice = kwargs.pop("voice", "alloy")
self.client = openai.AsyncOpenAI(api_key=api_key, **kwargs)
async def synthesize(
self,
text: str,
voice: Optional[str] = None,
language: Optional[str] = None,
format: Literal["mp3", "opus", "aac", "flac", "wav", "pcm"] = "mp3"
) -> bytes:
"""
UseOpenAI TTS APIText-to-speech synthesis
:param text:
:param voice:
:param language:
:param format:
:return:
"""
response = await self.client.audio.speech.create(
model=self.model,
voice=voice or self.voice,
input=text,
response_format=format
)
if response.response.status_code != 200:
raise Exception(
f"OpenAI TTS API ERROR，Code：{response.response.status_code}，Message：{response.response.text}")
return response.content
================================================
FILE: src/backend/bisheng/core/cache/__init__.py
================================================
from bisheng.core.cache.flow import InMemoryCache
from bisheng.core.cache.manager import cache_manager
__all__ = [
'cache_manager',
'InMemoryCache',
]
================================================
FILE: src/backend/bisheng/core/cache/base.py
================================================
import abc
class BaseCache(abc.ABC):
"""
Abstract base class for a cache.
"""
@abc.abstractmethod
def get(self, key):
"""
Retrieve an item from the cache.
Args:
key: The key of the item to retrieve.
Returns:
The value associated with the key, or None if the key is not found.
"""
@abc.abstractmethod
def set(self, key, value):
"""
Add an item to the cache.
Args:
key: The key of the item.
value: The value to cache.
"""
@abc.abstractmethod
def delete(self, key):
"""
Remove an item from the cache.
Args:
key: The key of the item to remove.
"""
@abc.abstractmethod
def clear(self):
"""
Clear all items from the cache.
"""
@abc.abstractmethod
def __contains__(self, key):
"""
Check if the key is in the cache.
Args:
key: The key of the item to check.
Returns:
True if the key is in the cache, False otherwise.
"""
@abc.abstractmethod
def __getitem__(self, key):
"""
Retrieve an item from the cache using the square bracket notation.
Args:
key: The key of the item to retrieve.
"""
@abc.abstractmethod
def __setitem__(self, key, value):
"""
Add an item to the cache using the square bracket notation.
Args:
key: The key of the item.
value: The value to cache.
"""
@abc.abstractmethod
def __delitem__(self, key):
"""
Remove an item from the cache using the square bracket notation.
Args:
key: The key of the item to remove.
"""
================================================
FILE: src/backend/bisheng/core/cache/flow.py
================================================
import threading
import time
from collections import OrderedDict
from bisheng.core.cache.base import BaseCache
class InMemoryCache(BaseCache):
"""
A simple in-memory cache using an OrderedDict.
This cache supports setting a maximum size and expiration time for cached items.
When the cache is full, it uses a Least Recently Used (LRU) eviction policy.
Thread-safe using a threading Lock.
Attributes:
max_size (int, optional): Maximum number of items to store in the cache.
expiration_time (int, optional): Time in seconds after which a cached item expires. Default is 1 hour.
Example:
cache = InMemoryCache(max_size=3, expiration_time=5)
# setting cache values
cache.set("a", 1)
cache.set("b", 2)
cache["c"] = 3
# getting cache values
a = cache.get("a")
b = cache["b"]
"""
def __init__(self, max_size=None, expiration_time=60 * 60):
"""
Initialize a new InMemoryCache instance.
Args:
max_size (int, optional): Maximum number of items to store in the cache.
expiration_time (int, optional): Time in seconds after which a cached item expires. Default is 1 hour.
"""
self._cache = OrderedDict()
self._lock = threading.Lock()
self.max_size = max_size
self.expiration_time = expiration_time
def get(self, key):
"""
Retrieve an item from the cache.
Args:
key: The key of the item to retrieve.
Returns:
The value associated with the key, or None if the key is not found or the item has expired.
"""
with self._lock:
if key in self._cache:
item = self._cache.pop(key)
if (
self.expiration_time is None
or time.time() - item['time'] < self.expiration_time
):
# Move the key to the end to make it recently used
self._cache[key] = item
return item['value']
else:
self.delete(key)
return None
def set(self, key, value):
"""
Add an item to the cache.
If the cache is full, the least recently used item is evicted.
Args:
key: The key of the item.
value: The value to cache.
"""
with self._lock:
if key in self._cache:
# Remove existing key before re-inserting to update order
self.delete(key)
elif self.max_size and len(self._cache) >= self.max_size:
# Remove least recently used item
self._cache.popitem(last=False)
self._cache[key] = {'value': value, 'time': time.time()}
def get_or_set(self, key, value):
"""
Retrieve an item from the cache. If the item does not exist, set it with the provided value.
Args:
key: The key of the item.
value: The value to cache if the item doesn't exist.
Returns:
The cached value associated with the key.
"""
with self._lock:
if key in self._cache:
return self.get(key)
self.set(key, value)
return value
def delete(self, key):
"""
Remove an item from the cache.
Args:
key: The key of the item to remove.
"""
# with self._lock:
self._cache.pop(key, None)
def clear(self):
"""
Clear all items from the cache.
"""
with self._lock:
self._cache.clear()
def __contains__(self, key):
"""Check if the key is in the cache."""
return key in self._cache
def __getitem__(self, key):
"""Retrieve an item from the cache using the square bracket notation."""
return self.get(key)
def __setitem__(self, key, value):
"""Add an item to the cache using the square bracket notation."""
self.set(key, value)
def __delitem__(self, key):
"""Remove an item from the cache using the square bracket notation."""
self.delete(key)
def __len__(self):
"""Return the number of items in the cache."""
return len(self._cache)
def __repr__(self):
"""Return a string representation of the InMemoryCache instance."""
return f'InMemoryCache(max_size={self.max_size}, expiration_time={self.expiration_time})'
================================================
FILE: src/backend/bisheng/core/cache/manager.py
================================================
from contextlib import contextmanager
from typing import Any, Awaitable, Callable, List, Optional
import pandas as pd
from bisheng.utils.util import get_cache_key
from PIL import Image
class Subject:
"""Base class for implementing the observer pattern."""
def __init__(self):
self.observers: List[Callable[[], None]] = []
def attach(self, observer: Callable[[], None]):
"""Attach an observer to the subject."""
self.observers.append(observer)
def detach(self, observer: Callable[[], None]):
"""Detach an observer from the subject."""
self.observers.remove(observer)
def notify(self):
"""Notify all observers about an event."""
for observer in self.observers:
if observer is None:
continue
observer()
class AsyncSubject:
"""Base class for implementing the async observer pattern."""
def __init__(self):
self.observers: List[Callable[[], Awaitable]] = []
def attach(self, observer: Callable[[], Awaitable]):
"""Attach an observer to the subject."""
self.observers.append(observer)
def detach(self, observer: Callable[[], Awaitable]):
"""Detach an observer from the subject."""
self.observers.remove(observer)
async def notify(self):
"""Notify all observers about an event."""
for observer in self.observers:
if observer is None:
continue
await observer()
class CacheManager(Subject):
"""Manages cache for different clients and notifies observers on changes."""
def __init__(self):
super().__init__()
self._cache = {}
self.current_client_id = None
self.current_chat_id = None
self.current_cache = {}
@contextmanager
def set_client_id(self, client_id: str, chat_id: str):
"""
Context manager to set the current client_id and associated cache.
Args:
client_id (str): The client identifier.
"""
previous_client_id = self.current_client_id
previous_chat_id = self.current_chat_id
self.current_client_id = client_id
self.current_chat_id = chat_id
self.current_cache = self._cache.setdefault(get_cache_key(client_id, chat_id), {})
try:
yield
finally:
self.current_client_id = previous_client_id
self.current_chat_id = previous_chat_id
self.current_cache = self._cache.get(get_cache_key(self.current_client_id, self.current_chat_id), {})
def add(self, name: str, obj: Any, obj_type: str, extension: Optional[str] = None):
"""
Add an object to the current client's cache.
Args:
name (str): The cache key.
obj (Any): The object to cache.
obj_type (str): The type of the object.
"""
object_extensions = {
'image': 'png',
'pandas': 'csv',
}
if obj_type in object_extensions:
_extension = object_extensions[obj_type]
else:
_extension = type(obj).__name__.lower()
self.current_cache[name] = {
'obj': obj,
'type': obj_type,
'extension': extension or _extension,
}
self.notify()
def add_pandas(self, name: str, obj: Any):
"""
Add a pandas DataFrame or Series to the current client's cache.
Args:
name (str): The cache key.
obj (Any): The pandas DataFrame or Series object.
"""
if isinstance(obj, (pd.DataFrame, pd.Series)):
self.add(name, obj.to_csv(), 'pandas', extension='csv')
else:
raise ValueError('Object is not a pandas DataFrame or Series')
def add_image(self, name: str, obj: Any, extension: str = 'png'):
"""
Add a PIL Image to the current client's cache.
Args:
name (str): The cache key.
obj (Any): The PIL Image object.
"""
if isinstance(obj, Image.Image):
self.add(name, obj, 'image', extension=extension)
else:
raise ValueError('Object is not a PIL Image')
def get(self, name: str):
"""
Get an object from the current client's cache.
Args:
name (str): The cache key.
Returns:
The cached object associated with the given cache key.
"""
return self.current_cache[name]
def get_last(self):
"""
Get the last added item in the current client's cache.
Returns:
The last added item in the cache.
"""
return list(self.current_cache.values())[-1]
cache_manager = CacheManager()
================================================
FILE: src/backend/bisheng/core/cache/redis_conn.py
================================================
import pickle
import typing
from typing import Dict, Optional
import redis
from loguru import logger
from redis import ConnectionPool, RedisCluster
from redis.asyncio import Redis as AsyncRedis
from redis.asyncio.client import Pipeline
from redis.asyncio.cluster import RedisCluster as AsyncRedisCluster
from redis.asyncio.sentinel import Sentinel as AsyncSentinel
from redis.backoff import ExponentialBackoff
from redis.cluster import ClusterNode
from redis.retry import Retry
from redis.sentinel import Sentinel
class RedisClient:
def __init__(self, redis_url, max_connections=100):
# # Sentry Mode
if isinstance(redis_url, Dict):
redis_conf = dict(redis_url)
mode = redis_conf.pop('mode', 'sentinel')
if mode == 'cluster':
# Cluster Mode
cluster_url = ''
if 'startup_nodes' in redis_conf:
first_node = redis_conf['startup_nodes'][0]
cluster_url = f'redis://{first_node["host"]}:{first_node["port"]}'
redis_conf['startup_nodes'] = [
ClusterNode(node.get('host'), node.get('port'))
for node in redis_conf['startup_nodes']
]
self.connection = RedisCluster.from_url(cluster_url, **redis_conf,
retry=Retry(ExponentialBackoff(), 6),
cluster_error_retry_attempts=1)
self.async_connection: typing.Union[AsyncRedisCluster, AsyncRedis] = AsyncRedisCluster.from_url(
cluster_url, **redis_conf, retry=Retry(ExponentialBackoff(), 6), cluster_error_retry_attempts=1)
return
hosts = [eval(x) for x in redis_conf.pop('sentinel_hosts')]
password = redis_conf.pop('sentinel_password', None)
master = redis_conf.pop('sentinel_master')
sentinel = Sentinel(sentinels=hosts, socket_timeout=0.1, sentinel_kwargs={'password': password})
async_sentinel = AsyncSentinel(sentinels=hosts, socket_timeout=0.1, sentinel_kwargs={'password': password})
# Get the connection of the master node
self.connection = sentinel.master_for(master, socket_timeout=0.1, **redis_conf)
self.async_connection: AsyncRedis = async_sentinel.master_for(master, socket_timeout=0.1, **redis_conf)
else:
# Singleplayer Mode
self.pool = ConnectionPool.from_url(redis_url, max_connections=max_connections)
self.async_pool = redis.asyncio.ConnectionPool.from_url(redis_url, max_connections=max_connections)
self.connection = redis.StrictRedis(connection_pool=self.pool)
self.async_connection: AsyncRedis = redis.asyncio.Redis.from_pool(self.async_pool)
def set(self, key, value, expiration=3600, enx=None):
try:
if pickled := pickle.dumps(value):
self.cluster_nodes(key)
if expiration:
result = self.connection.setex(key, expiration, pickled)
else:
result = self.connection.set(key, pickled)
if not result:
raise ValueError('RedisCache could not set the value.')
else:
logger.error(f'pickle error, value={value}')
except TypeError as exc:
raise TypeError('RedisCache only accepts values that can be pickled. ') from exc
async def aset(self, key, value, expiration=3600):
try:
if pickled := pickle.dumps(value):
await self.acluster_nodes(key)
if expiration:
result = await self.async_connection.setex(name=key, value=pickled, time=expiration)
else:
result = await self.async_connection.set(key, pickled)
if not result:
raise ValueError('RedisCache could not set the value.')
else:
logger.error(f'pickle error, value={value}')
except TypeError as exc:
raise TypeError('RedisCache only accepts values that can be pickled. ') from exc
def setNx(self, key, value, expiration=3600):
try:
if pickled := pickle.dumps(value):
self.cluster_nodes(key)
result = self.connection.setnx(key, pickled)
self.connection.expire(key, expiration)
if not result:
return False
return True
except TypeError as exc:
raise TypeError('RedisCache only accepts values that can be pickled. ') from exc
async def asetNx(self, key, value, expiration=3600):
try:
if pickled := pickle.dumps(value):
await self.acluster_nodes(key)
result = await self.async_connection.setnx(key, pickled)
await self.async_connection.expire(key, expiration)
if not result:
return False
return True
except TypeError as exc:
raise TypeError('RedisCache only accepts values that can be pickled. ') from exc
def setex(self, key, value, expiration=3600):
try:
if pickled := pickle.dumps(value):
self.cluster_nodes(key)
result = self.connection.setex(key, expiration, pickled)
if not result:
raise ValueError('RedisCache could not set the value.')
else:
logger.error(f'pickle error, value={value}')
except TypeError as exc:
raise TypeError('RedisCache only accepts values that can be pickled. ') from exc
async def asetex(self, key, value, expiration=3600):
try:
if pickled := pickle.dumps(value):
await self.acluster_nodes(key)
result = await self.async_connection.setex(key, expiration, pickled)
if not result:
raise ValueError('RedisCache could not set the value.')
else:
logger.error(f'pickle error, value={value}')
except TypeError as exc:
raise TypeError('RedisCache only accepts values that can be pickled. ') from exc
def mset(self, mapping: Dict[str, typing.Any], expiration: int = None) -> bool | None:
"""Bulk Settings"""
try:
if not mapping:
return True
serialized_mapping = {k: pickle.dumps(v) for k, v in mapping.items() if v is not None}
result = self.connection.mset(serialized_mapping)
if expiration:
# UsepipelineBatch Set Expiration Time
pipe = self.connection.pipeline()
for key in mapping.keys():
pipe.expire(key, expiration)
pipe.execute()
return bool(result)
except TypeError as exc:
raise TypeError('RedisCache only accepts values that can be pickled. ') from exc
async def amset(self, mapping: Dict[str, typing.Any], expiration: int = None) -> bool | None:
"""Asynchronous Batch Setup"""
try:
if not mapping:
return True
serialized_mapping = {k: pickle.dumps(v) for k, v in mapping.items() if v is not None}
result = await self.async_connection.mset(serialized_mapping)
if expiration:
# UsepipelineBatch Set Expiration Time
pipe: Pipeline = self.async_connection.pipeline()
for key in mapping.keys():
await pipe.expire(key, expiration)
await pipe.execute()
return bool(result)
except TypeError as exc:
raise TypeError('RedisCache only accepts values that can be pickled. ') from exc
def mget(self, keys: typing.List[str]) -> typing.List[typing.Any] | None:
"""Get in bulk"""
try:
if not keys:
return []
values = self.connection.mget(keys)
return [pickle.loads(v) for v in values if v is not None]
except TypeError as exc:
raise TypeError('RedisCache only accepts values that can be pickled. ') from exc
async def amget(self, keys: typing.List[str]) -> typing.List[typing.Any] | None:
"""Asynchronous Batch Acquisition"""
try:
if not keys:
return []
values = await self.async_connection.mget(keys)
return [pickle.loads(v) for v in values if v is not None]
except TypeError as exc:
raise TypeError('RedisCache only accepts values that can be pickled. ') from exc
async def akeys(self, pattern: str) -> typing.List[str]:
"""Get all keys matching patterns asynchronously"""
try:
await self.acluster_nodes(pattern)
keys = await self.async_connection.keys(pattern)
return [key.decode('utf-8') for key in keys]
except Exception as e:
raise e
def hsetkey(self, name, key, value, expiration=3600):
try:
self.cluster_nodes(key)
r = self.connection.hset(name, key, value)
if expiration:
self.connection.expire(name, expiration)
return r
except Exception as e:
raise e
async def ahsetkey(self, name, key, value, expiration=3600):
try:
await self.acluster_nodes(key)
r = await self.async_connection.hset(name, key, value)
if expiration:
await self.async_connection.expire(name, expiration)
return r
except Exception as e:
raise e
def hset(self, name,
key: Optional[str] = None,
value: Optional[str] = None,
mapping: Optional[dict] = None,
items: Optional[list] = None,
expiration: int = 3600):
try:
self.cluster_nodes(name)
r = self.connection.hset(name, key, value, mapping, items)
if expiration:
self.connection.expire(name, expiration)
return r
except Exception as e:
raise e
async def ahset(self, name,
key: Optional[str] = None,
value: Optional[str] = None,
mapping: Optional[dict] = None,
items: Optional[list] = None,
expiration: int = 3600):
try:
await self.acluster_nodes(name)
r = await self.async_connection.hset(name, key, value, mapping, items)
if expiration:
await self.async_connection.expire(name, expiration)
return r
except Exception as e:
raise e
def hget(self, name, key):
try:
self.cluster_nodes(name)
return self.connection.hget(name, key)
except Exception as e:
raise e
async def ahget(self, name, key):
try:
await self.acluster_nodes(name)
return await self.async_connection.hget(name, key)
except Exception as e:
raise e
def hgetall(self, name):
try:
self.cluster_nodes(name)
return self.connection.hgetall(name)
except Exception as e:
raise e
async def ahgetall(self, name):
try:
await self.acluster_nodes(name)
return await self.async_connection.hgetall(name)
except Exception as e:
raise e
def hdel(self, name, *keys):
try:
self.cluster_nodes(name)
return self.connection.hdel(name, *keys)
except Exception as e:
raise e
async def ahdel(self, name, *keys):
try:
await self.acluster_nodes(name)
return await self.async_connection.hdel(name, *keys)
except Exception as e:
raise e
def get(self, key):
try:
self.cluster_nodes(key)
value = self.connection.get(key)
return pickle.loads(value) if value else None
except Exception as e:
raise e
async def aget(self, key):
try:
await self.acluster_nodes(key)
value = await self.async_connection.get(key)
return pickle.loads(value) if value else None
except Exception as e:
# Handle the case where the value is None or not picklable
raise e
def incr(self, key, expiration=3600) -> int:
try:
self.cluster_nodes(key)
value = self.connection.incr(key)
if expiration:
self.connection.expire(key, expiration)
return value
except Exception as e:
raise e
async def aincr(self, key, expiration=3600) -> int:
try:
await self.acluster_nodes(key)
value = await self.async_connection.incr(key)
if expiration:
await self.async_connection.expire(key, expiration)
return value
except Exception as e:
raise e
def expire_key(self, key, expiration: int):
try:
self.cluster_nodes(key)
self.connection.expire(key, expiration)
except Exception as e:
raise e
async def aexpire_key(self, key, expiration: int):
try:
await self.acluster_nodes(key)
await self.async_connection.expire(key, expiration)
except Exception as e:
raise e
def delete(self, key):
try:
self.cluster_nodes(key)
return self.connection.delete(key)
except Exception as e:
raise e
async def adelete(self, key):
try:
await self.acluster_nodes(key)
return await self.async_connection.delete(key)
except Exception as e:
raise e
async def alpush(self, key, value, expiration=3600):
try:
await self.acluster_nodes(key)
ret = await self.async_connection.lpush(key, value)
if expiration:
await self.aexpire_key(key, expiration)
return ret
except Exception as e:
raise e
async def ablpop(self, key, timeout=0):
try:
await self.acluster_nodes(key)
value = await self.async_connection.blpop(key, timeout)
return pickle.loads(value[1]) if value and value[1] else None
except Exception as e:
raise e
async def alrange(self, key, start=0, end=-1):
try:
await self.acluster_nodes(key)
values = await self.async_connection.lrange(key, start, end)
return [pickle.loads(v) for v in values if v is not None]
except Exception as e:
raise e
async def alrem(self, key, value):
try:
await self.acluster_nodes(key)
value = pickle.dumps(value) if not isinstance(value, bytes) else value
return await self.async_connection.lrem(key, 0, value)
except Exception as e:
raise e
def rpush(self, key, value, expiration=3600):
try:
self.cluster_nodes(key)
ret = self.connection.rpush(key, value)
if expiration:
self.expire_key(key, expiration)
return ret
except Exception as e:
raise e
async def arpush(self, key, value, expiration=3600):
try:
await self.acluster_nodes(key)
value = pickle.dumps(value) if not isinstance(value, bytes) else value
ret = await self.async_connection.rpush(key, value)
if expiration:
await self.aexpire_key(key, expiration)
return ret
except Exception as e:
raise e
def lpop(self, key, count: int = None):
try:
self.cluster_nodes(key)
return self.connection.lpop(key, count)
except Exception as e:
raise e
async def alpop(self, key, count: int = None):
try:
await self.acluster_nodes(key)
return await self.async_connection.lpop(key, count)
except Exception as e:
raise e
def publish(self, key, value):
try:
self.cluster_nodes(key)
return self.connection.publish(key, value)
except Exception as e:
raise e
async def apublish(self, key, value):
try:
await self.acluster_nodes(key)
return await self.async_connection.publish(key, value)
except Exception as e:
raise e
def exists(self, key):
try:
self.cluster_nodes(key)
return self.connection.exists(key)
except Exception as e:
raise e
async def aexists(self, key):
try:
await self.acluster_nodes(key)
return await self.async_connection.exists(key)
except Exception as e:
raise e
def close(self):
self.connection.close()
async def aclose(self):
"""Asynchronous close method for the Redis connection."""
if hasattr(self, 'async_connection') and self.async_connection:
await self.async_connection.close()
else:
logger.warning("No async connection to close.")
# ==================== PipelineSupport ====================
def pipeline(self, transaction: bool = True) -> redis.client.Pipeline:
"""DapatkanpipelineObjects"""
return self.connection.pipeline(transaction=transaction)
def async_pipeline(self, transaction: bool = True) -> Pipeline:
"""Get AsynchronouspipelineObjects"""
return self.async_connection.pipeline(transaction=transaction)
async def allen(self, key: str) -> int:
"""Check if the key is in the cache using the 'in' operator."""
await self.acluster_nodes(key)
return await self.async_connection.llen(key)
def __contains__(self, key):
"""Check if the key is in the cache."""
self.cluster_nodes(key)
return False if key is None else self.connection.exists(key)
def __getitem__(self, key):
"""Retrieve an item from the cache using the square bracket notation."""
self.cluster_nodes(key)
return self.connection.get(key)
def __setitem__(self, key, value):
"""Add an item to the cache using the square bracket notation."""
self.cluster_nodes(key)
self.connection.set(key, value)
def __delitem__(self, key):
"""Remove an item from the cache using the square bracket notation."""
self.cluster_nodes(key)
self.connection.delete(key)
def cluster_nodes(self, key):
if isinstance(self.connection,
RedisCluster) and self.connection.get_default_node() is None:
target = self.connection.get_node_from_key(key)
self.connection.set_default_node(target)
async def acluster_nodes(self, key):
if isinstance(self.async_connection,
AsyncRedisCluster) and self.async_connection.get_default_node() is None:
target = self.async_connection.get_node_from_key(key)
self.async_connection.set_default_node(target)
================================================
FILE: src/backend/bisheng/core/cache/redis_manager.py
================================================
import logging
from typing import Optional, Union, Dict
from bisheng.core.context import BaseContextManager
from bisheng.core.cache.redis_conn import RedisClient
logger = logging.getLogger(__name__)
class RedisManager(BaseContextManager[RedisClient]):
"""Redis Global Manager
Responsible for management Redis Global lifecycle of connectivity, providing a unified access interface
Supports connection pool monitoring, health checks, and easy operations management
"""
name: str = "redis"
def __init__(
self,
redis_url: Optional[Union[str, Dict]] = None,
**kwargs
):
super().__init__(self.name, **kwargs)
self.redis_url = redis_url
if not self.redis_url:
raise ValueError("Redis URL is required. Please provide via parameter.")
async def _async_initialize(self) -> RedisClient:
"""Inisialisasi Redis Connection Manager"""
return RedisClient(self.redis_url)
def _sync_initialize(self) -> RedisClient:
"""Synchronization Initialization"""
return RedisClient(self.redis_url)
def _sync_cleanup(self) -> None:
"""Synchronous Cleanup Redis reasourse"""
if self._instance:
self._instance.close()
async def _async_cleanup(self) -> None:
"""Cleaned Redis reasourse"""
if self._instance:
await self._instance.aclose()
async def get_redis_client() -> RedisClient:
"""Dapatkan Redis Client Instance"""
from bisheng.core.context.manager import app_context
try:
return await app_context.async_get_instance(RedisManager.name)
except KeyError:
logger.warning(f"RedisManager not found in app_context. Registering a new instance.")
try:
from bisheng.common.services.config_service import settings
app_context.register_context(RedisManager(
redis_url=settings.redis_url
))
return await app_context.async_get_instance(RedisManager.name)
except Exception as e:
logger.error(f"Failed to register RedisManager: {e}")
raise
def get_redis_client_sync() -> RedisClient:
"""Synchronous fetch Redis Client Instance"""
from bisheng.core.context.manager import app_context
try:
return app_context.sync_get_instance(RedisManager.name)
except KeyError:
logger.warning(f"RedisManager not found in app_context. Registering a new instance.")
try:
from bisheng.common.services.config_service import settings
app_context.register_context(RedisManager(
redis_url=settings.redis_url
))
return app_context.sync_get_instance(RedisManager.name)
except Exception as e:
logger.error(f"Failed to register RedisManager: {e}")
raise
================================================
FILE: src/backend/bisheng/core/cache/utils.py
================================================
import asyncio
import base64
import contextlib
import functools
import hashlib
import json
import os
import shutil
import tempfile
from collections import OrderedDict
from io import BytesIO
from pathlib import Path
from typing import Any, Dict, Union, BinaryIO
from urllib.parse import unquote, urlparse
from uuid import uuid4
import aiofiles
import cchardet
import requests
from appdirs import user_cache_dir
from fastapi import UploadFile
from urllib3.util import parse_url
from bisheng.core.external.http_client.http_client_manager import get_http_client
from bisheng.core.storage.minio.minio_manager import get_minio_storage, get_minio_storage_sync
CACHE: Dict[str, Any] = {}
CACHE_DIR = user_cache_dir('bisheng', 'bisheng')
def create_cache_folder(func):
def wrapper(*args, **kwargs):
# Get the destination folder
cache_path = Path(CACHE_DIR) / PREFIX
# Create the destination folder if it doesn't exist
os.makedirs(cache_path, exist_ok=True)
return func(*args, **kwargs)
return wrapper
def create_cache_folder_async(func):
async def wrapper(*args, **kwargs):
# Get the destination folder
cache_path = Path(CACHE_DIR) / PREFIX
# Create the destination folder if it doesn't exist
os.makedirs(cache_path, exist_ok=True)
return await func(*args, **kwargs)
return wrapper
def memoize_dict(maxsize=128):
cache = OrderedDict()
def decorator(func):
@functools.wraps(func)
def wrapper(*args, **kwargs):
hashed = compute_dict_hash(args[0])
key = (func.__name__, hashed, frozenset(kwargs.items()))
if key not in cache:
result = func(*args, **kwargs)
cache[key] = result
if len(cache) > maxsize:
cache.popitem(last=False)
else:
result = cache[key]
return result
def clear_cache():
cache.clear()
wrapper.clear_cache = clear_cache # type: ignore
wrapper.cache = cache # type: ignore
return wrapper
return decorator
PREFIX = 'bisheng_cache'
@create_cache_folder
def clear_old_cache_files(max_cache_size: int = 3):
cache_dir = Path(tempfile.gettempdir()) / PREFIX
cache_files = list(cache_dir.glob('*.dill'))
if len(cache_files) > max_cache_size:
cache_files_sorted_by_mtime = sorted(cache_files,
key=lambda x: x.stat().st_mtime,
reverse=True)
for cache_file in cache_files_sorted_by_mtime[max_cache_size:]:
with contextlib.suppress(OSError):
os.remove(cache_file)
def compute_dict_hash(graph_data):
graph_data = filter_json(graph_data)
cleaned_graph_json = json.dumps(graph_data, sort_keys=True)
return hashlib.sha256(cleaned_graph_json.encode('utf-8')).hexdigest()
def filter_json(json_data):
filtered_data = json_data.copy()
# Remove 'viewport' and 'chatHistory' keys
if 'viewport' in filtered_data:
del filtered_data['viewport']
if 'chatHistory' in filtered_data:
del filtered_data['chatHistory']
# Filter nodes
if 'nodes' in filtered_data:
for node in filtered_data['nodes']:
if 'position' in node:
del node['position']
if 'positionAbsolute' in node:
del node['positionAbsolute']
if 'selected' in node:
del node['selected']
if 'dragging' in node:
del node['dragging']
return filtered_data
@create_cache_folder
def save_binary_file(content: str, file_name: str, accepted_types: list[str]) -> str:
"""
Save a binary file to the specified folder.
Args:
content: The content of the file as a bytes object.
file_name: The name of the file, including its extension.
Returns:
The path to the saved file.
"""
if not any(file_name.endswith(suffix) for suffix in accepted_types):
raise ValueError(f'File {file_name} is not accepted')
# Get the destination folder
cache_path = Path(CACHE_DIR) / PREFIX
if not content:
raise ValueError('Please, reload the file in the loader.')
data = content.split(',')[1]
decoded_bytes = base64.b64decode(data)
# Create the full file path
file_path = os.path.join(cache_path, file_name)
# Save the binary content to the file
with open(file_path, 'wb') as file:
file.write(decoded_bytes)
return file_path
def detect_encoding_cchardet(file_bytes: bytes, num_bytes=1024):
"""UsecchardetEncoding of the test file"""
result = cchardet.detect(file_bytes)
encoding = result['encoding']
confidence = result['confidence']
return encoding, confidence
def convert_encoding_cchardet(content: bytes, target_encoding='utf-8') -> BytesIO:
"""
Convert file encoding to target_encoding using cchardet for detection.
Args:
content:
target_encoding:
Returns:
BytesIO:
"""
source_encoding, confidence = detect_encoding_cchardet(content)
if confidence is None or confidence < 0.5 or source_encoding.lower() == target_encoding:
return BytesIO(content)
try:
# decode using detected encoding
text = content.decode(source_encoding)
except (UnicodeDecodeError, LookupError):
# If decoding fails, replace errors
text = content.decode(target_encoding, errors='replace')
# encode to target encoding
return BytesIO(text.encode(target_encoding))
async def upload_file_to_minio(file: UploadFile, object_name, bucket_name: str) -> str:
minio_client = await get_minio_storage()
await minio_client.put_object(bucket_name=bucket_name, object_name=object_name, file=file.file)
return await minio_client.get_share_link(object_name, bucket_name)
@create_cache_folder_async
async def save_file_to_folder(file: UploadFile, folder_name: str, file_name: str) -> str:
"""
Save uploaded file tofolder_nameFolders
:param file:
:param folder_name:
:param file_name:
:return:
"""
cache_path = Path(CACHE_DIR)
folder_path = cache_path / folder_name
# Create the folder if it doesn't exist
os.makedirs(folder_path, exist_ok=True)
# Save the file to the specified folder
file_path = folder_path / file_name
async with aiofiles.open(file_path, 'wb') as out_file:
while content := await file.read(8192):
await out_file.write(content)
return str(file_path)
@create_cache_folder
async def save_uploaded_file(file: UploadFile, folder_name, file_name, bucket_name: str = None):
"""
Save an uploaded file to the specified folder with a hash of its content as the file name.
Args:
file: The uploaded file object.
folder_name: The name of the folder to save the file in.
file_name: The name of the file, including its extension.
bucket_name: The name of the bucket_name
Returns:
The path to the saved file.
"""
minio_client = await get_minio_storage()
if bucket_name is None:
bucket_name = minio_client.tmp_bucket
cache_path = Path(CACHE_DIR)
folder_path = cache_path / folder_name
# Create the folder if it doesn't exist
if not folder_path.exists():
folder_path.mkdir(parents=True, exist_ok=True)
file_ext = file_name.split('.')[-1].lower()
file_data_to_upload = None
is_converted_text = False
try:
if file_ext in ('txt', 'md', 'csv'):
raw_content = await file.read()
file_data_to_upload = await asyncio.to_thread(
convert_encoding_cchardet,
content=raw_content,
target_encoding='utf-8'
)
is_converted_text = True
else:
# For other file types, use the original uploaded file
file_data_to_upload = file.file
# Reset file pointer to the beginning
file_data_to_upload.seek(0)
is_converted_text = False
await minio_client.put_object_tmp(object_name=file_name, file=file_data_to_upload)
finally:
if is_converted_text and file_data_to_upload:
file_data_to_upload.close()
file_path = await minio_client.get_share_link(file_name, bucket_name, clear_host=False)
return file_path
@create_cache_folder
def save_download_file(file_input: Union[bytes, BinaryIO], folder_name: str, filename: str) -> str:
"""
Synchronous I/O intensive tasks:
Write data stream to a temporary file
Simultaneously calculate SHA256
Rename a file based on the hash
"""
# Convert to stream objects
if isinstance(file_input, bytes):
src_stream = BytesIO(file_input)
else:
src_stream = file_input
# Make sure the pointer is at the beginning
if hasattr(src_stream, 'seek'):
src_stream.seek(0)
# Prepare a temporary file (write a temporary random filename first to avoid not being able to determine the filename before the hash calculation is finished).
cache_path = Path(CACHE_DIR)
folder_path = cache_path / folder_name
# Create the folder if it doesn't exist
if not folder_path.exists():
folder_path.mkdir(exist_ok=True)
temp_filename = f"tmp_{uuid4().hex}"
temp_file_path = folder_path / temp_filename
sha256_hash = hashlib.sha256()
try:
# Write to temporary file and calculate SHA256 simultaneously
with open(temp_file_path, 'wb') as dst_file:
chunk_size = 65536 # 64KB
while True:
chunk = src_stream.read(chunk_size)
if not chunk:
break
sha256_hash.update(chunk)
dst_file.write(chunk)
# calculate final hash
file_hash = sha256_hash.hexdigest()
# Logic for handling filename length limits
safe_filename = filename
if len(filename) > 60:
safe_filename = filename[-60:]
final_file_name = f'{file_hash}_{safe_filename}'
final_file_path = folder_path / final_file_name
# Rename (Move) Temporary File to Final Path
# If the file already exists, decide whether to overwrite or skip it based on your needs. This example demonstrates overwriting.
if final_file_path.exists():
os.remove(temp_file_path)
return str(final_file_path)
shutil.move(str(temp_file_path), str(final_file_path))
return str(final_file_path)
except Exception as e:
# Clean up temporary file in case of error
if temp_file_path.exists():
os.remove(temp_file_path)
raise e
finally:
if isinstance(file_input, bytes):
src_stream.close()
def file_download(file_path: str):
"""download file and return path"""
# Try processing as a local file first (extracted URL Parameters)
# If the system mounts a storage volume, remove ? The signature parameters behind it are read directly
local_candidate = file_path.split('?')[0]
if os.path.isfile(local_candidate):
file_name = os.path.basename(local_candidate)
# Compatible with legacy logic: handles what might be included in the filename md5 Prefix
file_name = file_name.split('_', 1)[-1] if '_' in file_name else file_name
return local_candidate, file_name
# Legacy Logic: Check if it is standard URL (Bawa http/https)
if _is_valid_url(file_path):
minio_client = get_minio_storage_sync()
minio_share_host = minio_client.get_minio_share_host()
url_obj = urlparse(file_path)
filename = unquote(url_obj.path.split('/')[-1])
if file_path.startswith(minio_share_host):
# download file from minio sdk
bucket_name, object_name = url_obj.path.replace(minio_share_host, "", 1).lstrip("/").split('/', 1)
object_name = unquote(object_name)
file_content = minio_client.get_object_sync(bucket_name, object_name)
else:
# download file from http url
r = requests.get(file_path, verify=False)
if r.status_code != 200:
raise ValueError('Check the url of your file; returned status code %s' % r.status_code)
# OthersContent-Dispositionheader to find the filename
content_disposition = r.headers.get('Content-Disposition')
if content_disposition:
filename = unquote(content_disposition).split('filename=')[-1].strip("\"'")
file_content = r.content
file_path = save_download_file(file_content, 'bisheng', filename)
return file_path, filename
# Medical Treatment: MinIO Relative path (In / Starts with a signature parameter)
# For Input: /bisheng/original/82324.docx?X-Amz-Algorithm=...
# No in this case host, cannot be accessed _is_valid_url Branch
elif file_path.startswith('/') and 'X-Amz-Algorithm' in file_path:
try:
minio_client = get_minio_storage_sync()
# Use urlparse parsing, it automatically separates path And query
url_obj = urlparse(file_path)
# path similar to /bisheng/original/82324.docx
# Remove the beginning /, and then split the first / Get bucket And object
path_parts = url_obj.path.lstrip("/").split('/', 1)
if len(path_parts) == 2:
bucket_name, object_name = path_parts
# Call Synchronized minio Method download
object_name = unquote(object_name)
file_content = minio_client.get_object_sync(bucket_name, object_name)
filename = unquote(object_name.split('/')[-1])
file_path = save_download_file(file_content, 'bisheng', filename)
return file_path, filename
except Exception as e:
# If the parsing fails, print the log and let the program continue to throw down ValueError
print(f"Error handling relative MinIO path: {e}")
elif not os.path.isfile(file_path):
raise ValueError('File path %s is not a valid file or url' % file_path)
# This is the one that handles purely local file paths (the one with no parameters) and is usually handled by the topmost logic 1 Interception
file_name = os.path.basename(file_path)
file_name = file_name.split('_', 1)[-1] if '_' in file_name else file_name
return file_path, file_name
async def async_file_download(file_path: str):
"""download file and return path"""
# Try processing as a local file first (extracted URL Parameters)
# If the system mounts the storage volume, this will solve the problem directly
local_candidate = file_path.split('?')[0]
if os.path.isfile(local_candidate):
file_name = os.path.basename(local_candidate)
# Is it included under processing?md5Logic of (Keep original logic)
file_name = file_name.split('_', 1)[-1] if '_' in file_name else file_name
return local_candidate, file_name
# Check if it is standard URL
if _is_valid_url(file_path):
http_client = await get_http_client()
minio_client = await get_minio_storage()
minio_share_host = minio_client.get_minio_share_host()
url_obj = parse_url(file_path)
filename = unquote(url_obj.path.split('/')[-1])
if file_path.startswith(minio_share_host):
# download file from minio sdk
bucket_name, object_name = url_obj.path.replace(minio_share_host, "", 1).lstrip("/").split('/', 1)
object_name = unquote(object_name)
file_content = await minio_client.get_object(bucket_name, object_name)
else:
r = await http_client.get(url=file_path, data_type="binary")
if r.status_code != 200:
raise ValueError('Check the url of your file; returned status code %s' % r.status_code)
content_disposition = r.headers.get('Content-Disposition') if r.headers else None
if content_disposition:
filename = unquote(content_disposition).split('filename=')[-1].strip("\"'")
file_content = r.body
file_path = save_download_file(file_content, 'bisheng', filename)
return file_path, filename
# Medical Treatment: MinIO Relative path (In / Starts with a signature parameter)
# For Input: /bisheng/original/82324.docx?X-Amz-Algorithm=...
elif file_path.startswith("/") and "X-Amz-Algorithm" in file_path:
try:
minio_client = await get_minio_storage()
# Resolve Path /bucket/object_key
url_obj = urlparse(file_path)
# path turned into /bisheng/original/82324.docx, remove the opening / and split the first one /
path_parts = url_obj.path.lstrip("/").split('/', 1)
if len(path_parts) == 2:
bucket_name, object_name = path_parts
object_name = unquote(object_name)
# Directly usable after finished products leave the factory minio client Download without http Request
file_content = await minio_client.get_object(bucket_name, object_name)
filename = unquote(object_name.split('/')[-1])
file_path = save_download_file(file_content, 'bisheng', filename)
return file_path, filename
except Exception as e:
# If parsing or downloading fails, log or drop it below ValueError
print(f"Error handling relative MinIO path: {e}")
raise ValueError('File path %s is not a valid file or url' % file_path)
def _is_valid_url(url: str):
"""Check if the url is valid."""
parsed = urlparse(url)
return bool(parsed.netloc) and bool(parsed.scheme)
================================================
FILE: src/backend/bisheng/core/config/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/config/settings.py
================================================
import ast
import json
import os
import re
from typing import Dict, List, Optional, Union
from celery.schedules import crontab
from cryptography.fernet import Fernet
from loguru import logger
from pydantic import ConfigDict, BaseModel, Field, field_validator, model_validator
secret_key = 'TI31VYJ-ldAq-FXo5QNPKV_lqGTFfp-MIdbK2Hm5F1E='
def encrypt_token(token: str):
return Fernet(secret_key).encrypt(token.encode())
def decrypt_token(token: str):
return Fernet(secret_key).decrypt(token).decode()
class LoggerConf(BaseModel):
"""Looger Config"""
level: str = 'DEBUG'
format: str = '[{level.name} process-{process.id}-{thread.id} {name}:{line}] - trace={extra[trace_id]} {message}' # noqa
handlers: List[Dict] = Field(default_factory=list, description='Log Processor')
@classmethod
def parse_logger_sink(cls, sink: str) -> str:
match = re.search(r'\{(.+?)\}', sink)
if not match:
return sink
env_keys = {}
for one in match.groups():
env_keys[one] = os.getenv(one, '')
return sink.format(**env_keys)
@field_validator('handlers')
@classmethod
def set_handlers(cls, value):
if value is None:
value = []
for one in value:
one['sink'] = cls.parse_logger_sink(one['sink'])
if one.get('filter'):
one['filter'] = eval(one['filter'])
return value
class PasswordConf(BaseModel):
""" Password Config """
password_valid_period: Optional[int] = Field(default=0, description='Password overXDays must be modified, Login prompt to change password again')
login_error_time_window: Optional[int] = Field(default=0, description='Login error time window,minutes unit')
max_error_times: Optional[int] = Field(default=0, description='Maximum number of errors, after which the user will be banned')
class SystemLoginMethod(BaseModel):
""" System Login Method Config """
bisheng_pro: bool = Field(default=False, description='Whether it is a commercial version, Verify Configuredlicense')
admin_username: Optional[str] = Field(default=None, description='Admin username registered via web')
allow_multi_login: bool = Field(default=True, description='Whether to allow multi-sign-on')
class MilvusConf(BaseModel):
""" milvus Configure """
connection_args: Optional[dict] = Field(default=None, description='milvus Configure')
is_partition: Optional[bool] = Field(default=True, description='Is itpartitionMode',
deprecated="Not in SupportpartitionMode")
partition_suffix: Optional[str] = Field(default='1', description='partitionSuffix',
deprecated="Not in SupportpartitionMode")
@field_validator('connection_args', mode='before')
@classmethod
def convert_connection_args(cls, value):
if isinstance(value, str):
return json.loads(value)
return value
class ElasticsearchConf(BaseModel):
""" elasticsearch Configure """
elasticsearch_url: Optional[str] = Field(default=None, alias='url',
description='elasticsearchAccesses address')
ssl_verify: Optional[str | dict] = Field(default='{}', description='Additional Arguments')
@model_validator(mode='after')
def validate(self):
if isinstance(self.ssl_verify, str):
self.ssl_verify = ast.literal_eval(self.ssl_verify)
return self
class VectorStores(BaseModel):
""" Vector Storage Configuration """
milvus: MilvusConf = Field(default_factory=MilvusConf, description='milvus Configure')
elasticsearch: ElasticsearchConf = Field(default_factory=ElasticsearchConf, description='elasticsearch Configure')
class MinioConf(BaseModel):
""" minio Configure """
secure: Optional[bool] = Field(default=False, description="Apakah ingin digunakan?https", alias="schema")
cert_check: Optional[bool] = Field(default=False, description="Whether to calibrate the certificate")
endpoint: Optional[str] = Field(default="127.0.0.1:9000", description="minio Service address")
sharepoint: Optional[str] = Field(default="127.0.0.1:9000", description="minio Public access address")
share_schema: Optional[bool] = Field(default=False, description="minio Whether the public access address is usedhttps")
share_cert_check: Optional[bool] = Field(default=False, description="minio Whether the public access address verifies the certificate")
access_key: Optional[str] = Field(default="minioadmin", description="minio Username")
secret_key: Optional[str] = Field(default="minioadmin", description="minio Passwords")
public_bucket: Optional[str] = Field(default="bisheng",
description="Store permanent files by defaultbucket. Files can be permanently accessed by anonymous users")
tmp_bucket: Optional[str] = Field(default="tmp-dir", description="Ad hocbucket, stored files will have an expiration date")
class ObjectStore(BaseModel):
""" Object Storage Configuration """
type: str = Field(default='minio', description="Object Storage Type")
minio: Optional[MinioConf] = Field(default_factory=MinioConf, description="minio Configure")
class WorkflowConf(BaseModel):
""" Workflow Configuration """
max_steps: int = Field(default=50, description="Maximum number of steps a node can run")
timeout: int = Field(default=720, description="Node timeout (min）")
class CeleryConf(BaseModel):
""" Celery Configure """
task_routers: Optional[Dict] = Field(default_factory=dict, description='Task Routing Configuration')
beat_schedule: Optional[Dict] = Field(default_factory=dict, description='Timed Task Configuration')
@model_validator(mode='after')
def validate(self):
if not self.task_routers:
self.task_routers = {
"bisheng.worker.knowledge.*": {"queue": "knowledge_celery"}, # Knowledge Base Related Tasks
"bisheng.worker.workflow.*": {"queue": "workflow_celery"}, # Workflow Execution Related Tasks
}
if 'telemetry_mid_user_increment' not in self.beat_schedule:
self.beat_schedule['telemetry_mid_user_increment'] = {
'task': 'bisheng.worker.telemetry.mid_table.sync_mid_user_increment',
'schedule': crontab('*/30 0 * * *'), # 00:30 exec every day
}
if 'telemetry_mid_knowledge_increment' not in self.beat_schedule:
self.beat_schedule['telemetry_mid_knowledge_increment'] = {
'task': 'bisheng.worker.telemetry.mid_table.sync_mid_knowledge_increment',
'schedule': crontab('*/30 0 * * *'), # 00:30 exec every day
}
if 'telemetry_sync_mid_app_increment' not in self.beat_schedule:
self.beat_schedule['telemetry_sync_mid_app_increment'] = {
'task': 'bisheng.worker.telemetry.mid_table.sync_mid_app_increment',
'schedule': crontab('*/30 0 * * *'), # 00:30 exec every day
}
if 'telemetry_sync_mid_user_interact_dtl' not in self.beat_schedule:
self.beat_schedule['telemetry_sync_mid_user_interact_dtl'] = {
'task': 'bisheng.worker.telemetry.mid_table.sync_mid_user_interact_dtl',
'schedule': crontab('*/30 0 * * *'), # 00:30 exec every day
}
# convert str to crontab
for key, task_info in self.beat_schedule.items():
if isinstance(task_info['schedule'], str):
self.beat_schedule[key]['schedule'] = crontab(task_info['schedule'])
return self
class LinsightConf(BaseModel):
""" Inspiration Configuration """
debug: bool = Field(default=False, description='Whether to opendebugMode')
tool_buffer: int = Field(default=100000, description='Maximum Tool Execution Historytoken, you need to summarize your history after')
max_steps: int = Field(default=200, description='Maximum number of steps per task to prevent infinite loops')
retry_num: int = Field(default=3, description='Number of times the model call was retried during the execution of the Ideas task')
retry_sleep: int = Field(default=5, description='Interval between retries of model calls during execution of Invisible Tasks (seconds)')
max_file_num: int = Field(default=5, description='BuatSOPJampromptThe number of user-uploaded file information placed in the')
max_knowledge_num: int = Field(default=20, description='BuatSOPJampromptThe amount of knowledge base information placed in the')
waiting_list_url: str = Field(default=None, description='waiting list Jump link')
default_temperature: float = Field(default=0, description='Default Temperature at Model Request')
retry_temperature: float = Field(default=1, description='reactModejsonModel temperature when retrying after parsing failure')
file_content_length: int = Field(default=5000, description='The number of characters to read the contents of the file when splitting subtasks, which will be truncated when exceeded')
max_file_content_num: int = Field(default=3, description='Number of files to read when subtasking, in reverse order by modification time')
class CookieConf(BaseModel):
""" Cookie Configure """
max_age: Optional[int] = Field(default=None, description="Cookie Maximum survival time in seconds")
path: str = Field(default='/', description="Cookie Path Properties for")
domain: Optional[str] = Field(default=None, description="Cookie Domain Properties for")
secure: bool = Field(default=False, description="enabled secure Property")
httponly: bool = Field(default=True, description="enabled HttpOnly Property")
samesite: str = Field(default=None, description="SameSite property, optional value is 'lax', 'strict', 'none'")
jwt_token_expire_time: int = Field(default=86400, description="JwtTokenExpiration time in seconds")
jwt_iss: str = Field(default='bisheng', description="JwtTokenIssuer of")
class Etl4lmConf(BaseModel):
""" Etl4lm Configure """
url: str = Field(default='', description='etl4lmService Address')
timeout: int = Field(default=600, description='etl4lmService Request Timeout (sec)')
ocr_sdk_url: str = Field(default='', description='etl4lm ocr sdkService Address')
class KnowledgeConf(BaseModel):
""" Knowledge Configure """
etl4lm: Etl4lmConf
class Settings(BaseModel):
""" Application Settings """
model_config = ConfigDict(validate_assignment=True, arbitrary_types_allowed=True, extra='ignore')
chains: dict = {}
agents: dict = {}
prompts: dict = {}
llms: dict = {}
tools: dict = {}
memories: dict = {}
embeddings: dict = {}
knowledges: dict = {}
vectorstores: dict = {}
documentloaders: dict = {}
wrappers: dict = {}
retrievers: dict = {}
toolkits: dict = {}
textsplitters: dict = {}
utilities: dict = {}
input_output: dict = {}
output_parsers: dict = {}
autogen_roles: dict = {}
dev: bool = False
environment: Union[dict, str] = 'dev'
# ↑↑↑ before config for langchain flow, will be deprecated
debug: bool = False
database_url: Optional[str] = None
redis_url: Optional[Union[str, Dict]] = None
celery_redis_url: Optional[Union[str, Dict]] = None
redis: Optional[dict] = None
admin: dict = {}
cache: str = 'InMemoryCache'
remove_api_keys: bool = False
bisheng_rt: dict = {}
default_llm: dict = {}
jwt_secret: str = 'secret'
gpts: dict = {}
openai_conf: dict = {}
minio_conf: dict = {}
linsight_conf: LinsightConf = LinsightConf()
logger_conf: LoggerConf = LoggerConf()
password_conf: PasswordConf = PasswordConf()
system_login_method: SystemLoginMethod = SystemLoginMethod()
vector_stores: VectorStores = VectorStores()
object_storage: ObjectStore = ObjectStore()
workflow_conf: WorkflowConf = WorkflowConf()
celery_task: CeleryConf = CeleryConf()
cookie_conf: CookieConf = CookieConf()
telemetry_elasticsearch: ElasticsearchConf = ElasticsearchConf()
license_str: Optional[str] = None # license Contents
@field_validator('database_url')
@classmethod
def set_database_url(cls, value):
if not value:
logger.debug('No database_url provided, trying bisheng_DATABASE_URL env variable')
if bisheng_database_url := os.getenv('bisheng_DATABASE_URL'):
value = bisheng_database_url
else:
logger.debug('No DATABASE_URL env variable, using sqlite database')
value = 'sqlite:///./bisheng.db'
else:
# Encrypt password
import re
pattern = r'(?<=:)[^:]+(?=@)' # Match colon after to@Any character before the symbol
match = re.search(pattern, value)
if match:
password = match.group(0)
new_password = decrypt_token(password)
new_mysql_url = re.sub(pattern, f'{new_password}', value)
value = new_mysql_url
return value
@model_validator(mode='before')
@classmethod
def set_redis_url(cls, values):
if 'redis_url' in values:
if isinstance(values['redis_url'], dict):
for k, v in values['redis_url'].items():
if isinstance(v, str) and v.startswith('encrypt(') and v.endswith(')'):
v = v[8:-1]
values['redis_url'][k] = decrypt_token(v)
else:
import re
pattern = r'(?<=:)[^:]+(?=@)' # Match colon after to@Any character before the symbol
match = re.search(pattern, values['redis_url'])
if match:
password = match.group(0)
new_password = decrypt_token(password)
new_redis_url = re.sub(pattern, f'{new_password}', values['redis_url'])
values['redis_url'] = new_redis_url
return values
@model_validator(mode='before')
@classmethod
def set_celery_redis_url(cls, values):
if 'celery_redis_url' in values:
if isinstance(values['celery_redis_url'], dict):
for k, v in values['celery_redis_url'].items():
if isinstance(v, str) and v.startswith('encrypt(') and v.endswith(')'):
v = v[8:-1]
values['celery_redis_url'][k] = decrypt_token(v)
else:
import re
pattern = r'(?<=:)[^:]+(?=@)' # Match colon after to@Any character before the symbol
match = re.search(pattern, values['celery_redis_url'])
if match:
password = match.group(0)
new_password = decrypt_token(password)
new_redis_url = re.sub(pattern, f'{new_password}', values['celery_redis_url'])
values['celery_redis_url'] = new_redis_url
return values
@model_validator(mode='before')
@classmethod
def validate_lists(cls, values):
for key, value in values.items():
if key != 'dev' and not value:
values[key] = []
return values
def get_minio_conf(self) -> MinioConf:
return self.object_storage.minio
def get_vectors_conf(self) -> VectorStores:
return self.vector_stores
def get_search_conf(self) -> ElasticsearchConf:
return self.vector_stores.elasticsearch
def get_telemetry_conf(self) -> ElasticsearchConf:
if not self.telemetry_elasticsearch.elasticsearch_url:
return self.vector_stores.elasticsearch
return self.telemetry_elasticsearch
================================================
FILE: src/backend/bisheng/core/context/__init__.py
================================================
"""Application Context Management Module
Provide a common context management framework, including:
- Thread-safe lazy loading mechanism
- Automatic Caching and Instance Lifecycle Management
- Dependency injection and initialization sequence control
- Retry mechanism and timeout handling
- Health check and monitoring functions
- Convenient context manager support
Usage examples:
# basic use
from bisheng.core.context import initialize_app_context, get_context
await initialize_app_context()
db = await get_context('database').async_get_instance()
# Using the Context Manager
async with get_context('database').async_context() as db:
# Use db
pass
"""
from .base import (
# Core classes and statuses
BaseContextManager,
FunctionContextManager,
ContextRegistry,
ContextState,
# Exception Class
ContextError,
ContextInitializationError,
ContextTimeoutError,
ContextStateError,
)
from .manager import (
# APP MANAGER
ApplicationContextManager,
app_context,
# Initialization and Lifecycle Management
initialize_app_context,
close_app_context,
# Context Access and Management
get_context,
async_get_instance,
sync_get_instance,
register_context,
# Monitoring and Diagnostics
health_check,
)
__all__ = [
# Core Base Classes and States
'BaseContextManager',
'FunctionContextManager',
'ContextRegistry',
'ContextState',
# Exception Class
'ContextError',
'ContextInitializationError',
'ContextTimeoutError',
'ContextStateError',
# APP MANAGER
'ApplicationContextManager',
'app_context',
# PRODUCT LIFECYCLE MANAGEMENT
'initialize_app_context',
'close_app_context',
# Context Access and Management
'get_context',
'async_get_instance',
'sync_get_instance',
'register_context',
# Monitoring and Diagnostics
'health_check',
]
================================================
FILE: src/backend/bisheng/core/context/base.py
================================================
import asyncio
import time
from abc import ABC, abstractmethod
from contextlib import asynccontextmanager, contextmanager
from enum import Enum
from threading import Lock, Event
from typing import Any, Dict, Optional, TypeVar, Generic, Callable, Awaitable, Union
from loguru import logger
T = TypeVar('T')
class ContextState(Enum):
"""Context State Enumeration"""
UNINITIALIZED = "uninitialized" # Not Initialized
INITIALIZING = "initializing" # Initializing
READY = "ready" # Working on it...
ERROR = "error" # Error State
CLOSING = "closing" # closing
CLOSED = "closed" # Closed
class ContextError(Exception):
"""Context-dependent exception base class"""
pass
class ContextInitializationError(ContextError):
"""Context initialization exception"""
pass
class ContextTimeoutError(ContextError):
"""Context operation timeout exception"""
pass
class ContextStateError(ContextError):
"""Context status exception"""
pass
class BaseContextManager(ABC, Generic[T]):
"""Underlying Manager Abstract Class
Defines the interfaces that all context managers must implement
Provides thread-safe lazy loading, caching, and lifecycle management
"""
name: str
_default_timeout: float = 30.0 # Timeout default
_default_retry_count: int = 3 # Default number of retries
def __init__(self, name: str = None, timeout: float = None, retry_count: int = None, **kwargs):
self.name = name or getattr(self.__class__, 'name', self.__class__.__name__.lower())
self.timeout = timeout or self._default_timeout
self.retry_count = retry_count or self._default_retry_count
self.state = ContextState.UNINITIALIZED
self._instance: Optional[T] = None
self._error: Optional[Exception] = None
# Synchronous and asynchronous locks
self._sync_lock = Lock()
self._async_lock = asyncio.Lock()
# Synchronization Waiting Event
self._sync_ready_event = Event()
self._async_ready_event = asyncio.Event()
@abstractmethod
async def _async_initialize(self) -> T:
"""Asynchronous initialization resources (abstract methods)"""
pass
@abstractmethod
def _sync_initialize(self) -> T:
"""Synchronous initialization resources (abstract methods)"""
pass
@abstractmethod
async def _async_cleanup(self) -> None:
"""Asynchronous Cleanup Resource (Abstract Method)"""
pass
@abstractmethod
def _sync_cleanup(self) -> None:
"""Synchronous Cleanup Resource (Abstract Method)"""
pass
def _validate_state_for_access(self) -> None:
"""Verify that the status allows access to the instance"""
if self.state == ContextState.ERROR:
error_msg = f"Context '{self.name}' is in error state"
if self._error:
error_msg += f": {self._error}"
raise ContextStateError(error_msg)
if self.state == ContextState.CLOSED:
raise ContextStateError(f"Context '{self.name}' is closed and cannot be accessed")
async def _wait_for_initialization_async(self) -> None:
"""Asynchronous waiting for initialization to complete"""
try:
await asyncio.wait_for(self._async_ready_event.wait(), timeout=self.timeout)
except asyncio.TimeoutError:
raise ContextTimeoutError(f"Context '{self.name}' initialization timeout after {self.timeout}s")
def _wait_for_initialization_sync(self) -> None:
"""Sync waiting for initialization to complete"""
if not self._sync_ready_event.wait(timeout=self.timeout):
raise ContextTimeoutError(f"Context '{self.name}' initialization timeout after {self.timeout}s")
async def async_get_instance(self) -> T:
"""Asynchronous Get Context Instance
Returns:
T: Contextual instance after initialization
Raises:
ContextStateError: Context is in error state or closed
ContextTimeoutError: Initialization timeout
ContextInitializationError: Initialization failed
"""
# Quick Path: Instance Ready
if self.state == ContextState.READY and self._instance is not None:
return self._instance
self._validate_state_for_access()
async with self._async_lock:
# Dual Lock Check Mode
if self.state == ContextState.READY and self._instance is not None:
return self._instance
# If initializing, wait for completion
if self.state == ContextState.INITIALIZING:
await self._wait_for_initialization_async()
self._validate_state_for_access()
if self.state == ContextState.READY and self._instance is not None:
return self._instance
# Start Initialization
return await self._perform_initialization_async()
async def _perform_initialization_async(self) -> T:
"""Execute asynchronous initialization logic"""
self.state = ContextState.INITIALIZING
self._async_ready_event.clear()
last_error = None
for attempt in range(self.retry_count):
try:
logger.debug(f"Initializing context '{self.name}' (attempt {attempt + 1}/{self.retry_count})")
self._instance = await self._async_initialize()
self.state = ContextState.READY
self._async_ready_event.set()
self._sync_ready_event.set()
logger.debug(f"Context '{self.name}' initialized successfully")
return self._instance
except Exception as e:
last_error = e
if attempt < self.retry_count - 1:
wait_time = 2 ** attempt # Exponential withdrawal
logger.warning(
f"Context '{self.name}' init attempt {attempt + 1} failed: {e}, retrying in {wait_time}s")
await asyncio.sleep(wait_time)
else:
logger.error(f"Context '{self.name}' initialization failed after {self.retry_count} attempts: {e}")
# All retries failed
self.state = ContextState.ERROR
self._error = last_error
self._async_ready_event.set() # Failed to notify waiter
self._sync_ready_event.set()
raise ContextInitializationError(f"Failed to initialize context '{self.name}': {last_error}") from last_error
def sync_get_instance(self) -> T:
"""Synchronize Get Context Instance
Returns:
T: Contextual instance after initialization
Raises:
ContextStateError: Context is in error state, closed or initialization timed out
ContextInitializationError: Initialization failed
"""
# Quick Path: Instance Ready
if self.state == ContextState.READY and self._instance is not None:
return self._instance
self._validate_state_for_access()
with self._sync_lock:
# Dual Lock Check Mode
if self.state == ContextState.READY and self._instance is not None:
return self._instance
# If initializing, wait for completion
if self.state == ContextState.INITIALIZING:
self._wait_for_initialization_sync()
self._validate_state_for_access()
if self.state == ContextState.READY and self._instance is not None:
return self._instance
# Start Initialization
return self._perform_initialization_sync()
def _perform_initialization_sync(self) -> T:
"""Execute Synchronization Initialization Logic"""
self.state = ContextState.INITIALIZING
self._sync_ready_event.clear()
self._async_ready_event.clear()
last_error = None
for attempt in range(self.retry_count):
try:
logger.debug(f"Initializing context '{self.name}' (attempt {attempt + 1}/{self.retry_count})")
self._instance = self._sync_initialize()
self.state = ContextState.READY
self._sync_ready_event.set()
self._async_ready_event.set()
logger.debug(f"Context '{self.name}' initialized successfully")
return self._instance
except Exception as e:
last_error = e
if attempt < self.retry_count - 1:
wait_time = 2 ** attempt # Exponential withdrawal
logger.warning(
f"Context '{self.name}' init attempt {attempt + 1} failed: {e}, retrying in {wait_time}s")
time.sleep(wait_time)
else:
logger.error(f"Context '{self.name}' initialization failed after {self.retry_count} attempts: {e}")
# All retries failed
self.state = ContextState.ERROR
self._error = last_error
self._sync_ready_event.set() # Failed to notify waiter
self._async_ready_event.set()
raise ContextInitializationError(f"Failed to initialize context '{self.name}': {last_error}") from last_error
async def async_close(self) -> None:
"""Shutdown context manager asynchronously
Safely release resources and set status to off
"""
if self.state in [ContextState.CLOSING, ContextState.CLOSED]:
return
async with self._async_lock:
if self.state in [ContextState.CLOSING, ContextState.CLOSED]:
return
self.state = ContextState.CLOSING
try:
if self._instance is not None:
logger.debug(f"Closing context '{self.name}'")
await self._async_cleanup()
self._instance = None
logger.debug(f"Context '{self.name}' closed successfully")
except Exception as e:
logger.error(f"Error closing context '{self.name}': {e}")
# Mark as closed to avoid resource leakage even if cleanup fails
finally:
self.state = ContextState.CLOSED
self._error = None
# Ensure events are set to prevent unlimited wait time for waiters
self._async_ready_event.set()
self._sync_ready_event.set()
def sync_close(self) -> None:
"""Sync Close Context Manager
Safely release resources and set status to off
"""
if self.state in [ContextState.CLOSING, ContextState.CLOSED]:
return
with self._sync_lock:
if self.state in [ContextState.CLOSING, ContextState.CLOSED]:
return
self.state = ContextState.CLOSING
try:
if self._instance is not None:
logger.debug(f"Closing context '{self.name}'")
self._sync_cleanup()
self._instance = None
logger.debug(f"Context '{self.name}' closed successfully")
except Exception as e:
logger.error(f"Error closing context '{self.name}': {e}")
# Mark as closed to avoid resource leakage even if cleanup fails
finally:
self.state = ContextState.CLOSED
self._error = None
# Ensure events are set to prevent unlimited wait time for waiters
self._async_ready_event.set()
self._sync_ready_event.set()
async def async_reset(self) -> None:
"""Reset Context Manager
Close the current instance and reset it to the uninitialized state, it will be reinitialized on the next visit
"""
await self.async_close()
async with self._async_lock:
self.state = ContextState.UNINITIALIZED
self._error = None
self._async_ready_event.clear()
self._sync_ready_event.clear()
def sync_reset(self) -> None:
"""Reset Context Manager
Close the current instance and reset it to the uninitialized state, it will be reinitialized on the next visit
"""
self.sync_close()
with self._sync_lock:
self.state = ContextState.UNINITIALIZED
self._error = None
self._async_ready_event.clear()
self._sync_ready_event.clear()
def is_ready(self) -> bool:
"""Check to see if it's ready
Returns:
bool: True If the context is initialized and available,False Otherwise,
"""
return self.state == ContextState.READY and self._instance is not None
def get_state(self) -> ContextState:
"""Fetch the current status
Returns:
ContextState: Current Context State
"""
return self.state
def get_error(self) -> Optional[Exception]:
"""Get the last error message
Returns:
Optional[Exception]: If in error state, return error message, otherwise return None
"""
return self._error
def get_info(self) -> Dict[str, Any]:
"""Get Context Information
Returns:
Dict[str, Any]: Dictionary with contextual details
"""
return {
'name': self.name,
'state': self.state.value,
'is_ready': self.is_ready(),
'timeout': self.timeout,
'retry_count': self.retry_count,
'error': str(self._error) if self._error else None,
'has_instance': self._instance is not None
}
@contextmanager
def sync_context(self):
"""Synchronization Context Manager
Use with Statements automatically fetch and free resources
Example:
with my_context.sync_context() as instance:
# Use instance
pass
"""
instance = self.sync_get_instance()
try:
yield instance
finally:
# Note: This does not close automatically as it may be used elsewhere
pass
@asynccontextmanager
async def async_context(self):
"""Asynchronous Context Manager
Use async with Statements automatically fetch and free resources
Example:
async with my_context.async_context() as instance:
# Use instance
pass
"""
instance = await self.async_get_instance()
try:
yield instance
finally:
# Note: This does not close automatically as it may be used elsewhere
pass
def __repr__(self) -> str:
return f"<{self.__class__.__name__}(name='{self.name}', state='{self.state.value}')>"
class FunctionContextManager(BaseContextManager[T]):
"""Function-based context manager
Allows initialization and cleanup logic to be defined via functions
"""
def __init__(
self,
name: str,
init_func: Union[Callable[[], Awaitable[T]], Callable[[], T]],
cleanup_func: Optional[Union[Callable[[T], Awaitable[None]], Callable[[T], None]]] = None,
**kwargs
):
super().__init__(name, **kwargs)
self.init_func = init_func
self.cleanup_func = cleanup_func
self._is_async = asyncio.iscoroutinefunction(init_func)
async def _async_initialize(self) -> T:
"""Initialize the instance using the initialization function"""
if not self._is_async:
# Execute in thread pool if initialization function is not asynchronous
loop = asyncio.get_event_loop()
return await loop.run_in_executor(None, self.init_func)
return await self.init_func()
def _sync_initialize(self) -> T:
"""Synchronization Initialization"""
if self._is_async:
raise TypeError(f"Cannot call async init_func '{self.init_func.__name__}' in sync context")
return self.init_func()
async def _async_cleanup(self) -> None:
"""Clean up instances using the cleanup function"""
if self.cleanup_func and self._instance:
if asyncio.iscoroutinefunction(self.cleanup_func):
await self.cleanup_func(self._instance)
else:
loop = asyncio.get_event_loop()
await loop.run_in_executor(None, self.cleanup_func, self._instance)
def _sync_cleanup(self) -> None:
"""Synchronous Cleanup"""
if self.cleanup_func and self._instance:
if asyncio.iscoroutinefunction(self.cleanup_func):
raise TypeError(f"Cannot call async cleanup_func '{self.cleanup_func.__name__}' in sync context")
self.cleanup_func(self._instance)
class ContextRegistry:
"""Context Registry
Manage Enrollment and Lifecycle for Multiple Context Managers
"""
def __init__(self):
self._contexts: Dict[str, BaseContextManager] = {}
self._lock = Lock()
def register(self, context_manager: BaseContextManager) -> None:
"""Register Context Manager
Args:
context_manager: Context manager to register
Raises:
ValueError: If a context with the same name already exists
"""
with self._lock:
if context_manager.name in self._contexts:
logger.warning(f"Context '{context_manager.name}' is already registered")
return
self._contexts[context_manager.name] = context_manager
logger.debug(f"Registered context: '{context_manager.name}'")
def unregister(self, name: str) -> None:
"""Log out of the context manager
Args:
name: Context name to be logged out
"""
with self._lock:
if name in self._contexts:
context = self._contexts[name]
# Close context before deleting
try:
context.sync_close()
except Exception as e:
logger.warning(f"Error closing context '{name}' during unregister: {e}")
del self._contexts[name]
logger.debug(f"Unregistered context: '{name}'")
def get_context(self, name: str) -> BaseContextManager:
"""Get Context Manager
Args:
name: Context Name
Returns:
BaseContextManager: Corresponding context manager
Raises:
KeyError: If the context does not exist
"""
if name not in self._contexts:
raise KeyError(f"Context '{name}' not found. Available contexts: {list(self._contexts.keys())}")
return self._contexts[name]
def has_context(self, name: str) -> bool:
"""Check if there is a context with the specified name
Args:
name: Context Name
Returns:
bool: Return if present True, otherwise go back to False
"""
return name in self._contexts
async def async_get_instance(self, name: str) -> Any:
"""Asynchronous Get Context Instance"""
context = self.get_context(name)
return await context.async_get_instance()
def sync_get_instance(self, name: str) -> Any:
"""Synchronize Get Context Instance"""
context = self.get_context(name)
return context.sync_get_instance()
async def async_close_all(self) -> None:
"""Shut down all context managers asynchronously"""
contexts = list(self._contexts.values())
# Shut down all contexts in parallel to improve performance
close_tasks = [context.async_close() for context in contexts]
results = await asyncio.gather(*close_tasks, return_exceptions=True)
# Log errors during shutdown
for context, result in zip(contexts, results):
if isinstance(result, Exception):
logger.error(f"Error closing context '{context.name}': {result}")
def sync_close_all(self) -> None:
"""Synchronously close all context managers"""
for context in self._contexts.values():
try:
context.sync_close()
except Exception as e:
logger.error(f"Error closing context '{context.name}': {e}")
def get_all_contexts(self) -> Dict[str, BaseContextManager]:
"""Get a copy of all context managers
Returns:
Dict[str, BaseContextManager]: Contains a dictionary copy of all context managers
"""
return self._contexts.copy()
def get_ready_contexts(self) -> Dict[str, BaseContextManager]:
"""Get all ready context managers
Returns:
Dict[str, BaseContextManager]: Dictionary with all ready context managers
"""
return {
name: context
for name, context in self._contexts.items()
if context.is_ready()
}
def get_context_states(self) -> Dict[str, ContextState]:
"""Get the status of all contexts
Returns:
Dict[str, ContextState]: Context name to state mapping
"""
return {
name: context.get_state()
for name, context in self._contexts.items()
}
async def health_check(self) -> Dict[str, bool]:
"""Perform health checks for all contexts
Returns:
Dict[str, bool]: Context name to health state mapping
"""
results = {}
for name, context in self._contexts.items():
try:
# Prefer to use if there is a custom health check method in context
if hasattr(context, 'health_check') and callable(getattr(context, 'health_check')):
health_check_func = getattr(context, 'health_check')
if asyncio.iscoroutinefunction(health_check_func):
results[name] = await health_check_func()
else:
results[name] = health_check_func()
else:
# Otherwise use basic readiness checks
results[name] = context.is_ready()
except Exception as e:
logger.error(f"Health check failed for context '{name}': {e}")
results[name] = False
return results
def clear(self) -> None:
"""Clear all contexts (close before deleting)"""
with self._lock:
# Close all contexts first
self.sync_close_all()
# Empty registry
self._contexts.clear()
logger.debug("All contexts cleared from registry")
def __len__(self) -> int:
"""Returns the number of registered contexts"""
return len(self._contexts)
def __contains__(self, name: str) -> bool:
"""Support 'name' in registry Grammar"""
return name in self._contexts
def __iter__(self):
"""Support for iterative context names"""
return iter(self._contexts.keys())
def __repr__(self) -> str:
return f""
================================================
FILE: src/backend/bisheng/core/context/examples.py
================================================
"""
Context Manager Usage Example
Demonstrates the various usage scenarios and features of the optimized Context Manager
"""
import asyncio
from typing import Any, Dict
from bisheng.core.context import (
# Primary
BaseContextManager,
FunctionContextManager,
ContextState,
# Application Management
initialize_app_context,
get_context,
async_get_instance,
sync_get_instance,
register_context,
health_check,
close_app_context,
# Exception Class
ContextError,
ContextInitializationError,
)
# 1. Examples of basic use
async def basic_usage_example():
"""Examples of basic use"""
print("=== Examples of basic use ===")
# Initialize app context
config = {
'database': {
'url': 'sqlite:///example.db',
'engine_config': {'pool_size': 10}
}
}
await initialize_app_context(config)
# Get Database Instance
database_context = get_context('database')
db_instance = await database_context.async_get_instance()
print(f"Database instance: {db_instance}")
# Check context status
print(f"Database state: {database_context.get_state()}")
print(f"Database info: {database_context.get_info()}")
# 2. Context Manager Usage Example
async def context_manager_example():
"""Context Manager Usage Example"""
print("\n=== Context manager example ===")
# Using the Asynchronous Context Manager
database_context = get_context('database')
async with database_context.async_context() as db:
print(f"Using database: {type(db)}")
# Database instances are safe to use here
# Using the Synchronization Context Manager
with database_context.sync_context() as db:
print(f"Using database (sync): {type(db)}")
# 3. Example of a custom context manager
class CacheManager(BaseContextManager[Dict[str, Any]]):
"""Example: Cache Manager"""
name = "cache"
def __init__(self, max_size: int = 1000, **kwargs):
super().__init__(**kwargs)
self.max_size = max_size
async def _async_initialize(self) -> Dict[str, Any]:
"""Initialize Cache"""
cache = {}
print(f"Cache initialized with max_size: {self.max_size}")
return cache
def _sync_initialize(self) -> Dict[str, Any]:
"""Synchronous Initialization Cache"""
cache = {}
print(f"Cache initialized (sync) with max_size: {self.max_size}")
return cache
async def _async_cleanup(self) -> None:
"""Clear cache"""
if self._instance:
self._instance.clear()
print("Cache cleared")
def _sync_cleanup(self) -> None:
"""Synchronous Cleanup Cache"""
if self._instance:
self._instance.clear()
print("Cache cleared (sync)")
async def health_check(self) -> bool:
"""Cache Health Check"""
try:
instance = await self.async_get_instance()
return isinstance(instance, dict)
except Exception:
return False
# 4. Functional context manager example
async def init_redis_connection():
"""impersonation Redis Connection Initialization"""
print("Connecting to Redis...")
await asyncio.sleep(0.1) # Analog network latency
return {"connection": "redis://localhost:6379", "status": "connected"}
async def cleanup_redis_connection(connection):
"""impersonation Redis Connection cleanup"""
print("Closing Redis connection...")
connection["status"] = "closed"
# 5. Complete usage examples
async def complete_example():
"""Complete usage examples"""
print("\n=== Full Usage Example ===")
try:
# 1. Register Custom Context Manager
cache_manager = CacheManager(max_size=500, timeout=10.0, retry_count=2)
register_context(
cache_manager,
dependencies=['database'], # Dependent database
initialize_order=20 # Initialize after database
)
# 2. Register Functional Context Manager
redis_manager = FunctionContextManager(
name="redis",
init_func=init_redis_connection,
cleanup_func=cleanup_redis_connection,
timeout=5.0
)
register_context(redis_manager, initialize_order=30)
# 3. Reinitialize to ensure that the context of the new registration is initialized
await initialize_app_context()
# 4. Get all contextual instances
contexts = {
'database': await async_get_instance('database'),
'cache': await async_get_instance('cache'),
'redis': await async_get_instance('redis')
}
print("All contexts initialized:")
for name, instance in contexts.items():
print(f" {name}: {type(instance)}")
# 5. Perform a health check
health_results = await health_check(include_details=True)
print("\nHealth check results:")
for name, result in health_results.items():
if isinstance(result, dict):
print(f" {name}: healthy={result.get('healthy')}, state={result.get('state')}")
else:
print(f" {name}: {result}")
# 6. Demonstrate error handling
try:
# Attempt to get non-existent context
await async_get_instance('nonexistent')
except KeyError as e:
print(f"\nExpected error: {e}")
# 7. Demo Reset Function
cache_context = get_context('cache')
print(f"\nCache state before reset: {cache_context.get_state()}")
await cache_context.async_reset()
print(f"Cache state after reset: {cache_context.get_state()}")
# Re-fetching will trigger a reinitialization
cache_instance = await cache_context.async_get_instance()
print(f"Cache state after re-init: {cache_context.get_state()}")
except ContextError as e:
print(f"Context error: {e}")
except Exception as e:
print(f"Unexpected error: {e}")
finally:
# Clean up resources
await close_app_context()
print("\nApplication context closed")
# 6. Examples of performance and concurrent testing
async def concurrency_example():
"""Example of concurrent access"""
print("\n=== Example of concurrent access ===")
await initialize_app_context()
async def worker(worker_id: int):
"""Work corridor"""
try:
# Get the same context instance concurrently
db = await async_get_instance('database')
print(f"Worker {worker_id} got database: {id(db)}")
# Simulate some work
await asyncio.sleep(0.1)
return f"Worker {worker_id} completed"
except Exception as e:
return f"Worker {worker_id} failed: {e}"
# Create multiple concurrent workflows
workers = [worker(i) for i in range(10)]
results = await asyncio.gather(*workers)
print("Concurrent access results:")
for result in results:
print(f" {result}")
await close_app_context()
# 7. Examples of monitoring and diagnostics
async def monitoring_example():
"""Examples of monitoring and diagnostics"""
print("\n=== Examples of monitoring and diagnostics ===")
await initialize_app_context()
# Register some test contexts
register_context(CacheManager(name="cache1"))
register_context(CacheManager(name="cache2"))
# Get app context
from bisheng.core.context.manager import app_context
context_info = app_context.get_context_info()
print("Application context info:")
print(f" Initialized: {context_info['initialized']}")
print(f" Context count: {context_info['context_count']}")
print(f" Initialization order: {context_info['initialization_order']}")
print(f" Dependencies: {context_info['dependencies']}")
# Get the status of each context
states = context_info['context_states']
print("\nContext states:")
for name, state in states.items():
print(f" {name}: {state.value if hasattr(state, 'value') else state}")
# Perform detailed health checks
detailed_health = await health_check(include_details=True)
print("\nDetailed health check:")
for name, details in detailed_health.items():
if isinstance(details, dict):
print(f" {name}:")
print(f" Healthy: {details.get('healthy')}")
print(f" State: {details.get('state')}")
print(f" Error: {details.get('error', 'None')}")
else:
print(f" {name}: {details}")
await close_app_context()
# The main function.
async def main():
"""Run all examples"""
print("Post Context Manager Optimization Demo\n")
await basic_usage_example()
await context_manager_example()
await complete_example()
await concurrency_example()
await monitoring_example()
print("\n=== All sample runs completed ===")
if __name__ == "__main__":
asyncio.run(main())
================================================
FILE: src/backend/bisheng/core/context/manager.py
================================================
"""App Context Manager
Global context manager with integrated lazy loading and caching
Provides easy dependency injection and lifecycle management
"""
import asyncio
from contextlib import asynccontextmanager, contextmanager
from typing import Optional, Dict, Any, TypeVar, List, Union
from loguru import logger
from bisheng.core.config.settings import Settings
from bisheng.core.context.base import (
ContextRegistry,
BaseContextManager,
ContextState,
ContextError
)
T = TypeVar('T')
class ApplicationContextManager:
"""App Context Manager
Responsible for managing the entire application infrastructure service lifecycle and providing a unified access interface
Supports dependency injection, bulk operations, and health checks
"""
def __init__(self):
self._registry = ContextRegistry()
self._initialized = False
self._initialization_lock = asyncio.Lock()
self._initialization_order: List[str] = []
self._dependencies: Dict[str, List[str]] = {}
async def initialize(self, config: Settings) -> None:
"""Initialize app context
Args:
config: Optional configuration dictionary for passing to individual context managers
Raises:
ContextError: Thrown on initialization failure
"""
async with self._initialization_lock:
if self._initialized:
logger.debug("Application context already initialized")
return
try:
# Register default context manager
self._register_default_contexts(config or {})
# Initialize all contexts in dependency order
await self._initialize_contexts_in_order()
self._initialized = True
logger.info("Application context initialized successfully")
except Exception as e:
logger.exception(f"Failed to initialize application context: {e}")
# Clean Up Initialized Resources
await self.async_close()
raise ContextError(f"Application context initialization failed: {e}") from e
def _register_default_contexts(self, config: Settings) -> None:
"""Register default context manager"""
try:
from bisheng.core.database.manager import DatabaseManager
self.register_context(DatabaseManager(database_url=config.database_url))
from bisheng.core.cache.redis_manager import RedisManager
self.register_context(RedisManager(redis_url=config.redis_url))
from bisheng.core.storage.minio.minio_manager import MinioManager
self.register_context(MinioManager(minio_config=config.object_storage.minio))
from bisheng.core.search.elasticsearch.manager import EsConnManager, statistics_es_name
self.register_context(EsConnManager(es_hosts=config.get_search_conf().elasticsearch_url,
**config.get_search_conf().ssl_verify))
self.register_context(EsConnManager(es_hosts=config.get_telemetry_conf().elasticsearch_url,
name=statistics_es_name,
**config.get_telemetry_conf().ssl_verify))
from bisheng.core.external.http_client.http_client_manager import HttpClientManager
self.register_context(HttpClientManager())
from bisheng.core.prompts.manager import PromptManager
self.register_context(PromptManager())
logger.debug("Default contexts registered")
except ImportError as e:
logger.warning(f"Failed to import default context managers: {e}")
except Exception as e:
logger.error(f"Failed to register default contexts: {e}")
raise
async def _initialize_contexts_in_order(self) -> None:
"""Initialize context in dependency order"""
# If no initialization order is defined, initialize in registration order
if not self._initialization_order:
self._initialization_order = list(self._registry.get_all_contexts().keys())
initialized = set()
for context_name in self._initialization_order:
if context_name not in initialized:
await self._initialize_context_with_dependencies(context_name, initialized)
async def _initialize_context_with_dependencies(self, context_name: str, initialized: set) -> None:
"""Recursive initialization context and its dependencies"""
if context_name in initialized:
return
# Initialize dependencies first
dependencies = self._dependencies.get(context_name, [])
for dep_name in dependencies:
if dep_name not in initialized:
await self._initialize_context_with_dependencies(dep_name, initialized)
# Then initialize the current context
try:
context = self._registry.get_context(context_name)
await context.async_get_instance() # This will trigger initialization
initialized.add(context_name)
logger.debug(f"Initialized context: '{context_name}'")
except Exception as e:
logger.error(f"Failed to initialize context '{context_name}': {e}")
raise
async def async_get_instance(self, name: str) -> T:
"""Gets the context instance of the specified name asynchronously"""
context = self.get_context(name)
return await context.async_get_instance()
def sync_get_instance(self, name: str) -> T:
"""Synchronously get a contextual instance of the specified name"""
context = self.get_context(name)
return context.sync_get_instance()
def get_context(self, name: str) -> BaseContextManager:
"""Gets the context instance of the specified name"""
return self._registry.get_context(name)
def register_context(
self,
context: BaseContextManager,
dependencies: Optional[List[str]] = None,
initialize_order: Optional[int] = None
) -> None:
"""Register a new context manager
Args:
context: Context manager to register
dependencies: A list of other context names that the context depends on
initialize_order: Initialization order (smaller numbers initialize earlier)
Raises:
ValueError: If the context name already exists
"""
self._registry.register(context)
# Record dependencies
if dependencies:
self._dependencies[context.name] = dependencies
# Update initialization order
if initialize_order is not None:
# Insert to Specified Location
if context.name in self._initialization_order:
self._initialization_order.remove(context.name)
insert_pos = 0
for i, existing_name in enumerate(self._initialization_order):
existing_context = self._registry.get_context(existing_name)
if getattr(existing_context, '_initialize_order', float('inf')) > initialize_order:
insert_pos = i
break
insert_pos = i + 1
self._initialization_order.insert(insert_pos, context.name)
context._initialize_order = initialize_order
elif context.name not in self._initialization_order:
self._initialization_order.append(context.name)
logger.debug(f"Registered context '{context.name}' with dependencies: {dependencies or []}")
def unregister_context(self, name: str) -> None:
"""Log out of the context manager"""
self._registry.unregister(name)
async def health_check(self, include_details: bool = False) -> Union[Dict[str, bool], Dict[str, Dict[str, Any]]]:
"""Perform a health check
Args:
include_details: Does it contain detailed health check information?
Returns:
Union[Dict[str, bool], Dict[str, Dict[str, Any]]]: Health Check Results
"""
results = await self._registry.health_check()
if not include_details:
return results
# Include details
detailed_results = {}
for name, is_healthy in results.items():
try:
context = self._registry.get_context(name)
detailed_results[name] = {
'healthy': is_healthy,
'state': context.get_state().value,
'error': str(context.get_error()) if context.get_error() else None,
'info': context.get_info() if hasattr(context, 'get_info') else {}
}
except Exception as e:
detailed_results[name] = {
'healthy': False,
'error': f"Failed to get context info: {e}"
}
return detailed_results
async def async_close(self) -> None:
"""Close app context
Close all contexts in reverse order from initialization
"""
if not self._initialized:
return
try:
# Close context in reverse order to ensure dependencies are handled correctly
await self._close_contexts_in_reverse_order()
# Cleanup status
self._initialized = False
self._initialization_order.clear()
self._dependencies.clear()
logger.info("Application context closed successfully")
except Exception as e:
logger.error(f"Error closing application context: {e}")
raise
async def _close_contexts_in_reverse_order(self) -> None:
"""Close all contexts in reverse order"""
close_order = list(reversed(self._initialization_order))
for context_name in close_order:
try:
if self._registry.has_context(context_name):
context = self._registry.get_context(context_name)
await context.async_close()
logger.debug(f"Closed context: '{context_name}'")
except Exception as e:
logger.error(f"Error closing context '{context_name}': {e}")
# Ensure all contexts are closed
await self._registry.async_close_all()
def is_initialized(self) -> bool:
"""Check if initialized"""
return self._initialized
def get_registry(self) -> ContextRegistry:
"""Get Context Registry
Returns:
ContextRegistry: Context Registry Instance
"""
return self._registry
def get_context_info(self) -> Dict[str, Any]:
"""Get app context details
Returns:
Dict[str, Any]: Dictionary with app context details
"""
return {
'initialized': self._initialized,
'context_count': len(self._registry),
'initialization_order': self._initialization_order.copy(),
'dependencies': self._dependencies.copy(),
'context_states': self._registry.get_context_states()
}
@contextmanager
def sync_context(self, *context_names: str):
"""Synchronize context manager for batch fetching multiple contexts
Args:
*context_names: List of context names to get
Example:
with app_context.sync_context('database', 'cache') as (db, cache):
# Use db And cache
pass
"""
instances = []
try:
for name in context_names:
instance = self.sync_get_instance(name)
instances.append(instance)
if len(instances) == 1:
yield instances[0]
else:
yield tuple(instances)
finally:
# Note: This does not close automatically as it may be used elsewhere
pass
@asynccontextmanager
async def async_context(self, *context_names: str):
"""Asynchronous context manager for batch fetching multiple contexts
Args:
*context_names: List of context names to get
Example:
async with app_context.async_context('database', 'cache') as (db, cache):
# Use db And cache
pass
"""
instances = []
try:
for name in context_names:
instance = await self.async_get_instance(name)
instances.append(instance)
if len(instances) == 1:
yield instances[0]
else:
yield tuple(instances)
finally:
# Note: This does not close automatically as it may be used elsewhere
pass
async def reset_context(self, context_name: str) -> None:
"""Reset Specified Context
Args:
context_name: Context name to reset
Raises:
KeyError: If the context does not exist
"""
context = self._registry.get_context(context_name)
await context.async_reset()
logger.info(f"Context '{context_name}' reset successfully")
async def restart_context(self, context_name: str) -> None:
"""Restart the specified context (initialize immediately after reset)
Args:
context_name: Context name to restart
Raises:
KeyError: If the context does not exist
"""
await self.reset_context(context_name)
await self.async_get_instance(context_name) # Trigger reinitialization
logger.info(f"Context '{context_name}' restarted successfully")
def list_contexts(self, state_filter: Optional[ContextState] = None) -> List[str]:
"""List all context names
Args:
state_filter: Optional status filter that returns only the context of the specified status
Returns:
List[str]: Context Name List
"""
if state_filter is None:
return list(self._registry.get_all_contexts().keys())
return [
name for name, context in self._registry.get_all_contexts().items()
if context.get_state() == state_filter
]
# Global App Context Instance
app_context = ApplicationContextManager()
async def initialize_app_context(config: Settings) -> None:
"""
Initialize global app context
:param config:
:return:
"""
await app_context.initialize(config)
def get_context(name: str) -> BaseContextManager:
"""Convenient way to get context
Args:
name: Context Name
Returns:
BaseContextManager: Corresponding context manager
Raises:
KeyError: If the context does not exist
"""
return app_context.get_context(name)
async def async_get_instance(name: str) -> Any:
"""Convenient way to get contextual instances asynchronously
Args:
name: Context Name
Returns:
Any: Contextual instances
Raises:
KeyError: If the context does not exist
ContextError: If initialization fails
"""
return await app_context.async_get_instance(name)
def sync_get_instance(name: str) -> Any:
"""Convenient way to get contextual instances synchronously
Args:
name: Context Name
Returns:
Any: Contextual instances
Raises:
KeyError: If the context does not exist
ContextError: If initialization fails
"""
return app_context.sync_get_instance(name)
async def close_app_context() -> None:
"""Convenient way to turn off global app context"""
await app_context.async_close()
def register_context(
context: BaseContextManager,
dependencies: Optional[List[str]] = None,
initialize_order: Optional[int] = None
) -> None:
"""Convenient way to register a context
Args:
context: Context manager to register
dependencies: A list of other context names that the context depends on
initialize_order: Initialization order (smaller numbers initialize earlier)
Example:
# Register a cache context that depends on the database
register_context(cache_manager, dependencies=['database'], initialize_order=10)
"""
app_context.register_context(context, dependencies, initialize_order)
async def health_check(include_details: bool = False) -> Union[Dict[str, bool], Dict[str, Dict[str, Any]]]:
"""Convenient way to perform a health check
Args:
include_details: Does it contain detailed health check information?
Returns:
Union[Dict[str, bool], Dict[str, Dict[str, Any]]]: Health Check Results
"""
return await app_context.health_check(include_details)
================================================
FILE: src/backend/bisheng/core/database/__init__.py
================================================
"""Database Infrastructure Module
Provides database connection management, session management, transaction management, and context management capabilities
"""
from .connection import DatabaseConnectionManager # noqa: F401
from .manager import (
get_database_connection,
sync_get_database_connection,
get_async_db_session,
get_sync_db_session
)
__all__ = [
'DatabaseConnectionManager',
'get_database_connection',
'sync_get_database_connection',
'get_async_db_session',
'get_sync_db_session'
]
================================================
FILE: src/backend/bisheng/core/database/alembic/README.md
================================================
Generic single-database configuration.
* 创建迁移脚本
```bash
alembic revision -m "描述信息"
```
* 自动生成迁移脚本，基于模型和数据库的差异，差异较大时建议先手动创建迁移脚本再修改
```bash
alembic revision --autogenerate -m "描述信息"
```
* 应用所有未应用的迁移脚本
```bash
alembic upgrade head
```
* 应用到指定版本
```bash
alembic upgrade <版本号>
```
* 回滚到上一个版本
```bash
alembic downgrade -1
```
* 回滚到指定版本
```bash
alembic downgrade <版本号>
```
* 查看当前数据库版本
```bash
alembic current
```
* 查看迁移历史
```bash
alembic history
```
* 查看未应用的迁移脚本
```bash
alembic heads
```
* 生成迁移脚本的SQL文件，而不是直接应用到数据库
```bash
alembic upgrade head --sql > upgrade.sql
```
* 配置文件及目录说明
* alembic.ini: 主配置文件，包含数据库连接字符串等全局配置
* bisheng/core/database/alembic/env.py: 环境配置文件，定义了如何连接数据库和加载模型
* bisheng/core/database/alembic/versions/: 存放迁移脚本的目录
* bisheng/core/database/alembic/script.py.mako: 迁移脚本模板文件
* bisheng/core/database/alembic/README.md: 本说明文件
================================================
FILE: src/backend/bisheng/core/database/alembic/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/database/alembic/env.py
================================================
from logging.config import fileConfig
from alembic import context
from bisheng.common.models.base import SQLModelSerializable
# this is the Alembic Config object, which provides
# access to the values within the .ini file in use.
config = context.config
# Interpret the config file for Python logging.
# This line sets up loggers basically.
if config.config_file_name is not None:
fileConfig(config.config_file_name)
# add your model's MetaData object here
# for 'autogenerate' support
# from myapp import mymodel
# target_metadata = mymodel.Base.metadata
target_metadata = SQLModelSerializable.metadata
# other values from the config, defined by the needs of env.py,
# can be acquired:
# my_important_option = config.get_main_option("my_important_option")
# ... etc.
def run_migrations_online() -> None:
"""Run migrations in 'online' mode.
In this scenario we need to create an Engine
and associate a connection with the context.
"""
from bisheng.core.database.manager import sync_get_database_connection
database_conn_manager = sync_get_database_connection()
with database_conn_manager.engine.connect() as connection:
context.configure(
connection=connection, target_metadata=target_metadata
)
with context.begin_transaction():
context.run_migrations()
run_migrations_online()
================================================
FILE: src/backend/bisheng/core/database/alembic/script.py.mako
================================================
"""${message}
Revision ID: ${up_revision}
Revises: ${down_revision | comma,n}
Create Date: ${create_date}
"""
from typing import Sequence, Union
from alembic import op
import sqlalchemy as sa
${imports if imports else ""}
# revision identifiers, used by Alembic.
revision: str = ${repr(up_revision)}
down_revision: Union[str, Sequence[str], None] = ${repr(down_revision)}
branch_labels: Union[str, Sequence[str], None] = ${repr(branch_labels)}
depends_on: Union[str, Sequence[str], None] = ${repr(depends_on)}
def upgrade() -> None:
"""Upgrade schema."""
${upgrades if upgrades else "pass"}
def downgrade() -> None:
"""Downgrade schema."""
${downgrades if downgrades else "pass"}
================================================
FILE: src/backend/bisheng/core/database/alembic/versions/v2_3_0_beta1_9ba42685e830.py
================================================
"""Add a metadata_fields field to the knowledge table.
Revision ID: 9ba42685e830
Revises:
Create Date: 2025-11-11 16:00:42.582363
"""
from typing import Sequence, Union
from alembic import op
import sqlalchemy as sa
# revision identifiers, used by Alembic.
revision: str = '9ba42685e830'
down_revision: Union[str, Sequence[str], None] = None
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
"""Upgrade schema."""
op.add_column('knowledge', sa.Column('metadata_fields', sa.JSON, nullable=True, comment='Metadata Field Configuration for Knowledge Base'))
# knowledgefile OF TABLE)extra_meta to user_metadata Data field
op.add_column('knowledgefile', sa.Column('user_metadata', sa.JSON, nullable=True, comment='User-defined metadata'))
# Taking the original extra_meta Data migration of fields to user_metadata Data field
op.execute('UPDATE knowledgefile SET user_metadata = extra_meta')
op.drop_column('knowledgefile', 'extra_meta')
op.add_column('knowledgefile',sa.Column('updater_id', sa.INT, nullable=True, comment='Updated ByID'))
def downgrade() -> None:
"""Downgrade schema."""
op.drop_column('knowledge', 'metadata_fields')
# Recovery knowledgefile OF TABLE) extra_meta Data field
op.add_column('knowledgefile',
sa.Column('extra_meta', sa.VARCHAR(255), nullable=True, comment='User-defined metadata'))
op.execute('UPDATE knowledgefile SET extra_meta = user_metadata')
op.drop_column('knowledgefile', 'user_metadata')
op.drop_column('knowledgefile', 'updater_id')
================================================
FILE: src/backend/bisheng/core/database/connection.py
================================================
"""Database Connection Management Module"""
import logging
from typing import Optional, Dict, Any, Generator
from contextlib import asynccontextmanager, contextmanager
from sqlalchemy import create_engine, Engine
from sqlalchemy.exc import OperationalError
from sqlalchemy.ext.asyncio import create_async_engine, AsyncEngine
from sqlalchemy.orm import sessionmaker
from sqlalchemy.pool import StaticPool
from sqlmodel import Session, SQLModel
from sqlmodel.ext.asyncio.session import AsyncSession
from sqlalchemy.ext.asyncio import async_sessionmaker
logger = logging.getLogger(__name__)
class DatabaseConnectionManager:
"""Database Connection Manager
Responsible for managing database engine creation, connection pool configuration and lifecycle management
"""
def __init__(self, database_url: str, **engine_kwargs):
self.database_url = database_url
self.async_database_url = self._convert_to_async_url(database_url)
self.engine_kwargs = engine_kwargs
self._engine: Optional[Engine] = None
self._async_engine: Optional[AsyncEngine] = None
self._async_session_maker: Optional[async_sessionmaker] = None
def _convert_to_async_url(self, url: str) -> str:
"""Database will be synchronizedURLConvert to AsynchronousURL"""
if "pymysql" in url:
return url.replace("pymysql", "aiomysql")
elif "psycopg2" in url:
return url.replace("psycopg2", "asyncpg")
return url
def _get_default_engine_config(self) -> Dict[str, Any]:
"""Get default engine configuration"""
config = {
'pool_size': 100,
'max_overflow': 20,
'pool_timeout': 30,
'pool_pre_ping': True,
'pool_recycle': 3600, # 1 hour
}
# SQLiteSPECIAL CONFIGURATION
if self.database_url.startswith("sqlite"):
config.update({
'connect_args': {'check_same_thread': False},
'poolclass': StaticPool,
'pool_size': 1,
'max_overflow': 0,
})
# MySQLSPECIAL CONFIGURATION
elif "mysql" in self.database_url:
if 'connect_args' not in config:
config['connect_args'] = {}
config['connect_args']['charset'] = 'utf8mb4'
return config
@property
def engine(self) -> Engine:
"""Get Synchronization Database Engine"""
if self._engine is None:
config = self._get_default_engine_config()
config.update(self.engine_kwargs)
self._engine = create_engine(
self.database_url,
**config
)
logger.debug(f"Created sync database engine for {self.database_url}")
return self._engine
@property
def async_engine(self) -> AsyncEngine:
"""Get Asynchronous Database Engine"""
if self._async_engine is None:
config = self._get_default_engine_config()
config.update(self.engine_kwargs)
# Remove Synchronization Engine Specific Configuration
config.pop('poolclass', None)
self._async_engine = create_async_engine(
self.async_database_url,
**config
)
logger.debug(f"Created async database engine for {self.async_database_url}")
return self._async_engine
@contextmanager
def create_session(self) -> Generator[Session, Any, None]:
"""Create a sync session"""
session_maker = sessionmaker(
bind=self.engine,
class_=Session,
expire_on_commit=False,
autoflush=True,
autocommit=False
)
with session_maker() as session:
try:
yield session
except Exception as e:
session.rollback()
logger.error(f"Database session rolled back due to error: {e}")
raise
finally:
session.close()
@asynccontextmanager
async def async_session(self):
"""Asynchronous Session Context Manager"""
session_maker = async_sessionmaker(
bind=self.async_engine,
class_=AsyncSession,
expire_on_commit=False,
autoflush=True,
autocommit=False
)
async with session_maker() as session:
try:
yield session
except Exception as e:
await session.rollback()
logger.error(f"Database session rolled back due to error: {e}")
raise
finally:
await session.close()
async def create_db_and_tables(self) -> None:
"""Creation of databases and tables"""
async with self.async_engine.begin() as conn:
try:
await conn.run_sync(SQLModel.metadata.create_all)
except OperationalError as oe:
logger.warning(f"Table creation skipped due to OperationalError: {oe}")
except Exception as exc:
logger.error(f"Error creating tables: {exc}")
raise RuntimeError("Error creating tables") from exc
logger.info('Database and tables created successfully')
async def close(self):
"""Close database connection"""
if self._async_engine:
await self._async_engine.dispose()
logger.debug("Async database engine disposed")
def close_sync(self):
"""Synchronously close database connections"""
if self._engine:
self._engine.dispose()
logger.debug("Sync database engine disposed")
def __del__(self):
"""Destructor ensures release of resources"""
if self._engine:
try:
self._engine.dispose()
except Exception:
pass
================================================
FILE: src/backend/bisheng/core/database/manager.py
================================================
"""Database Global Manager
Provides global management and easy access to database context
Supports health checks, connection pool monitoring, and transaction management
"""
import logging
from typing import Dict, Any, Optional, AsyncGenerator
from contextlib import asynccontextmanager, contextmanager
from sqlalchemy import text
from sqlmodel.ext.asyncio.session import AsyncSession
from bisheng.core.context import BaseContextManager
from bisheng.core.database import DatabaseConnectionManager
logger = logging.getLogger(__name__)
class DatabaseManager(BaseContextManager[DatabaseConnectionManager]):
"""Database Global Manager
Responsible for managing the global lifecycle of database connections and providing a unified access interface
Supports connection pool monitoring, health checks, and easy session management
"""
name: str = "database"
def __init__(
self,
database_url: Optional[str] = None,
engine_config: Optional[Dict[str, Any]] = None,
**kwargs
):
super().__init__(self.name, **kwargs)
self.database_url = database_url
if not self.database_url:
raise ValueError("Database URL is required. Please provide via parameter or settings.database_url")
self.engine_config = engine_config or {}
async def _async_initialize(self) -> DatabaseConnectionManager:
"""Initialize Database Connection Manager"""
return DatabaseConnectionManager(
self.database_url,
**self.engine_config
)
def _sync_initialize(self) -> DatabaseConnectionManager:
"""Synchronization Initialization"""
return DatabaseConnectionManager(
self.database_url,
**self.engine_config
)
async def _async_cleanup(self) -> None:
"""Clean up database resources"""
if self._instance:
await self._instance.close()
def _sync_cleanup(self) -> None:
"""Synchronously clean up database resources"""
if self._instance:
self._instance.close_sync()
async def health_check(self) -> bool:
"""Database Health Check
Returns:
bool: True If the database connection is normal,False Otherwise,
"""
try:
database_instance = await self.async_get_instance()
# Perform simple queries using asynchronous sessions
async with database_instance.async_session() as session:
await session.exec(text("SELECT 1"))
logger.debug("Database health check passed")
return True
except Exception as e:
logger.error(f"Database health check failed: {e}")
return False
async def create_tables_if_not_exists(self) -> None:
"""Create a database table (if it does not exist)
This is a convenient method for initializing the database structure
"""
try:
database_instance = await self.async_get_instance()
await database_instance.create_db_and_tables()
logger.info("Database tables created or verified successfully")
except Exception as e:
logger.error(f"Failed to create database tables: {e}")
raise
async def get_database_connection() -> DatabaseConnectionManager:
"""Get Global Database Connection Manager Instance
Returns:
DatabaseConnectionManager: Database Connection Manager Instance
Raises:
ContextError: If the database context is not registered or initialization fails
"""
from bisheng.core.context.manager import app_context
try:
return await app_context.async_get_instance(DatabaseManager.name)
except KeyError:
logger.warning(f"Database context not found, registering default instance")
try:
from bisheng.common.services.config_service import settings
app_context.register_context(DatabaseManager(settings.database_url))
return await app_context.async_get_instance(DatabaseManager.name)
except Exception as e:
logger.error(f"Failed to register and initialize database context: {e}")
raise
def sync_get_database_connection() -> DatabaseConnectionManager:
"""Get global database connection manager instance synchronously
Returns:
DatabaseConnectionManager: Database Connection Manager Instance
Raises:
ContextError: If the database context is not registered or initialization fails
"""
from bisheng.core.context.manager import app_context
try:
return app_context.sync_get_instance(DatabaseManager.name)
except KeyError:
logger.warning(f"Database context not found, registering default instance")
try:
from bisheng.common.services.config_service import settings
app_context.register_context(DatabaseManager(settings.database_url))
return app_context.sync_get_instance(DatabaseManager.name)
except Exception as e:
logger.error(f"Failed to register and initialize database context: {e}")
raise
@asynccontextmanager
async def get_async_db_session() -> AsyncGenerator[AsyncSession, None]:
"""A convenient way to get asynchronous database sessions
Example:
async with get_async_db_session() as session:
result = await session.execute("SELECT * FROM users")
Yields:
AsyncSession: Asynchronous database session
"""
db_manager = await get_database_connection()
async with db_manager.async_session() as session:
yield session
@contextmanager
def get_sync_db_session():
"""Convenient way to get synchronized database sessions
Example:
with get_sync_db_session() as session:
result = session.execute("SELECT * FROM users")
Yields:
Session: Synchronize database sessions
"""
db_manager = sync_get_database_connection()
with db_manager.create_session() as session:
yield session
================================================
FILE: src/backend/bisheng/core/external/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/external/http_client/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/external/http_client/client.py
================================================
import logging
from typing import Dict, Optional, Union, Any, AsyncGenerator, Literal
import aiohttp
from aiohttp import ClientSession, ClientTimeout, TCPConnector
from pydantic import BaseModel
logger = logging.getLogger(__name__)
class ResponseData(BaseModel):
status_code: int
body: Optional[Union[Dict[str, Any], str, bytes]] = None
error: Optional[str] = None
headers: Optional[Dict[str, str]] = None
class AsyncHttpClient(object):
def __init__(self, timeout: int = 120, keep_alive: bool = True, limit_per_host: int = 100,
headers: Dict[str, str] = None, loop=None):
"""
Asynchronous HTTP client using aiohttp.
:param timeout: Timeout in seconds for the requests.
:param max_retries: Maximum number of retries for failed requests.
:param keep_alive: Whether to keep the connection alive.
:param limit_per_host: Maximum number of connections per host.
:param headers:
:param loop:
"""
self.headers = headers or self._get_default_headers()
self.timeout = timeout
self.keep_alive = keep_alive
self.limit_per_host = limit_per_host
self.loop = loop
self._aiohttp_client: Optional[ClientSession] = None
@staticmethod
def _get_default_headers() -> Dict[str, str]:
return {
"Accept": "application/json",
"Content-Type": "application/json",
"User-Agent": "aiohttp-client",
}
def add_custom_header(self, key: str, value: str) -> None:
self.headers[key] = value
def clear_custom_headers(self) -> None:
self.headers = self._get_default_headers()
async def get_aiohttp_client(self, loop=None) -> ClientSession:
loop = loop or self.loop
if self._aiohttp_client is None:
timeout = ClientTimeout(total=self.timeout)
connector = TCPConnector(keepalive_timeout=self.timeout, limit_per_host=self.limit_per_host, loop=loop)
self._aiohttp_client = ClientSession(timeout=timeout, connector=connector,
headers=self.headers, loop=loop)
return self._aiohttp_client
async def close_aiohttp_client(self) -> None:
if self._aiohttp_client:
await self._aiohttp_client.close()
self._aiohttp_client = None
async def _execute_request(
self,
method: str,
url: str,
headers: Dict[str, str],
params: Optional[Dict[str, Union[str, int]]] = None,
body: Optional[Union[Dict[str, Any], str, bytes, Any]] = None,
data_type: Literal["json", "text", "binary"] = "json",
destroy_session: bool = False,
clear_headers: bool = False,
**kwargs,
) -> ResponseData:
"""
Execute an HTTP request.
:param method: HTTP method (GET, POST, PUT, DELETE, etc.)
:param url: URL for the request.
:param headers: Headers for the request.
:param params: Query parameters for the request.
:param body: Body of the request.
:param data_type: Type of data to return (json, text, binary).
:param destroy_session: Whether to close the session after the request.
:param clear_headers: Whether to clear custom headers after the request.
:return: Response as a dictionary with status_code, body, and error.
"""
client = await self.get_aiohttp_client()
response: ResponseData = ResponseData(status_code=0)
headers = headers or self.headers
try:
async with client.request(
method=method,
url=url,
headers=headers,
params=params,
json=body if isinstance(body, dict) else None,
data=body if isinstance(body, str) or isinstance(body, bytes) else None,
**kwargs,
) as http_response:
response.status_code = http_response.status
response.headers = dict(http_response.headers)
if data_type == "json":
response.body = await http_response.json()
elif data_type == "text":
response.body = await http_response.text()
elif data_type == "binary":
response.body = await http_response.read()
http_response.raise_for_status()
except aiohttp.ClientResponseError as cre:
logger.error(f"Client response error during {method} request to {url}: {cre}")
response.error = str(cre)
response.status_code = cre.status
except Exception as error:
logger.error(f"Error during {method} request to {url}: {error}")
response.error = str(error)
response.status_code = 500
finally:
if destroy_session:
await self.close_aiohttp_client()
if clear_headers:
self.clear_custom_headers()
return response
async def get(self, url: str, params: Optional[Dict[str, Union[str, int]]] = None,
data_type: Literal["json", "text", "binary"] = "json", destroy_session: bool = False,
headers: Dict = None, clear_headers: bool = False, **kwargs) -> ResponseData:
return await self._execute_request("GET", url=url, headers=headers, params=params, data_type=data_type,
destroy_session=destroy_session, clear_headers=clear_headers, **kwargs)
async def post(self, url: str, body: Optional[Union[Dict[str, Any], str, bytes, Any]] = None,
data_type: Literal["json", "text", "binary"] = "json", destroy_session: bool = False,
headers: Dict = None, clear_headers: bool = False, **kwargs) -> ResponseData:
return await self._execute_request("POST", url=url, headers=headers, body=body, data_type=data_type,
destroy_session=destroy_session, clear_headers=clear_headers, **kwargs)
async def put(self, url: str, body: Optional[Union[Dict[str, Any], str, bytes, Any]] = None,
data_type: Literal["json", "text", "binary"] = "json", destroy_session: bool = False,
headers: Dict = None, clear_headers: bool = False, **kwargs) -> ResponseData:
return await self._execute_request("PUT", url=url, headers=headers, body=body, data_type=data_type,
destroy_session=destroy_session, clear_headers=clear_headers, **kwargs)
async def patch(self, url: str, body: Optional[Union[Dict[str, Any], str, bytes, Any]] = None,
data_type: Literal["json", "text", "binary"] = "json", destroy_session: bool = False,
headers: Dict = None, clear_headers: bool = False, **kwargs) -> ResponseData:
return await self._execute_request("PATCH", url=url, headers=headers, body=body, data_type=data_type,
destroy_session=destroy_session, clear_headers=clear_headers, **kwargs)
async def delete(self, url: str, params: Optional[Dict[str, Union[str, int]]] = None,
data_type: Literal["json", "text", "binary"] = "json", destroy_session: bool = False,
headers: Dict = None, clear_headers: bool = False, **kwargs) -> ResponseData:
return await self._execute_request("DELETE", url=url, headers=headers, params=params, data_type=data_type,
destroy_session=destroy_session, clear_headers=clear_headers, **kwargs)
async def stream(
self, method: Literal["GET", "POST", "PUT", "PATCH"],
url: str,
headers: Dict[str, str] = None,
params: Optional[Dict[str, Union[str, int]]] = None,
body: Optional[Union[Dict[str, Any], str]] = None, chunk_size: int = 1024 * 1024 * 10,
destroy_session: bool = False,
clear_headers: bool = False,
**kwargs
) -> AsyncGenerator[bytes, None]:
"""
Streaming requests
"""
client = await self.get_aiohttp_client()
try:
async with client.request(method=method,
url=url,
headers=headers or self.headers,
params=params,
json=body if isinstance(body, dict) else None,
data=body if isinstance(body, str) else None, **kwargs) as response:
response.raise_for_status()
async for chunk in response.content.iter_chunked(chunk_size):
yield chunk
except Exception as error:
logger.error(f"Error during streaming from {url}: {error}")
raise error
finally:
if destroy_session:
await self.close_aiohttp_client()
if clear_headers:
self.clear_custom_headers()
================================================
FILE: src/backend/bisheng/core/external/http_client/http_client_manager.py
================================================
import logging
from bisheng.core.context import BaseContextManager
from bisheng.core.context.base import T
from bisheng.core.external.http_client.client import AsyncHttpClient
logger = logging.getLogger(__name__)
class HttpClientManager(BaseContextManager[AsyncHttpClient]):
"""HTTP Client Manager
Inherited From BaseContextManagerProvided by HTTP Client lifecycle management
Supports lazy loading and caching
"""
name = 'http_client'
def __init__(self):
super().__init__()
async def _async_initialize(self) -> AsyncHttpClient:
"""Async Initialization HTTP Client"""
http_client = AsyncHttpClient()
await http_client.get_aiohttp_client()
return http_client
def _sync_initialize(self) -> T:
pass
async def _async_cleanup(self) -> None:
"""Asynchronous Cleanup HTTP Client"""
if self._instance:
await self._instance.close_aiohttp_client()
def _sync_cleanup(self) -> None:
"""Synchronous Cleanup HTTP Client"""
pass
async def get_http_client() -> AsyncHttpClient:
"""Dapatkan HTTP Client Instance"""
from bisheng.core.context.manager import app_context
try:
return await app_context.async_get_instance(HttpClientManager.name)
except KeyError:
logger.warning(f"HttpClientManager not found in context, registering new instance.")
try:
from bisheng.common.services.config_service import settings
app_context.register_context(HttpClientManager())
return await app_context.async_get_instance(HttpClientManager.name)
except Exception as e:
logger.error(f"Failed to register HttpClientManager: {e}")
raise
================================================
FILE: src/backend/bisheng/core/logger.py
================================================
import logging
import sys
import uuid
from contextvars import ContextVar
from pathlib import Path
from typing import Optional
from loguru import logger
from bisheng.core.config.settings import LoggerConf
trace_id_var: ContextVar[Optional[str]] = ContextVar("trace_id", default=None)
def trace_id_generator() -> str:
return str(uuid.uuid4().hex)
class TraceIdFilter:
def __init__(self, filter_func=None):
self.filter_func = filter_func
def __call__(self, record):
if record["extra"].get("trace_id") is None:
record["extra"]["trace_id"] = trace_id_var.get()
if self.filter_func is not None:
return self.filter_func(record)
return True
class InterceptHandler(logging.Handler):
"""Interception Criteria logging logs, transferring to Loguru Medical Treatment:"""
def emit(self, record):
# Dapatkan loguru counterpart's level
try:
level = logger.level(record.levelname).name
except ValueError:
level = record.levelno
# Get call stack depth
frame, depth = logging.currentframe(), 2
while frame and frame.f_code.co_filename == logging.__file__:
frame = frame.f_back
depth += 1
# Forward to loguru
logger.opt(depth=depth, exception=record.exc_info).log(level, record.getMessage())
def set_logger_config(logger_config: LoggerConf):
"""
Configuration Logs
:param logger_config:
:return:
"""
logger.remove()
# Configure Root Logger
logging.root.handlers = [InterceptHandler()]
logging.root.setLevel(logger_config.level)
# Set specific loggers to INFO level to reduce verbosity
logging.getLogger("uvicorn.access").setLevel(logging.INFO)
logging.getLogger("elastic_transport").setLevel(logging.INFO)
logging.getLogger("elasticsearch").setLevel(logging.INFO)
logging.getLogger("urllib3").setLevel(logging.INFO)
logging.getLogger("python_multipart").setLevel(logging.INFO)
logging.getLogger("aiohttp_retry").setLevel(logging.INFO)
# Block all existing loggers
for name in logging.root.manager.loggerDict.keys():
logging_logger = logging.getLogger(name)
logging_logger.handlers = [InterceptHandler()]
logging_logger.propagate = False
# Add Default Console Log Processor
logger.add(
sys.stdout,
format=logger_config.format,
level=logger_config.level,
filter=TraceIdFilter(),
enqueue=True,
backtrace=True,
diagnose=False
)
# Add additional log processors
for handler in logger_config.handlers:
log_file = Path(handler['sink'])
log_file.parent.mkdir(parents=True, exist_ok=True)
filter_func = handler.pop('filter', None)
logger.add(**handler, filter=TraceIdFilter(filter_func))
logger.debug(f'Logger set up with log level: {logger_config.level}')
================================================
FILE: src/backend/bisheng/core/middleware/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/prompts/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/prompts/manager.py
================================================
import logging
from abc import ABC
from bisheng.core.context import BaseContextManager
from bisheng.core.prompts.prompt_loader import PromptLoader
logger = logging.getLogger(__name__)
class PromptManager(BaseContextManager[PromptLoader]):
"""Prompt Manager
Responsible for loading, caching, and managing prompts
"""
name = 'prompts'
def __init__(self, **kwargs):
super().__init__(**kwargs)
async def _async_initialize(self) -> PromptLoader:
"""Asynchronous initialization prompt word loader"""
logger.info("Initializing PromptLoader...")
prompt_loader = PromptLoader()
logger.info("PromptLoader initialized.")
return prompt_loader
def _sync_initialize(self) -> PromptLoader:
"""Synchronous Initialization Prompt Loader"""
logger.info("Initializing PromptLoader...")
prompt_loader = PromptLoader()
logger.info("PromptLoader initialized.")
return prompt_loader
async def _async_cleanup(self) -> None:
"""Clean up database resources"""
pass
def _sync_cleanup(self) -> None:
"""Synchronously clean up database resources"""
pass
async def get_prompt_manager() -> PromptLoader:
"""Get a prompt word manager instance (asynchronous)"""
from bisheng.core.context.manager import app_context
try:
return await app_context.async_get_instance(PromptManager.name)
except KeyError:
logger.warning(f"PromptManager not found in context. Registering...")
try:
from bisheng.common.services.config_service import settings
app_context.register_context(PromptManager())
return await app_context.async_get_instance(PromptManager.name)
except Exception as e:
logger.error(f"Failed to register PromptManager: {e}")
raise
def get_prompt_manager_sync() -> PromptLoader:
"""Get a prompt word manager instance (sync method)"""
from bisheng.core.context.manager import app_context
try:
return app_context.sync_get_instance(PromptManager.name)
except KeyError:
logger.warning(f"PromptManager not found in context. Registering...")
try:
from bisheng.common.services.config_service import settings
app_context.register_context(PromptManager())
return app_context.sync_get_instance(PromptManager.name)
except Exception as e:
logger.error(f"Failed to register PromptManager: {e}")
raise
================================================
FILE: src/backend/bisheng/core/prompts/prompt_loader.py
================================================
import copy
import logging
import os
from enum import Enum
from string import Template
from typing import Union, Dict
import yaml
from pydantic import BaseModel, Field
logger = logging.getLogger(__name__)
class PromptTypeEnum(str, Enum):
"""
PromptType Enumeration
"""
PROMPT = "prompt"
CHATPROMPT = "chat_prompt"
class ChatPromptSchema(BaseModel):
system: str = Field(default='', description='SystemPromptContents')
user: str = Field(default='', description='UsersPromptContents')
# StandardPrompt Schema
class PromptSchema(BaseModel):
"""
StandardPrompt Schema
"""
description: str = Field(default='', description='PromptDescription')
type: PromptTypeEnum = Field(default=PromptTypeEnum.PROMPT, description='PromptType')
prompt: Union[str, ChatPromptSchema] = Field(..., description='PromptContents')
# embeddedPromptLoader..
class PromptLoader(object):
def __init__(self):
"""
InisialisasiPromptLoader..
"""
self.prompt_yaml_dir = os.path.join(os.path.dirname(__file__), 'yaml')
self.prompts_storage = {}
self._load_all()
# analyzingprompts
@staticmethod
def _parse_prompt(prompts_data: dict) -> Dict[str, PromptSchema]:
"""
analyzingPromptDATA
:param prompts_data: PromptDATA
:return: After parsingPromptObjects
"""
parsed_prompts = {}
for prompt_name, prompt_data in prompts_data.items():
if not isinstance(prompt_data, dict):
raise ValueError(f"Invalid prompt format for {prompt_name}: Expected a dictionary.")
prompt_schema = PromptSchema(**prompt_data)
parsed_prompts[prompt_name] = prompt_schema
return parsed_prompts
def _load_all(self):
for root, _, files in os.walk(self.prompt_yaml_dir):
for file in files:
if not file.endswith('.yaml') and not file.endswith('.yml'):
continue
file_path = os.path.join(root, file)
namespace = os.path.splitext(file)[0]
with open(file_path, 'r', encoding='utf-8') as f:
try:
prompt_data = yaml.safe_load(f)
if isinstance(prompt_data, dict):
self.prompts_storage[namespace] = self._parse_prompt(prompt_data.get('prompts', {}))
else:
raise ValueError(f"Invalid YAML format in {file_path}: Expected a dictionary.")
except yaml.YAMLError as e:
logger.error(f"Error parsing YAML file {file_path}: {e}")
continue
# Get the specifiedPrompt
def get_prompt(self, namespace: str, prompt_name: str) -> PromptSchema:
"""
Get the specifiedPrompt
:param namespace: namespace
:param prompt_name: PromptPart Name
:return: PromptObjects
"""
if namespace in self.prompts_storage:
return copy.deepcopy(self.prompts_storage[namespace].get(prompt_name, None))
else:
raise KeyError(f"Namespace '{namespace}' not found in prompts storage.")
def render_prompt(self, namespace: str, prompt_name: str, **kwargs) -> PromptSchema:
"""
Render specifiedPrompt
:param namespace: namespace
:param prompt_name: PromptPart Name
:param kwargs: Rendering parameters
:return: RenderedPromptString
"""
prompt_obj = self.get_prompt(namespace, prompt_name)
if prompt_obj.type == PromptTypeEnum.PROMPT:
prompt_obj.prompt = Template(prompt_obj.prompt).safe_substitute(**kwargs)
elif prompt_obj.type == PromptTypeEnum.CHATPROMPT:
prompt_obj.prompt.system = Template(prompt_obj.prompt.system).safe_substitute(**kwargs)
prompt_obj.prompt.user = Template(prompt_obj.prompt.user).safe_substitute(**kwargs)
else:
raise ValueError(f"Unsupported prompt type: {prompt_obj.type}")
return prompt_obj
================================================
FILE: src/backend/bisheng/core/prompts/yaml/gen_title.yaml
================================================
version: v1
description: 标题生成
prompts:
linsight:
description: "Linsight任务标题生成"
type: chat_prompt
prompt:
system: |
# 任务
你是一位资深信息架构师，请根据下面用户在会话中提供的「用户目标」，生成一个用于会话列表展示的对话标题（≤10 个汉字或英文单词）。
# 约束
仅输出标题内容，不要解释、不要添加多余文本或 Markdown。
user: |
用户目标:
${USER_GOAL}
call_reason_summary:
description: "人工介入原因总结"
type: chat_prompt
prompt:
system: |
# 任务
你是一位资深信息架构师,请根据下面用户在会话中提供的「问题原因」，生成摘要信息（≤10 个汉字或英文单词）。
# 约束
仅输出标题内容，不要解释、不要添加多余文本或 Markdown。
user: |
问题原因:
${USER_GOAL}
================================================
FILE: src/backend/bisheng/core/prompts/yaml/qa.yaml
================================================
version: v1
description: 标准问答（QA）生成器
prompts:
simple_qa_chat:
description: "生成简单的问答"
type: chat_prompt
prompt:
system: |
你是一个问答生成器，基于用户提供的上下文信息，生成简洁明了的问答对。
user: |
基于以下上下文信息，生成一个问答对：
上下文信息：${context}
请回答以下问题：${question}
simple_qa:
description: "生成标准的问答"
type: prompt
prompt: |
你是一个问答生成器，基于用户提供的上下文信息，生成简洁明了的问答对。
上下文信息：
${context}
请回答以下问题：
${question}
================================================
FILE: src/backend/bisheng/core/prompts/yaml/sop.yaml
================================================
version: v1
description: SOP
prompts:
gen_sop_summary:
description: "根据SOP详情生成摘要信息"
type: chat_prompt
prompt:
system: |
# 任务
1. 你是一位资深信息架构师，请根据下面用户提供的 SOP 详细内容，生成用于在表格中展示的 SOP 描述字段，控制在 50 字以内。
2. 生成结果请严格使用 JSON 格式表示。
# JSON 格式模板
```json
{
"sop_description": "生成结果"
}
```
user: |
SOP 详细内容：
${sop_detail}
================================================
FILE: src/backend/bisheng/core/prompts/yaml/workstation.yaml
================================================
version: v1
description: 工作台个人知识库
prompts:
personal_knowledge:
description: "个人知识库提示词"
type: prompt
prompt: |
# task
你是一个知识库问答助手：
1.按照用户使用的语言回答问题，并且答案要严谨专业。
2.你需要依据以下【参考文本】中的内容来回答，当【参考文本】中有明确与用户问题相关的内容时才进行回答，不可根据自己的知识来回答。若相关内容中包含图片内容的，请直接输出图片源文件，若占位符（例如：），回答时请保留这些占位符，不要修改或删除。
4. 如果文档中包含表格，请不要做任何修改，或返回表格格式的markdown
3.由于【参考文本】可能包含多个来自不同信息源的信息，所以根据这些不同的信息源可能得出有差异甚至冲突的答案，当发现这种情况时，这些答案都列举出来；如果没有冲突或差异，则只需要给出一个最终结果。
4.若【参考文本】中内容与用户问题不相关则回复“没有找到相关内容”。
# 参考文本
```
${retrieved_file_content}
```
# 用户问题
```
${question}
```
================================================
FILE: src/backend/bisheng/core/search/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/search/elasticsearch/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/search/elasticsearch/es_connection.py
================================================
from typing import Optional
from elasticsearch import Elasticsearch, AsyncElasticsearch
class ESConnection:
def __init__(self, es_hosts: str, **kwargs):
self.es_hosts = es_hosts
self.es_kwargs = kwargs
self._sync_es_connection: Optional[Elasticsearch] = None
self._es_connection: Optional[AsyncElasticsearch] = None
def _create_es_connection(self) -> 'AsyncElasticsearch':
return AsyncElasticsearch(
hosts=self.es_hosts,
**self.es_kwargs
)
def _create_sync_es_connection(self) -> 'Elasticsearch':
return Elasticsearch(
hosts=self.es_hosts,
**self.es_kwargs
)
@property
def es_connection(self) -> 'AsyncElasticsearch':
if self._es_connection is None:
self._es_connection = self._create_es_connection()
return self._es_connection
@property
def sync_es_connection(self) -> 'Elasticsearch':
if self._sync_es_connection is None:
self._sync_es_connection = self._create_sync_es_connection()
return self._sync_es_connection
async def close(self):
if self._es_connection is not None:
await self._es_connection.close()
self._es_connection = None
def sync_close(self):
if self._sync_es_connection is not None:
self._sync_es_connection.close()
self._sync_es_connection = None
================================================
FILE: src/backend/bisheng/core/search/elasticsearch/manager.py
================================================
import logging
from elasticsearch import AsyncElasticsearch, Elasticsearch
from bisheng.core.context import BaseContextManager
from bisheng.core.search.elasticsearch.es_connection import ESConnection
logger = logging.getLogger(__name__)
statistics_es_name = "statistics_elasticsearch"
class EsConnManager((BaseContextManager[ESConnection])):
name: str = "elasticsearch"
def __init__(self, es_hosts: str, name: str = None, **kwargs):
if name:
self.name = name
super().__init__(self.name, **kwargs)
self.es_hosts = es_hosts
self.kwargs = kwargs
async def _async_initialize(self) -> ESConnection:
"""Inisialisasi Elasticsearch Connection Manager"""
return ESConnection(self.es_hosts, **self.kwargs)
def _sync_initialize(self) -> ESConnection:
"""Synchronization Initialization"""
return ESConnection(self.es_hosts, **self.kwargs)
async def _async_cleanup(self) -> None:
"""Cleaned Elasticsearch reasourse"""
if self._instance:
await self._instance.close()
def _sync_cleanup(self) -> None:
"""Synchronous Cleanup Elasticsearch reasourse"""
if self._instance:
self._instance.sync_close()
async def get_es_connection() -> AsyncElasticsearch:
"""Dapatkan Elasticsearch Connection Instance"""
from bisheng.core.context.manager import app_context
try:
return (await app_context.async_get_instance(EsConnManager.name)).es_connection
except KeyError:
logger.warning(f"EsConnManager not found in app_context. Registering a new instance.")
try:
from bisheng.common.services.config_service import settings
app_context.register_context(
EsConnManager(es_hosts=settings.get_telemetry_conf().elasticsearch_url,
**settings.get_telemetry_conf().ssl_verify
))
return (await app_context.async_get_instance(EsConnManager.name)).es_connection
except Exception as e:
logger.error(f"Failed to register EsConnManager: {e}")
raise KeyError(f"EsConnManager not found in app_context and failed to register a new instance.") from e
async def get_statistics_es_connection() -> AsyncElasticsearch:
"""Get statistics Elasticsearch Connection Instance"""
from bisheng.core.context.manager import app_context
try:
return (await app_context.async_get_instance(statistics_es_name)).es_connection
except KeyError:
logger.warning(f"Statistics EsConnManager not found in app_context. Registering a new instance.")
try:
from bisheng.common.services.config_service import settings
app_context.register_context(
EsConnManager(es_hosts=settings.get_telemetry_conf().elasticsearch_url,
name=statistics_es_name,
**settings.get_telemetry_conf().statistics_ssl_verify))
return (await app_context.async_get_instance(statistics_es_name)).es_connection
except Exception as e:
logger.error(f"Failed to register Statistics EsConnManager: {e}")
raise KeyError(
f"Statistics EsConnManager not found in app_context and failed to register a new instance.") from e
def get_es_connection_sync() -> Elasticsearch:
"""Synchronous fetch Elasticsearch Connection Instance"""
from bisheng.core.context.manager import app_context
try:
return app_context.sync_get_instance(EsConnManager.name).sync_es_connection
except KeyError:
logger.warning(f"EsConnManager not found in app_context. Registering a new instance.")
try:
from bisheng.common.services.config_service import settings
app_context.register_context(
EsConnManager(es_hosts=settings.get_telemetry_conf().elasticsearch_url,
**settings.get_telemetry_conf().ssl_verify
))
return app_context.sync_get_instance(EsConnManager.name).sync_es_connection
except Exception as e:
logger.error(f"Failed to register EsConnManager: {e}")
raise KeyError(f"EsConnManager not found in app_context and failed to register a new instance.") from e
def get_statistics_es_connection_sync() -> Elasticsearch:
"""Sync Fetch Stats Elasticsearch Connection Instance"""
from bisheng.core.context.manager import app_context
try:
return app_context.sync_get_instance(statistics_es_name).sync_es_connection
except KeyError:
logger.warning(f"Statistics EsConnManager not found in app_context. Registering a new instance.")
try:
from bisheng.common.services.config_service import settings
app_context.register_context(
EsConnManager(es_hosts=settings.get_telemetry_conf().elasticsearch_url,
name=statistics_es_name,
**settings.get_telemetry_conf().ssl_verify))
return app_context.sync_get_instance(statistics_es_name).sync_es_connection
except Exception as e:
logger.error(f"Failed to register Statistics EsConnManager: {e}")
raise KeyError(
f"Statistics EsConnManager not found in app_context and failed to register a new instance.") from e
================================================
FILE: src/backend/bisheng/core/storage/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/storage/base.py
================================================
from abc import ABC, abstractmethod
from pathlib import Path
from typing import Union, BinaryIO
class BaseStorage(ABC):
"""Base class for storage backends."""
@abstractmethod
async def create_bucket(self, bucket_name: str) -> None:
"""Create a new storage bucket."""
pass
@abstractmethod
def create_bucket_sync(self, bucket_name: str) -> None:
"""Create a new storage bucket (synchronous)."""
pass
@abstractmethod
async def check_bucket_exists(self, bucket_name: str) -> bool:
"""Check if a storage bucket exists."""
pass
@abstractmethod
def check_bucket_exists_sync(self, bucket_name: str) -> bool:
"""Check if a storage bucket exists (synchronous)."""
pass
@abstractmethod
async def get_all_buckets(self) -> list:
"""Get a list of all storage buckets."""
pass
@abstractmethod
def get_all_buckets_sync(self) -> list:
"""Get a list of all storage buckets (synchronous)."""
pass
@abstractmethod
async def remove_bucket(self, bucket_name: str) -> None:
"""Remove a storage bucket."""
pass
@abstractmethod
def remove_bucket_sync(self, bucket_name: str) -> None:
"""Remove a storage bucket (synchronous)."""
pass
@abstractmethod
async def put_object(self, bucket_name: str, object_name: str, file: Union[bytes, BinaryIO, Path, str],
content_type: str = None, **kwargs) -> None:
"""Upload an object to a storage bucket."""
pass
@abstractmethod
def put_object_sync(self, bucket_name: str, object_name: str, file: Union[bytes, BinaryIO, Path, str],
content_type: str = None, **kwargs) -> None:
"""Upload an object to a storage bucket (synchronous)."""
pass
@abstractmethod
async def get_object(self, bucket_name: str, object_name: str) -> bytes:
"""Download an object from a storage bucket."""
pass
@abstractmethod
def get_object_sync(self, bucket_name: str, object_name: str) -> bytes:
"""Download an object from a storage bucket (synchronous)."""
pass
@abstractmethod
async def object_exists(self, bucket_name: str, object_name: str) -> bool:
"""Check if an object exists in a storage bucket."""
pass
def object_exists_sync(self, bucket_name: str, object_name: str) -> bool:
"""Check if an object exists in a storage bucket (synchronous)."""
pass
@abstractmethod
async def copy_object(self, source_bucket: str, source_object: str,
dest_bucket: str, dest_object: str) -> None:
"""Copy an object from one storage bucket to another."""
pass
@abstractmethod
def copy_object_sync(self, source_bucket: str, source_object: str,
dest_bucket: str, dest_object: str) -> None:
"""Copy an object from one storage bucket to another (synchronous)."""
pass
@abstractmethod
async def remove_object(self, bucket_name: str, object_name: str) -> None:
"""Remove an object from a storage bucket."""
pass
@abstractmethod
def remove_object_sync(self, bucket_name: str, object_name: str) -> None:
"""Remove an object from a storage bucket (synchronous)."""
pass
================================================
FILE: src/backend/bisheng/core/storage/minio/__init__.py
================================================
================================================
FILE: src/backend/bisheng/core/storage/minio/minio_manager.py
================================================
import logging
from abc import ABC
from typing import Union
from bisheng.core.config.settings import MinioConf
from bisheng.core.context import BaseContextManager
from bisheng.core.storage.minio.minio_storage import MinioStorage
logger = logging.getLogger(__name__)
class MinioManager(BaseContextManager[MinioStorage]):
"""Minio Global Manager
Responsible for management Minio Global lifecycle of storage, providing a unified access interface
Supports connection pool monitoring, health checks, and easy storage management
"""
name: str = "minio"
def __init__(
self,
minio_config: Union[MinioConf, dict],
**kwargs
):
super().__init__(self.name, **kwargs)
self.minio_config = minio_config
if isinstance(self.minio_config, dict):
self.minio_config = MinioConf(**self.minio_config)
async def _async_initialize(self) -> MinioStorage:
"""Inisialisasi Minio Storage Manager"""
return MinioStorage(
self.minio_config
)
def _sync_initialize(self) -> MinioStorage:
"""Synchronization Initialization"""
return MinioStorage(
self.minio_config
)
def _sync_cleanup(self) -> None:
"""Synchronous Cleanup Minio reasourse"""
if self._instance:
self._instance.close_sync()
async def _async_cleanup(self) -> None:
"""Cleaned Minio reasourse"""
if self._instance:
await self._instance.close()
async def get_minio_storage() -> MinioStorage:
"""Dapatkan Minio Storage Instance"""
from bisheng.core.context.manager import app_context
try:
return await app_context.async_get_instance(MinioManager.name)
except KeyError:
logger.warning(f"MinioManager not found in app_context. Registering a new instance.")
try:
from bisheng.common.services.config_service import settings
app_context.register_context(MinioManager(
minio_config=settings.object_storage.minio
))
return await app_context.async_get_instance(MinioManager.name)
except Exception as e:
logger.error(f"Failed to register MinioManager: {e}")
raise
def get_minio_storage_sync() -> MinioStorage:
"""Synchronous fetch Minio Storage Instance"""
from bisheng.core.context.manager import app_context
try:
return app_context.sync_get_instance(MinioManager.name)
except KeyError:
logger.warning(f"MinioManager not found in app_context. Registering a new instance.")
try:
from bisheng.common.services.config_service import settings
app_context.register_context(MinioManager(
minio_config=settings.object_storage.minio
))
return app_context.sync_get_instance(MinioManager.name)
except Exception as e:
logger.error(f"Failed to register MinioManager: {e}")
raise
================================================
FILE: src/backend/bisheng/core/storage/minio/minio_storage.py
================================================
import json
import os
from abc import ABC
from datetime import timedelta
from io import BytesIO
from pathlib import Path
from typing import BinaryIO, Union, Optional
import minio
import miniopy_async
import miniopy_async.commonconfig as miniopy_async_commonconfig
from loguru import logger
from minio.commonconfig import Filter
from minio.lifecycleconfig import LifecycleConfig, Rule, Expiration
from bisheng.core.config.settings import MinioConf
from bisheng.core.storage.base import BaseStorage
class MinioStorage(BaseStorage, ABC):
"""MinIO storage backend implementation."""
def __init__(self, minio_config: MinioConf):
self.minio_config = minio_config
self.bucket = minio_config.public_bucket
self.tmp_bucket = minio_config.tmp_bucket
self.minio_client_sync = minio.Minio(
endpoint=minio_config.endpoint,
access_key=minio_config.access_key,
secret_key=minio_config.secret_key,
secure=minio_config.secure,
cert_check=minio_config.cert_check
)
self.minio_client = miniopy_async.Minio(
endpoint=minio_config.endpoint,
access_key=minio_config.access_key,
secret_key=minio_config.secret_key,
secure=minio_config.secure,
cert_check=minio_config.cert_check
)
self._init_bucket_conf()
def _init_bucket_conf(self):
# create need bucket
self.create_bucket_sync(bucket_name=self.bucket)
self.create_bucket_sync(bucket_name=self.tmp_bucket)
# set knowledge chunk images anonymous read policy
anonymous_read_policy = {
"Version": "2012-10-17",
"Statement": [{
"Effect": "Allow",
"Principal": {
"AWS": ["*"]
},
"Action": ["s3:GetObject"],
"Resource": [f"arn:aws:s3:::{self.bucket}/knowledge/images/*",
f"arn:aws:s3:::{self.bucket}/tmp/images/*"]
}]
}
try:
policy = self.minio_client_sync.get_bucket_policy(self.bucket)
except Exception as e:
if str(e).find('NoSuchBucketPolicy') == -1:
raise e
self.minio_client_sync.set_bucket_policy(self.bucket, json.dumps(anonymous_read_policy))
# set tmp bucket lifecycle
if not self.minio_client_sync.get_bucket_lifecycle(self.tmp_bucket):
lifecycle_conf = LifecycleConfig([
Rule(
'Enabled',
rule_filter=Filter(prefix='*'),
rule_id='rule1',
expiration=Expiration(days=3),
),
], )
self.minio_client_sync.set_bucket_lifecycle(self.tmp_bucket, lifecycle_conf)
async def create_bucket(self, bucket_name: str) -> None:
if not await self.minio_client.bucket_exists(bucket_name):
await self.minio_client.make_bucket(bucket_name)
def create_bucket_sync(self, bucket_name: str) -> None:
if not self.minio_client_sync.bucket_exists(bucket_name):
self.minio_client_sync.make_bucket(bucket_name)
async def check_bucket_exists(self, bucket_name: str) -> bool:
return await self.minio_client.bucket_exists(bucket_name)
def check_bucket_exists_sync(self, bucket_name: str) -> bool:
return self.minio_client_sync.bucket_exists(bucket_name)
async def get_all_buckets(self) -> list:
return await self.minio_client.list_buckets()
def get_all_buckets_sync(self) -> list:
return self.minio_client_sync.list_buckets()
async def remove_bucket(self, bucket_name: str) -> None:
if await self.minio_client.bucket_exists(bucket_name):
await self.minio_client.remove_bucket(bucket_name)
def remove_bucket_sync(self, bucket_name: str) -> None:
if self.minio_client_sync.bucket_exists(bucket_name):
self.minio_client_sync.remove_bucket(bucket_name)
async def put_object(self, *, bucket_name: Optional[str] = None, object_name: str,
file: Union[bytes, BinaryIO, Path, str],
content_type: str = "application/octet-stream", **kwargs) -> None:
if bucket_name is None:
bucket_name = self.bucket
#
if isinstance(file, (Path, str)):
file_path = str(file)
await self.minio_client.fput_object(
bucket_name=bucket_name,
object_name=object_name,
file_path=file_path,
content_type=content_type,
**kwargs
)
return
data_stream = file
if isinstance(file, bytes):
data_stream = BytesIO(file)
if 'length' not in kwargs:
try:
if hasattr(data_stream, "getbuffer"):
kwargs['length'] = data_stream.getbuffer().nbytes
elif hasattr(data_stream, "fileno"):
kwargs['length'] = os.fstat(data_stream.fileno()).st_size
else:
data_stream.seek(0, 2)
kwargs['length'] = data_stream.tell()
data_stream.seek(0)
except Exception as e:
raise ValueError(f"Could not determine file length for upload: {str(e)}")
if hasattr(data_stream, "seek") and callable(data_stream.seek):
data_stream.seek(0)
await self.minio_client.put_object(
bucket_name=bucket_name,
object_name=object_name,
data=data_stream,
content_type=content_type,
**kwargs
)
def put_object_sync(self, *, bucket_name: Optional[str] = None, object_name: str,
file: Union[bytes, BinaryIO, Path, str],
content_type: str = "application/octet-stream", **kwargs) -> None:
if bucket_name is None:
bucket_name = self.bucket
if isinstance(file, (Path, str)):
file_path = str(file)
self.minio_client_sync.fput_object(
bucket_name=bucket_name,
object_name=object_name,
file_path=file_path,
content_type=content_type,
**kwargs
)
return
data_stream = file
if isinstance(file, bytes):
data_stream = BytesIO(file)
if 'length' not in kwargs:
try:
if hasattr(data_stream, "getbuffer"):
kwargs['length'] = data_stream.getbuffer().nbytes
elif hasattr(data_stream, "fileno"):
try:
kwargs['length'] = os.fstat(data_stream.fileno()).st_size
except Exception:
data_stream.seek(0, 2)
kwargs['length'] = data_stream.tell()
data_stream.seek(0)
else:
data_stream.seek(0, 2)
kwargs['length'] = data_stream.tell()
data_stream.seek(0)
except Exception as e:
raise ValueError(f"Could not determine file length for upload: {str(e)}")
if hasattr(data_stream, "seek") and callable(data_stream.seek):
data_stream.seek(0)
self.minio_client_sync.put_object(
bucket_name=bucket_name,
object_name=object_name,
data=data_stream,
content_type=content_type,
**kwargs
)
async def put_object_tmp(self, object_name: str, file: Union[bytes, BinaryIO, Path, str],
content_type: str = "application/octet-stream", **kwargs) -> None:
await self.put_object(
bucket_name=self.tmp_bucket,
object_name=object_name,
file=file,
content_type=content_type,
**kwargs
)
def put_object_tmp_sync(self, object_name: str, file: Union[bytes, BinaryIO, Path, str],
content_type: str = "application/octet-stream", **kwargs) -> None:
self.put_object_sync(
bucket_name=self.tmp_bucket,
object_name=object_name,
file=file,
content_type=content_type,
**kwargs
)
async def get_object(self, bucket_name: Optional[str] = None, object_name: str = None) -> bytes | None:
if bucket_name is None:
bucket_name = self.bucket
if object_name is None:
raise ValueError("get_object: object_name must be provided")
response = await self.minio_client.get_object(bucket_name, object_name)
try:
data = await response.read()
return data
except Exception:
raise
finally:
response.close()
def get_object_sync(self, bucket_name: Optional[str] = None, object_name: str = None) -> bytes | None:
if bucket_name is None:
bucket_name = self.bucket
if object_name is None:
raise ValueError("get_object_sync: object_name must be provided")
response = self.minio_client_sync.get_object(bucket_name, object_name)
try:
data = response.read()
return data
except Exception:
raise
finally:
response.close()
response.release_conn()
async def object_exists(self, bucket_name: Optional[str] = None, object_name: str = None) -> bool:
if not bucket_name:
bucket_name = self.bucket
if not object_name:
logger.warning("object_exists_sync: object_name must be provided")
return False
try:
await self.minio_client.stat_object(bucket_name, object_name)
return True
except Exception as e:
if 'code: NoSuchKey' in str(e):
return False
raise e
def object_exists_sync(self, bucket_name: Optional[str] = None, object_name: str = None) -> bool:
if not bucket_name:
bucket_name = self.bucket
if not object_name:
logger.warning("object_exists_sync: object_name must be provided")
return False
try:
self.minio_client_sync.stat_object(bucket_name, object_name)
return True
except Exception as e:
if 'code: NoSuchKey' in str(e):
return False
raise e
async def copy_object(self, source_bucket: str = None, source_object: str = None,
dest_bucket: str = None, dest_object: str = None) -> None:
if source_bucket is None:
source_bucket = self.tmp_bucket
if dest_bucket is None:
dest_bucket = self.bucket
source = miniopy_async_commonconfig.CopySource(
bucket_name=source_bucket,
object_name=source_object
)
await self.minio_client.copy_object(
bucket_name=dest_bucket,
object_name=dest_object,
source=source
)
def copy_object_sync(self, source_bucket: str = None, source_object: str = None,
dest_bucket: str = None, dest_object: str = None) -> None:
if source_bucket is None:
source_bucket = self.tmp_bucket
if dest_bucket is None:
dest_bucket = self.bucket
source = minio.commonconfig.CopySource(
bucket_name=source_bucket,
object_name=source_object
)
self.minio_client_sync.copy_object(
bucket_name=dest_bucket,
object_name=dest_object,
source=source
)
async def remove_object(self, bucket_name: Optional[str] = None, object_name: str = None) -> None:
if bucket_name is None:
bucket_name = self.bucket
if object_name is None:
raise ValueError("remove_object: object_name must be provided")
await self.minio_client.remove_object(bucket_name, object_name)
def remove_object_sync(self, bucket_name: Optional[str] = None, object_name: str = None) -> None:
if bucket_name is None:
bucket_name = self.bucket
if object_name is None:
raise ValueError("remove_object_sync: object_name must be provided")
self.minio_client_sync.remove_object(bucket_name, object_name)
async def get_share_link(self, object_name, bucket=None, clear_host: bool = True, expire_days: int = 7) -> str:
"""
DapatkanminioFile sharing link
:param object_name:
:param bucket:
:param clear_host: Do you want to removehostAddress: urlVia FrontendnginxProxy Access
:param expire_days: Link expiration time, days
:return:
"""
if bucket is None:
bucket = self.bucket
# filepath "/" There will be at the beginningnginxQuestions
if object_name[0] == '/':
object_name = object_name[1:]
share_link = await self.minio_client.presigned_get_object(bucket, object_name,
expires=timedelta(days=expire_days))
if clear_host:
share_link = self.clear_minio_share_host(share_link)
return share_link
def get_share_link_sync(self, object_name, bucket=None, clear_host: bool = True, expire_days: int = 7) -> str:
"""
Synchronous fetchminioFile sharing link, Default RemovalhostAddress: urlwill go through the front endnginxProxy Access
:param object_name:
:param bucket:
:param clear_host: Do you want to removehostAddress: urlVia FrontendnginxProxy Access
:param expire_days: Link expiration time, days
:return:
"""
if bucket is None:
bucket = self.bucket
# filepath "/" There will be at the beginningnginxQuestions
if object_name[0] == '/':
object_name = object_name[1:]
share_link = self.minio_client_sync.presigned_get_object(bucket, object_name,
expires=timedelta(days=expire_days))
if clear_host:
share_link = self.clear_minio_share_host(share_link)
return share_link
def get_minio_share_host(self) -> str:
"""
Dapatkanminio share host
"""
minio_share = self.minio_config.sharepoint
if self.minio_config.share_schema:
return f'https://{minio_share}'
return f'http://{minio_share}'
def clear_minio_share_host(self, file_url: str):
"""
TODO The logical solution is to deploy ahttpsright of privacyminioUse with front-end
to be erasedurlhitting the nail on the headminio shareAddress: Let the front end throughnginxProxy to access resources
"""
share_host = self.get_minio_share_host()
return file_url.replace(share_host, '')
async def close(self) -> None:
"""Close Minio Client link"""
await self.minio_client.close_session()
def close_sync(self) -> None:
"""Sync off Minio Client link"""
del self.minio_client_sync
================================================
FILE: src/backend/bisheng/core/vectorstore/__init__.py
================================================
from langchain_elasticsearch import ElasticsearchStore, AsyncElasticsearchStore
from langchain_elasticsearch.vectorstores import BM25Strategy, AsyncBM25Strategy
from langchain_milvus import Milvus
__all__ = [
'Milvus',
'ElasticsearchStore',
'AsyncElasticsearchStore',
'BM25Strategy',
'AsyncBM25Strategy'
]
================================================
FILE: src/backend/bisheng/core/vectorstore/ensemble_retriever.py
================================================
import asyncio
from typing import List, Dict, Any, Sequence, Coroutine, Optional
from pydantic import Field, BaseModel, PrivateAttr, ConfigDict
from langchain_core.retrievers import BaseRetriever
from langchain_core.vectorstores import VectorStore
from langchain_core.documents import Document
from langchain_core.callbacks import (
CallbackManagerForRetrieverRun,
AsyncCallbackManagerForRetrieverRun
)
from langchain_core.runnables import RunnableConfig
from bisheng.core.ai.rerank.rrf_rerank import RRFRerank
# Input Retriever Parameter Model
class VectorRetrieverParams(BaseModel):
model_config = ConfigDict(arbitrary_types_allowed=True)
vector_store: VectorStore = Field(..., description="Vector Library Instance")
search_kwargs: Dict = Field(default_factory=dict, description="Retrieve Parameters")
weight: float = Field(default=1.0, description="Retriever Weight")
class RRFMultiVectorRetriever(BaseRetriever):
"""
predicated upon RRF (Reciprocal Rank Fusion) Multiple vector retriever.
Sync enabled/Asynchronous calls that support character-length-based context truncation.
"""
vector_store_params: List[VectorRetrieverParams] = Field(
..., description="Multiple vector libraries and their retrieval parameters and weights"
)
top_k: int = Field(default=5, description="The final number of documents returned. Note: Recommended settings >0 Value of")
rrf_c: int = Field(default=60, description="RRF constant c, smooth ranking weights")
remove_zero_score: bool = Field(default=True, description="Remove or not RRF Score is 0 Documents")
max_context_length: int = Field(default=0, description="Returns the maximum total character length of a document,0Indicates unlimited")
_rrf_rerank: RRFRerank = PrivateAttr()
_retrievers: List[BaseRetriever] = PrivateAttr(default_factory=list)
def __init__(self, **data: Any):
super().__init__(**data)
weights = []
self._retrievers = []
for param in self.vector_store_params:
weights.append(param.weight)
# Convert To retriever
self._retrievers.append(
param.vector_store.as_retriever(search_kwargs=param.search_kwargs)
)
# Inisialisasi RRF Reorder
self._rrf_rerank = RRFRerank(
retrievers=self._retrievers,
weights=weights,
c=self.rrf_c,
remove_zero_score=self.remove_zero_score
)
def _get_relevant_documents(
self, query: str, *, run_manager: CallbackManagerForRetrieverRun, **kwargs
) -> Sequence[Document]:
docs_lists = []
# Pass-through callback configuration
run_config: RunnableConfig = {"callbacks": run_manager.get_child()}
for retriever in self._retrievers:
try:
tmp_docs = retriever.invoke(query, config=run_config, **kwargs)
docs_lists.append(tmp_docs)
except Exception as e:
# Log exceptions but continue to execute other retrievers
run_manager.on_text(f"Retriever {retriever} failed with error: {e}", end="\n")
docs_lists.append([])
# Recall RRF Reorder
reranked_docs = self._rrf_rerank.compress_documents(query=query, documents=docs_lists)
return self._post_process_documents(reranked_docs)
async def _aget_relevant_documents(
self, query: str, *, run_manager: AsyncCallbackManagerForRetrieverRun, **kwargs
) -> Sequence[Document]:
tasks: List[Coroutine] = []
run_config: RunnableConfig = {"callbacks": run_manager.get_child()}
for retriever in self._retrievers:
tasks.append(
retriever.ainvoke(query, config=run_config, **kwargs)
)
# Concurrent execution
results = await asyncio.gather(*tasks, return_exceptions=True)
docs_lists = []
for idx, result in enumerate(results):
if isinstance(result, Exception):
# Log exceptions but continue to execute other retrievers
await run_manager.on_text(f"Retriever {self._retrievers[idx]} failed with error: {result}", end="\n")
docs_lists.append([])
else:
docs_lists.append(result)
# RRF Rearrangements
reranked_docs = self._rrf_rerank.compress_documents(query=query, documents=docs_lists)
return self._post_process_documents(reranked_docs)
def _post_process_documents(self, documents: Sequence[Document]) -> List[Document]:
"""cut off TopK And Length filtering"""
# Top K cut off
if self.top_k > 0:
documents = documents[:self.top_k]
# Length filtering
return self._filter_by_context_length(list(documents))
def _filter_by_context_length(self, documents: List[Document]) -> List[Document]:
"""according max_context_length Filter Documents List"""
if self.max_context_length <= 0:
return documents
curr_size = 0
filtered_docs = []
for doc in documents:
doc_size = len(doc.page_content)
# Determine if the limit has been exceeded after joining the document
if curr_size + doc_size <= self.max_context_length:
filtered_docs.append(doc)
curr_size += doc_size
else:
break
return filtered_docs
================================================
FILE: src/backend/bisheng/core/vectorstore/multi_retriever.py
================================================
import asyncio
from typing import List, Dict
from langchain_core.callbacks import CallbackManagerForRetrieverRun
from langchain_core.documents import Document
from langchain_core.retrievers import BaseRetriever
from langchain_core.vectorstores import VectorStore
class MultiRetriever(BaseRetriever):
""" A retriever that combines multiple retrievers. """
vectors: List[VectorStore]
search_kwargs: List[Dict]
finally_k: int = 0
def _get_relevant_documents(
self, query: str, *, run_manager: CallbackManagerForRetrieverRun, **kwargs
) -> list[Document]:
docs: List[tuple[Document, float]] = []
for index, vector in enumerate(self.vectors):
kwargs_ = self.search_kwargs[index] | kwargs
tmp_docs = vector.similarity_search_with_score(query, **kwargs_)
docs.extend(tmp_docs)
return self.parse_doc_with_score(docs)
async def _aget_relevant_documents(
self, query: str, *, run_manager: CallbackManagerForRetrieverRun, **kwargs
) -> list[Document]:
tasks = []
for index, vector in enumerate(self.vectors):
kwargs_ = self.search_kwargs[index] | kwargs
tasks.append(vector.asimilarity_search_with_score(query, **kwargs_))
results = await asyncio.gather(*tasks)
docs: List[tuple[Document, float]] = []
for tmp_docs in results:
docs.extend(tmp_docs)
return self.parse_doc_with_score(docs)
def parse_doc_with_score(self, docs_with_score: List[tuple[Document, float]]) -> List[Document]:
""" parse documents with score to documents only """
docs_with_score.sort(key=lambda x: x[1])
if self.finally_k > 0:
docs_with_score = docs_with_score[:self.finally_k]
return [doc for doc, _ in docs_with_score]
================================================
FILE: src/backend/bisheng/custom/__init__.py
================================================
================================================
FILE: src/backend/bisheng/custom/customs.py
================================================
from bisheng.template import frontend_node
# These should always be instantiated
CUSTOM_NODES = {
# "prompts": {
# "ZeroShotPrompt": frontend_node.prompts.ZeroShotPromptNode(),
# },
'tools': {
'PythonFunctionTool': frontend_node.tools.PythonFunctionToolNode(),
'PythonFunction': frontend_node.tools.PythonFunctionNode(),
'Tool': frontend_node.tools.ToolNode(),
},
'agents': {
'JsonAgent': frontend_node.agents.JsonAgentNode(),
# "CSVAgent": frontend_node.agents.CSVAgentNode(),
'AgentInitializer': frontend_node.agents.InitializeAgentNode(),
'VectorStoreAgent': frontend_node.agents.VectorStoreAgentNode(),
'VectorStoreRouterAgent': frontend_node.agents.VectorStoreRouterAgentNode(),
# "SQLAgent": frontend_node.agents.SQLAgentNode(),
},
'utilities': {
'SQLDatabase': frontend_node.agents.SQLDatabaseNode(),
},
'memories': {
'PostgresChatMessageHistory':
frontend_node.memories.PostgresChatMessageHistoryFrontendNode(),
'MongoDBChatMessageHistory':
frontend_node.memories.MongoDBChatMessageHistoryFrontendNode(),
},
'chains': {
'SeriesCharacterChain': frontend_node.chains.SeriesCharacterChainNode(),
'TimeTravelGuideChain': frontend_node.chains.TimeTravelGuideChainNode(),
'MidJourneyPromptChain': frontend_node.chains.MidJourneyPromptChainNode(),
'load_qa_chain': frontend_node.chains.CombineDocsChainNode(),
'load_summarize_chain': frontend_node.chains.SummarizeDocsChain(),
},
'embeddings': {
'OpenAIProxyEmbedding': frontend_node.embeddings.OpenAIProxyEmbedding(),
},
'input_output': {
'InputNode': frontend_node.input_output.InputNode(),
'OutputNode': frontend_node.input_output.OutputNode(),
'InputFileNode': frontend_node.input_output.InputFileNode(),
},
# 'output_parsers': {
# 'RouterOutputParser': frontend_node.output_parsers.RouterOutputParserFrontendNode(),
# }
}
def get_custom_nodes(node_type: str):
"""Get custom nodes."""
return CUSTOM_NODES.get(node_type, {})
================================================
FILE: src/backend/bisheng/database/__init__.py
================================================
================================================
FILE: src/backend/bisheng/database/base.py
================================================
import uuid
from sqlalchemy import func, Select
from sqlmodel.ext.asyncio.session import AsyncSession
from sqlmodel import Session
def get_count(session: Session, q: Select) -> int:
"""
Number of fetch query results
:param session:
:param q:
:return:
"""
count_q = q.with_only_columns(func.count()).order_by(None).select_from(q.get_final_froms()[0])
iterator = session.exec(count_q)
for count in iterator:
return count
return 0
async def async_get_count(session: AsyncSession, q: Select) -> int:
"""
Get the number of asynchronous query results
:param session:
:param q:
:return:
"""
count_q = q.with_only_columns(func.count()).order_by(None).select_from(q.get_final_froms()[0])
iterator = await session.exec(count_q)
for count in iterator:
return count
return 0
def uuid_hex() -> str:
"""
Generate oneUUIDhexadecimal string
:return: UUIDhexadecimal string
"""
return uuid.uuid4().hex
================================================
FILE: src/backend/bisheng/database/constants.py
================================================
from enum import Enum
# Default Normal User Role'sID
DefaultRole = 2
# Super Admin RoleID
AdminRole = 1
# Some of the basiccategoryType
class MessageCategory(Enum):
QUESTION = 'question' # User Questions
ANSWER = 'answer' # Answers
================================================
FILE: src/backend/bisheng/database/data/component.json
================================================
[{"单文档问答":{"id":"RetrievalQA-3568b","type":"InputFileNode","node":{"output_types":[],"display_name":"单文档问答","documentation":"","base_classes":["BaseRetrievalQA","RetrievalQA","Chain","function"],"description":"Chain for question-answering against an index.","template":{"token_max_CombineDocsChain-Pud2p":{"l2":false,"info":"当前只对stuff 生效","list":false,"name":"token_max","show":true,"type":"int","value":-1,"l2_name":"token_max","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"token_max","proxy":{"id":"CombineDocsChain-Pud2p","field":"token_max"}},"chain_type_CombineDocsChain-Pud2p":{"l2":false,"info":"","list":true,"name":"chain_type","show":true,"type":"str","value":"stuff","l2_name":"chain_type","options":["stuff","map_reduce","map_rerank","refine"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"CombineDocsChain-Pud2p","field":"chain_type"},"display_name":"Chain Type"},"tags_RetrievalQA-qH6Mk":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-qH6Mk","field":"tags"},"display_name":"Tags"},"memory_RetrievalQA-qH6Mk":{"l2":false,"info":"","list":false,"name":"memory","show":true,"type":"BaseMemory","l2_name":"memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-qH6Mk","field":"memory"},"display_name":"Memory"},"verbose_RetrievalQA-qH6Mk":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-qH6Mk","field":"verbose"},"display_name":"Verbose"},"metadata_RetrievalQA-qH6Mk":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-qH6Mk","field":"metadata"},"display_name":"Metadata"},"callbacks_RetrievalQA-qH6Mk":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-qH6Mk","field":"callbacks"},"display_name":"Callbacks"},"input_key_RetrievalQA-qH6Mk":{"l2":false,"info":"","list":false,"name":"input_key","show":true,"type":"str","value":"query","l2_name":"input_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-qH6Mk","field":"input_key"},"display_name":"Input Key"},"input_node_RetrievalQA-qH6Mk":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question","proxy":{"id":"RetrievalQA-qH6Mk","field":"input_node"}},"output_key_RetrievalQA-qH6Mk":{"l2":false,"info":"","list":false,"name":"output_key","show":true,"type":"str","value":"result","l2_name":"output_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-qH6Mk","field":"output_key"},"display_name":"Output Key"},"return_source_documents_RetrievalQA-qH6Mk":{"l2":false,"info":"","list":false,"name":"return_source_documents","show":true,"type":"bool","value":true,"l2_name":"return_source_documents","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-qH6Mk","field":"return_source_documents"},"display_name":"Return Source Documents"},"proxy_url_OpenAIProxyEmbedding-yvld7":{"l2":false,"info":"","list":false,"name":"proxy_url","show":false,"type":"str","value":"","l2_name":"proxy_url","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"http://43.133.35.137:8080","proxy":{"id":"OpenAIProxyEmbedding-yvld7","field":"proxy_url"},"display_name":"Proxy Url"},"file_path_InputFileNode-jdLkB":{"l2":false,"info":"","list":false,"name":"file_path","show":true,"type":"file","value":"","l2_name":"file_path","advanced":true,"password":false,"required":false,"suffixes":[".html",".md",".txt",".jpg",".png",".jpeg",".csv",".doc",".docx",".pdf",".ppt",".pptx",".xlsx",".tiff"],"fileTypes":["html","md","txt","jpg","png","jpeg","csv","doc","docx","pdf","ppt","pptx","tiff","xlsx"],"file_path":null,"multiline":false,"placeholder":"","display_name":"输入内容","proxy":{"id":"InputFileNode-jdLkB","field":"file_path"}},"n_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"l2_name":"n","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"n"},"display_name":"N"},"tags_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"tags"},"display_name":"Tags"},"cache_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"cache","show":false,"type":"bool","l2_name":"cache","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"cache"},"display_name":"Cache"},"top_p_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"l2_name":"top_p","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"top_p"},"display_name":"Top P"},"client_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"client","show":false,"type":"Any","l2_name":"client","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"client"},"display_name":"Client"},"headers_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","l2_name":"headers","advanced":true,"password":false,"required":false,"multiline":true,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"headers"},"display_name":"Headers"},"verbose_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"verbose"},"display_name":"Verbose"},"metadata_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"metadata"},"display_name":"Metadata"},"callbacks_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"callbacks"},"display_name":"Callbacks"},"streaming_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"l2_name":"streaming","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"streaming"},"display_name":"Streaming"},"max_tokens_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","l2_name":"max_tokens","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"max_tokens"},"display_name":"Max Tokens"},"model_name_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-3.5-turbo-16k-0613","l2_name":"model_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"model_name"},"display_name":"Model Name"},"max_retries_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"l2_name":"max_retries","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"max_retries"},"display_name":"Max Retries"},"temperature_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"l2_name":"temperature","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"temperature"},"display_name":"Temperature"},"model_kwargs_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"model_kwargs","show":false,"type":"code","l2_name":"model_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"model_kwargs"},"display_name":"Model Kwargs"},"elemai_api_key_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","l2_name":"elemai_api_key","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"elemai_api_key"},"display_name":"Elemai Api Key"},"elemai_base_url_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","l2_name":"elemai_base_url","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"elemai_base_url"},"display_name":"Elemai Base Url"},"request_timeout_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"request_timeout","show":false,"type":"float","l2_name":"request_timeout","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"request_timeout"},"display_name":"Request Timeout"},"tiktoken_model_name_ProxyChatLLM-rnE4s":{"l2":false,"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","l2_name":"tiktoken_model_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-rnE4s","field":"tiktoken_model_name"},"display_name":"Tiktoken Model Name"},"drop_old_Milvus-T3kRH":{"l2":false,"info":"","list":false,"name":"drop_old","show":false,"type":"bool","value":false,"l2_name":"drop_old","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-T3kRH","field":"drop_old"},"display_name":"Drop Old"},"metadatas_Milvus-T3kRH":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-T3kRH","field":"metadatas"},"display_name":"Metadatas"},"index_params_Milvus-T3kRH":{"l2":false,"info":"","list":false,"name":"index_params","show":false,"type":"code","l2_name":"index_params","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-T3kRH","field":"index_params"},"display_name":"Index Params"},"search_kwargs_Milvus-T3kRH":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":false,"type":"code","value":"{}","l2_name":"search_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-T3kRH","field":"search_kwargs"},"display_name":"Search Kwargs"},"search_params_Milvus-T3kRH":{"l2":false,"info":"","list":false,"name":"search_params","show":false,"type":"code","l2_name":"search_params","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-T3kRH","field":"search_params"},"display_name":"Search Params"},"collection_name_Milvus-T3kRH":{"l2":false,"info":"","list":false,"name":"collection_name","show":true,"type":"str","value":"","l2_name":"collection_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","collection_id":"","proxy":{"id":"Milvus-T3kRH","field":"collection_name"},"display_name":"Collection Name"},"connection_args_Milvus-T3kRH":{"l2":false,"info":"","list":false,"name":"connection_args","show":false,"type":"code","value":"","l2_name":"connection_args","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-T3kRH","field":"connection_args"},"display_name":"Connection Args"},"consistency_level_Milvus-T3kRH":{"l2":false,"info":"","list":false,"name":"consistency_level","show":false,"type":"str","value":"Session","l2_name":"consistency_level","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-T3kRH","field":"consistency_level"},"display_name":"Consistency Level"},"metadata_ElemUnstructuredLoaderV0-LAGXM":{"info":"","list":false,"name":"metadata","show":false,"type":"code","value":"{}","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Metadata","proxy":{"id":"ElemUnstructuredLoaderV0-LAGXM","field":"metadata"}},"unstructured_api_url_ElemUnstructuredLoaderV0-LAGXM":{"info":"","list":false,"name":"unstructured_api_url","show":true,"type":"str","value":"","advanced":false,"password":true,"required":true,"multiline":false,"placeholder":"","display_name":"unstructured_api_url","proxy":{"id":"ElemUnstructuredLoaderV0-LAGXM","field":"unstructured_api_url"}},"chunk_size_RecursiveCharacterTextSplitter-a7fc9":{"info":"","list":false,"name":"chunk_size","show":true,"type":"int","value":1000,"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Chunk Size","proxy":{"id":"RecursiveCharacterTextSplitter-a7fc9","field":"chunk_size"}},"separators_RecursiveCharacterTextSplitter-a7fc9":{"info":"","list":false,"name":"separators","show":true,"type":"str","value":"\\n\\n","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Separator","proxy":{"id":"RecursiveCharacterTextSplitter-a7fc9","field":"separators"}},"chunk_overlap_RecursiveCharacterTextSplitter-a7fc9":{"info":"","list":false,"name":"chunk_overlap","show":true,"type":"int","value":200,"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Chunk Overlap","proxy":{"id":"RecursiveCharacterTextSplitter-a7fc9","field":"chunk_overlap"}},"separator_type_RecursiveCharacterTextSplitter-a7fc9":{"info":"","list":true,"name":"separator_type","show":true,"type":"str","value":"Text","options":["Text","cobol","cpp","csharp","go","html","java","js","kotlin","latex","markdown","php","proto","python","rst","ruby","rust","scala","sol","swift","ts"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Separator Type","proxy":{"id":"RecursiveCharacterTextSplitter-a7fc9","field":"separator_type"}}},"flow":{"data":{"nodes":[{"width":384,"height":376,"id":"CombineDocsChain-Pud2p","type":"genericNode","position":{"x":1222.7195691813163,"y":-12.131560981144517},"data":{"id":"CombineDocsChain-Pud2p","node":{"template":{"llm":{"l2":false,"info":"","list":false,"name":"llm","show":true,"type":"BaseLanguageModel","l2_name":"llm","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"LLM"},"_type":"load_qa_chain","token_max":{"l2":false,"info":"当前只对stuff 生效","list":false,"name":"token_max","show":true,"type":"int","value":-1,"l2_name":"token_max","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"token_max"},"chain_type":{"l2":false,"info":"","list":true,"name":"chain_type","show":true,"type":"str","value":"stuff","l2_name":"chain_type","options":["stuff","map_reduce","map_rerank","refine"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""}},"description":"Load question answering chain.","base_classes":["BaseCombineDocumentsChain","function"],"display_name":"CombineDocsChain","output_types":[],"custom_fields":{},"documentation":""},"type":"CombineDocsChain","value":null},"selected":true,"positionAbsolute":{"x":1222.7195691813163,"y":-12.131560981144517}},{"width":384,"height":316,"id":"RetrievalQA-qH6Mk","type":"genericNode","position":{"x":1898.9908793385102,"y":150.74108594665665},"data":{"id":"RetrievalQA-qH6Mk","node":{"template":{"tags":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"RetrievalQA","memory":{"l2":false,"info":"","list":false,"name":"memory","show":true,"type":"BaseMemory","l2_name":"memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"verbose":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"metadata":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_key":{"l2":false,"info":"","list":false,"name":"input_key","show":true,"type":"str","value":"query","l2_name":"input_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":""},"retriever":{"l2":false,"info":"","list":false,"name":"retriever","show":true,"type":"BaseRetriever","l2_name":"retriever","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"input_node":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question"},"output_key":{"l2":false,"info":"","list":false,"name":"output_key","show":true,"type":"str","value":"result","l2_name":"output_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":""},"combine_documents_chain":{"l2":false,"info":"","list":false,"name":"combine_documents_chain","show":true,"type":"BaseCombineDocumentsChain","l2_name":"combine_documents_chain","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"return_source_documents":{"l2":false,"info":"","list":false,"name":"return_source_documents","show":true,"type":"bool","value":true,"l2_name":"return_source_documents","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Chain for question-answering against an index.","base_classes":["BaseRetrievalQA","RetrievalQA","Chain","function"],"display_name":"RetrievalQA","output_types":[],"custom_fields":{},"documentation":"https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"},"type":"RetrievalQA","value":null},"selected":true,"positionAbsolute":{"x":1898.9908793385102,"y":150.74108594665665}},{"width":384,"height":172,"id":"OpenAIProxyEmbedding-yvld7","type":"genericNode","position":{"x":585.3522336200281,"y":1118.8504631949986},"data":{"id":"OpenAIProxyEmbedding-yvld7","node":{"template":{"_type":"proxy_embedding","proxy_url":{"l2":false,"info":"","list":false,"name":"proxy_url","show":false,"type":"str","value":"","l2_name":"proxy_url","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"http://43.133.35.137:8080"}},"description":" 使用自建的embedding服务使用openai进行embed ","base_classes":["Embeddings"],"display_name":"OpenAIProxyEmbedding","output_types":[],"custom_fields":{},"documentation":""},"type":"OpenAIProxyEmbedding","value":null},"selected":true,"positionAbsolute":{"x":585.3522336200281,"y":1118.8504631949986}},{"width":384,"height":254,"id":"InputFileNode-jdLkB","type":"genericNode","position":{"x":-67.64298275986698,"y":405.6605420525051},"data":{"id":"InputFileNode-jdLkB","node":{"template":{"_type":"InputFileNode","file_path":{"l2":false,"info":"","list":false,"name":"file_path","show":true,"type":"file","value":"","l2_name":"file_path","advanced":false,"password":false,"required":false,"suffixes":[".html",".md",".txt",".jpg",".png",".jpeg",".csv",".doc",".docx",".pdf",".ppt",".pptx",".xlsx",".tiff"],"fileTypes":["html","md","txt","jpg","png","jpeg","csv","doc","docx","pdf","ppt","pptx","tiff","xlsx"],"file_path":null,"multiline":false,"placeholder":"","display_name":"输入内容"}},"description":"输入节点，用来自动对接输入","base_classes":["fileNode"],"display_name":"InputFileNode","output_types":[],"custom_fields":{},"documentation":""},"type":"InputFileNode","value":null},"selected":true,"positionAbsolute":{"x":-67.64298275986698,"y":405.6605420525051}},{"width":384,"height":500,"id":"ProxyChatLLM-rnE4s","type":"genericNode","position":{"x":598.1704660724836,"y":-139.8508856646713},"data":{"id":"ProxyChatLLM-rnE4s","node":{"template":{"n":{"l2":false,"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"l2_name":"n","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tags":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"ProxyChatLLM","cache":{"l2":false,"info":"","list":false,"name":"cache","show":false,"type":"bool","l2_name":"cache","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"top_p":{"l2":false,"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"l2_name":"top_p","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"client":{"l2":false,"info":"","list":false,"name":"client","show":false,"type":"Any","l2_name":"client","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"headers":{"l2":false,"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","l2_name":"headers","advanced":false,"password":false,"required":false,"multiline":true,"placeholder":""},"verbose":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"metadata":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"streaming":{"l2":false,"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"l2_name":"streaming","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_tokens":{"l2":false,"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","l2_name":"max_tokens","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":""},"model_name":{"l2":false,"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-3.5-turbo-16k-0613","l2_name":"model_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_retries":{"l2":false,"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"l2_name":"max_retries","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"temperature":{"l2":false,"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"l2_name":"temperature","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"model_kwargs":{"l2":false,"info":"","list":false,"name":"model_kwargs","show":true,"type":"code","l2_name":"model_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"elemai_api_key":{"l2":false,"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","l2_name":"elemai_api_key","advanced":false,"password":true,"required":false,"multiline":false,"placeholder":""},"elemai_base_url":{"l2":false,"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","l2_name":"elemai_base_url","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"request_timeout":{"l2":false,"info":"","list":false,"name":"request_timeout","show":false,"type":"float","l2_name":"request_timeout","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tiktoken_model_name":{"l2":false,"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","l2_name":"tiktoken_model_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Wrapper around proxy Chat large language models.","base_classes":["BaseLanguageModel","ProxyChatLLM","BaseChatModel","BaseLLM"],"display_name":"ProxyChatLLM","output_types":[],"custom_fields":{},"documentation":""},"type":"ProxyChatLLM","value":null},"selected":true,"positionAbsolute":{"x":598.1704660724836,"y":-139.8508856646713}},{"width":384,"height":346,"id":"Milvus-T3kRH","type":"genericNode","position":{"x":1356.1418476973313,"y":809.6096821830637},"data":{"id":"Milvus-T3kRH","node":{"template":{"_type":"Milvus","drop_old":{"l2":false,"info":"","list":false,"name":"drop_old","show":false,"type":"bool","value":false,"l2_name":"drop_old","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"documents":{"l2":false,"info":"","list":true,"name":"documents","show":true,"type":"Document","l2_name":"documents","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Documents"},"embedding":{"l2":false,"info":"","list":false,"name":"embedding","show":true,"type":"Embeddings","l2_name":"embedding","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Embedding"},"metadatas":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"index_params":{"l2":false,"info":"","list":false,"name":"index_params","show":false,"type":"code","l2_name":"index_params","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"search_kwargs":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":true,"type":"code","value":"{}","l2_name":"search_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"search_params":{"l2":false,"info":"","list":false,"name":"search_params","show":false,"type":"code","l2_name":"search_params","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"collection_name":{"l2":false,"info":"","list":false,"name":"collection_name","show":true,"type":"str","value":"","l2_name":"collection_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","collection_id":""},"connection_args":{"l2":false,"info":"","list":false,"name":"connection_args","show":true,"type":"code","value":"","l2_name":"connection_args","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"consistency_level":{"l2":false,"info":"","list":false,"name":"consistency_level","show":false,"type":"str","value":"Session","l2_name":"consistency_level","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Create a Milvus collection, indexes it with HNSW, and insert data.","base_classes":["VectorStore","Milvus","BaseRetriever","VectorStoreRetriever"],"display_name":"Milvus","output_types":[],"custom_fields":{},"documentation":"http://192.168.106.116:8030"},"type":"Milvus","value":null},"selected":true,"positionAbsolute":{"x":1356.1418476973313,"y":809.6096821830637}},{"width":384,"height":392,"id":"ElemUnstructuredLoaderV0-LAGXM","type":"genericNode","position":{"x":370.3915867302851,"y":393.863205643299},"data":{"id":"ElemUnstructuredLoaderV0-LAGXM","node":{"template":{"_type":"ElemUnstructuredLoaderV0","metadata":{"info":"","list":false,"name":"metadata","show":true,"type":"code","value":"{}","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Metadata"},"file_path":{"info":"","list":false,"name":"file_path","show":true,"type":"fileNode","value":"","advanced":false,"password":false,"required":false,"suffixes":[".html",".md",".txt",".jpg",".png",".jpeg",".csv",".doc",".docx",".pdf",".ppt",".pptx",".xlsx",".tiff"],"fileTypes":["html","md","txt","jpg","png","jpeg","csv","doc","docx","pdf","ppt","pptx","tiff","xlsx"],"multiline":false,"placeholder":""},"unstructured_api_url":{"info":"","list":false,"name":"unstructured_api_url","show":true,"type":"str","value":"","advanced":false,"password":true,"required":true,"multiline":false,"placeholder":"","display_name":"unstructured_api_url"}},"description":"Loads a PDF with pypdf and chunks at character level. dummy version","base_classes":["Document"],"display_name":"ElemUnstructuredLoaderV0","output_types":["Document"],"custom_fields":{},"documentation":""},"type":"ElemUnstructuredLoaderV0","value":null},"selected":true,"positionAbsolute":{"x":370.3915867302851,"y":393.863205643299}},{"width":384,"height":540,"id":"RecursiveCharacterTextSplitter-a7fc9","type":"genericNode","position":{"x":840.0634079971517,"y":439.4498034537745},"data":{"id":"RecursiveCharacterTextSplitter-a7fc9","node":{"template":{"_type":"RecursiveCharacterTextSplitter","documents":{"info":"","list":false,"name":"documents","show":true,"type":"Document","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"chunk_size":{"info":"","list":false,"name":"chunk_size","show":true,"type":"int","value":1000,"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Chunk Size"},"separators":{"info":"","list":false,"name":"separators","show":true,"type":"str","value":"\\n\\n","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Separator"},"chunk_overlap":{"info":"","list":false,"name":"chunk_overlap","show":true,"type":"int","value":200,"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Chunk Overlap"},"separator_type":{"info":"","list":true,"name":"separator_type","show":true,"type":"str","value":"Text","options":["Text","cobol","cpp","csharp","go","html","java","js","kotlin","latex","markdown","php","proto","python","rst","ruby","rust","scala","sol","swift","ts"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Separator Type"}},"description":"Splitting text by recursively look at characters.","base_classes":["Document"],"display_name":"RecursiveCharacterTextSplitter","output_types":["Document"],"custom_fields":{},"documentation":"https://python.langchain.com/docs/modules/data_connection/document_transformers/text_splitters/recursive_text_splitter"},"type":"RecursiveCharacterTextSplitter","value":null},"selected":true,"positionAbsolute":{"x":840.0634079971517,"y":439.4498034537745}}],"edges":[{"source":"CombineDocsChain-Pud2p","target":"RetrievalQA-qH6Mk","sourceHandle":"CombineDocsChain|CombineDocsChain-Pud2p|BaseCombineDocumentsChain|function","targetHandle":"BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-qH6Mk","id":"reactflow__edge-CombineDocsChain-Pud2pCombineDocsChain|CombineDocsChain-Pud2p|BaseCombineDocumentsChain|function-RetrievalQA-qH6MkBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-qH6Mk","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"ProxyChatLLM-rnE4s","target":"CombineDocsChain-Pud2p","sourceHandle":"ProxyChatLLM|ProxyChatLLM-rnE4s|BaseLanguageModel|ProxyChatLLM|BaseChatModel|BaseLLM","targetHandle":"BaseLanguageModel|llm|CombineDocsChain-Pud2p","id":"reactflow__edge-ProxyChatLLM-rnE4sProxyChatLLM|ProxyChatLLM-rnE4s|BaseLanguageModel|ProxyChatLLM|BaseChatModel|BaseLLM-CombineDocsChain-Pud2pBaseLanguageModel|llm|CombineDocsChain-Pud2p","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"Milvus-T3kRH","target":"RetrievalQA-qH6Mk","sourceHandle":"Milvus|Milvus-T3kRH|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever","targetHandle":"BaseRetriever|retriever|RetrievalQA-qH6Mk","id":"reactflow__edge-Milvus-T3kRHMilvus|Milvus-T3kRH|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever-RetrievalQA-qH6MkBaseRetriever|retriever|RetrievalQA-qH6Mk","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"OpenAIProxyEmbedding-yvld7","target":"Milvus-T3kRH","sourceHandle":"OpenAIProxyEmbedding|OpenAIProxyEmbedding-yvld7|Embeddings","targetHandle":"Embeddings|embedding|Milvus-T3kRH","id":"reactflow__edge-OpenAIProxyEmbedding-yvld7OpenAIProxyEmbedding|OpenAIProxyEmbedding-yvld7|Embeddings-Milvus-T3kRHEmbeddings|embedding|Milvus-T3kRH","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"InputFileNode-jdLkB","target":"ElemUnstructuredLoaderV0-LAGXM","sourceHandle":"InputFileNode|InputFileNode-jdLkB|fileNode","targetHandle":"fileNode|file_path|ElemUnstructuredLoaderV0-LAGXM","id":"reactflow__edge-InputFileNode-jdLkBInputFileNode|InputFileNode-jdLkB|fileNode-ElemUnstructuredLoaderV0-LAGXMfileNode|file_path|ElemUnstructuredLoaderV0-LAGXM","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"ElemUnstructuredLoaderV0-LAGXM","target":"RecursiveCharacterTextSplitter-a7fc9","sourceHandle":"ElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-LAGXM|Document","targetHandle":"Document|documents|RecursiveCharacterTextSplitter-a7fc9","id":"reactflow__edge-ElemUnstructuredLoaderV0-LAGXMElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-LAGXM|Document-RecursiveCharacterTextSplitter-a7fc9Document|documents|RecursiveCharacterTextSplitter-a7fc9","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"RecursiveCharacterTextSplitter-a7fc9","target":"Milvus-T3kRH","sourceHandle":"RecursiveCharacterTextSplitter|RecursiveCharacterTextSplitter-a7fc9|Document","targetHandle":"Document|documents|Milvus-T3kRH","id":"reactflow__edge-RecursiveCharacterTextSplitter-a7fc9RecursiveCharacterTextSplitter|RecursiveCharacterTextSplitter-a7fc9|Document-Milvus-T3kRHDocument|documents|Milvus-T3kRH","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true}],"viewport":{"zoom":1,"x":0,"y":0}},"is_component":false,"name":"","description":"","id":"2bec4","status":0,"write":false,"guide_word":""}}}},{"知识库问答_严谨版":{"id":"RetrievalQA-011e4","type":"RetrievalQA","node":{"output_types":[],"display_name":"知识库问答_严谨版","documentation":"","base_classes":["RetrievalQA","BaseRetrievalQA","Chain","function"],"description":"Chain for question-answering against an index.","template":{"n_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"l2_name":"n","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"n"},"display_name":"N"},"tags_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"tags"},"display_name":"Tags"},"cache_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"cache","show":false,"type":"bool","l2_name":"cache","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"cache"},"display_name":"Cache"},"top_p_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"l2_name":"top_p","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"top_p"},"display_name":"Top P"},"client_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"client","show":false,"type":"Any","l2_name":"client","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"client"},"display_name":"Client"},"headers_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","l2_name":"headers","advanced":true,"password":false,"required":false,"multiline":true,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"headers"},"display_name":"Headers"},"verbose_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"verbose"},"display_name":"Verbose"},"metadata_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"metadata"},"display_name":"Metadata"},"callbacks_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"callbacks"},"display_name":"Callbacks"},"streaming_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"l2_name":"streaming","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"streaming"},"display_name":"Streaming"},"max_tokens_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","l2_name":"max_tokens","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"max_tokens"},"display_name":"Max Tokens"},"model_name_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-3.5-turbo-16k-0613","l2_name":"model_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"model_name"},"display_name":"Model Name"},"max_retries_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"l2_name":"max_retries","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"max_retries"},"display_name":"Max Retries"},"temperature_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"l2_name":"temperature","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"temperature"},"display_name":"Temperature"},"model_kwargs_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"model_kwargs","show":false,"type":"code","l2_name":"model_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"model_kwargs"},"display_name":"Model Kwargs"},"elemai_api_key_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","l2_name":"elemai_api_key","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"elemai_api_key"},"display_name":"Elemai Api Key"},"elemai_base_url_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","l2_name":"elemai_base_url","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"elemai_base_url"},"display_name":"Elemai Base Url"},"request_timeout_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"request_timeout","show":false,"type":"float","l2_name":"request_timeout","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"request_timeout"},"display_name":"Request Timeout"},"tiktoken_model_name_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","l2_name":"tiktoken_model_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"tiktoken_model_name"},"display_name":"Tiktoken Model Name"},"tags_RetrievalQA-ypS4m":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-ypS4m","field":"tags"},"display_name":"Tags"},"memory_RetrievalQA-ypS4m":{"l2":false,"info":"","list":false,"name":"memory","show":true,"type":"BaseMemory","l2_name":"memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-ypS4m","field":"memory"},"display_name":"Memory"},"verbose_RetrievalQA-ypS4m":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-ypS4m","field":"verbose"},"display_name":"Verbose"},"metadata_RetrievalQA-ypS4m":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-ypS4m","field":"metadata"},"display_name":"Metadata"},"callbacks_RetrievalQA-ypS4m":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-ypS4m","field":"callbacks"},"display_name":"Callbacks"},"input_key_RetrievalQA-ypS4m":{"l2":false,"info":"","list":false,"name":"input_key","show":true,"type":"str","value":"query","l2_name":"input_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-ypS4m","field":"input_key"},"display_name":"Input Key"},"input_node_RetrievalQA-ypS4m":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question","proxy":{"id":"RetrievalQA-ypS4m","field":"input_node"}},"output_key_RetrievalQA-ypS4m":{"l2":false,"info":"","list":false,"name":"output_key","show":true,"type":"str","value":"result","l2_name":"output_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-ypS4m","field":"output_key"},"display_name":"Output Key"},"return_source_documents_RetrievalQA-ypS4m":{"l2":false,"info":"","list":false,"name":"return_source_documents","show":true,"type":"bool","value":true,"l2_name":"return_source_documents","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-ypS4m","field":"return_source_documents"},"display_name":"Return Source Documents"},"token_max_CombineDocsChain-bMaXk":{"l2":false,"info":"只对Stuff类型生效","list":false,"name":"token_max","show":true,"type":"int","value":7000,"l2_name":"token_max","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"token_max","proxy":{"id":"CombineDocsChain-bMaXk","field":"token_max"}},"chain_type_CombineDocsChain-bMaXk":{"l2":false,"info":"","list":true,"name":"chain_type","show":true,"type":"str","value":"stuff","l2_name":"chain_type","options":["stuff","map_reduce","map_rerank","refine"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"CombineDocsChain-bMaXk","field":"chain_type"},"display_name":"Chain Type"},"document_prompt_CombineDocsChain-bMaXk":{"info":"","list":false,"name":"document_prompt","type":"BasePromptTemplate","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"CombineDocsChain-bMaXk","field":"document_prompt"},"display_name":"Document Prompt"},"context_PromptTemplate-gM5zg":{"info":"","list":false,"name":"context","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"context","proxy":{"id":"PromptTemplate-gM5zg","field":"context"}},"question_PromptTemplate-gM5zg":{"info":"","list":false,"name":"question","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"question","proxy":{"id":"PromptTemplate-gM5zg","field":"question"}},"template_PromptTemplate-gM5zg":{"l2":false,"info":"","list":false,"name":"template","show":true,"type":"prompt","value":"参考文本：\n{context}\n\n----------------------\n# Role：客服\n\n# Background :\n- 你是一名客服人员。你的日常工作是根据【参考文本】的内容回复用户的各类咨询问题。\n\n# Goals:\n- 以上【参考文本】中是从企业知识库中查找到的相关信息，只能结合以上信息进行回答，若以上内容为空或其中没有找到能回答【用户问题】的内容时，则回复“没有找到相关内容”，不能根据你自己的知识自己发挥。\n- 特别要注意区分【用户问题】与【参考文本】中不同的日期、人名、公司名这些关键信息。\n- 用中文回答问题，并且答案要严谨、专业、清晰、可读性好。\n- 拥有排版审美, 会利用序号, 缩进, 分隔线和换行符等等来美化信息排版。\n\n----------------------\n\n用户问题: {question}\n你的回答:","l2_name":"template","advanced":false,"password":false,"required":true,"multiline":true,"placeholder":"","proxy":{"id":"PromptTemplate-gM5zg","field":"template"},"display_name":"Template"},"output_parser_PromptTemplate-gM5zg":{"l2":false,"info":"","list":false,"name":"output_parser","show":true,"type":"BaseOutputParser","l2_name":"output_parser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-gM5zg","field":"output_parser"},"display_name":"Output Parser"},"input_variables_PromptTemplate-gM5zg":{"l2":false,"info":"","list":true,"name":"input_variables","show":false,"type":"str","value":["context","question"],"l2_name":"input_variables","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-gM5zg","field":"input_variables"},"display_name":"Input Variables"},"template_format_PromptTemplate-gM5zg":{"l2":false,"info":"","list":false,"name":"template_format","show":false,"type":"str","value":"f-string","l2_name":"template_format","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-gM5zg","field":"template_format"},"display_name":"Template Format"},"partial_variables_PromptTemplate-gM5zg":{"l2":false,"info":"","list":false,"name":"partial_variables","show":false,"type":"code","l2_name":"partial_variables","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-gM5zg","field":"partial_variables"},"display_name":"Partial Variables"},"validate_template_PromptTemplate-gM5zg":{"l2":false,"info":"","list":false,"name":"validate_template","show":false,"type":"bool","value":true,"l2_name":"validate_template","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-gM5zg","field":"validate_template"},"display_name":"Validate Template"},"ids_ElasticKeywordsSearch-dVhq9":{"l2":false,"info":"","list":true,"name":"ids","show":false,"type":"str","l2_name":"ids","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ElasticKeywordsSearch-dVhq9","field":"ids"},"display_name":"Ids"},"documents_ElasticKeywordsSearch-dVhq9":{"l2":false,"info":"","list":true,"name":"documents","show":true,"type":"Document","l2_name":"documents","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Documents","proxy":{"id":"ElasticKeywordsSearch-dVhq9","field":"documents"}},"embedding_ElasticKeywordsSearch-dVhq9":{"l2":false,"info":"","list":false,"name":"embedding","show":false,"type":"Embeddings","l2_name":"embedding","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Embedding","proxy":{"id":"ElasticKeywordsSearch-dVhq9","field":"embedding"}},"metadatas_ElasticKeywordsSearch-dVhq9":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ElasticKeywordsSearch-dVhq9","field":"metadatas"},"display_name":"Metadatas"},"index_name_ElasticKeywordsSearch-dVhq9":{"l2":true,"info":"","list":false,"name":"index_name","show":true,"type":"str","value":"col_1701960496_3a537dc9","l2_name":"知识库","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ElasticKeywordsSearch-dVhq9","field":"index_name"},"display_name":"Index Name"},"ssl_verify_ElasticKeywordsSearch-dVhq9":{"l2":false,"info":"","list":false,"name":"ssl_verify","show":true,"type":"str","value":"","l2_name":"ssl_verify","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"ssl_verify","proxy":{"id":"ElasticKeywordsSearch-dVhq9","field":"ssl_verify"}},"search_kwargs_ElasticKeywordsSearch-dVhq9":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":false,"type":"code","value":"{\"k\":6,\"query_strategy\":\"match_phrase\",\"must_or_should\":\"must\"}","l2_name":"search_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ElasticKeywordsSearch-dVhq9","field":"search_kwargs"},"display_name":"Search Kwargs"},"refresh_indices_ElasticKeywordsSearch-dVhq9":{"l2":false,"info":"","list":false,"name":"refresh_indices","show":false,"type":"bool","value":true,"l2_name":"refresh_indices","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ElasticKeywordsSearch-dVhq9","field":"refresh_indices"},"display_name":"Refresh Indices"},"elasticsearch_url_ElasticKeywordsSearch-dVhq9":{"l2":false,"info":"","list":false,"name":"elasticsearch_url","show":true,"type":"str","value":"","l2_name":"elasticsearch_url","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"ES_connection_url","proxy":{"id":"ElasticKeywordsSearch-dVhq9","field":"elasticsearch_url"}},"question_PromptTemplate-4iBFj":{"info":"","list":false,"name":"question","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"question","proxy":{"id":"PromptTemplate-4iBFj","field":"question"}},"template_PromptTemplate-4iBFj":{"l2":false,"info":"","list":false,"name":"template","show":true,"type":"prompt","value":"分析给定Question，提取Question中包含的KeyWords，输出列表形式\n\nExamples:\nQuestion: 数据项素2022年营收是多少？\nKeyWords: ['数据项素', '2022', '营收, '营业收入']\nQuestion: 深圳出差住宿标准是什么？\nKeyWords: ['深圳', '出差', '住宿标准', '报销标准']\n\n\n----------------\nQuestion: {question}\nKeyWords: ","l2_name":"template","advanced":false,"password":false,"required":true,"multiline":true,"placeholder":"","proxy":{"id":"PromptTemplate-4iBFj","field":"template"},"display_name":"Template"},"output_parser_PromptTemplate-4iBFj":{"l2":false,"info":"","list":false,"name":"output_parser","show":true,"type":"BaseOutputParser","l2_name":"output_parser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-4iBFj","field":"output_parser"},"display_name":"Output Parser"},"input_variables_PromptTemplate-4iBFj":{"l2":false,"info":"","list":true,"name":"input_variables","show":false,"type":"str","value":["question"],"l2_name":"input_variables","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-4iBFj","field":"input_variables"},"display_name":"Input Variables"},"template_format_PromptTemplate-4iBFj":{"l2":false,"info":"","list":false,"name":"template_format","show":false,"type":"str","value":"f-string","l2_name":"template_format","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-4iBFj","field":"template_format"},"display_name":"Template Format"},"partial_variables_PromptTemplate-4iBFj":{"l2":false,"info":"","list":false,"name":"partial_variables","show":false,"type":"code","l2_name":"partial_variables","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-4iBFj","field":"partial_variables"},"display_name":"Partial Variables"},"validate_template_PromptTemplate-4iBFj":{"l2":false,"info":"","list":false,"name":"validate_template","show":false,"type":"bool","value":true,"l2_name":"validate_template","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-4iBFj","field":"validate_template"},"display_name":"Validate Template"}},"flow":{"data":{"nodes":[{"width":384,"height":500,"id":"ProxyChatLLM-UYoB4","type":"genericNode","position":{"x":-2220.2561641720736,"y":-777.5185403158035},"data":{"id":"ProxyChatLLM-UYoB4","node":{"template":{"n":{"l2":false,"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"l2_name":"n","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tags":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"ProxyChatLLM","cache":{"l2":false,"info":"","list":false,"name":"cache","show":false,"type":"bool","l2_name":"cache","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"top_p":{"l2":false,"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"l2_name":"top_p","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"client":{"l2":false,"info":"","list":false,"name":"client","show":false,"type":"Any","l2_name":"client","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"headers":{"l2":false,"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","l2_name":"headers","advanced":false,"password":false,"required":false,"multiline":true,"placeholder":""},"verbose":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"metadata":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"streaming":{"l2":false,"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"l2_name":"streaming","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_tokens":{"l2":false,"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","l2_name":"max_tokens","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":""},"model_name":{"l2":false,"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-3.5-turbo-16k-0613","l2_name":"model_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_retries":{"l2":false,"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"l2_name":"max_retries","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"temperature":{"l2":false,"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"l2_name":"temperature","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"model_kwargs":{"l2":false,"info":"","list":false,"name":"model_kwargs","show":true,"type":"code","l2_name":"model_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"elemai_api_key":{"l2":false,"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","l2_name":"elemai_api_key","advanced":false,"password":true,"required":false,"multiline":false,"placeholder":""},"elemai_base_url":{"l2":false,"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","l2_name":"elemai_base_url","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"request_timeout":{"l2":false,"info":"","list":false,"name":"request_timeout","show":false,"type":"float","l2_name":"request_timeout","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tiktoken_model_name":{"l2":false,"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","l2_name":"tiktoken_model_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Wrapper around proxy Chat large language models.","base_classes":["BaseChatModel","ProxyChatLLM","BaseLanguageModel","BaseLLM"],"display_name":"ProxyChatLLM","output_types":[],"custom_fields":{},"documentation":""},"type":"ProxyChatLLM","value":null},"selected":true,"positionAbsolute":{"x":-2220.2561641720736,"y":-777.5185403158035}},{"width":384,"height":316,"id":"RetrievalQA-ypS4m","type":"genericNode","position":{"x":-523.0150819181391,"y":97.4334598716227},"data":{"id":"RetrievalQA-ypS4m","node":{"template":{"tags":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"RetrievalQA","memory":{"l2":false,"info":"","list":false,"name":"memory","show":true,"type":"BaseMemory","l2_name":"memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"verbose":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"metadata":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_key":{"l2":false,"info":"","list":false,"name":"input_key","show":true,"type":"str","value":"query","l2_name":"input_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":""},"retriever":{"l2":false,"info":"","list":false,"name":"retriever","show":true,"type":"BaseRetriever","l2_name":"retriever","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"input_node":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question"},"output_key":{"l2":false,"info":"","list":false,"name":"output_key","show":true,"type":"str","value":"result","l2_name":"output_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":""},"combine_documents_chain":{"l2":false,"info":"","list":false,"name":"combine_documents_chain","show":true,"type":"BaseCombineDocumentsChain","l2_name":"combine_documents_chain","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"return_source_documents":{"l2":false,"info":"","list":false,"name":"return_source_documents","show":true,"type":"bool","value":true,"l2_name":"return_source_documents","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Chain for question-answering against an index.","base_classes":["RetrievalQA","BaseRetrievalQA","Chain","function"],"display_name":"RetrievalQA","output_types":[],"custom_fields":{},"documentation":"https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"},"type":"RetrievalQA","value":null},"selected":true,"positionAbsolute":{"x":-523.0150819181391,"y":97.4334598716227}},{"width":384,"height":412,"id":"CombineDocsChain-bMaXk","type":"genericNode","position":{"x":-1130.4006392228448,"y":-400.50282859447026},"data":{"id":"CombineDocsChain-bMaXk","node":{"template":{"llm":{"l2":false,"info":"","list":false,"name":"llm","show":true,"type":"BaseLanguageModel","l2_name":"llm","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"LLM"},"_type":"load_qa_chain","prompt":{"l2":false,"info":"只对Stuff类型生效","list":false,"name":"prompt","show":true,"type":"BasePromptTemplate","l2_name":"prompt","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"prompt"},"token_max":{"l2":false,"info":"只对Stuff类型生效","list":false,"name":"token_max","show":true,"type":"int","value":7000,"l2_name":"token_max","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"token_max"},"chain_type":{"l2":false,"info":"","list":true,"name":"chain_type","show":true,"type":"str","value":"stuff","l2_name":"chain_type","options":["stuff","map_reduce","map_rerank","refine"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"document_prompt":{"info":"","list":false,"name":"document_prompt","type":"BasePromptTemplate","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Load question answering chain.","base_classes":["BaseCombineDocumentsChain","function"],"display_name":"CombineDocsChain","output_types":[],"custom_fields":{},"documentation":""},"type":"CombineDocsChain","value":null},"selected":true,"positionAbsolute":{"x":-1130.4006392228448,"y":-400.50282859447026}},{"width":384,"height":454,"id":"PromptTemplate-gM5zg","type":"genericNode","position":{"x":-1632.2039543868736,"y":-814.266749101787},"data":{"id":"PromptTemplate-gM5zg","node":{"name":"","template":{"_type":"PromptTemplate","context":{"info":"","list":false,"name":"context","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"context"},"question":{"info":"","list":false,"name":"question","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"question"},"template":{"l2":false,"info":"","list":false,"name":"template","show":true,"type":"prompt","value":"参考文本：\n{context}\n\n----------------------\n# Role：客服\n\n# Background :\n- 你是一名客服人员。你的日常工作是根据【参考文本】的内容回复用户的各类咨询问题。\n\n# Goals:\n- 以上【参考文本】中是从企业知识库中查找到的相关信息，只能结合以上信息进行回答，若以上内容为空或其中没有找到能回答【用户问题】的内容时，则回复“没有找到相关内容”，不能根据你自己的知识自己发挥。\n- 特别要注意区分【用户问题】与【参考文本】中不同的日期、人名、公司名这些关键信息。\n- 用中文回答问题，并且答案要严谨、专业、清晰、可读性好。\n- 拥有排版审美, 会利用序号, 缩进, 分隔线和换行符等等来美化信息排版。\n\n----------------------\n\n用户问题: {question}\n你的回答:","l2_name":"template","advanced":false,"password":false,"required":true,"multiline":true,"placeholder":""},"output_parser":{"l2":false,"info":"","list":false,"name":"output_parser","show":true,"type":"BaseOutputParser","l2_name":"output_parser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_variables":{"l2":false,"info":"","list":true,"name":"input_variables","show":false,"type":"str","value":["context","question"],"l2_name":"input_variables","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"template_format":{"l2":false,"info":"","list":false,"name":"template_format","show":false,"type":"str","value":"f-string","l2_name":"template_format","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"partial_variables":{"l2":false,"info":"","list":false,"name":"partial_variables","show":false,"type":"code","l2_name":"partial_variables","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"validate_template":{"l2":false,"info":"","list":false,"name":"validate_template","show":false,"type":"bool","value":true,"l2_name":"validate_template","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Schema to represent a prompt for an LLM.","base_classes":["StringPromptTemplate","BasePromptTemplate","PromptTemplate"],"display_name":"PromptTemplate","output_types":[],"custom_fields":{"":["context","question"],"template":["context","question"]},"documentation":"https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/","field_formatters":{"formatters":{"openai_api_key":{}},"base_formatters":{"dict":{},"list":{},"show":{},"union":{},"kwargs":{},"default":{},"headers":{},"optional":{},"password":{},"multiline":{},"model_fields":{"MODEL_DICT":{"OpenAI":["text-davinci-003","text-davinci-002","text-curie-001","text-babbage-001","text-ada-001"],"Anthropic":["claude-v1","claude-v1-100k","claude-instant-v1","claude-instant-v1-100k","claude-v1.3","claude-v1.3-100k","claude-v1.2","claude-v1.0","claude-instant-v1.1","claude-instant-v1.1-100k","claude-instant-v1.0"],"ChatOpenAI":["gpt-3.5-turbo-0613","gpt-3.5-turbo","gpt-3.5-turbo-16k-0613","gpt-3.5-turbo-16k","gpt-4-0613","gpt-4-32k-0613","gpt-4","gpt-4-32k"],"ChatAnthropic":["claude-v1","claude-v1-100k","claude-instant-v1","claude-instant-v1-100k","claude-v1.3","claude-v1.3-100k","claude-v1.2","claude-v1.0","claude-instant-v1.1","claude-instant-v1.1-100k","claude-instant-v1.0"]}},"dict_code_file":{}}}},"type":"PromptTemplate","value":null},"selected":true,"positionAbsolute":{"x":-1632.2039543868736,"y":-814.266749101787}},{"width":384,"height":628,"id":"ElasticKeywordsSearch-dVhq9","type":"genericNode","position":{"x":-1620.6487761012545,"y":-186.12742019985436},"data":{"id":"ElasticKeywordsSearch-dVhq9","node":{"template":{"ids":{"l2":false,"info":"","list":true,"name":"ids","show":false,"type":"str","l2_name":"ids","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"llm":{"l2":false,"info":"","list":false,"name":"llm","show":true,"type":"BaseLLM","l2_name":"llm","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"LLM"},"_type":"ElasticKeywordsSearch","prompt":{"l2":false,"info":"","list":false,"name":"prompt","show":true,"type":"BasePromptTemplate","l2_name":"prompt","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"prompt"},"documents":{"l2":false,"info":"","list":true,"name":"documents","show":true,"type":"Document","l2_name":"documents","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Documents"},"embedding":{"l2":false,"info":"","list":false,"name":"embedding","show":false,"type":"Embeddings","l2_name":"embedding","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Embedding"},"metadatas":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"index_name":{"l2":true,"info":"","list":false,"name":"index_name","show":true,"type":"str","value":"col_1701960496_3a537dc9","l2_name":"知识库","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"ssl_verify":{"l2":false,"info":"","list":false,"name":"ssl_verify","show":true,"type":"str","value":"","l2_name":"ssl_verify","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"ssl_verify"},"search_kwargs":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":true,"type":"code","value":"{\"k\":6,\"query_strategy\":\"match_phrase\",\"must_or_should\":\"must\"}","l2_name":"search_kwargs","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"refresh_indices":{"l2":false,"info":"","list":false,"name":"refresh_indices","show":false,"type":"bool","value":true,"l2_name":"refresh_indices","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"elasticsearch_url":{"l2":false,"info":"","list":false,"name":"elasticsearch_url","show":true,"type":"str","value":"","l2_name":"elasticsearch_url","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"ES_connection_url"}},"description":"Construct ElasticKeywordsSearch wrapper from raw documents.","base_classes":["ElasticKeywordsSearch","VectorStore","BaseRetriever","VectorStoreRetriever"],"display_name":"ElasticKeywordsSearch","output_types":[],"custom_fields":{},"documentation":"http://192.168.106.116:8030"},"type":"ElasticKeywordsSearch","value":null},"selected":true,"positionAbsolute":{"x":-1620.6487761012545,"y":-186.12742019985436}},{"width":384,"height":372,"id":"PromptTemplate-4iBFj","type":"genericNode","position":{"x":-2302.5673221463094,"y":-146.52073590180217},"data":{"id":"PromptTemplate-4iBFj","node":{"name":"","template":{"_type":"PromptTemplate","question":{"info":"","list":false,"name":"question","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"question"},"template":{"l2":false,"info":"","list":false,"name":"template","show":true,"type":"prompt","value":"分析给定Question，提取Question中包含的KeyWords，输出列表形式\n\nExamples:\nQuestion: 数据项素2022年营收是多少？\nKeyWords: ['数据项素', '2022', '营收, '营业收入']\nQuestion: 深圳出差住宿标准是什么？\nKeyWords: ['深圳', '出差', '住宿标准', '报销标准']\n\n\n----------------\nQuestion: {question}\nKeyWords: ","l2_name":"template","advanced":false,"password":false,"required":true,"multiline":true,"placeholder":""},"output_parser":{"l2":false,"info":"","list":false,"name":"output_parser","show":true,"type":"BaseOutputParser","l2_name":"output_parser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_variables":{"l2":false,"info":"","list":true,"name":"input_variables","show":false,"type":"str","value":["question"],"l2_name":"input_variables","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"template_format":{"l2":false,"info":"","list":false,"name":"template_format","show":false,"type":"str","value":"f-string","l2_name":"template_format","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"partial_variables":{"l2":false,"info":"","list":false,"name":"partial_variables","show":false,"type":"code","l2_name":"partial_variables","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"validate_template":{"l2":false,"info":"","list":false,"name":"validate_template","show":false,"type":"bool","value":true,"l2_name":"validate_template","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Schema to represent a prompt for an LLM.","base_classes":["StringPromptTemplate","BasePromptTemplate","PromptTemplate"],"display_name":"PromptTemplate","output_types":[],"custom_fields":{"":["question"],"template":["question"]},"documentation":"https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/","field_formatters":{"formatters":{"openai_api_key":{}},"base_formatters":{"dict":{},"list":{},"show":{},"union":{},"kwargs":{},"default":{},"headers":{},"optional":{},"password":{},"multiline":{},"model_fields":{"MODEL_DICT":{"OpenAI":["text-davinci-003","text-davinci-002","text-curie-001","text-babbage-001","text-ada-001"],"Anthropic":["claude-v1","claude-v1-100k","claude-instant-v1","claude-instant-v1-100k","claude-v1.3","claude-v1.3-100k","claude-v1.2","claude-v1.0","claude-instant-v1.1","claude-instant-v1.1-100k","claude-instant-v1.0"],"ChatOpenAI":["gpt-3.5-turbo-0613","gpt-3.5-turbo","gpt-3.5-turbo-16k-0613","gpt-3.5-turbo-16k","gpt-4-0613","gpt-4-32k-0613","gpt-4","gpt-4-32k"],"ChatAnthropic":["claude-v1","claude-v1-100k","claude-instant-v1","claude-instant-v1-100k","claude-v1.3","claude-v1.3-100k","claude-v1.2","claude-v1.0","claude-instant-v1.1","claude-instant-v1.1-100k","claude-instant-v1.0"]}},"dict_code_file":{}}}},"type":"PromptTemplate","value":null},"selected":true,"positionAbsolute":{"x":-2302.5673221463094,"y":-146.52073590180217}}],"edges":[{"source":"CombineDocsChain-bMaXk","target":"RetrievalQA-ypS4m","sourceHandle":"CombineDocsChain|CombineDocsChain-bMaXk|BaseCombineDocumentsChain|function","targetHandle":"BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-ypS4m","id":"reactflow__edge-CombineDocsChain-bMaXkCombineDocsChain|CombineDocsChain-bMaXk|BaseCombineDocumentsChain|function-RetrievalQA-ypS4mBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-ypS4m","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"PromptTemplate-gM5zg","target":"CombineDocsChain-bMaXk","sourceHandle":"PromptTemplate|PromptTemplate-gM5zg|StringPromptTemplate|BasePromptTemplate|PromptTemplate","targetHandle":"BasePromptTemplate|prompt|CombineDocsChain-bMaXk","id":"reactflow__edge-PromptTemplate-gM5zgPromptTemplate|PromptTemplate-gM5zg|StringPromptTemplate|BasePromptTemplate|PromptTemplate-CombineDocsChain-bMaXkBasePromptTemplate|prompt|CombineDocsChain-bMaXk","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"ElasticKeywordsSearch-dVhq9","target":"RetrievalQA-ypS4m","sourceHandle":"ElasticKeywordsSearch|ElasticKeywordsSearch-dVhq9|ElasticKeywordsSearch|VectorStore|BaseRetriever|VectorStoreRetriever","targetHandle":"BaseRetriever|retriever|RetrievalQA-ypS4m","id":"reactflow__edge-ElasticKeywordsSearch-dVhq9ElasticKeywordsSearch|ElasticKeywordsSearch-dVhq9|ElasticKeywordsSearch|VectorStore|BaseRetriever|VectorStoreRetriever-RetrievalQA-ypS4mBaseRetriever|retriever|RetrievalQA-ypS4m","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"PromptTemplate-4iBFj","target":"ElasticKeywordsSearch-dVhq9","sourceHandle":"PromptTemplate|PromptTemplate-4iBFj|StringPromptTemplate|BasePromptTemplate|PromptTemplate","targetHandle":"BasePromptTemplate|prompt|ElasticKeywordsSearch-dVhq9","id":"reactflow__edge-PromptTemplate-4iBFjPromptTemplate|PromptTemplate-4iBFj|StringPromptTemplate|BasePromptTemplate|PromptTemplate-ElasticKeywordsSearch-dVhq9BasePromptTemplate|prompt|ElasticKeywordsSearch-dVhq9","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"ProxyChatLLM-UYoB4","target":"CombineDocsChain-bMaXk","sourceHandle":"ProxyChatLLM|ProxyChatLLM-UYoB4|BaseChatModel|ProxyChatLLM|BaseLanguageModel|BaseLLM","targetHandle":"BaseLanguageModel|llm|CombineDocsChain-bMaXk","id":"reactflow__edge-ProxyChatLLM-UYoB4ProxyChatLLM|ProxyChatLLM-UYoB4|BaseChatModel|ProxyChatLLM|BaseLanguageModel|BaseLLM-CombineDocsChain-bMaXkBaseLanguageModel|llm|CombineDocsChain-bMaXk","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"ProxyChatLLM-UYoB4","target":"ElasticKeywordsSearch-dVhq9","sourceHandle":"ProxyChatLLM|ProxyChatLLM-UYoB4|BaseChatModel|ProxyChatLLM|BaseLanguageModel|BaseLLM","targetHandle":"BaseLLM|llm|ElasticKeywordsSearch-dVhq9","id":"reactflow__edge-ProxyChatLLM-UYoB4ProxyChatLLM|ProxyChatLLM-UYoB4|BaseChatModel|ProxyChatLLM|BaseLanguageModel|BaseLLM-ElasticKeywordsSearch-dVhq9BaseLLM|llm|ElasticKeywordsSearch-dVhq9","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true}],"viewport":{"zoom":1,"x":0,"y":0}},"is_component":false,"name":"","description":"","id":"e4a9c","status":0,"write":false,"guide_word":""}}}},{"知识库问答_增强版":{"id":"ConversationalRetrievalChain-f19da","type":"ConversationalRetrievalChain","node":{"output_types":[],"display_name":"知识库问答_增强版","documentation":"","base_classes":["BaseConversationalRetrievalChain","Chain","ConversationalRetrievalChain","function"],"description":"Convenience method to load chain from LLM and retriever.","template":{"n_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"l2_name":"n","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"n"},"display_name":"N"},"tags_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"tags"},"display_name":"Tags"},"cache_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"cache","show":false,"type":"bool","l2_name":"cache","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"cache"},"display_name":"Cache"},"top_p_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"l2_name":"top_p","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"top_p"},"display_name":"Top P"},"client_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"client","show":false,"type":"Any","l2_name":"client","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"client"},"display_name":"Client"},"headers_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","l2_name":"headers","advanced":true,"password":false,"required":false,"multiline":true,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"headers"},"display_name":"Headers"},"verbose_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"verbose"},"display_name":"Verbose"},"metadata_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"metadata"},"display_name":"Metadata"},"callbacks_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"callbacks"},"display_name":"Callbacks"},"streaming_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"l2_name":"streaming","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"streaming"},"display_name":"Streaming"},"max_tokens_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","l2_name":"max_tokens","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"max_tokens"},"display_name":"Max Tokens"},"model_name_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-3.5-turbo-16k-0613","l2_name":"model_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"model_name"},"display_name":"Model Name"},"max_retries_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"l2_name":"max_retries","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"max_retries"},"display_name":"Max Retries"},"temperature_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"l2_name":"temperature","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"temperature"},"display_name":"Temperature"},"model_kwargs_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"model_kwargs","show":false,"type":"code","l2_name":"model_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"model_kwargs"},"display_name":"Model Kwargs"},"elemai_api_key_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","l2_name":"elemai_api_key","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"elemai_api_key"},"display_name":"Elemai Api Key"},"elemai_base_url_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","l2_name":"elemai_base_url","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"elemai_base_url"},"display_name":"Elemai Base Url"},"request_timeout_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"request_timeout","show":false,"type":"float","l2_name":"request_timeout","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"request_timeout"},"display_name":"Request Timeout"},"tiktoken_model_name_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","l2_name":"tiktoken_model_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"tiktoken_model_name"},"display_name":"Tiktoken Model Name"},"tags_MixEsVectorRetriever-J35CZ":{"l2":false,"info":"","list":true,"name":"tags","show":true,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"MixEsVectorRetriever-J35CZ","field":"tags"},"display_name":"Tags"},"metadata_MixEsVectorRetriever-J35CZ":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"MixEsVectorRetriever-J35CZ","field":"metadata"},"display_name":"Metadata"},"combine_strategy_MixEsVectorRetriever-J35CZ":{"l2":false,"info":"","list":true,"name":"combine_strategy","show":true,"type":"str","value":"mix","l2_name":"combine_strategy","options":["keyword_front","vector_front","mix"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"MixEsVectorRetriever-J35CZ","field":"combine_strategy"},"display_name":"Combine Strategy"},"drop_old_Milvus-cyR5W":{"l2":false,"info":"","list":false,"name":"drop_old","show":false,"type":"bool","value":false,"l2_name":"drop_old","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-cyR5W","field":"drop_old"},"display_name":"Drop Old"},"documents_Milvus-cyR5W":{"l2":false,"info":"","list":true,"name":"documents","show":true,"type":"Document","l2_name":"documents","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Documents","proxy":{"id":"Milvus-cyR5W","field":"documents"}},"embedding_Milvus-cyR5W":{"l2":false,"info":"","list":false,"name":"embedding","show":false,"type":"Embeddings","l2_name":"embedding","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Embedding","proxy":{"id":"Milvus-cyR5W","field":"embedding"}},"metadatas_Milvus-cyR5W":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-cyR5W","field":"metadatas"},"display_name":"Metadatas"},"index_params_Milvus-cyR5W":{"l2":false,"info":"","list":false,"name":"index_params","show":false,"type":"code","l2_name":"index_params","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-cyR5W","field":"index_params"},"display_name":"Index Params"},"search_kwargs_Milvus-cyR5W":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":false,"type":"code","value":"{}","l2_name":"search_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-cyR5W","field":"search_kwargs"},"display_name":"Search Kwargs"},"search_params_Milvus-cyR5W":{"l2":false,"info":"","list":false,"name":"search_params","show":false,"type":"code","l2_name":"search_params","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-cyR5W","field":"search_params"},"display_name":"Search Params"},"collection_name_Milvus-cyR5W":{"l2":true,"info":"","list":false,"name":"collection_name","show":true,"type":"str","value":"col_1700798314_e3f656ab","l2_name":"collection_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-cyR5W","field":"collection_name"},"display_name":"Collection Name"},"connection_args_Milvus-cyR5W":{"l2":false,"info":"","list":false,"name":"connection_args","show":false,"type":"code","value":"","l2_name":"connection_args","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-cyR5W","field":"connection_args"},"display_name":"Connection Args"},"consistency_level_Milvus-cyR5W":{"l2":false,"info":"","list":false,"name":"consistency_level","show":false,"type":"str","value":"Session","l2_name":"consistency_level","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-cyR5W","field":"consistency_level"},"display_name":"Consistency Level"},"ids_ElasticKeywordsSearch-31Et9":{"l2":false,"info":"","list":true,"name":"ids","show":false,"type":"str","l2_name":"ids","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ElasticKeywordsSearch-31Et9","field":"ids"},"display_name":"Ids"},"llm_ElasticKeywordsSearch-31Et9":{"l2":false,"info":"","list":false,"name":"llm","type":"BaseLLM","l2_name":"llm","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"LLM","proxy":{"id":"ElasticKeywordsSearch-31Et9","field":"llm"}},"prompt_ElasticKeywordsSearch-31Et9":{"l2":false,"info":"","list":false,"name":"prompt","type":"BasePromptTemplate","l2_name":"prompt","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"prompt","proxy":{"id":"ElasticKeywordsSearch-31Et9","field":"prompt"}},"documents_ElasticKeywordsSearch-31Et9":{"l2":false,"info":"","list":true,"name":"documents","show":true,"type":"Document","l2_name":"documents","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Documents","proxy":{"id":"ElasticKeywordsSearch-31Et9","field":"documents"}},"embedding_ElasticKeywordsSearch-31Et9":{"l2":false,"info":"","list":false,"name":"embedding","show":false,"type":"Embeddings","l2_name":"embedding","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Embedding","proxy":{"id":"ElasticKeywordsSearch-31Et9","field":"embedding"}},"metadatas_ElasticKeywordsSearch-31Et9":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ElasticKeywordsSearch-31Et9","field":"metadatas"},"display_name":"Metadatas"},"index_name_ElasticKeywordsSearch-31Et9":{"l2":true,"info":"","list":false,"name":"index_name","show":true,"type":"str","value":"col_1700798314_e3f656ab","l2_name":"index_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ElasticKeywordsSearch-31Et9","field":"index_name"},"display_name":"Index Name"},"ssl_verify_ElasticKeywordsSearch-31Et9":{"l2":false,"info":"","list":false,"name":"ssl_verify","show":true,"type":"str","value":"","l2_name":"ssl_verify","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"ssl_verify","proxy":{"id":"ElasticKeywordsSearch-31Et9","field":"ssl_verify"}},"search_kwargs_ElasticKeywordsSearch-31Et9":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":false,"type":"code","value":"{}","l2_name":"search_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ElasticKeywordsSearch-31Et9","field":"search_kwargs"},"display_name":"Search Kwargs"},"refresh_indices_ElasticKeywordsSearch-31Et9":{"l2":false,"info":"","list":false,"name":"refresh_indices","show":false,"type":"bool","value":true,"l2_name":"refresh_indices","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ElasticKeywordsSearch-31Et9","field":"refresh_indices"},"display_name":"Refresh Indices"},"elasticsearch_url_ElasticKeywordsSearch-31Et9":{"l2":false,"info":"","list":false,"name":"elasticsearch_url","show":true,"type":"str","value":"","l2_name":"elasticsearch_url","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"ES_connection_url","proxy":{"id":"ElasticKeywordsSearch-31Et9","field":"elasticsearch_url"}},"verbose_ConversationalRetrievalChain-SmY3w":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationalRetrievalChain-SmY3w","field":"verbose"},"display_name":"Verbose"},"callbacks_ConversationalRetrievalChain-SmY3w":{"l2":false,"info":"","list":false,"name":"callbacks","show":false,"type":"Callbacks","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationalRetrievalChain-SmY3w","field":"callbacks"},"display_name":"Callbacks"},"chain_type_ConversationalRetrievalChain-SmY3w":{"l2":false,"info":"","list":true,"name":"chain_type","show":true,"type":"str","value":"stuff","l2_name":"chain_type","options":["stuff","map_reduce","map_rerank","refine"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"ConversationalRetrievalChain-SmY3w","field":"chain_type"},"display_name":"Chain Type"},"input_node_ConversationalRetrievalChain-SmY3w":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question","proxy":{"id":"ConversationalRetrievalChain-SmY3w","field":"input_node"}},"condense_question_llm_ConversationalRetrievalChain-SmY3w":{"l2":false,"info":"","list":false,"name":"condense_question_llm","show":false,"type":"BaseLanguageModel","l2_name":"condense_question_llm","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationalRetrievalChain-SmY3w","field":"condense_question_llm"},"display_name":"Condense Question LLM"},"return_source_documents_ConversationalRetrievalChain-SmY3w":{"l2":false,"info":"","list":false,"name":"return_source_documents","show":true,"type":"bool","value":true,"l2_name":"return_source_documents","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Return source documents","proxy":{"id":"ConversationalRetrievalChain-SmY3w","field":"return_source_documents"}},"condense_question_prompt_ConversationalRetrievalChain-SmY3w":{"l2":false,"info":"","list":false,"name":"condense_question_prompt","show":true,"type":"BasePromptTemplate","value":{"_type":"prompt","template":"Given the following conversation and a follow up question, rephrase the follow up question to be a standalone question, in its original language.\n\nChat History:\n{chat_history}\nFollow Up Input: {question}\nStandalone question:","output_parser":null,"input_variables":["chat_history","question"],"template_format":"f-string","partial_variables":{},"validate_template":true},"l2_name":"condense_question_prompt","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationalRetrievalChain-SmY3w","field":"condense_question_prompt"},"display_name":"Condense Question Prompt"},"ai_prefix_ConversationBufferMemory-K1tss":{"l2":false,"info":"","list":false,"name":"ai_prefix","show":false,"type":"str","value":"AI","l2_name":"ai_prefix","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-K1tss","field":"ai_prefix"},"display_name":"Ai Prefix"},"input_key_ConversationBufferMemory-K1tss":{"l2":false,"info":"The variable to be used as Chat Input when more than one variable is available.","list":false,"name":"input_key","show":true,"type":"str","value":"","l2_name":"input_key","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-K1tss","field":"input_key"},"display_name":"Input Key"},"memory_key_ConversationBufferMemory-K1tss":{"l2":false,"info":"","list":false,"name":"memory_key","show":true,"type":"str","value":"chat_history","l2_name":"memory_key","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-K1tss","field":"memory_key"},"display_name":"Memory Key"},"output_key_ConversationBufferMemory-K1tss":{"l2":false,"info":"The variable to be used as Chat Output (e.g. answer in a ConversationalRetrievalChain)","list":false,"name":"output_key","show":true,"type":"str","value":"answer","l2_name":"output_key","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-K1tss","field":"output_key"},"display_name":"Output Key"},"chat_memory_ConversationBufferMemory-K1tss":{"l2":false,"info":"","list":false,"name":"chat_memory","show":true,"type":"BaseChatMessageHistory","l2_name":"chat_memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-K1tss","field":"chat_memory"},"display_name":"Chat Memory"},"human_prefix_ConversationBufferMemory-K1tss":{"l2":false,"info":"","list":false,"name":"human_prefix","show":false,"type":"str","value":"Human","l2_name":"human_prefix","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-K1tss","field":"human_prefix"},"display_name":"Human Prefix"},"return_messages_ConversationBufferMemory-K1tss":{"l2":false,"info":"","list":false,"name":"return_messages","show":true,"type":"bool","l2_name":"return_messages","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-K1tss","field":"return_messages"},"display_name":"Return Messages"},"template_PromptTemplate-bs0vj":{"l2":false,"info":"","list":false,"name":"template","show":true,"type":"prompt","value":"参考文本：\n{context}\n\n----------------------\n\n1.用中文回答用户问题，并且答案要严谨专业。\n2.你需要依据以上【参考文本】中的内容来回答，当【参考文本】中有明确与用户问题相关的内容时才进行回答，不可根据自己的知识来回答。\n3.由于【参考文本】可能包含多个来自不同信息源的信息，所以根据这些不同的信息源可能得出有差异甚至冲突的答案，当发现这种情况时，这些答案都列举出来；如果没有冲突或差异，则只需要给出一个最终结果。\n4.若【参考文本】中内容与用户问题不相关则回复“没有找到相关内容”。\n\n----------------------\n\n用户问题: {question}\n你的回答:","l2_name":"template","advanced":false,"password":false,"required":true,"multiline":true,"placeholder":"","proxy":{"id":"PromptTemplate-bs0vj","field":"template"},"display_name":"Template"},"output_parser_PromptTemplate-bs0vj":{"l2":false,"info":"","list":false,"name":"output_parser","show":true,"type":"BaseOutputParser","l2_name":"output_parser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-bs0vj","field":"output_parser"},"display_name":"Output Parser"},"input_variables_PromptTemplate-bs0vj":{"l2":false,"info":"","list":true,"name":"input_variables","show":false,"type":"str","value":["context","question"],"l2_name":"input_variables","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-bs0vj","field":"input_variables"},"display_name":"Input Variables"},"template_format_PromptTemplate-bs0vj":{"l2":false,"info":"","list":false,"name":"template_format","show":false,"type":"str","value":"f-string","l2_name":"template_format","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-bs0vj","field":"template_format"},"display_name":"Template Format"},"partial_variables_PromptTemplate-bs0vj":{"l2":false,"info":"","list":false,"name":"partial_variables","show":false,"type":"code","l2_name":"partial_variables","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-bs0vj","field":"partial_variables"},"display_name":"Partial Variables"},"validate_template_PromptTemplate-bs0vj":{"l2":false,"info":"","list":false,"name":"validate_template","show":false,"type":"bool","value":true,"l2_name":"validate_template","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-bs0vj","field":"validate_template"},"display_name":"Validate Template"}},"flow":{"data":{"nodes":[{"width":384,"height":500,"id":"ProxyChatLLM-UYoB4","type":"genericNode","position":{"x":-3019.932873785603,"y":-759.0579388411045},"data":{"id":"ProxyChatLLM-UYoB4","node":{"template":{"n":{"l2":false,"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"l2_name":"n","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tags":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"ProxyChatLLM","cache":{"l2":false,"info":"","list":false,"name":"cache","show":false,"type":"bool","l2_name":"cache","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"top_p":{"l2":false,"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"l2_name":"top_p","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"client":{"l2":false,"info":"","list":false,"name":"client","show":false,"type":"Any","l2_name":"client","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"headers":{"l2":false,"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","l2_name":"headers","advanced":false,"password":false,"required":false,"multiline":true,"placeholder":""},"verbose":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"metadata":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"streaming":{"l2":false,"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"l2_name":"streaming","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_tokens":{"l2":false,"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","l2_name":"max_tokens","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":""},"model_name":{"l2":false,"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-3.5-turbo-16k-0613","l2_name":"model_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_retries":{"l2":false,"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"l2_name":"max_retries","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"temperature":{"l2":false,"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"l2_name":"temperature","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"model_kwargs":{"l2":false,"info":"","list":false,"name":"model_kwargs","show":true,"type":"code","l2_name":"model_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"elemai_api_key":{"l2":false,"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","l2_name":"elemai_api_key","advanced":false,"password":true,"required":false,"multiline":false,"placeholder":""},"elemai_base_url":{"l2":false,"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","l2_name":"elemai_base_url","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"request_timeout":{"l2":false,"info":"","list":false,"name":"request_timeout","show":false,"type":"float","l2_name":"request_timeout","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tiktoken_model_name":{"l2":false,"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","l2_name":"tiktoken_model_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Wrapper around proxy Chat large language models.","base_classes":["ProxyChatLLM","BaseLanguageModel","BaseChatModel","BaseLLM"],"display_name":"ProxyChatLLM","output_types":[],"custom_fields":{},"documentation":""},"type":"ProxyChatLLM","value":null},"selected":true,"positionAbsolute":{"x":-3019.932873785603,"y":-759.0579388411045}},{"width":384,"height":350,"id":"MixEsVectorRetriever-J35CZ","type":"genericNode","position":{"x":-2407.3760484096842,"y":-129.80633940439208},"data":{"id":"MixEsVectorRetriever-J35CZ","node":{"template":{"tags":{"l2":false,"info":"","list":true,"name":"tags","show":true,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"MixEsVectorRetriever","metadata":{"l2":false,"info":"","list":false,"name":"metadata","show":true,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"combine_strategy":{"l2":false,"info":"","list":true,"name":"combine_strategy","show":true,"type":"str","value":"mix","l2_name":"combine_strategy","options":["keyword_front","vector_front","mix"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"vector_retriever":{"l2":false,"info":"","list":false,"name":"vector_retriever","show":true,"type":"BaseRetriever","l2_name":"vector_retriever","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"keyword_retriever":{"l2":false,"info":"","list":false,"name":"keyword_retriever","show":true,"type":"BaseRetriever","l2_name":"keyword_retriever","advanced":false,"password":true,"required":true,"multiline":false,"placeholder":"","value":""}},"description":"This class ensemble the results of es retriever and vector retriever.","base_classes":["BaseRetriever","MixEsVectorRetriever"],"display_name":"MixEsVectorRetriever","output_types":[],"custom_fields":{},"documentation":""},"type":"MixEsVectorRetriever","value":null},"selected":true,"positionAbsolute":{"x":-2407.3760484096842,"y":-129.80633940439208}},{"width":384,"height":392,"id":"Milvus-cyR5W","type":"genericNode","position":{"x":-3010.2766638860544,"y":280.83609057857933},"data":{"id":"Milvus-cyR5W","node":{"template":{"_type":"Milvus","drop_old":{"l2":false,"info":"","list":false,"name":"drop_old","show":false,"type":"bool","value":false,"l2_name":"drop_old","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"documents":{"l2":false,"info":"","list":true,"name":"documents","show":true,"type":"Document","l2_name":"documents","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Documents"},"embedding":{"l2":false,"info":"","list":false,"name":"embedding","show":false,"type":"Embeddings","l2_name":"embedding","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Embedding"},"metadatas":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"index_params":{"l2":false,"info":"","list":false,"name":"index_params","show":false,"type":"code","l2_name":"index_params","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"search_kwargs":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":true,"type":"code","value":"{}","l2_name":"search_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"search_params":{"l2":false,"info":"","list":false,"name":"search_params","show":false,"type":"code","l2_name":"search_params","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"collection_name":{"l2":true,"info":"","list":false,"name":"collection_name","show":true,"type":"str","value":"col_1700798314_e3f656ab","l2_name":"collection_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"connection_args":{"l2":false,"info":"","list":false,"name":"connection_args","show":true,"type":"code","value":"","l2_name":"connection_args","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"consistency_level":{"l2":false,"info":"","list":false,"name":"consistency_level","show":false,"type":"str","value":"Session","l2_name":"consistency_level","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Create a Milvus collection, indexes it with HNSW, and insert data.","base_classes":["VectorStore","Milvus","BaseRetriever","VectorStoreRetriever"],"display_name":"Milvus","output_types":[],"custom_fields":{},"documentation":"http://192.168.106.116:8030"},"type":"Milvus","value":null},"selected":true,"positionAbsolute":{"x":-3010.2766638860544,"y":280.83609057857933}},{"width":384,"height":474,"id":"ElasticKeywordsSearch-31Et9","type":"genericNode","position":{"x":-3007.9415392561414,"y":-220.71638201134618},"data":{"id":"ElasticKeywordsSearch-31Et9","node":{"template":{"ids":{"l2":false,"info":"","list":true,"name":"ids","show":false,"type":"str","l2_name":"ids","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"llm":{"l2":false,"info":"","list":false,"name":"llm","type":"BaseLLM","l2_name":"llm","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"LLM"},"_type":"ElasticKeywordsSearch","prompt":{"l2":false,"info":"","list":false,"name":"prompt","type":"BasePromptTemplate","l2_name":"prompt","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"prompt"},"documents":{"l2":false,"info":"","list":true,"name":"documents","show":true,"type":"Document","l2_name":"documents","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Documents"},"embedding":{"l2":false,"info":"","list":false,"name":"embedding","show":false,"type":"Embeddings","l2_name":"embedding","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Embedding"},"metadatas":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"index_name":{"l2":true,"info":"","list":false,"name":"index_name","show":true,"type":"str","value":"col_1700798314_e3f656ab","l2_name":"index_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"ssl_verify":{"l2":false,"info":"","list":false,"name":"ssl_verify","show":true,"type":"str","value":"","l2_name":"ssl_verify","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"ssl_verify"},"search_kwargs":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":true,"type":"code","value":"{}","l2_name":"search_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"refresh_indices":{"l2":false,"info":"","list":false,"name":"refresh_indices","show":false,"type":"bool","value":true,"l2_name":"refresh_indices","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"elasticsearch_url":{"l2":false,"info":"","list":false,"name":"elasticsearch_url","show":true,"type":"str","value":"","l2_name":"elasticsearch_url","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"ES_connection_url"}},"description":"Construct ElasticKeywordsSearch wrapper from raw documents.","base_classes":["VectorStore","ElasticKeywordsSearch","BaseRetriever","VectorStoreRetriever"],"display_name":"ElasticKeywordsSearch","output_types":[],"custom_fields":{},"documentation":"http://192.168.106.116:8030"},"type":"ElasticKeywordsSearch","value":null},"selected":true,"positionAbsolute":{"x":-3007.9415392561414,"y":-220.71638201134618}},{"width":384,"height":494,"id":"ConversationalRetrievalChain-SmY3w","type":"genericNode","position":{"x":-1547.0269680478611,"y":-543.0897871904267},"data":{"id":"ConversationalRetrievalChain-SmY3w","node":{"template":{"llm":{"l2":false,"info":"","list":false,"name":"llm","show":true,"type":"BaseLanguageModel","l2_name":"llm","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"_type":"ConversationalRetrievalChain","memory":{"l2":false,"info":"","list":false,"name":"memory","show":true,"type":"BaseChatMemory","l2_name":"memory","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"verbose":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"l2":false,"info":"","list":false,"name":"callbacks","show":false,"type":"Callbacks","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"retriever":{"l2":false,"info":"","list":false,"name":"retriever","show":true,"type":"BaseRetriever","l2_name":"retriever","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"chain_type":{"l2":false,"info":"","list":true,"name":"chain_type","show":true,"type":"str","value":"stuff","l2_name":"chain_type","options":["stuff","map_reduce","map_rerank","refine"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"input_node":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question"},"condense_question_llm":{"l2":false,"info":"","list":false,"name":"condense_question_llm","show":false,"type":"BaseLanguageModel","l2_name":"condense_question_llm","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"return_source_documents":{"l2":false,"info":"","list":false,"name":"return_source_documents","show":true,"type":"bool","value":true,"l2_name":"return_source_documents","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Return source documents"},"condense_question_prompt":{"l2":false,"info":"","list":false,"name":"condense_question_prompt","show":true,"type":"BasePromptTemplate","value":{"_type":"prompt","template":"Given the following conversation and a follow up question, rephrase the follow up question to be a standalone question, in its original language.\n\nChat History:\n{chat_history}\nFollow Up Input: {question}\nStandalone question:","output_parser":null,"input_variables":["chat_history","question"],"template_format":"f-string","partial_variables":{},"validate_template":true},"l2_name":"condense_question_prompt","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"combine_docs_chain_kwargs":{"l2":false,"info":"","list":false,"name":"combine_docs_chain_kwargs","show":true,"type":"BasePromptTemplate","l2_name":"combine_docs_chain_kwargs","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"prompt"}},"description":"Convenience method to load chain from LLM and retriever.","base_classes":["BaseConversationalRetrievalChain","Chain","ConversationalRetrievalChain","function"],"display_name":"ConversationalRetrievalChain","output_types":[],"custom_fields":{},"documentation":"https://python.langchain.com/docs/modules/chains/popular/chat_vector_db"},"type":"ConversationalRetrievalChain","value":null},"selected":true,"positionAbsolute":{"x":-1547.0269680478611,"y":-543.0897871904267}},{"width":384,"height":526,"id":"ConversationBufferMemory-K1tss","type":"genericNode","position":{"x":-2419.6151156542605,"y":-708.8327872944353},"data":{"id":"ConversationBufferMemory-K1tss","node":{"template":{"_type":"ConversationBufferMemory","ai_prefix":{"l2":false,"info":"","list":false,"name":"ai_prefix","show":false,"type":"str","value":"AI","l2_name":"ai_prefix","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_key":{"l2":false,"info":"The variable to be used as Chat Input when more than one variable is available.","list":false,"name":"input_key","show":true,"type":"str","value":"","l2_name":"input_key","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"memory_key":{"l2":false,"info":"","list":false,"name":"memory_key","show":true,"type":"str","value":"chat_history","l2_name":"memory_key","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"output_key":{"l2":false,"info":"The variable to be used as Chat Output (e.g. answer in a ConversationalRetrievalChain)","list":false,"name":"output_key","show":true,"type":"str","value":"answer","l2_name":"output_key","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"chat_memory":{"l2":false,"info":"","list":false,"name":"chat_memory","show":true,"type":"BaseChatMessageHistory","l2_name":"chat_memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"human_prefix":{"l2":false,"info":"","list":false,"name":"human_prefix","show":false,"type":"str","value":"Human","l2_name":"human_prefix","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"return_messages":{"l2":false,"info":"","list":false,"name":"return_messages","show":true,"type":"bool","l2_name":"return_messages","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Buffer for storing conversation memory.","base_classes":["BaseMemory","ConversationBufferMemory","BaseChatMemory"],"display_name":"ConversationBufferMemory","output_types":[],"custom_fields":{},"documentation":"https://python.langchain.com/docs/modules/memory/how_to/buffer"},"type":"ConversationBufferMemory","value":null},"selected":true,"positionAbsolute":{"x":-2419.6151156542605,"y":-708.8327872944353}},{"width":384,"height":290,"id":"PromptTemplate-bs0vj","type":"genericNode","position":{"x":-2424.4899097749667,"y":-1236.9354837042674},"data":{"id":"PromptTemplate-bs0vj","node":{"name":"","template":{"_type":"PromptTemplate","template":{"l2":false,"info":"","list":false,"name":"template","show":true,"type":"prompt","value":"参考文本：\n{context}\n\n----------------------\n\n1.用中文回答用户问题，并且答案要严谨专业。\n2.你需要依据以上【参考文本】中的内容来回答，当【参考文本】中有明确与用户问题相关的内容时才进行回答，不可根据自己的知识来回答。\n3.由于【参考文本】可能包含多个来自不同信息源的信息，所以根据这些不同的信息源可能得出有差异甚至冲突的答案，当发现这种情况时，这些答案都列举出来；如果没有冲突或差异，则只需要给出一个最终结果。\n4.若【参考文本】中内容与用户问题不相关则回复“没有找到相关内容”。\n\n----------------------\n\n用户问题: {question}\n你的回答:","l2_name":"template","advanced":false,"password":false,"required":true,"multiline":true,"placeholder":""},"output_parser":{"l2":false,"info":"","list":false,"name":"output_parser","show":true,"type":"BaseOutputParser","l2_name":"output_parser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_variables":{"l2":false,"info":"","list":true,"name":"input_variables","show":false,"type":"str","value":["context","question"],"l2_name":"input_variables","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"template_format":{"l2":false,"info":"","list":false,"name":"template_format","show":false,"type":"str","value":"f-string","l2_name":"template_format","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"partial_variables":{"l2":false,"info":"","list":false,"name":"partial_variables","show":false,"type":"code","l2_name":"partial_variables","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"validate_template":{"l2":false,"info":"","list":false,"name":"validate_template","show":false,"type":"bool","value":true,"l2_name":"validate_template","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Schema to represent a prompt for an LLM.","base_classes":["BasePromptTemplate","PromptTemplate","StringPromptTemplate"],"display_name":"PromptTemplate","output_types":[],"custom_fields":{"":["context","question"],"template":["context","question"]},"documentation":"https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/","field_formatters":{"formatters":{"openai_api_key":{}},"base_formatters":{"dict":{},"list":{},"show":{},"union":{},"kwargs":{},"default":{},"headers":{},"optional":{},"password":{},"multiline":{},"model_fields":{"MODEL_DICT":{"OpenAI":["text-davinci-003","text-davinci-002","text-curie-001","text-babbage-001","text-ada-001"],"Anthropic":["claude-v1","claude-v1-100k","claude-instant-v1","claude-instant-v1-100k","claude-v1.3","claude-v1.3-100k","claude-v1.2","claude-v1.0","claude-instant-v1.1","claude-instant-v1.1-100k","claude-instant-v1.0"],"ChatOpenAI":["gpt-3.5-turbo-0613","gpt-3.5-turbo","gpt-3.5-turbo-16k-0613","gpt-3.5-turbo-16k","gpt-4-0613","gpt-4-32k-0613","gpt-4","gpt-4-32k"],"ChatAnthropic":["claude-v1","claude-v1-100k","claude-instant-v1","claude-instant-v1-100k","claude-v1.3","claude-v1.3-100k","claude-v1.2","claude-v1.0","claude-instant-v1.1","claude-instant-v1.1-100k","claude-instant-v1.0"]}},"dict_code_file":{}}}},"type":"PromptTemplate","value":null},"selected":true,"positionAbsolute":{"x":-2424.4899097749667,"y":-1236.9354837042674}}],"edges":[{"source":"Milvus-cyR5W","target":"MixEsVectorRetriever-J35CZ","sourceHandle":"Milvus|Milvus-cyR5W|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever","targetHandle":"BaseRetriever|vector_retriever|MixEsVectorRetriever-J35CZ","id":"reactflow__edge-Milvus-cyR5WMilvus|Milvus-cyR5W|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever-MixEsVectorRetriever-J35CZBaseRetriever|vector_retriever|MixEsVectorRetriever-J35CZ","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"ElasticKeywordsSearch-31Et9","target":"MixEsVectorRetriever-J35CZ","sourceHandle":"ElasticKeywordsSearch|ElasticKeywordsSearch-31Et9|VectorStore|ElasticKeywordsSearch|BaseRetriever|VectorStoreRetriever","targetHandle":"BaseRetriever|keyword_retriever|MixEsVectorRetriever-J35CZ","id":"reactflow__edge-ElasticKeywordsSearch-31Et9ElasticKeywordsSearch|ElasticKeywordsSearch-31Et9|VectorStore|ElasticKeywordsSearch|BaseRetriever|VectorStoreRetriever-MixEsVectorRetriever-J35CZBaseRetriever|keyword_retriever|MixEsVectorRetriever-J35CZ","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"MixEsVectorRetriever-J35CZ","target":"ConversationalRetrievalChain-SmY3w","sourceHandle":"MixEsVectorRetriever|MixEsVectorRetriever-J35CZ|BaseRetriever|MixEsVectorRetriever","targetHandle":"BaseRetriever|retriever|ConversationalRetrievalChain-SmY3w","id":"reactflow__edge-MixEsVectorRetriever-J35CZMixEsVectorRetriever|MixEsVectorRetriever-J35CZ|BaseRetriever|MixEsVectorRetriever-ConversationalRetrievalChain-SmY3wBaseRetriever|retriever|ConversationalRetrievalChain-SmY3w","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"ProxyChatLLM-UYoB4","target":"ConversationalRetrievalChain-SmY3w","sourceHandle":"ProxyChatLLM|ProxyChatLLM-UYoB4|ProxyChatLLM|BaseLanguageModel|BaseChatModel|BaseLLM","targetHandle":"BaseLanguageModel|llm|ConversationalRetrievalChain-SmY3w","id":"reactflow__edge-ProxyChatLLM-UYoB4ProxyChatLLM|ProxyChatLLM-UYoB4|ProxyChatLLM|BaseLanguageModel|BaseChatModel|BaseLLM-ConversationalRetrievalChain-SmY3wBaseLanguageModel|llm|ConversationalRetrievalChain-SmY3w","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"ConversationBufferMemory-K1tss","target":"ConversationalRetrievalChain-SmY3w","sourceHandle":"ConversationBufferMemory|ConversationBufferMemory-K1tss|BaseMemory|ConversationBufferMemory|BaseChatMemory","targetHandle":"BaseChatMemory|memory|ConversationalRetrievalChain-SmY3w","id":"reactflow__edge-ConversationBufferMemory-K1tssConversationBufferMemory|ConversationBufferMemory-K1tss|BaseMemory|ConversationBufferMemory|BaseChatMemory-ConversationalRetrievalChain-SmY3wBaseChatMemory|memory|ConversationalRetrievalChain-SmY3w","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"PromptTemplate-bs0vj","target":"ConversationalRetrievalChain-SmY3w","sourceHandle":"PromptTemplate|PromptTemplate-bs0vj|BasePromptTemplate|PromptTemplate|StringPromptTemplate","targetHandle":"BasePromptTemplate|combine_docs_chain_kwargs|ConversationalRetrievalChain-SmY3w","id":"reactflow__edge-PromptTemplate-bs0vjPromptTemplate|PromptTemplate-bs0vj|BasePromptTemplate|PromptTemplate|StringPromptTemplate-ConversationalRetrievalChain-SmY3wBasePromptTemplate|combine_docs_chain_kwargs|ConversationalRetrievalChain-SmY3w","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true}],"viewport":{"zoom":1,"x":0,"y":0}},"is_component":false,"name":"","description":"","id":"89297","status":0,"write":false,"guide_word":""}}}},{"知识库问答":{"id":"RetrievalQA-527fa","type":"RetrievalQA","node":{"output_types":[],"display_name":"知识库问答","documentation":"","base_classes":["RetrievalQA","Chain","BaseRetrievalQA","function"],"description":"Chain for question-answering against an index.","template":{"drop_old_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"drop_old","show":false,"type":"bool","value":false,"l2_name":"drop_old","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"drop_old"},"display_name":"Drop Old"},"documents_Milvus-CWpae":{"l2":false,"info":"","list":true,"name":"documents","show":true,"type":"Document","l2_name":"documents","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Documents","proxy":{"id":"Milvus-CWpae","field":"documents"}},"embedding_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"embedding","show":false,"type":"Embeddings","l2_name":"embedding","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Embedding","proxy":{"id":"Milvus-CWpae","field":"embedding"}},"metadatas_Milvus-CWpae":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"metadatas"},"display_name":"Metadatas"},"index_params_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"index_params","show":false,"type":"code","l2_name":"index_params","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"index_params"},"display_name":"Index Params"},"search_kwargs_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":false,"type":"code","value":"{}","l2_name":"search_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"search_kwargs"},"display_name":"Search Kwargs"},"search_params_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"search_params","show":false,"type":"code","l2_name":"search_params","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"search_params"},"display_name":"Search Params"},"collection_name_Milvus-CWpae":{"l2":true,"info":"","list":false,"name":"collection_name","show":true,"type":"str","value":"毕昇文档20240309","l2_name":"选择知识库","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","collection_id":3520,"proxy":{"id":"Milvus-CWpae","field":"collection_name"},"display_name":"Collection Name"},"connection_args_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"connection_args","show":false,"type":"code","value":"","l2_name":"connection_args","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"connection_args"},"display_name":"Connection Args"},"consistency_level_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"consistency_level","show":false,"type":"str","value":"Session","l2_name":"consistency_level","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"consistency_level"},"display_name":"Consistency Level"},"tags_RetrievalQA-x5nap":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"tags"},"display_name":"Tags"},"memory_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"memory","show":true,"type":"BaseMemory","l2_name":"memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"memory"},"display_name":"Memory"},"verbose_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"verbose"},"display_name":"Verbose"},"metadata_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"metadata"},"display_name":"Metadata"},"callbacks_RetrievalQA-x5nap":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"callbacks"},"display_name":"Callbacks"},"input_key_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"input_key","show":true,"type":"str","value":"query","l2_name":"input_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"input_key"},"display_name":"Input Key"},"input_node_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question","proxy":{"id":"RetrievalQA-x5nap","field":"input_node"}},"output_key_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"output_key","show":true,"type":"str","value":"result","l2_name":"output_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"output_key"},"display_name":"Output Key"},"return_source_documents_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"return_source_documents","show":true,"type":"bool","value":true,"l2_name":"return_source_documents","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"return_source_documents"},"display_name":"Return Source Documents"},"n_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"l2_name":"n","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"n"},"display_name":"N"},"tags_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"tags"},"display_name":"Tags"},"cache_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"cache","show":false,"type":"bool","l2_name":"cache","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"cache"},"display_name":"Cache"},"top_p_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"l2_name":"top_p","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"top_p"},"display_name":"Top P"},"client_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"client","show":false,"type":"Any","l2_name":"client","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"client"},"display_name":"Client"},"headers_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","l2_name":"headers","advanced":true,"password":false,"required":false,"multiline":true,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"headers"},"display_name":"Headers"},"verbose_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"verbose"},"display_name":"Verbose"},"metadata_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"metadata"},"display_name":"Metadata"},"callbacks_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"callbacks"},"display_name":"Callbacks"},"streaming_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"l2_name":"streaming","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"streaming"},"display_name":"Streaming"},"max_tokens_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","l2_name":"max_tokens","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"max_tokens"},"display_name":"Max Tokens"},"model_name_ProxyChatLLM-UYoB4":{"l2":true,"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-3.5-turbo-16k-0613","l2_name":"填写模型名称（支持所有主流闭源模型）","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"model_name"},"display_name":"Model Name"},"max_retries_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"l2_name":"max_retries","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"max_retries"},"display_name":"Max Retries"},"temperature_ProxyChatLLM-UYoB4":{"l2":true,"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"l2_name":"温度（控制输出随机性）","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"temperature"},"display_name":"Temperature"},"model_kwargs_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"model_kwargs","show":false,"type":"code","l2_name":"model_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"model_kwargs"},"display_name":"Model Kwargs"},"elemai_api_key_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","l2_name":"elemai_api_key","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"elemai_api_key"},"display_name":"Elemai Api Key"},"elemai_base_url_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","l2_name":"elemai_base_url","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"elemai_base_url"},"display_name":"Elemai Base Url"},"request_timeout_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"request_timeout","show":false,"type":"float","l2_name":"request_timeout","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"request_timeout"},"display_name":"Request Timeout"},"tiktoken_model_name_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","l2_name":"tiktoken_model_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"tiktoken_model_name"},"display_name":"Tiktoken Model Name"},"token_max_CombineDocsChain-71a85":{"info":"只对Stuff类型生效","list":false,"name":"token_max","show":true,"type":"int","value":-1,"advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"token_max","proxy":{"id":"CombineDocsChain-71a85","field":"token_max"}},"chain_type_CombineDocsChain-71a85":{"info":"","list":true,"name":"chain_type","show":true,"type":"str","value":"stuff","options":["stuff","map_reduce","map_rerank","refine"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"CombineDocsChain-71a85","field":"chain_type"},"display_name":"Chain Type"},"source_PromptTemplate-3c164":{"info":"","list":false,"name":"source","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"source","proxy":{"id":"PromptTemplate-3c164","field":"source"}},"template_PromptTemplate-3c164":{"l2":false,"info":"","list":false,"name":"template","show":true,"type":"prompt","value":"title: {source}\n{page_content}\n----------------","l2_name":"提示词模板","advanced":false,"password":false,"required":true,"multiline":true,"placeholder":"","proxy":{"id":"PromptTemplate-3c164","field":"template"},"display_name":"Template"},"input_types_PromptTemplate-3c164":{"info":"","list":false,"name":"input_types","show":false,"type":"code","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-3c164","field":"input_types"},"display_name":"Input Types"},"page_content_PromptTemplate-3c164":{"info":"","list":false,"name":"page_content","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"page_content","proxy":{"id":"PromptTemplate-3c164","field":"page_content"}},"output_parser_PromptTemplate-3c164":{"info":"","list":false,"name":"output_parser","show":true,"type":"BaseOutputParser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-3c164","field":"output_parser"},"display_name":"Output Parser"},"input_variables_PromptTemplate-3c164":{"info":"","list":true,"name":"input_variables","show":false,"type":"str","value":["source","page_content"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-3c164","field":"input_variables"},"display_name":"Input Variables"},"template_format_PromptTemplate-3c164":{"info":"","list":false,"name":"template_format","show":false,"type":"Union[Literal['f-string'], Literal['jinja2']]","value":"f-string","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-3c164","field":"template_format"},"display_name":"Template Format"},"partial_variables_PromptTemplate-3c164":{"info":"","list":false,"name":"partial_variables","show":false,"type":"code","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-3c164","field":"partial_variables"},"display_name":"Partial Variables"},"validate_template_PromptTemplate-3c164":{"info":"","list":false,"name":"validate_template","show":false,"type":"bool","value":false,"advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-3c164","field":"validate_template"},"display_name":"Validate Template"},"context_PromptTemplate-ced29":{"info":"","list":false,"name":"context","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"context","proxy":{"id":"PromptTemplate-ced29","field":"context"}},"question_PromptTemplate-ced29":{"info":"","list":false,"name":"question","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"question","proxy":{"id":"PromptTemplate-ced29","field":"question"}},"template_PromptTemplate-ced29":{"l2":true,"info":"","list":false,"name":"template","show":true,"type":"prompt","value":"参考文本：\n{context}\n\n----------------------\n# Role：企业客服\n\n# Background :\n你是一家名叫XX公司的客服人员。你的日常工作是回复公司内用户的各类咨询问题，所以当用户提到“公司”时，指的是XX公司。\n\n# Goals:\n- 以上【参考文本】中是从企业知识库中查找到的相关信息，只能结合以上信息进行回答，若以上内容为空或其中没有找到能回答【用户问题】的内容时，则回复“没有找到相关内容”，不能根据你自己的知识自己发挥。\n- 特别要注意区分【用户问题】与【参考文本】中不同的日期、人名、公司名这些关键信息。\n- 用中文回答问题，并且答案要严谨、专业、清晰、可读性好。\n- 拥有排版审美, 会利用序号, 缩进, 分隔线和换行符等等来美化信息排版。\n\n----------------------\n\n用户问题: {question}\n你的回答:","l2_name":"提示词模板","advanced":false,"password":false,"required":true,"multiline":true,"placeholder":"","proxy":{"id":"PromptTemplate-ced29","field":"template"},"display_name":"Template"},"input_types_PromptTemplate-ced29":{"info":"","list":false,"name":"input_types","show":false,"type":"code","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-ced29","field":"input_types"},"display_name":"Input Types"},"output_parser_PromptTemplate-ced29":{"info":"","list":false,"name":"output_parser","show":true,"type":"BaseOutputParser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-ced29","field":"output_parser"},"display_name":"Output Parser"},"input_variables_PromptTemplate-ced29":{"info":"","list":true,"name":"input_variables","show":false,"type":"str","value":["context","question"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-ced29","field":"input_variables"},"display_name":"Input Variables"},"template_format_PromptTemplate-ced29":{"info":"","list":false,"name":"template_format","show":false,"type":"Union[Literal['f-string'], Literal['jinja2']]","value":"f-string","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-ced29","field":"template_format"},"display_name":"Template Format"},"partial_variables_PromptTemplate-ced29":{"info":"","list":false,"name":"partial_variables","show":false,"type":"code","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-ced29","field":"partial_variables"},"display_name":"Partial Variables"},"validate_template_PromptTemplate-ced29":{"info":"","list":false,"name":"validate_template","show":false,"type":"bool","value":false,"advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-ced29","field":"validate_template"},"display_name":"Validate Template"}},"flow":{"data":{"nodes":[{"width":384,"height":310,"id":"Milvus-CWpae","type":"genericNode","position":{"x":-2861.8694801119937,"y":-286.57134522726807},"data":{"id":"Milvus-CWpae","node":{"l2_name":"知识库","template":{"_type":"Milvus","drop_old":{"l2":false,"info":"","list":false,"name":"drop_old","show":false,"type":"bool","value":false,"l2_name":"drop_old","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"documents":{"l2":false,"info":"","list":true,"name":"documents","show":true,"type":"Document","l2_name":"documents","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Documents"},"embedding":{"l2":false,"info":"","list":false,"name":"embedding","show":false,"type":"Embeddings","l2_name":"embedding","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Embedding"},"metadatas":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"index_params":{"l2":false,"info":"","list":false,"name":"index_params","show":false,"type":"code","l2_name":"index_params","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"search_kwargs":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":true,"type":"code","value":"{}","l2_name":"search_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"search_params":{"l2":false,"info":"","list":false,"name":"search_params","show":false,"type":"code","l2_name":"search_params","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"collection_name":{"l2":true,"info":"","list":false,"name":"collection_name","show":true,"type":"str","value":"毕昇文档20240309","l2_name":"选择知识库","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","collection_id":3520},"connection_args":{"l2":false,"info":"","list":false,"name":"connection_args","show":true,"type":"code","value":"","l2_name":"connection_args","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"consistency_level":{"l2":false,"info":"","list":false,"name":"consistency_level","show":false,"type":"str","value":"Session","l2_name":"consistency_level","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Create a Milvus collection, indexes it with HNSW, and insert data.","base_classes":["Milvus","VectorStore","BaseRetriever","VectorStoreRetriever"],"display_name":"Milvus","output_types":[],"custom_fields":{},"documentation":"http://192.168.106.116:8030"},"type":"Milvus","value":null},"selected":true,"positionAbsolute":{"x":-2861.8694801119937,"y":-286.57134522726807}},{"width":384,"height":316,"id":"RetrievalQA-x5nap","type":"genericNode","position":{"x":-2198.289678633326,"y":-664.9918874506891},"data":{"id":"RetrievalQA-x5nap","node":{"template":{"tags":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"RetrievalQA","memory":{"l2":false,"info":"","list":false,"name":"memory","show":true,"type":"BaseMemory","l2_name":"memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"verbose":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"metadata":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_key":{"l2":false,"info":"","list":false,"name":"input_key","show":true,"type":"str","value":"query","l2_name":"input_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":""},"retriever":{"l2":false,"info":"","list":false,"name":"retriever","show":true,"type":"BaseRetriever","l2_name":"retriever","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"input_node":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question"},"output_key":{"l2":false,"info":"","list":false,"name":"output_key","show":true,"type":"str","value":"result","l2_name":"output_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":""},"combine_documents_chain":{"l2":false,"info":"","list":false,"name":"combine_documents_chain","show":true,"type":"BaseCombineDocumentsChain","l2_name":"combine_documents_chain","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"return_source_documents":{"l2":false,"info":"","list":false,"name":"return_source_documents","show":true,"type":"bool","value":true,"l2_name":"return_source_documents","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Chain for question-answering against an index.","base_classes":["RetrievalQA","Chain","BaseRetrievalQA","function"],"display_name":"RetrievalQA","output_types":[],"custom_fields":{},"documentation":"https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"},"type":"RetrievalQA","value":null},"selected":true,"positionAbsolute":{"x":-2198.289678633326,"y":-664.9918874506891}},{"width":384,"height":500,"id":"ProxyChatLLM-UYoB4","type":"genericNode","position":{"x":-3490.7895099718235,"y":-440.7836760027338},"data":{"id":"ProxyChatLLM-UYoB4","node":{"l2_name":"模型配置","template":{"n":{"l2":false,"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"l2_name":"n","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tags":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"ProxyChatLLM","cache":{"l2":false,"info":"","list":false,"name":"cache","show":false,"type":"bool","l2_name":"cache","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"top_p":{"l2":false,"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"l2_name":"top_p","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"client":{"l2":false,"info":"","list":false,"name":"client","show":false,"type":"Any","l2_name":"client","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"headers":{"l2":false,"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","l2_name":"headers","advanced":false,"password":false,"required":false,"multiline":true,"placeholder":""},"verbose":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"metadata":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"streaming":{"l2":false,"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"l2_name":"streaming","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_tokens":{"l2":false,"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","l2_name":"max_tokens","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":""},"model_name":{"l2":true,"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-3.5-turbo-16k-0613","l2_name":"填写模型名称（支持所有主流闭源模型）","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_retries":{"l2":false,"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"l2_name":"max_retries","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"temperature":{"l2":true,"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"l2_name":"温度（控制输出随机性）","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"model_kwargs":{"l2":false,"info":"","list":false,"name":"model_kwargs","show":true,"type":"code","l2_name":"model_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"elemai_api_key":{"l2":false,"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","l2_name":"elemai_api_key","advanced":false,"password":true,"required":false,"multiline":false,"placeholder":""},"elemai_base_url":{"l2":false,"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","l2_name":"elemai_base_url","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"request_timeout":{"l2":false,"info":"","list":false,"name":"request_timeout","show":false,"type":"float","l2_name":"request_timeout","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tiktoken_model_name":{"l2":false,"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","l2_name":"tiktoken_model_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Wrapper around proxy Chat large language models.","base_classes":["ProxyChatLLM","BaseLanguageModel","BaseChatModel","BaseLLM"],"display_name":"ProxyChatLLM","output_types":[],"custom_fields":{},"documentation":""},"type":"ProxyChatLLM","value":null},"selected":true,"positionAbsolute":{"x":-3490.7895099718235,"y":-440.7836760027338}},{"width":384,"height":448,"id":"CombineDocsChain-71a85","type":"genericNode","position":{"x":-2872.4473624642733,"y":-941.8372718902958},"data":{"id":"CombineDocsChain-71a85","node":{"template":{"llm":{"info":"","list":false,"name":"llm","show":true,"type":"BaseLanguageModel","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"LLM"},"_type":"load_qa_chain","prompt":{"info":"只对Stuff类型生效","list":false,"name":"prompt","show":true,"type":"BasePromptTemplate","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"prompt"},"token_max":{"info":"只对Stuff类型生效","list":false,"name":"token_max","show":true,"type":"int","value":-1,"advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"token_max"},"chain_type":{"info":"","list":true,"name":"chain_type","show":true,"type":"str","value":"stuff","options":["stuff","map_reduce","map_rerank","refine"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"document_prompt":{"info":"","list":false,"name":"document_prompt","show":true,"type":"BasePromptTemplate","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Load question answering chain.","base_classes":["BaseCombineDocumentsChain","function"],"display_name":"CombineDocsChain","output_types":[],"custom_fields":{},"documentation":""},"type":"CombineDocsChain","value":null},"selected":true,"positionAbsolute":{"x":-2872.4473624642733,"y":-941.8372718902958}},{"width":384,"height":454,"id":"PromptTemplate-3c164","type":"genericNode","position":{"x":-3470.0105149154433,"y":-1425.3602052103988},"data":{"id":"PromptTemplate-3c164","node":{"name":"","l2_name":"召回段落拼接提示词","template":{"_type":"PromptTemplate","source":{"info":"","list":false,"name":"source","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"source"},"template":{"l2":false,"info":"","list":false,"name":"template","show":true,"type":"prompt","value":"title: {source}\n{page_content}\n----------------","l2_name":"提示词模板","advanced":false,"password":false,"required":true,"multiline":true,"placeholder":""},"input_types":{"info":"","list":false,"name":"input_types","show":false,"type":"code","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"page_content":{"info":"","list":false,"name":"page_content","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"page_content"},"output_parser":{"info":"","list":false,"name":"output_parser","show":true,"type":"BaseOutputParser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_variables":{"info":"","list":true,"name":"input_variables","show":false,"type":"str","value":["source","page_content"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"template_format":{"info":"","list":false,"name":"template_format","show":false,"type":"Union[Literal['f-string'], Literal['jinja2']]","value":"f-string","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"partial_variables":{"info":"","list":false,"name":"partial_variables","show":false,"type":"code","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"validate_template":{"info":"","list":false,"name":"validate_template","show":false,"type":"bool","value":false,"advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"A prompt template for a language model.","base_classes":["Runnable","StringPromptTemplate","RunnableSerializable","PromptTemplate","Generic","BasePromptTemplate"],"display_name":"PromptTemplate","output_types":[],"custom_fields":{"":["source","page_content"]},"documentation":"https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/","field_formatters":{}},"type":"PromptTemplate","value":null},"selected":true,"positionAbsolute":{"x":-3470.0105149154433,"y":-1425.3602052103988}},{"width":384,"height":454,"id":"PromptTemplate-ced29","type":"genericNode","position":{"x":-3479.600564127818,"y":-925.8998699556519},"data":{"id":"PromptTemplate-ced29","node":{"name":"","l2_name":"大模型问答提示词","template":{"_type":"PromptTemplate","context":{"info":"","list":false,"name":"context","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"context"},"question":{"info":"","list":false,"name":"question","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser","VariableNode"],"placeholder":"","display_name":"question"},"template":{"l2":true,"info":"","list":false,"name":"template","show":true,"type":"prompt","value":"参考文本：\n{context}\n\n----------------------\n# Role：企业客服\n\n# Background :\n你是一家名叫XX公司的客服人员。你的日常工作是回复公司内用户的各类咨询问题，所以当用户提到“公司”时，指的是XX公司。\n\n# Goals:\n- 以上【参考文本】中是从企业知识库中查找到的相关信息，只能结合以上信息进行回答，若以上内容为空或其中没有找到能回答【用户问题】的内容时，则回复“没有找到相关内容”，不能根据你自己的知识自己发挥。\n- 特别要注意区分【用户问题】与【参考文本】中不同的日期、人名、公司名这些关键信息。\n- 用中文回答问题，并且答案要严谨、专业、清晰、可读性好。\n- 拥有排版审美, 会利用序号, 缩进, 分隔线和换行符等等来美化信息排版。\n\n----------------------\n\n用户问题: {question}\n你的回答:","l2_name":"提示词模板","advanced":false,"password":false,"required":true,"multiline":true,"placeholder":""},"input_types":{"info":"","list":false,"name":"input_types","show":false,"type":"code","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"output_parser":{"info":"","list":false,"name":"output_parser","show":true,"type":"BaseOutputParser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_variables":{"info":"","list":true,"name":"input_variables","show":false,"type":"str","value":["context","question"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"template_format":{"info":"","list":false,"name":"template_format","show":false,"type":"Union[Literal['f-string'], Literal['jinja2']]","value":"f-string","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"partial_variables":{"info":"","list":false,"name":"partial_variables","show":false,"type":"code","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"validate_template":{"info":"","list":false,"name":"validate_template","show":false,"type":"bool","value":false,"advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"A prompt template for a language model.","base_classes":["Runnable","StringPromptTemplate","RunnableSerializable","PromptTemplate","Generic","BasePromptTemplate"],"display_name":"PromptTemplate","output_types":[],"custom_fields":{"":["context","question"]},"documentation":"https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/","field_formatters":{}},"type":"PromptTemplate","value":null},"selected":true,"positionAbsolute":{"x":-3479.600564127818,"y":-925.8998699556519}}],"edges":[{"source":"Milvus-CWpae","target":"RetrievalQA-x5nap","sourceHandle":"Milvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever","targetHandle":"BaseRetriever|retriever|RetrievalQA-x5nap","id":"reactflow__edge-Milvus-CWpaeMilvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever-RetrievalQA-x5napBaseRetriever|retriever|RetrievalQA-x5nap","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"ProxyChatLLM-UYoB4","target":"CombineDocsChain-71a85","sourceHandle":"ProxyChatLLM|ProxyChatLLM-UYoB4|ProxyChatLLM|BaseLanguageModel|BaseChatModel|BaseLLM","targetHandle":"BaseLanguageModel|llm|CombineDocsChain-71a85","id":"reactflow__edge-ProxyChatLLM-UYoB4ProxyChatLLM|ProxyChatLLM-UYoB4|ProxyChatLLM|BaseLanguageModel|BaseChatModel|BaseLLM-CombineDocsChain-71a85BaseLanguageModel|llm|CombineDocsChain-71a85","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"CombineDocsChain-71a85","target":"RetrievalQA-x5nap","sourceHandle":"CombineDocsChain|CombineDocsChain-71a85|BaseCombineDocumentsChain|function","targetHandle":"BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap","id":"reactflow__edge-CombineDocsChain-71a85CombineDocsChain|CombineDocsChain-71a85|BaseCombineDocumentsChain|function-RetrievalQA-x5napBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"PromptTemplate-3c164","target":"CombineDocsChain-71a85","sourceHandle":"PromptTemplate|PromptTemplate-3c164|Runnable|StringPromptTemplate|RunnableSerializable|PromptTemplate|Generic|BasePromptTemplate","targetHandle":"BasePromptTemplate|document_prompt|CombineDocsChain-71a85","id":"reactflow__edge-PromptTemplate-3c164PromptTemplate|PromptTemplate-3c164|Runnable|StringPromptTemplate|RunnableSerializable|PromptTemplate|Generic|BasePromptTemplate-CombineDocsChain-71a85BasePromptTemplate|document_prompt|CombineDocsChain-71a85","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"PromptTemplate-ced29","target":"CombineDocsChain-71a85","sourceHandle":"PromptTemplate|PromptTemplate-ced29|Runnable|StringPromptTemplate|RunnableSerializable|PromptTemplate|Generic|BasePromptTemplate","targetHandle":"BasePromptTemplate|prompt|CombineDocsChain-71a85","id":"reactflow__edge-PromptTemplate-ced29PromptTemplate|PromptTemplate-ced29|Runnable|StringPromptTemplate|RunnableSerializable|PromptTemplate|Generic|BasePromptTemplate-CombineDocsChain-71a85BasePromptTemplate|prompt|CombineDocsChain-71a85","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true}],"viewport":{"zoom":1,"x":0,"y":0}},"is_component":false,"name":"","description":"","id":"72f43","status":0,"write":false,"guide_word":""}}}},{"表格数据问答":{"id":"RetrievalQA-c61ba","type":"InputFileNode","node":{"output_types":[],"display_name":"表格数据问答","documentation":"","base_classes":["RetrievalQA","Chain","BaseRetrievalQA","function"],"description":"Chain for question-answering against an index.","template":{"token_max_CombineDocsChain-RCPNA":{"l2":false,"info":"当前只对stuff 生效","list":false,"name":"token_max","show":true,"type":"int","value":-1,"l2_name":"token_max","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"token_max","proxy":{"id":"CombineDocsChain-RCPNA","field":"token_max"}},"chain_type_CombineDocsChain-RCPNA":{"l2":false,"info":"","list":true,"name":"chain_type","show":true,"type":"str","value":"stuff","l2_name":"chain_type","options":["stuff","map_reduce","map_rerank","refine"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"CombineDocsChain-RCPNA","field":"chain_type"},"display_name":"Chain Type"},"drop_old_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"drop_old","show":false,"type":"bool","value":false,"l2_name":"drop_old","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"drop_old"},"display_name":"Drop Old"},"metadatas_Milvus-CWpae":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"metadatas"},"display_name":"Metadatas"},"index_params_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"index_params","show":false,"type":"code","l2_name":"index_params","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"index_params"},"display_name":"Index Params"},"search_kwargs_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":false,"type":"code","value":"{}","l2_name":"search_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"search_kwargs"},"display_name":"Search Kwargs"},"search_params_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"search_params","show":false,"type":"code","l2_name":"search_params","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"search_params"},"display_name":"Search Params"},"collection_name_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"collection_name","show":true,"type":"str","value":"","l2_name":"选择知识库","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","collection_id":"","proxy":{"id":"Milvus-CWpae","field":"collection_name"},"display_name":"Collection Name"},"connection_args_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"connection_args","show":false,"type":"code","value":"","l2_name":"connection_args","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"connection_args"},"display_name":"Connection Args"},"consistency_level_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"consistency_level","show":false,"type":"str","value":"Session","l2_name":"consistency_level","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"consistency_level"},"display_name":"Consistency Level"},"tags_RetrievalQA-x5nap":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"tags"},"display_name":"Tags"},"memory_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"memory","show":true,"type":"BaseMemory","l2_name":"memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"memory"},"display_name":"Memory"},"verbose_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"verbose"},"display_name":"Verbose"},"metadata_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"metadata"},"display_name":"Metadata"},"callbacks_RetrievalQA-x5nap":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"callbacks"},"display_name":"Callbacks"},"input_key_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"input_key","show":true,"type":"str","value":"query","l2_name":"input_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"input_key"},"display_name":"Input Key"},"input_node_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question","proxy":{"id":"RetrievalQA-x5nap","field":"input_node"}},"output_key_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"output_key","show":true,"type":"str","value":"result","l2_name":"output_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"output_key"},"display_name":"Output Key"},"return_source_documents_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"return_source_documents","show":true,"type":"bool","value":true,"l2_name":"return_source_documents","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"return_source_documents"},"display_name":"Return Source Documents"},"n_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"l2_name":"n","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"n"},"display_name":"N"},"tags_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"tags"},"display_name":"Tags"},"cache_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"cache","show":false,"type":"bool","l2_name":"cache","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"cache"},"display_name":"Cache"},"top_p_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"l2_name":"top_p","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"top_p"},"display_name":"Top P"},"client_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"client","show":false,"type":"Any","l2_name":"client","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"client"},"display_name":"Client"},"headers_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","l2_name":"headers","advanced":true,"password":false,"required":false,"multiline":true,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"headers"},"display_name":"Headers"},"verbose_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"verbose"},"display_name":"Verbose"},"metadata_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"metadata"},"display_name":"Metadata"},"callbacks_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"callbacks"},"display_name":"Callbacks"},"streaming_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"l2_name":"streaming","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"streaming"},"display_name":"Streaming"},"max_tokens_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","l2_name":"max_tokens","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"max_tokens"},"display_name":"Max Tokens"},"model_name_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-4","l2_name":"model_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"model_name"},"display_name":"Model Name"},"max_retries_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"l2_name":"max_retries","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"max_retries"},"display_name":"Max Retries"},"temperature_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"l2_name":"temperature","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"temperature"},"display_name":"Temperature"},"model_kwargs_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"model_kwargs","show":false,"type":"code","l2_name":"model_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"model_kwargs"},"display_name":"Model Kwargs"},"elemai_api_key_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","l2_name":"elemai_api_key","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"elemai_api_key"},"display_name":"Elemai Api Key"},"elemai_base_url_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","l2_name":"elemai_base_url","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"elemai_base_url"},"display_name":"Elemai Base Url"},"request_timeout_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"request_timeout","show":false,"type":"float","l2_name":"request_timeout","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"request_timeout"},"display_name":"Request Timeout"},"tiktoken_model_name_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","l2_name":"tiktoken_model_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"tiktoken_model_name"},"display_name":"Tiktoken Model Name"},"proxy_url_OpenAIProxyEmbedding-1771b":{"info":"","list":false,"name":"proxy_url","show":false,"type":"str","value":"","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"http://43.133.35.137:8080","proxy":{"id":"OpenAIProxyEmbedding-1771b","field":"proxy_url"},"display_name":"Proxy Url"},"metadata_CSVLoader-854c0":{"info":"","list":false,"name":"metadata","show":false,"type":"code","value":"{}","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Metadata","proxy":{"id":"CSVLoader-854c0","field":"metadata"}},"file_path_InputFileNode-e4363":{"info":"","list":false,"name":"file_path","show":true,"type":"file","value":"","advanced":true,"password":false,"required":false,"suffixes":[".csv"],"fileTypes":["csv"],"file_path":null,"multiline":false,"placeholder":"","proxy":{"id":"InputFileNode-e4363","field":"file_path"},"display_name":"File Path"},"file_type_InputFileNode-e4363":{"info":"Tips for which file should upload","list":false,"name":"file_type","show":true,"type":"str","value":"CSV文件","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"提示上传文件类型","display_name":"Name","proxy":{"id":"InputFileNode-e4363","field":"file_type"}}},"flow":{"data":{"nodes":[{"width":384,"height":376,"id":"CombineDocsChain-RCPNA","type":"genericNode","position":{"x":1039.4135881298942,"y":116.97346377366182},"data":{"id":"CombineDocsChain-RCPNA","node":{"template":{"llm":{"l2":false,"info":"","list":false,"name":"llm","show":true,"type":"BaseLanguageModel","l2_name":"llm","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"LLM"},"_type":"load_qa_chain","token_max":{"l2":false,"info":"当前只对stuff 生效","list":false,"name":"token_max","show":true,"type":"int","value":-1,"l2_name":"token_max","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"token_max"},"chain_type":{"l2":false,"info":"","list":true,"name":"chain_type","show":true,"type":"str","value":"stuff","l2_name":"chain_type","options":["stuff","map_reduce","map_rerank","refine"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""}},"description":"Load question answering chain.","base_classes":["BaseCombineDocumentsChain","function"],"display_name":"CombineDocsChain","output_types":[],"custom_fields":{},"documentation":""},"type":"CombineDocsChain","value":null},"selected":true,"positionAbsolute":{"x":1039.4135881298942,"y":116.97346377366182}},{"width":384,"height":346,"id":"Milvus-CWpae","type":"genericNode","position":{"x":1052.1050968084205,"y":659.5536672641022},"data":{"id":"Milvus-CWpae","node":{"l2_name":"知识库","template":{"_type":"Milvus","drop_old":{"l2":false,"info":"","list":false,"name":"drop_old","show":false,"type":"bool","value":false,"l2_name":"drop_old","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"documents":{"l2":false,"info":"","list":true,"name":"documents","show":true,"type":"Document","l2_name":"documents","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Documents"},"embedding":{"l2":false,"info":"","list":false,"name":"embedding","show":true,"type":"Embeddings","l2_name":"embedding","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Embedding"},"metadatas":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"index_params":{"l2":false,"info":"","list":false,"name":"index_params","show":false,"type":"code","l2_name":"index_params","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"search_kwargs":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":true,"type":"code","value":"{}","l2_name":"search_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"search_params":{"l2":false,"info":"","list":false,"name":"search_params","show":false,"type":"code","l2_name":"search_params","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"collection_name":{"l2":false,"info":"","list":false,"name":"collection_name","show":true,"type":"str","value":"","l2_name":"选择知识库","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","collection_id":""},"connection_args":{"l2":false,"info":"","list":false,"name":"connection_args","show":true,"type":"code","value":"","l2_name":"connection_args","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"consistency_level":{"l2":false,"info":"","list":false,"name":"consistency_level","show":false,"type":"str","value":"Session","l2_name":"consistency_level","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Create a Milvus collection, indexes it with HNSW, and insert data.","base_classes":["Milvus","VectorStore","BaseRetriever","VectorStoreRetriever"],"display_name":"Milvus","output_types":[],"custom_fields":{},"documentation":"http://192.168.106.116:8030"},"type":"Milvus","value":null},"selected":true,"positionAbsolute":{"x":1052.1050968084205,"y":659.5536672641022}},{"width":384,"height":316,"id":"RetrievalQA-x5nap","type":"genericNode","position":{"x":1715.6848982870883,"y":279.846110701463},"data":{"id":"RetrievalQA-x5nap","node":{"template":{"tags":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"RetrievalQA","memory":{"l2":false,"info":"","list":false,"name":"memory","show":true,"type":"BaseMemory","l2_name":"memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"verbose":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"metadata":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_key":{"l2":false,"info":"","list":false,"name":"input_key","show":true,"type":"str","value":"query","l2_name":"input_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":""},"retriever":{"l2":false,"info":"","list":false,"name":"retriever","show":true,"type":"BaseRetriever","l2_name":"retriever","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"input_node":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question"},"output_key":{"l2":false,"info":"","list":false,"name":"output_key","show":true,"type":"str","value":"result","l2_name":"output_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":""},"combine_documents_chain":{"l2":false,"info":"","list":false,"name":"combine_documents_chain","show":true,"type":"BaseCombineDocumentsChain","l2_name":"combine_documents_chain","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"return_source_documents":{"l2":false,"info":"","list":false,"name":"return_source_documents","show":true,"type":"bool","value":true,"l2_name":"return_source_documents","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Chain for question-answering against an index.","base_classes":["RetrievalQA","Chain","BaseRetrievalQA","function"],"display_name":"RetrievalQA","output_types":[],"custom_fields":{},"documentation":"https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"},"type":"RetrievalQA","value":null},"selected":true,"positionAbsolute":{"x":1715.6848982870883,"y":279.846110701463}},{"width":384,"height":500,"id":"ProxyChatLLM-UYoB4","type":"genericNode","position":{"x":437.5688719308881,"y":20},"data":{"id":"ProxyChatLLM-UYoB4","node":{"template":{"n":{"l2":false,"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"l2_name":"n","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tags":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"ProxyChatLLM","cache":{"l2":false,"info":"","list":false,"name":"cache","show":false,"type":"bool","l2_name":"cache","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"top_p":{"l2":false,"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"l2_name":"top_p","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"client":{"l2":false,"info":"","list":false,"name":"client","show":false,"type":"Any","l2_name":"client","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"headers":{"l2":false,"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","l2_name":"headers","advanced":false,"password":false,"required":false,"multiline":true,"placeholder":""},"verbose":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"metadata":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"streaming":{"l2":false,"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"l2_name":"streaming","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_tokens":{"l2":false,"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","l2_name":"max_tokens","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":""},"model_name":{"l2":false,"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-4","l2_name":"model_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_retries":{"l2":false,"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"l2_name":"max_retries","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"temperature":{"l2":false,"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"l2_name":"temperature","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"model_kwargs":{"l2":false,"info":"","list":false,"name":"model_kwargs","show":true,"type":"code","l2_name":"model_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"elemai_api_key":{"l2":false,"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","l2_name":"elemai_api_key","advanced":false,"password":true,"required":false,"multiline":false,"placeholder":""},"elemai_base_url":{"l2":false,"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","l2_name":"elemai_base_url","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"request_timeout":{"l2":false,"info":"","list":false,"name":"request_timeout","show":false,"type":"float","l2_name":"request_timeout","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tiktoken_model_name":{"l2":false,"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","l2_name":"tiktoken_model_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Wrapper around proxy Chat large language models.","base_classes":["ProxyChatLLM","BaseLanguageModel","BaseChatModel","BaseLLM"],"display_name":"ProxyChatLLM","output_types":[],"custom_fields":{},"documentation":""},"type":"ProxyChatLLM","value":null},"selected":true,"positionAbsolute":{"x":437.5688719308881,"y":20}},{"width":384,"height":172,"id":"OpenAIProxyEmbedding-1771b","type":"genericNode","position":{"x":540.1525954325598,"y":1040.7357784004414},"data":{"id":"OpenAIProxyEmbedding-1771b","node":{"template":{"_type":"proxy_embedding","proxy_url":{"info":"","list":false,"name":"proxy_url","show":false,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"http://43.133.35.137:8080"}},"description":" 使用自建的embedding服务使用openai进行embed ","base_classes":["Embeddings"],"display_name":"OpenAIProxyEmbedding","output_types":[],"custom_fields":{},"documentation":""},"type":"OpenAIProxyEmbedding","value":null},"selected":true,"positionAbsolute":{"x":540.1525954325598,"y":1040.7357784004414}},{"width":384,"height":290,"id":"CSVLoader-854c0","type":"genericNode","position":{"x":521.107515963895,"y":664.4805878718827},"data":{"id":"CSVLoader-854c0","node":{"template":{"_type":"CSVLoader","metadata":{"info":"","list":false,"name":"metadata","show":true,"type":"code","value":"{}","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Metadata"},"file_path":{"info":"","list":false,"name":"file_path","show":true,"type":"fileNode","value":"","advanced":false,"password":false,"required":false,"suffixes":[".csv"],"fileTypes":["csv"],"multiline":false,"placeholder":""}},"description":"Load a `CSV` file into a list of Documents.","base_classes":["Document"],"display_name":"CSVLoader","output_types":["Document"],"custom_fields":{},"documentation":"https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/csv"},"type":"CSVLoader","value":null},"selected":true,"positionAbsolute":{"x":521.107515963895,"y":664.4805878718827}},{"width":384,"height":336,"id":"InputFileNode-e4363","type":"genericNode","position":{"x":18,"y":635.9129686688856},"data":{"id":"InputFileNode-e4363","node":{"template":{"_type":"InputFileNode","file_path":{"info":"","list":false,"name":"file_path","show":true,"type":"file","value":"","advanced":false,"password":false,"required":false,"suffixes":[".csv"],"fileTypes":["csv"],"file_path":null,"multiline":false,"placeholder":""},"file_type":{"info":"Tips for which file should upload","list":false,"name":"file_type","show":true,"type":"str","value":"CSV文件","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"提示上传文件类型","display_name":"Name"}},"description":"输入节点，用来自动对接输入","base_classes":["fileNode"],"display_name":"InputFileNode","output_types":[],"custom_fields":{},"documentation":""},"type":"InputFileNode","value":null},"selected":true,"positionAbsolute":{"x":18,"y":635.9129686688856}}],"edges":[{"source":"CombineDocsChain-RCPNA","target":"RetrievalQA-x5nap","sourceHandle":"CombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function","targetHandle":"BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap","id":"reactflow__edge-CombineDocsChain-RCPNACombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function-RetrievalQA-x5napBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"Milvus-CWpae","target":"RetrievalQA-x5nap","sourceHandle":"Milvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever","targetHandle":"BaseRetriever|retriever|RetrievalQA-x5nap","id":"reactflow__edge-Milvus-CWpaeMilvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever-RetrievalQA-x5napBaseRetriever|retriever|RetrievalQA-x5nap","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"ProxyChatLLM-UYoB4","target":"CombineDocsChain-RCPNA","sourceHandle":"ProxyChatLLM|ProxyChatLLM-UYoB4|ProxyChatLLM|BaseLanguageModel|BaseChatModel|BaseLLM","targetHandle":"BaseLanguageModel|llm|CombineDocsChain-RCPNA","id":"reactflow__edge-ProxyChatLLM-UYoB4ProxyChatLLM|ProxyChatLLM-UYoB4|ProxyChatLLM|BaseLanguageModel|BaseChatModel|BaseLLM-CombineDocsChain-RCPNABaseLanguageModel|llm|CombineDocsChain-RCPNA","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"OpenAIProxyEmbedding-1771b","target":"Milvus-CWpae","sourceHandle":"OpenAIProxyEmbedding|OpenAIProxyEmbedding-1771b|Embeddings","targetHandle":"Embeddings|embedding|Milvus-CWpae","id":"reactflow__edge-OpenAIProxyEmbedding-1771bOpenAIProxyEmbedding|OpenAIProxyEmbedding-1771b|Embeddings-Milvus-CWpaeEmbeddings|embedding|Milvus-CWpae","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"CSVLoader-854c0","target":"Milvus-CWpae","sourceHandle":"CSVLoader|CSVLoader-854c0|Document","targetHandle":"Document|documents|Milvus-CWpae","id":"reactflow__edge-CSVLoader-854c0CSVLoader|CSVLoader-854c0|Document-Milvus-CWpaeDocument|documents|Milvus-CWpae","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"InputFileNode-e4363","target":"CSVLoader-854c0","sourceHandle":"InputFileNode|InputFileNode-e4363|fileNode","targetHandle":"fileNode|file_path|CSVLoader-854c0","id":"reactflow__edge-InputFileNode-e4363InputFileNode|InputFileNode-e4363|fileNode-CSVLoader-854c0fileNode|file_path|CSVLoader-854c0","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true}],"viewport":{"zoom":1,"x":0,"y":0}},"is_component":false,"name":"","description":"","id":"7db0b","status":0,"write":false,"guide_word":""}}}},{"输入网址进行问答":{"id":"RetrievalQA-731c5","type":"RetrievalQA","node":{"output_types":[],"display_name":"输入网址进行问答","documentation":"","base_classes":["RetrievalQA","Chain","BaseRetrievalQA","function"],"description":"Chain for question-answering against an index.","template":{"token_max_CombineDocsChain-RCPNA":{"l2":false,"info":"当前只对stuff 生效","list":false,"name":"token_max","show":true,"type":"int","value":-1,"l2_name":"token_max","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"token_max","proxy":{"id":"CombineDocsChain-RCPNA","field":"token_max"}},"chain_type_CombineDocsChain-RCPNA":{"l2":false,"info":"","list":true,"name":"chain_type","show":true,"type":"str","value":"stuff","l2_name":"chain_type","options":["stuff","map_reduce","map_rerank","refine"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"CombineDocsChain-RCPNA","field":"chain_type"},"display_name":"Chain Type"},"drop_old_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"drop_old","show":false,"type":"bool","value":false,"l2_name":"drop_old","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"drop_old"},"display_name":"Drop Old"},"metadatas_Milvus-CWpae":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"metadatas"},"display_name":"Metadatas"},"index_params_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"index_params","show":false,"type":"code","l2_name":"index_params","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"index_params"},"display_name":"Index Params"},"search_kwargs_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":false,"type":"code","value":"{}","l2_name":"search_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"search_kwargs"},"display_name":"Search Kwargs"},"search_params_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"search_params","show":false,"type":"code","l2_name":"search_params","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"search_params"},"display_name":"Search Params"},"collection_name_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"collection_name","show":true,"type":"str","value":"","l2_name":"选择知识库","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","collection_id":"","proxy":{"id":"Milvus-CWpae","field":"collection_name"},"display_name":"Collection Name"},"connection_args_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"connection_args","show":false,"type":"code","value":"","l2_name":"connection_args","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"connection_args"},"display_name":"Connection Args"},"consistency_level_Milvus-CWpae":{"l2":false,"info":"","list":false,"name":"consistency_level","show":false,"type":"str","value":"Session","l2_name":"consistency_level","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"Milvus-CWpae","field":"consistency_level"},"display_name":"Consistency Level"},"tags_RetrievalQA-x5nap":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"tags"},"display_name":"Tags"},"memory_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"memory","show":true,"type":"BaseMemory","l2_name":"memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"memory"},"display_name":"Memory"},"verbose_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"verbose"},"display_name":"Verbose"},"metadata_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"metadata"},"display_name":"Metadata"},"callbacks_RetrievalQA-x5nap":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"callbacks"},"display_name":"Callbacks"},"input_key_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"input_key","show":true,"type":"str","value":"query","l2_name":"input_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"input_key"},"display_name":"Input Key"},"input_node_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question","proxy":{"id":"RetrievalQA-x5nap","field":"input_node"}},"output_key_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"output_key","show":true,"type":"str","value":"result","l2_name":"output_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"output_key"},"display_name":"Output Key"},"return_source_documents_RetrievalQA-x5nap":{"l2":false,"info":"","list":false,"name":"return_source_documents","show":true,"type":"bool","value":true,"l2_name":"return_source_documents","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"RetrievalQA-x5nap","field":"return_source_documents"},"display_name":"Return Source Documents"},"n_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"l2_name":"n","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"n"},"display_name":"N"},"tags_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"tags"},"display_name":"Tags"},"cache_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"cache","show":false,"type":"bool","l2_name":"cache","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"cache"},"display_name":"Cache"},"top_p_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"l2_name":"top_p","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"top_p"},"display_name":"Top P"},"client_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"client","show":false,"type":"Any","l2_name":"client","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"client"},"display_name":"Client"},"headers_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","l2_name":"headers","advanced":true,"password":false,"required":false,"multiline":true,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"headers"},"display_name":"Headers"},"verbose_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"verbose"},"display_name":"Verbose"},"metadata_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"metadata"},"display_name":"Metadata"},"callbacks_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"callbacks"},"display_name":"Callbacks"},"streaming_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"l2_name":"streaming","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"streaming"},"display_name":"Streaming"},"max_tokens_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","l2_name":"max_tokens","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"max_tokens"},"display_name":"Max Tokens"},"model_name_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-4","l2_name":"model_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"model_name"},"display_name":"Model Name"},"max_retries_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"l2_name":"max_retries","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"max_retries"},"display_name":"Max Retries"},"temperature_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"l2_name":"temperature","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"temperature"},"display_name":"Temperature"},"model_kwargs_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"model_kwargs","show":false,"type":"code","l2_name":"model_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"model_kwargs"},"display_name":"Model Kwargs"},"elemai_api_key_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","l2_name":"elemai_api_key","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"elemai_api_key"},"display_name":"Elemai Api Key"},"elemai_base_url_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","l2_name":"elemai_base_url","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"elemai_base_url"},"display_name":"Elemai Base Url"},"request_timeout_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"request_timeout","show":false,"type":"float","l2_name":"request_timeout","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"request_timeout"},"display_name":"Request Timeout"},"tiktoken_model_name_ProxyChatLLM-UYoB4":{"l2":false,"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","l2_name":"tiktoken_model_name","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-UYoB4","field":"tiktoken_model_name"},"display_name":"Tiktoken Model Name"},"metadata_WebBaseLoader-af6db":{"info":"","list":false,"name":"metadata","show":false,"type":"code","value":"{}","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Metadata","proxy":{"id":"WebBaseLoader-af6db","field":"metadata"}},"web_path_WebBaseLoader-af6db":{"l2":true,"info":"","list":false,"name":"web_path","show":true,"type":"str","value":"https://dataelem.feishu.cn/wiki/ZxW6wZyAJicX4WkG0NqcWsbynde","l2_name":"网址","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Web Page","proxy":{"id":"WebBaseLoader-af6db","field":"web_path"}},"proxy_url_OpenAIProxyEmbedding-1771b":{"info":"","list":false,"name":"proxy_url","show":false,"type":"str","value":"","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"http://43.133.35.137:8080","proxy":{"id":"OpenAIProxyEmbedding-1771b","field":"proxy_url"},"display_name":"Proxy Url"}},"flow":{"data":{"nodes":[{"width":384,"height":376,"id":"CombineDocsChain-RCPNA","type":"genericNode","position":{"x":-1187.1604122631927,"y":-662.1420645633651},"data":{"id":"CombineDocsChain-RCPNA","node":{"template":{"llm":{"l2":false,"info":"","list":false,"name":"llm","show":true,"type":"BaseLanguageModel","l2_name":"llm","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"LLM"},"_type":"load_qa_chain","token_max":{"l2":false,"info":"当前只对stuff 生效","list":false,"name":"token_max","show":true,"type":"int","value":-1,"l2_name":"token_max","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"token_max"},"chain_type":{"l2":false,"info":"","list":true,"name":"chain_type","show":true,"type":"str","value":"stuff","l2_name":"chain_type","options":["stuff","map_reduce","map_rerank","refine"],"advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""}},"description":"Load question answering chain.","base_classes":["BaseCombineDocumentsChain","function"],"display_name":"CombineDocsChain","output_types":[],"custom_fields":{},"documentation":""},"type":"CombineDocsChain","value":null},"selected":true,"positionAbsolute":{"x":-1187.1604122631927,"y":-662.1420645633651}},{"width":384,"height":346,"id":"Milvus-CWpae","type":"genericNode","position":{"x":-1174.4689035846664,"y":-119.56186107292496},"data":{"id":"Milvus-CWpae","node":{"l2_name":"知识库","template":{"_type":"Milvus","drop_old":{"l2":false,"info":"","list":false,"name":"drop_old","show":false,"type":"bool","value":false,"l2_name":"drop_old","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"documents":{"l2":false,"info":"","list":true,"name":"documents","show":true,"type":"Document","l2_name":"documents","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Documents"},"embedding":{"l2":false,"info":"","list":false,"name":"embedding","show":true,"type":"Embeddings","l2_name":"embedding","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Embedding"},"metadatas":{"l2":false,"info":"","list":true,"name":"metadatas","show":false,"type":"code","l2_name":"metadatas","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"index_params":{"l2":false,"info":"","list":false,"name":"index_params","show":false,"type":"code","l2_name":"index_params","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"search_kwargs":{"l2":false,"info":"","list":false,"name":"search_kwargs","show":true,"type":"code","value":"{}","l2_name":"search_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"search_params":{"l2":false,"info":"","list":false,"name":"search_params","show":false,"type":"code","l2_name":"search_params","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"collection_name":{"l2":false,"info":"","list":false,"name":"collection_name","show":true,"type":"str","value":"","l2_name":"选择知识库","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","collection_id":""},"connection_args":{"l2":false,"info":"","list":false,"name":"connection_args","show":true,"type":"code","value":"","l2_name":"connection_args","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"consistency_level":{"l2":false,"info":"","list":false,"name":"consistency_level","show":false,"type":"str","value":"Session","l2_name":"consistency_level","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Create a Milvus collection, indexes it with HNSW, and insert data.","base_classes":["Milvus","VectorStore","BaseRetriever","VectorStoreRetriever"],"display_name":"Milvus","output_types":[],"custom_fields":{},"documentation":"http://192.168.106.116:8030"},"type":"Milvus","value":null},"selected":true,"positionAbsolute":{"x":-1174.4689035846664,"y":-119.56186107292496}},{"width":384,"height":316,"id":"RetrievalQA-x5nap","type":"genericNode","position":{"x":-510.8891021059987,"y":-499.2694176355641},"data":{"id":"RetrievalQA-x5nap","node":{"template":{"tags":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"RetrievalQA","memory":{"l2":false,"info":"","list":false,"name":"memory","show":true,"type":"BaseMemory","l2_name":"memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"verbose":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"metadata":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_key":{"l2":false,"info":"","list":false,"name":"input_key","show":true,"type":"str","value":"query","l2_name":"input_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":""},"retriever":{"l2":false,"info":"","list":false,"name":"retriever","show":true,"type":"BaseRetriever","l2_name":"retriever","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"input_node":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question"},"output_key":{"l2":false,"info":"","list":false,"name":"output_key","show":true,"type":"str","value":"result","l2_name":"output_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":""},"combine_documents_chain":{"l2":false,"info":"","list":false,"name":"combine_documents_chain","show":true,"type":"BaseCombineDocumentsChain","l2_name":"combine_documents_chain","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"return_source_documents":{"l2":false,"info":"","list":false,"name":"return_source_documents","show":true,"type":"bool","value":true,"l2_name":"return_source_documents","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Chain for question-answering against an index.","base_classes":["RetrievalQA","Chain","BaseRetrievalQA","function"],"display_name":"RetrievalQA","output_types":[],"custom_fields":{},"documentation":"https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"},"type":"RetrievalQA","value":null},"selected":true,"positionAbsolute":{"x":-510.8891021059987,"y":-499.2694176355641}},{"width":384,"height":500,"id":"ProxyChatLLM-UYoB4","type":"genericNode","position":{"x":-1789.0051284621989,"y":-759.1155283370271},"data":{"id":"ProxyChatLLM-UYoB4","node":{"template":{"n":{"l2":false,"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"l2_name":"n","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tags":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"ProxyChatLLM","cache":{"l2":false,"info":"","list":false,"name":"cache","show":false,"type":"bool","l2_name":"cache","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"top_p":{"l2":false,"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"l2_name":"top_p","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"client":{"l2":false,"info":"","list":false,"name":"client","show":false,"type":"Any","l2_name":"client","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"headers":{"l2":false,"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","l2_name":"headers","advanced":false,"password":false,"required":false,"multiline":true,"placeholder":""},"verbose":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"metadata":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"streaming":{"l2":false,"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"l2_name":"streaming","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_tokens":{"l2":false,"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","l2_name":"max_tokens","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":""},"model_name":{"l2":false,"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-4","l2_name":"model_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_retries":{"l2":false,"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"l2_name":"max_retries","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"temperature":{"l2":false,"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"l2_name":"temperature","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"model_kwargs":{"l2":false,"info":"","list":false,"name":"model_kwargs","show":true,"type":"code","l2_name":"model_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"elemai_api_key":{"l2":false,"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","l2_name":"elemai_api_key","advanced":false,"password":true,"required":false,"multiline":false,"placeholder":""},"elemai_base_url":{"l2":false,"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","l2_name":"elemai_base_url","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"request_timeout":{"l2":false,"info":"","list":false,"name":"request_timeout","show":false,"type":"float","l2_name":"request_timeout","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tiktoken_model_name":{"l2":false,"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","l2_name":"tiktoken_model_name","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Wrapper around proxy Chat large language models.","base_classes":["ProxyChatLLM","BaseLanguageModel","BaseChatModel","BaseLLM"],"display_name":"ProxyChatLLM","output_types":[],"custom_fields":{},"documentation":""},"type":"ProxyChatLLM","value":null},"selected":true,"positionAbsolute":{"x":-1789.0051284621989,"y":-759.1155283370271}},{"width":384,"height":356,"id":"WebBaseLoader-af6db","type":"genericNode","position":{"x":-1699.1181246063038,"y":-135.15223886710186},"data":{"id":"WebBaseLoader-af6db","node":{"l2_name":"输入网址","template":{"_type":"WebBaseLoader","metadata":{"info":"","list":false,"name":"metadata","show":true,"type":"code","value":"{}","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Metadata"},"web_path":{"l2":true,"info":"","list":false,"name":"web_path","show":true,"type":"str","value":"https://dataelem.feishu.cn/wiki/ZxW6wZyAJicX4WkG0NqcWsbynde","l2_name":"网址","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","display_name":"Web Page"}},"description":"Load HTML pages using `urllib` and parse them with `BeautifulSoup'.","base_classes":["Document"],"display_name":"WebBaseLoader","output_types":["Document"],"custom_fields":{},"documentation":"https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/web_base"},"type":"WebBaseLoader","value":null},"selected":true,"positionAbsolute":{"x":-1699.1181246063038,"y":-135.15223886710186}},{"width":384,"height":172,"id":"OpenAIProxyEmbedding-1771b","type":"genericNode","position":{"x":-1686.4214049605273,"y":261.6202500634144},"data":{"id":"OpenAIProxyEmbedding-1771b","node":{"template":{"_type":"proxy_embedding","proxy_url":{"info":"","list":false,"name":"proxy_url","show":false,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"http://43.133.35.137:8080"}},"description":" 使用自建的embedding服务使用openai进行embed ","base_classes":["Embeddings"],"display_name":"OpenAIProxyEmbedding","output_types":[],"custom_fields":{},"documentation":""},"type":"OpenAIProxyEmbedding","value":null},"selected":true,"positionAbsolute":{"x":-1686.4214049605273,"y":261.6202500634144}}],"edges":[{"source":"CombineDocsChain-RCPNA","target":"RetrievalQA-x5nap","sourceHandle":"CombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function","targetHandle":"BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap","id":"reactflow__edge-CombineDocsChain-RCPNACombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function-RetrievalQA-x5napBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"Milvus-CWpae","target":"RetrievalQA-x5nap","sourceHandle":"Milvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever","targetHandle":"BaseRetriever|retriever|RetrievalQA-x5nap","id":"reactflow__edge-Milvus-CWpaeMilvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever-RetrievalQA-x5napBaseRetriever|retriever|RetrievalQA-x5nap","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"ProxyChatLLM-UYoB4","target":"CombineDocsChain-RCPNA","sourceHandle":"ProxyChatLLM|ProxyChatLLM-UYoB4|ProxyChatLLM|BaseLanguageModel|BaseChatModel|BaseLLM","targetHandle":"BaseLanguageModel|llm|CombineDocsChain-RCPNA","id":"reactflow__edge-ProxyChatLLM-UYoB4ProxyChatLLM|ProxyChatLLM-UYoB4|ProxyChatLLM|BaseLanguageModel|BaseChatModel|BaseLLM-CombineDocsChain-RCPNABaseLanguageModel|llm|CombineDocsChain-RCPNA","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"WebBaseLoader-af6db","target":"Milvus-CWpae","sourceHandle":"WebBaseLoader|WebBaseLoader-af6db|Document","targetHandle":"Document|documents|Milvus-CWpae","id":"reactflow__edge-WebBaseLoader-af6dbWebBaseLoader|WebBaseLoader-af6db|Document-Milvus-CWpaeDocument|documents|Milvus-CWpae","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"OpenAIProxyEmbedding-1771b","target":"Milvus-CWpae","sourceHandle":"OpenAIProxyEmbedding|OpenAIProxyEmbedding-1771b|Embeddings","targetHandle":"Embeddings|embedding|Milvus-CWpae","id":"reactflow__edge-OpenAIProxyEmbedding-1771bOpenAIProxyEmbedding|OpenAIProxyEmbedding-1771b|Embeddings-Milvus-CWpaeEmbeddings|embedding|Milvus-CWpae","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true}],"viewport":{"zoom":1,"x":0,"y":0}},"is_component":false,"name":"","description":"","id":"ddc2b","status":0,"write":false,"guide_word":""}}}},{"角色扮演":{"id":"LLMChain-2dd3b","type":"LLMChain","node":{"output_types":[],"display_name":"角色扮演","documentation":"","base_classes":["Chain","LLMChain","function"],"description":"Chain to run queries against LLMs.","template":{"tags_LLMChain-wRz0c":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"LLMChain-wRz0c","field":"tags"},"display_name":"Tags"},"verbose_LLMChain-wRz0c":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"LLMChain-wRz0c","field":"verbose"},"display_name":"Verbose"},"metadata_LLMChain-wRz0c":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"LLMChain-wRz0c","field":"metadata"},"display_name":"Metadata"},"callbacks_LLMChain-wRz0c":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"LLMChain-wRz0c","field":"callbacks"},"display_name":"Callbacks"},"input_node_LLMChain-wRz0c":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question","proxy":{"id":"LLMChain-wRz0c","field":"input_node"}},"llm_kwargs_LLMChain-wRz0c":{"l2":false,"info":"","list":false,"name":"llm_kwargs","show":false,"type":"code","l2_name":"llm_kwargs","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"LLMChain-wRz0c","field":"llm_kwargs"},"display_name":"LLM Kwargs"},"output_key_LLMChain-wRz0c":{"l2":false,"info":"","list":false,"name":"output_key","show":true,"type":"str","value":"text","l2_name":"output_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"LLMChain-wRz0c","field":"output_key"},"display_name":"Output Key"},"output_parser_LLMChain-wRz0c":{"l2":false,"info":"","list":false,"name":"output_parser","show":false,"type":"BaseLLMOutputParser","l2_name":"output_parser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"LLMChain-wRz0c","field":"output_parser"},"display_name":"Output Parser"},"return_final_only_LLMChain-wRz0c":{"l2":false,"info":"","list":false,"name":"return_final_only","show":false,"type":"bool","value":true,"l2_name":"return_final_only","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"LLMChain-wRz0c","field":"return_final_only"},"display_name":"Return Final Only"},"input_PromptTemplate-ZoUVT":{"info":"","list":false,"name":"input","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser"],"placeholder":"","display_name":"input","proxy":{"id":"PromptTemplate-ZoUVT","field":"input"}},"template_PromptTemplate-ZoUVT":{"l2":false,"info":"","list":false,"name":"template","show":true,"type":"prompt","value":"System：\n你将扮演一个科技公司的面试官，面试者是一名产品经理，提出 3 个犀利的问题。\n请注意：\n- 每次只问一个问题\n- 面试者回答问题后请简短点评一下，然后问下一个问题，不要试图纠正候选人的错误；\n- 如果你认为用户连续几次回答的都不对，就少问一点；\n- 问完最后一个问题后，你可以问这样一个问题：上一份工作为什么离职？用户回答该问题后，请表示理解与支持，并停止与面试者的对话。若面试者继续说话你可以礼貌拒绝。\n\n以下是你们的对话记录\n\n{chat_history}\n\n面试者：{input}\n\n你的回答：","l2_name":"template","advanced":false,"password":false,"required":true,"multiline":true,"placeholder":"","proxy":{"id":"PromptTemplate-ZoUVT","field":"template"},"display_name":"Template"},"chat_history_PromptTemplate-ZoUVT":{"info":"","list":false,"name":"chat_history","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser"],"placeholder":"","display_name":"chat_history","proxy":{"id":"PromptTemplate-ZoUVT","field":"chat_history"}},"output_parser_PromptTemplate-ZoUVT":{"l2":false,"info":"","list":false,"name":"output_parser","show":false,"type":"BaseOutputParser","l2_name":"output_parser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-ZoUVT","field":"output_parser"},"display_name":"Output Parser"},"input_variables_PromptTemplate-ZoUVT":{"l2":false,"info":"","list":true,"name":"input_variables","show":false,"type":"str","value":["chat_history","input"],"l2_name":"input_variables","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-ZoUVT","field":"input_variables"},"display_name":"Input Variables"},"template_format_PromptTemplate-ZoUVT":{"l2":false,"info":"","list":false,"name":"template_format","show":false,"type":"str","value":"f-string","l2_name":"template_format","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-ZoUVT","field":"template_format"},"display_name":"Template Format"},"partial_variables_PromptTemplate-ZoUVT":{"l2":false,"info":"","list":false,"name":"partial_variables","show":false,"type":"code","l2_name":"partial_variables","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-ZoUVT","field":"partial_variables"},"display_name":"Partial Variables"},"validate_template_PromptTemplate-ZoUVT":{"l2":false,"info":"","list":false,"name":"validate_template","show":false,"type":"bool","value":true,"l2_name":"validate_template","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"PromptTemplate-ZoUVT","field":"validate_template"},"display_name":"Validate Template"},"ai_prefix_ConversationBufferMemory-lI5Pt":{"l2":false,"info":"","list":false,"name":"ai_prefix","show":false,"type":"str","value":"AI","l2_name":"ai_prefix","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-lI5Pt","field":"ai_prefix"},"display_name":"Ai Prefix"},"input_key_ConversationBufferMemory-lI5Pt":{"l2":false,"info":"The variable to be used as Chat Input when more than one variable is available.","list":false,"name":"input_key","show":true,"type":"str","value":"","l2_name":"input_key","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-lI5Pt","field":"input_key"},"display_name":"Input Key"},"memory_key_ConversationBufferMemory-lI5Pt":{"l2":false,"info":"","list":false,"name":"memory_key","show":true,"type":"str","value":"chat_history","l2_name":"memory_key","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-lI5Pt","field":"memory_key"},"display_name":"Memory Key"},"output_key_ConversationBufferMemory-lI5Pt":{"l2":false,"info":"The variable to be used as Chat Output (e.g. answer in a ConversationalRetrievalChain)","list":false,"name":"output_key","show":true,"type":"str","value":"","l2_name":"output_key","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-lI5Pt","field":"output_key"},"display_name":"Output Key"},"chat_memory_ConversationBufferMemory-lI5Pt":{"l2":false,"info":"","list":false,"name":"chat_memory","show":true,"type":"BaseChatMessageHistory","l2_name":"chat_memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-lI5Pt","field":"chat_memory"},"display_name":"Chat Memory"},"human_prefix_ConversationBufferMemory-lI5Pt":{"l2":false,"info":"","list":false,"name":"human_prefix","show":false,"type":"str","value":"Human","l2_name":"human_prefix","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-lI5Pt","field":"human_prefix"},"display_name":"Human Prefix"},"return_messages_ConversationBufferMemory-lI5Pt":{"l2":false,"info":"","list":false,"name":"return_messages","show":true,"type":"bool","l2_name":"return_messages","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ConversationBufferMemory-lI5Pt","field":"return_messages"},"display_name":"Return Messages"},"n_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"n"},"display_name":"N"},"tags_ProxyChatLLM-C0lLH":{"info":"","list":true,"name":"tags","show":false,"type":"str","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"tags"},"display_name":"Tags"},"cache_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"cache","show":false,"type":"bool","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"cache"},"display_name":"Cache"},"top_p_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"top_p"},"display_name":"Top P"},"client_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"client","show":false,"type":"Any","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"client"},"display_name":"Client"},"headers_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","advanced":true,"password":false,"required":false,"multiline":true,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"headers"},"display_name":"Headers"},"verbose_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"verbose"},"display_name":"Verbose"},"metadata_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"metadata","show":false,"type":"code","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"metadata"},"display_name":"Metadata"},"callbacks_ProxyChatLLM-C0lLH":{"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"callbacks"},"display_name":"Callbacks"},"streaming_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"streaming"},"display_name":"Streaming"},"max_tokens_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"max_tokens"},"display_name":"Max Tokens"},"model_name_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-3.5-turbo-16k-0613","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"model_name"},"display_name":"Model Name"},"max_retries_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"max_retries"},"display_name":"Max Retries"},"temperature_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"temperature"},"display_name":"Temperature"},"model_kwargs_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"model_kwargs","show":false,"type":"code","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"model_kwargs"},"display_name":"Model Kwargs"},"elemai_api_key_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"elemai_api_key"},"display_name":"Elemai Api Key"},"elemai_base_url_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"elemai_base_url"},"display_name":"Elemai Base Url"},"request_timeout_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"request_timeout","show":false,"type":"float","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"request_timeout"},"display_name":"Request Timeout"},"tiktoken_model_name_ProxyChatLLM-C0lLH":{"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":"","proxy":{"id":"ProxyChatLLM-C0lLH","field":"tiktoken_model_name"},"display_name":"Tiktoken Model Name"}},"flow":{"data":{"nodes":[{"width":384,"height":316,"id":"LLMChain-wRz0c","type":"genericNode","position":{"x":621.1034022791711,"y":113.49419945635839},"data":{"id":"LLMChain-wRz0c","node":{"template":{"llm":{"l2":false,"info":"","list":false,"name":"llm","show":true,"type":"BaseLanguageModel","l2_name":"llm","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"tags":{"l2":false,"info":"","list":true,"name":"tags","show":false,"type":"str","l2_name":"tags","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"LLMChain","memory":{"l2":false,"info":"","list":false,"name":"memory","show":true,"type":"BaseMemory","l2_name":"memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"prompt":{"l2":false,"info":"","list":false,"name":"prompt","show":true,"type":"BasePromptTemplate","l2_name":"prompt","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"verbose":{"l2":false,"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"l2_name":"verbose","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"metadata":{"l2":false,"info":"","list":false,"name":"metadata","show":false,"type":"code","l2_name":"metadata","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"l2":false,"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","l2_name":"callbacks","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_node":{"l2":false,"info":"","list":false,"name":"input_node","show":true,"type":"input","l2_name":"input_node","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":"","display_name":"Preset Question"},"llm_kwargs":{"l2":false,"info":"","list":false,"name":"llm_kwargs","show":false,"type":"code","l2_name":"llm_kwargs","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"output_key":{"l2":false,"info":"","list":false,"name":"output_key","show":true,"type":"str","value":"text","l2_name":"output_key","advanced":true,"password":false,"required":true,"multiline":false,"placeholder":""},"output_parser":{"l2":false,"info":"","list":false,"name":"output_parser","show":false,"type":"BaseLLMOutputParser","l2_name":"output_parser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"return_final_only":{"l2":false,"info":"","list":false,"name":"return_final_only","show":false,"type":"bool","value":true,"l2_name":"return_final_only","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Chain to run queries against LLMs.","base_classes":["Chain","LLMChain","function"],"display_name":"LLMChain","output_types":[],"custom_fields":{},"documentation":"https://python.langchain.com/docs/modules/chains/foundational/llm_chain"},"type":"LLMChain","value":null},"selected":true,"positionAbsolute":{"x":621.1034022791711,"y":113.49419945635839}},{"width":384,"height":418,"id":"PromptTemplate-ZoUVT","type":"genericNode","position":{"x":-152.59469279870132,"y":532.7389820028246},"data":{"id":"PromptTemplate-ZoUVT","node":{"name":"","template":{"_type":"PromptTemplate","input":{"info":"","list":false,"name":"input","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser"],"placeholder":"","display_name":"input"},"template":{"l2":false,"info":"","list":false,"name":"template","show":true,"type":"prompt","value":"System：\n你将扮演一个科技公司的面试官，面试者是一名产品经理，提出 3 个犀利的问题。\n请注意：\n- 每次只问一个问题\n- 面试者回答问题后请简短点评一下，然后问下一个问题，不要试图纠正候选人的错误；\n- 如果你认为用户连续几次回答的都不对，就少问一点；\n- 问完最后一个问题后，你可以问这样一个问题：上一份工作为什么离职？用户回答该问题后，请表示理解与支持，并停止与面试者的对话。若面试者继续说话你可以礼貌拒绝。\n\n以下是你们的对话记录\n\n{chat_history}\n\n面试者：{input}\n\n你的回答：","l2_name":"template","advanced":false,"password":false,"required":true,"multiline":true,"placeholder":""},"chat_history":{"info":"","list":false,"name":"chat_history","show":true,"type":"str","value":"","advanced":false,"password":false,"required":false,"multiline":true,"input_types":["Document","BaseOutputParser"],"placeholder":"","display_name":"chat_history"},"output_parser":{"l2":false,"info":"","list":false,"name":"output_parser","show":false,"type":"BaseOutputParser","l2_name":"output_parser","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_variables":{"l2":false,"info":"","list":true,"name":"input_variables","show":false,"type":"str","value":["chat_history","input"],"l2_name":"input_variables","advanced":false,"password":false,"required":true,"multiline":false,"placeholder":""},"template_format":{"l2":false,"info":"","list":false,"name":"template_format","show":false,"type":"str","value":"f-string","l2_name":"template_format","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"partial_variables":{"l2":false,"info":"","list":false,"name":"partial_variables","show":false,"type":"code","l2_name":"partial_variables","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"validate_template":{"l2":false,"info":"","list":false,"name":"validate_template","show":false,"type":"bool","value":true,"l2_name":"validate_template","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Schema to represent a prompt for an LLM.","base_classes":["PromptTemplate","StringPromptTemplate","BasePromptTemplate"],"display_name":"PromptTemplate","output_types":[],"custom_fields":{"template":["chat_history","input"]},"documentation":"https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/","field_formatters":{"formatters":{"openai_api_key":{}},"base_formatters":{"dict":{},"list":{},"show":{},"union":{},"kwargs":{},"default":{},"headers":{},"optional":{},"password":{},"multiline":{},"model_fields":{"MODEL_DICT":{"OpenAI":["text-davinci-003","text-davinci-002","text-curie-001","text-babbage-001","text-ada-001"],"Anthropic":["claude-v1","claude-v1-100k","claude-instant-v1","claude-instant-v1-100k","claude-v1.3","claude-v1.3-100k","claude-v1.2","claude-v1.0","claude-instant-v1.1","claude-instant-v1.1-100k","claude-instant-v1.0"],"ChatOpenAI":["gpt-3.5-turbo-0613","gpt-3.5-turbo","gpt-3.5-turbo-16k-0613","gpt-3.5-turbo-16k","gpt-4-0613","gpt-4-32k-0613","gpt-4","gpt-4-32k"],"ChatAnthropic":["claude-v1","claude-v1-100k","claude-instant-v1","claude-instant-v1-100k","claude-v1.3","claude-v1.3-100k","claude-v1.2","claude-v1.0","claude-instant-v1.1","claude-instant-v1.1-100k","claude-instant-v1.0"]}},"dict_code_file":{}}}},"type":"PromptTemplate","value":null},"selected":true,"positionAbsolute":{"x":-152.59469279870132,"y":532.7389820028246}},{"width":384,"height":526,"id":"ConversationBufferMemory-lI5Pt","type":"genericNode","position":{"x":-556.5758769509117,"y":-30.001495256074236},"data":{"id":"ConversationBufferMemory-lI5Pt","node":{"template":{"_type":"ConversationBufferMemory","ai_prefix":{"l2":false,"info":"","list":false,"name":"ai_prefix","show":false,"type":"str","value":"AI","l2_name":"ai_prefix","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"input_key":{"l2":false,"info":"The variable to be used as Chat Input when more than one variable is available.","list":false,"name":"input_key","show":true,"type":"str","value":"","l2_name":"input_key","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"memory_key":{"l2":false,"info":"","list":false,"name":"memory_key","show":true,"type":"str","value":"chat_history","l2_name":"memory_key","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"output_key":{"l2":false,"info":"The variable to be used as Chat Output (e.g. answer in a ConversationalRetrievalChain)","list":false,"name":"output_key","show":true,"type":"str","value":"","l2_name":"output_key","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"chat_memory":{"l2":false,"info":"","list":false,"name":"chat_memory","show":true,"type":"BaseChatMessageHistory","l2_name":"chat_memory","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"human_prefix":{"l2":false,"info":"","list":false,"name":"human_prefix","show":false,"type":"str","value":"Human","l2_name":"human_prefix","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"return_messages":{"l2":false,"info":"","list":false,"name":"return_messages","show":true,"type":"bool","l2_name":"return_messages","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Buffer for storing conversation memory.","base_classes":["BaseChatMemory","ConversationBufferMemory","BaseMemory"],"display_name":"ConversationBufferMemory","output_types":[],"custom_fields":{},"documentation":"https://python.langchain.com/docs/modules/memory/how_to/buffer"},"type":"ConversationBufferMemory","value":null},"selected":true,"positionAbsolute":{"x":-556.5758769509117,"y":-30.001495256074236}},{"width":384,"height":500,"id":"ProxyChatLLM-C0lLH","type":"genericNode","position":{"x":-26.90943659439222,"y":-434.6186906908237},"data":{"id":"ProxyChatLLM-C0lLH","node":{"template":{"n":{"info":"","list":false,"name":"n","show":false,"type":"int","value":1,"advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tags":{"info":"","list":true,"name":"tags","show":false,"type":"str","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"_type":"ProxyChatLLM","cache":{"info":"","list":false,"name":"cache","show":false,"type":"bool","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"top_p":{"info":"","list":false,"name":"top_p","show":true,"type":"float","value":0.9,"advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"client":{"info":"","list":false,"name":"client","show":false,"type":"Any","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"headers":{"info":"","list":false,"name":"headers","show":false,"type":"code","value":"","advanced":false,"password":false,"required":false,"multiline":true,"placeholder":""},"verbose":{"info":"","list":false,"name":"verbose","show":false,"type":"bool","value":false,"advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"metadata":{"info":"","list":false,"name":"metadata","show":false,"type":"code","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"callbacks":{"info":"","list":true,"name":"callbacks","show":false,"type":"langchain.callbacks.base.BaseCallbackHandler","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"streaming":{"info":"","list":false,"name":"streaming","show":false,"type":"bool","value":false,"advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_tokens":{"info":"","list":false,"name":"max_tokens","show":true,"type":"int","value":"","advanced":true,"password":true,"required":false,"multiline":false,"placeholder":""},"model_name":{"info":"","list":false,"name":"model_name","show":true,"type":"str","value":"gpt-3.5-turbo-16k-0613","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"max_retries":{"info":"","list":false,"name":"max_retries","show":false,"type":"int","value":6,"advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"temperature":{"info":"","list":false,"name":"temperature","show":true,"type":"float","value":0.7,"advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"model_kwargs":{"info":"","list":false,"name":"model_kwargs","show":true,"type":"code","advanced":true,"password":false,"required":false,"multiline":false,"placeholder":""},"elemai_api_key":{"info":"","list":false,"name":"elemai_api_key","show":true,"type":"str","value":"","advanced":false,"password":true,"required":false,"multiline":false,"placeholder":""},"elemai_base_url":{"info":"","list":false,"name":"elemai_base_url","show":true,"type":"str","value":"http://192.168.106.12:6001/v1/chatcompletion","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"request_timeout":{"info":"","list":false,"name":"request_timeout","show":false,"type":"float","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""},"tiktoken_model_name":{"info":"","list":false,"name":"tiktoken_model_name","show":false,"type":"str","advanced":false,"password":false,"required":false,"multiline":false,"placeholder":""}},"description":"Wrapper around proxy Chat large language models.","base_classes":["ProxyChatLLM","BaseLanguageModel","BaseChatModel","BaseLLM"],"display_name":"ProxyChatLLM","output_types":[],"custom_fields":{},"documentation":""},"type":"ProxyChatLLM","value":null},"selected":true,"positionAbsolute":{"x":-26.90943659439222,"y":-434.6186906908237}}],"edges":[{"source":"PromptTemplate-ZoUVT","target":"LLMChain-wRz0c","sourceHandle":"PromptTemplate|PromptTemplate-ZoUVT|PromptTemplate|StringPromptTemplate|BasePromptTemplate","targetHandle":"BasePromptTemplate|prompt|LLMChain-wRz0c","id":"reactflow__edge-PromptTemplate-ZoUVTPromptTemplate|PromptTemplate-ZoUVT|PromptTemplate|StringPromptTemplate|BasePromptTemplate-LLMChain-wRz0cBasePromptTemplate|prompt|LLMChain-wRz0c","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"ConversationBufferMemory-lI5Pt","target":"LLMChain-wRz0c","sourceHandle":"ConversationBufferMemory|ConversationBufferMemory-lI5Pt|BaseChatMemory|ConversationBufferMemory|BaseMemory","targetHandle":"BaseMemory|memory|LLMChain-wRz0c","id":"reactflow__edge-ConversationBufferMemory-lI5PtConversationBufferMemory|ConversationBufferMemory-lI5Pt|BaseChatMemory|ConversationBufferMemory|BaseMemory-LLMChain-wRz0cBaseMemory|memory|LLMChain-wRz0c","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true},{"source":"ProxyChatLLM-C0lLH","target":"LLMChain-wRz0c","sourceHandle":"ProxyChatLLM|ProxyChatLLM-C0lLH|ProxyChatLLM|BaseLanguageModel|BaseChatModel|BaseLLM","targetHandle":"BaseLanguageModel|llm|LLMChain-wRz0c","id":"reactflow__edge-ProxyChatLLM-C0lLHProxyChatLLM|ProxyChatLLM-C0lLH|ProxyChatLLM|BaseLanguageModel|BaseChatModel|BaseLLM-LLMChain-wRz0cBaseLanguageModel|llm|LLMChain-wRz0c","style":{"stroke":"#555"},"className":"stroke-gray-900 ","animated":false,"selected":true}],"viewport":{"zoom":1,"x":0,"y":0}},"is_component":false,"name":"","description":"","id":"32a69","status":0,"write":false,"guide_word":""}}}}]
================================================
FILE: src/backend/bisheng/database/data/sft_model.json
================================================
[
{
"model_name": "Baichuan2-7B-Chat"
},
{
"model_name": "Baichuan2-13B-Chat"
},
{
"model_name": "chatglm2-6b"
},
{
"model_name": "chatglm3-6b"
},
{
"model_name": "internlm-chat-7b-8k"
},
{
"model_name": "internlm-20b-chat"
},
{
"model_name": "Llama-2-7b-chat-hf"
},
{
"model_name": "Llama-2-13b-chat-hf"
},
{
"model_name": "Qwen-7B-Chat"
},
{
"model_name": "Qwen-14B-Chat"
}
]
================================================
FILE: src/backend/bisheng/database/data/t_gpts_tools.json
================================================
[
{
"name": "时间",
"logo": null,
"desc": "返回当前时间，如果未指定时区，将使用 UTC+8 时间",
"tool_key": "get_current_time",
"type": 1,
"extra": "{}",
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:36:00",
"update_time": "2024-03-29 14:36:00",
"id": 1,
"api_params": []
},
{
"name": "计算器",
"logo": null,
"desc": "使用 Python 内置工具进行数学表达式计算",
"tool_key": "calculator",
"type": 2,
"extra": "{}",
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:37:11",
"update_time": "2024-03-29 14:37:11",
"id": 2,
"api_params": []
},
{
"name": "论文获取",
"logo": null,
"desc": "从 Arxiv 网站检索论文的工具，输入为检索关键词。",
"tool_key": "arxiv",
"type": 3,
"extra": "{}",
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:39:37",
"update_time": "2024-03-29 14:39:37",
"id": 3,
"api_params": [
{
"name": "query",
"description": "search query to look up",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "Dalle3绘画",
"logo": null,
"desc": "根据提示词生成图像。",
"tool_key": "dalle_image_generator",
"type": 4,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:40:32",
"update_time": "2024-03-29 14:40:32",
"id": 4,
"api_params": [
{
"name": "query",
"description": "Description about image.",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "Bing web搜索",
"logo": null,
"desc": "使用 query 进行 Bing 搜索并返回返回结果。",
"tool_key": "bing_search",
"type": 5,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:41:16",
"update_time": "2024-03-29 14:41:16",
"id": 5,
"api_params": [
{
"name": "query",
"description": "query to look up in Bing search",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "代码执行器",
"logo": null,
"desc": "可以执行 Python 和 shell 代码，并在 600 秒内返回结果，适合处理数学、编程、图片和文件等问题场景。",
"tool_key": "bisheng_code_interpreter",
"type": 6,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:42:17",
"update_time": "2024-03-29 14:42:17",
"id": 6,
"api_params": [
{
"name": "python_code",
"description": "The pure python script to be evaluated. \\nThe contents will be in main.py. \\nIt should not be in markdown format.",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "人员所有公司",
"logo": null,
"desc": "可以通过公司名称或ID和人名获取企业人员的所有相关公司，包括其担任法人、股东、董监高的公司信息",
"tool_key": "tianyancha_all_companys_by_company",
"type": 7,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:44:34",
"update_time": "2024-03-29 14:44:34",
"id": 7,
"api_params": [
{
"name": "query",
"description": "human who you want to search",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "name",
"description": "company name which human worked",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "搜索企业",
"logo": null,
"desc": "可以通过关键词获取企业列表，企业列表包括公司名称或ID、类型、成立日期、经营状态、统一社会信用代码等字段的详细信息",
"tool_key": "tianyancha_search_company",
"type": 7,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:50:07",
"update_time": "2024-03-29 14:50:07",
"id": 9,
"api_params": [
{
"name": "query",
"description": "搜索关键字（公司名称、公司id、注册号或社会统一信用代码）",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "企业基本信息",
"logo": null,
"desc": "可以通过公司名称或ID获取企业基本信息，企业基本信息包括公司名称或ID、类型、成立日期、经营状态、注册资本、法人、工商注册号、统一社会信用代码、组织机构代码、纳税人识别号等字段信息",
"tool_key": "tianyancha_get_company_baseinfo",
"type": 7,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:51:06",
"update_time": "2024-03-29 14:51:06",
"id": 10,
"api_params": [
{
"name": "query",
"description": "搜索关键字（公司名称、公司id、注册号或社会统一信用代码）",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "企业知识产权信息",
"logo": null,
"desc": "可以通过公司名称或ID获取包含商标、专利、作品著作权、软件著作权、网站备案等维度的相关信息",
"tool_key": "tianyancha_ip_rights",
"type": 7,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:51:33",
"update_time": "2024-03-29 14:51:33",
"id": 11,
"api_params": [
{
"name": "query",
"description": "搜索关键字（公司名称、公司id、注册号或社会统一信用代码）",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "企业司法风险",
"logo": null,
"desc": "可以通过公司名称或ID获取包含法律诉讼、法院公告、开庭公告、失信人、被执行人、立案信息、送达公告等维度的相关信息",
"tool_key": "tianyancha_judicial_risk",
"type": 7,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:52:13",
"update_time": "2024-03-29 14:52:13",
"id": 12,
"api_params": [
{
"name": "query",
"description": "搜索关键字（公司名称、公司id、注册号或社会统一信用代码）",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "企业法律诉讼",
"logo": null,
"desc": "可以通过公司名称或ID获取企业法律诉讼信息，法律诉讼包括案件名称、案由、案件身份、案号等字段的详细信息",
"tool_key": "tianyancha_law_suit_case",
"type": 7,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:52:36",
"update_time": "2024-03-29 14:52:36",
"id": 13,
"api_params": [
{
"name": "query",
"description": "搜索关键字（公司名称、公司id、注册号或社会统一信用代码）",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "企业工商信息",
"logo": null,
"desc": "可以通过公司名称或ID获取包含企业基本信息、主要人员、股东信息、对外投资、分支机构等维度的相关信息",
"tool_key": "tianyancha_ic_info",
"type": 7,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:53:06",
"update_time": "2024-03-29 14:53:06",
"id": 14,
"api_params": [
{
"name": "query",
"description": "搜索关键字（公司名称、公司id、注册号或社会统一信用代码）",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "企业工商信息变更记录",
"logo": null,
"desc": "可以通过公司名称或ID获取企业变更记录，变更记录包括工商变更事项、变更前后信息等字段的详细信息",
"tool_key": "tianyancha_company_change_info",
"type": 7,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:54:02",
"update_time": "2024-03-29 14:54:02",
"id": 15,
"api_params": [
{
"name": "query",
"description": "搜索关键字（公司名称、公司id、注册号或社会统一信用代码）",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "企业股东",
"logo": null,
"desc": "可以通过公司名称或ID获取企业股东信息，股东信息包括股东名、出资比例、出资金额、股东总数等字段的详细信息",
"tool_key": "tianyancha_company_holders",
"type": 7,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:54:28",
"update_time": "2024-03-29 14:54:28",
"id": 16,
"api_params": [
{
"name": "query",
"description": "搜索关键字（公司名称、公司id、注册号或社会统一信用代码）",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "企业天眼风险",
"logo": null,
"desc": "可以通过关键字（公司名称、公司id、注册号或社会统一信用代码）获取企业相关天眼风险列表，包括企业自身/周边/预警风险信息。",
"tool_key": "tianyancha_riskinfo",
"type": 7,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:54:57",
"update_time": "2024-03-29 14:54:57",
"id": 17,
"api_params": [
{
"name": "query",
"description": "搜索关键字（公司名称、公司id、注册号或社会统一信用代码）",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "股票实时行情",
"logo": null,
"desc": "查询中国A股（沪深北交易所）股票或指数的实时行情数据，返回：现价、涨跌额、涨跌幅、成交量、成交额。",
"tool_key": "sina_realtime_info",
"type": 8,
"extra": null,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:56:15",
"update_time": "2024-03-29 14:56:15",
"id": 18,
"api_params": [
{
"name": "prefix",
"description": "前缀。如果是\"stock_symbol\"传入的为股票代码，则需要传入s_;\n如果\"stock_symbol\"传入的为指数代码，则为空。",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "stock_exchange",
"description": "交易所简写。股票上市的交易所，或者发布行情指数的交易所。可选项有\"sh\"(上海证券交易所)、\" sz\"( 深圳证券交易所)、\"bj\"( 北京证券交易所)",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "stock_symbol",
"description": "6位数字的股票或者指数代码。\\n参考信息：\\n- 如果问题中未给出，可能需要上网查询。\\n- 上交所股票通常以 6 开头，深交所股票通常以 0、3 开头，北交所股票通常以 8 开头。\\n- 上交所行情指数通常以 000 开头，深交所指数通常以 399 开头。同一个指数可能会同时在两个交易所发布，例如沪深 300 有\"sh000300\"和\"sz399300\"两个代码。",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "股票历史行情",
"logo": null,
"desc": "查询中国A股（沪深北交易所）股票或指数的的历史行情数据，返回：时间、开盘价、最高价、最低价、收盘价、成交量（股）。",
"tool_key": "sina_history_KLine",
"type": 8,
"extra": null,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:56:38",
"update_time": "2024-03-29 14:56:38",
"id": 19,
"api_params": [
{
"name": "date",
"description": "需要查询的时间，按照”2024-03-26“格式，传入日期",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "stock_exchange",
"description": "交易所简写。股票上市的交易所，或者发布行情指数的交易所。可选项有\"sh\"(上海证券交易所)、\" sz\"( 深圳证券交易所)、\"bj\"( 北京证券交易所)",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "stock_symbol",
"description": "6位数字的股票或者指数代码。\\n参考信息：\\n- 如果问题中未给出，可能需要上网查询。\\n- 上交所股票通常以 6 开头，深交所股票通常以 0、3 开头，北交所股票通常以 8 开头。\\n- 上交所行情指数通常以 000 开头，深交所指数通常以 399 开头。同一个指数可能会同时在两个交易所发布，例如沪深 300 有\"sh000300\"和\"sz399300\"两个代码。",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "社会融资规模增量",
"logo": null,
"desc": "中国社会融资规模增量月度统计数据。返回数据包括：月份，社会融资规模增量(单位：亿元)，社融增量分项数据。分项数据具体包括：人民币贷款，外币贷款，委托贷款，信托贷款，未贴现银行承兑汇票，企业债券，非金融企业境内股票融资。",
"tool_key": "macro_china_shrzgm",
"type": 8,
"extra": null,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:59:06",
"update_time": "2024-03-29 14:59:06",
"id": 20,
"api_params": [
{
"name": "start_date",
"description": "开始月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "end_date",
"description": "结束月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "GDP",
"logo": null,
"desc": " 中国国内生产总值（GDP）季度统计数据。返回数据包括：季度，当年累计GDP 绝对值及同比增长情况，第一、二、三产业 GDP 绝对值以及同比增长情况。",
"tool_key": "macro_china_gdp_yearly",
"type": 8,
"extra": null,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 14:59:38",
"update_time": "2024-03-29 14:59:38",
"id": 21,
"api_params": [
{
"name": "start_date",
"description": "开始月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "end_date",
"description": "结束月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "CPI",
"logo": null,
"desc": "中国居民消费价格指数(CPI，上年同月=100)月度统计数据。返回数据包括：月份，全国当月 CPI，全国当月同比增长，全国当月环比增长，全国当年 CPI 累计值；城市当月 CPI，城市当月同比增长，城市当月环比增长，城市当年 CPI 累计值；农村当月 CPI，农村当月同比增长，农村当月环比增长，农村当年 CPI 累计值。",
"tool_key": "macro_china_cpi",
"type": 8,
"extra": null,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 15:00:00",
"update_time": "2024-03-29 15:00:00",
"id": 22,
"api_params": [
{
"name": "start_date",
"description": "开始月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "end_date",
"description": "结束月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "PPI",
"logo": null,
"desc": "中国工业品出厂价格指数（PPI）月度统计数据。返回数据包括：月份，当月 PPI，当月同比增长，当年 PPI 累计值。",
"tool_key": "macro_china_ppi",
"type": 8,
"extra": null,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 15:00:24",
"update_time": "2024-03-29 15:00:24",
"id": 23,
"api_params": [
{
"name": "start_date",
"description": "开始月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "end_date",
"description": "结束月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "货币供应量",
"logo": null,
"desc": "中国货币供应量月度统计数据。返回数据包括：月份，当月 M0、M1、M2数量以及同、环比情况。",
"tool_key": "macro_china_money_supply",
"type": 8,
"extra": null,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 15:00:54",
"update_time": "2024-03-29 15:00:54",
"id": 24,
"api_params": [
{
"name": "start_date",
"description": "开始月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "end_date",
"description": "结束月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "社会消费品零售总额",
"logo": null,
"desc": "中国社会消费品零售总额月度统计数据。返回数据包括：月份，当月总额以及同、环比，当年累计总额以及同比情况。",
"tool_key": "macro_china_consumer_goods_retail",
"type": 8,
"extra": null,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-03-29 15:01:47",
"update_time": "2024-03-29 15:01:47",
"id": 25,
"api_params": [
{
"name": "start_date",
"description": "开始月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "end_date",
"description": "结束月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "PMI",
"logo": null,
"desc": "中国 PMI （采购经理人指数）月度统计数据。返回数据包括：月份制造业 PMI，制造业 PMI 同比增长，非制造业 PMI，非制造业 PMI 同比增长。",
"tool_key": "macro_china_pmi",
"type": 8,
"extra": null,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-04-15 12:01:11",
"update_time": "2024-04-15 12:01:11",
"id": 26,
"api_params": [
{
"name": "start_date",
"description": "开始月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "end_date",
"description": "结束月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "中美国债收益率",
"logo": null,
"desc": "本接口返回指定时间段[start_date,end_date]内交易日的中美两国的 2 年、5 年、10 年、30 年、10 年-2 年收益率数据。start_date表示起始日期，end_date表示结束日期，日期格式例如 2024-04-07",
"tool_key": "macro_bond_zh_us_rate",
"type": 8,
"extra": null,
"is_preset": 1,
"is_delete": 0,
"user_id": null,
"create_time": "2024-04-15 12:03:27",
"update_time": "2024-04-15 12:03:27",
"id": 27,
"api_params": [
{
"name": "start_date",
"description": "开始月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "end_date",
"description": "结束月份, 使用YYYY-MM-DD 方式表示",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
]
},
{
"name": "发送钉钉群消息",
"logo": null,
"desc": "通过钉钉群机器人，快速将消息推送到指定钉钉群组中",
"api_params": [
{
"in": "query",
"name": "url",
"schema": {
"type": "string"
},
"required": true,
"description": "自定义机器人的Wehhook地址"
},
{
"in": "query",
"name": "message",
"schema": {
"type": "string"
},
"required": true,
"description": "发送的文本消息内容"
}
],
"tool_key": "ding_send_message",
"type": 9,
"extra": null,
"is_preset": 1,
"is_delete": 0,
"user_id": 3,
"create_time": "2024-05-08 14:36:50",
"update_time": "2024-05-08 14:36:50",
"id": 28
},
{
"name": "单页面爬取|对应 Scrape 模式",
"logo": null,
"desc": "爬取并返回指定 URL 页面的内容，不会爬取子页面。",
"api_params": [
{
"in": "path",
"name": "target_url",
"schema": {
"type": "string"
},
"required": true,
"description": "要爬取的网站 url"
}
],
"tool_key": "fire_search_scrape",
"type": 10,
"extra": null,
"is_preset": 1,
"is_delete": 0,
"user_id": 3,
"create_time": "2024-05-08 14:36:50",
"update_time": "2025-02-27 19:05:31",
"id": 29
},
{
"name": "获取单网页",
"logo": null,
"desc": "爬取指定URL（支持pdf），并将其转换为适合大模型处理的markdown格式",
"api_params": [
{
"in": "path",
"name": "target_url",
"schema": {
"type": "string"
},
"required": true,
"description": "要获取的目标网页"
}
],
"tool_key": "jina_get_markdown",
"type": 11,
"is_preset": 1,
"is_delete": 0,
"user_id": 3,
"create_time": "2024-05-08 14:36:50",
"update_time": "2025-02-14 15:06:45",
"id": 30
},
{
"name": "Stable diffusion",
"logo": null,
"desc": "使用Stable Diffusion模型，根据用户提示词生成图像",
"api_params": [
{
"in": "path",
"name": "prompt",
"schema": {
"type": "string"
},
"required": true,
"description": "提示词生成图片描述词"
},
{
"in": "path",
"name": "negative_prompt",
"schema": {
"type": "string"
},
"required": false,
"description": "不希望图片包含的内容"
}
],
"tool_key": "silicon_stable_diffusion",
"type": 12,
"is_preset": 1,
"is_delete": 0,
"user_id": 3,
"create_time": "2024-05-08 14:36:50",
"update_time": "2025-02-20 11:47:25",
"id": 31
},
{
"name": "Flux",
"logo": null,
"desc": "使用Flux模型，根据用户提示词生成图像",
"api_params": [
{
"in": "path",
"name": "prompt",
"schema": {
"type": "string"
},
"required": true,
"description": "提示词生成图片描述词（建议使用英文）"
}
],
"tool_key": "silicon_flux",
"type": 12,
"is_preset": 1,
"is_delete": 0,
"user_id": 3,
"create_time": "2024-05-08 14:36:50",
"update_time": "2025-02-20 11:47:25",
"id": 32
},
{
"name": "向指定用户或者群聊发送消息",
"logo": null,
"desc": "向指定用户或者群聊发送消息",
"api_params": [
{
"in": "query",
"name": "message",
"schema": {
"type": "string"
},
"required": true,
"description": "发送的文本消息内容"
},
{
"in": "query",
"name": "receive_id",
"schema": {
"type": "string"
},
"required": true,
"description": "消息接收者的id"
},
{
"in": "query",
"name": "receive_id_type",
"schema": {
"type": "string"
},
"required": true,
"description": "用户id类型，可选值：open_id（标识一个用户在某个应用中的身份）；union_id（标识一个用户在某个应用开发商下的身份）；user_id（标识一个用户在某个租户内的身份）；email（以用户的真实邮箱来标识用户）；chat_id（以群 ID 来标识群聊）"
}
],
"tool_key": "feishu_send_message",
"type": 13,
"is_preset": 1,
"is_delete": 0,
"user_id": 3,
"create_time": "2024-05-08 14:36:50",
"update_time": "2025-02-24 14:35:04",
"id": 33
},
{
"name": "发送企业微信群消息",
"logo": null,
"desc": "通过企业微信群机器人，快速将消息推送到指定企业微信群组中",
"api_params": [
{
"in": "query",
"name": "url",
"schema": {
"type": "string"
},
"required": true,
"description": "微信机器人的webhook地址"
},
{
"in": "query",
"name": "message",
"schema": {
"type": "string"
},
"required": true,
"description": "发送的消息内容"
}
],
"tool_key": "wechat_send_message",
"type": 15,
"extra": "{}",
"is_preset": 1,
"is_delete": 0,
"user_id": 3,
"create_time": "2024-05-08 14:36:50",
"update_time": "2025-02-08 10:55:32",
"id": 34
},
{
"name": "发送邮件",
"logo": null,
"desc": "给单个或多个邮箱发送邮件（多个邮箱账号使用\",\"分隔）",
"api_params": [
{
"in": "query",
"name": "receiver",
"schema": {
"type": "string"
},
"required": true,
"description": "接受邮件的邮箱地址，确定邮件发送对象"
},
{
"in": "query",
"name": "subject",
"schema": {
"type": "string"
},
"required": false,
"description": "邮件主题"
},
{
"in": "query",
"name": "content",
"schema": {
"type": "string"
},
"required": true,
"description": "邮件的正文内容"
}
],
"tool_key": "email_send_email",
"type": 14,
"is_preset": 1,
"is_delete": 0,
"user_id": 3,
"create_time": "2024-05-08 14:36:50",
"update_time": "2025-03-03 22:10:20",
"id": 35
},
{
"name": "深度爬取|对应 Crawl 模式",
"logo": null,
"desc": "爬取并返回指定 URL 以及所有可访问子页面的内容。",
"api_params": [
{
"in": "path",
"name": "target_url",
"schema": {
"type": "string"
},
"required": true,
"description": "要爬取网站的起始 url。"
}
],
"tool_key": "fire_search_crawl",
"type": 10,
"is_preset": 1,
"is_delete": 0,
"user_id": 3,
"create_time": "2024-05-08 14:36:50",
"update_time": "2025-02-27 19:05:31",
"id": 36
},
{
"name": "获取指定单聊、群聊的历史消息",
"logo": null,
"desc": "支持在单聊或群聊中快速获取相关内容",
"api_params": [
{
"in": "query",
"name": "container_id",
"schema": {
"type": "string"
},
"required": true,
"description": "单聊或群聊的id，或话题 id"
},
{
"in": "query",
"name": "container_id_type",
"schema": {
"type": "string"
},
"required": true,
"description": "容器类型。 可选值有： chat：包含单聊（p2p）和群聊（group）； thread：话题 。"
},
{
"in": "query",
"name": "start_time",
"schema": {
"type": "string"
},
"required": false,
"description": "待查询历史信息的起始时间，秒级时间戳。 注意：thread 容器类型暂不支持获取指定时间范围内的消息。"
},
{
"in": "query",
"name": "end_time",
"schema": {
"type": "string"
},
"required": false,
"description": "待查询历史信息的结束时间，秒级时间戳。注意：thread 容器类型暂不支持获取指定时间范围内的消息。"
},
{
"in": "query",
"name": "page_size",
"schema": {
"type": "string"
},
"required": false,
"description": "分页大小，单次请求所返回的数据条目数，默认值20，取值范围1~50。"
},
{
"in": "query",
"name": "sort_type",
"schema": {
"type": "string"
},
"required": false,
"description": "可选值有：ByCreateTimeAsc（按消息创建时间升序排列）；ByCreateTimeDesc（按消息创建时间降序排列）"
},
{
"in": "query",
"name": "page_token",
"schema": {
"type": "string"
},
"required": false,
"description": "分页标记，第一次请求不填，表示从头开始遍历；分页查询结果还有更多项时会同时返回新的 page_token，下次遍历可采用该 page_token 获取查询结果"
}
],
"tool_key": "feishu_get_chat_messages",
"type": 13,
"is_preset": 1,
"is_delete": 0,
"user_id": 3,
"create_time": "2024-05-08 14:36:50",
"update_time": "2025-02-24 14:35:04",
"id": 37
},
{
"name": "联网搜索",
"logo": null,
"desc": "使用 query 进行联网检索并返回结果。",
"api_params": [
{
"name": "query",
"description": "Search query",
"in": "query",
"required": true,
"schema": {
"type": "string"
}
}
],
"tool_key": "web_search",
"type": 16,
"is_preset": 1,
"is_delete": 0,
"user_id": 1,
"id": 38
}
]
================================================
FILE: src/backend/bisheng/database/data/t_gpts_tools_type.json
================================================
[
{
"id": 1,
"name": "时间",
"logo": null,
"description": "获取当前时间",
"auth_method": null,
"api_key": null,
"auth_type": null,
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:25:26",
"update_time": "2024-04-30 10:25:26",
"openapi_schema": null
},
{
"id": 2,
"name": "计算器",
"logo": null,
"description": "使用 Python 内置工具进行数学表达式计算",
"auth_method": null,
"api_key": null,
"auth_type": null,
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:28:30",
"update_time": "2024-04-30 10:28:30",
"openapi_schema": null
},
{
"id": 3,
"name": "论文获取",
"logo": null,
"description": "从 Arxiv 网站检索论文的工具，输入为检索关键词。",
"auth_method": null,
"api_key": null,
"auth_type": null,
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:28:49",
"update_time": "2024-04-30 10:28:49",
"openapi_schema": null
},
{
"id": 4,
"name": "Dalle3绘画",
"logo": "",
"description": "OpenAI 文生图模型",
"auth_method": null,
"api_key": null,
"auth_type": null,
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:24:30",
"update_time": "2024-04-30 10:24:30",
"openapi_schema": null
},
{
"id": 5,
"name": "Bing web搜索",
"logo": "",
"description": "BIng 搜索引擎，可联网检索互联网信息，例如天气、汇率、时事等",
"auth_method": null,
"api_key": null,
"auth_type": null,
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:23:20",
"update_time": "2024-04-30 10:23:20",
"openapi_schema": null
},
{
"id": 6,
"name": "代码执行器",
"logo": null,
"description": "通过执行代码完成图表绘制、文件处理等编程类操作",
"auth_method": null,
"api_key": null,
"auth_type": null,
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:26:07",
"update_time": "2024-04-30 10:26:07",
"openapi_schema": null
},
{
"id": 7,
"name": "天眼查",
"logo": null,
"description": "企业信息查询",
"auth_method": null,
"api_key": null,
"auth_type": null,
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:26:40",
"update_time": "2024-04-30 10:26:40",
"openapi_schema": null
},
{
"id": 8,
"name": "经济金融数据",
"logo": null,
"description": "包含股票、基金、期货等行情数据和宏观经济、公司财务等基本面数据的金融大数据平台",
"auth_method": null,
"api_key": null,
"auth_type": null,
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:27:12",
"update_time": "2024-04-30 10:27:12",
"openapi_schema": null
},
{
"id": 9,
"name": "钉钉",
"logo": "",
"description": "钉钉群机器人发送消息工具",
"server_host": "",
"auth_method": 0,
"api_key": "",
"auth_type": "basic",
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:26:40",
"update_time": "2025-01-10 21:22:37",
"openapi_schema": null
},
{
"id": 10,
"name": "Firecrawl",
"logo": "",
"description": "指定 URL 爬取网页内容，并将其转换为 Markdown 格式",
"server_host": "",
"auth_method": 0,
"api_key": "",
"auth_type": "basic",
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:26:40",
"update_time": "2025-02-27 19:05:31",
"openapi_schema": null
},
{
"id": 11,
"name": "Jina AI",
"logo": "",
"description": "将目标网址（支持 PDF）内容转换为大模型可处理的 Markdown 格式",
"server_host": "",
"auth_method": 0,
"api_key": "",
"auth_type": "basic",
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:26:40",
"update_time": "2025-02-14 15:06:45",
"openapi_schema": null
},
{
"id": 12,
"name": "SiliconFlow",
"logo": "",
"description": "基于文本提示生成高质量图像，支持 Flux 和 Stable Diffusion 模型",
"server_host": "",
"auth_method": 0,
"api_key": "",
"auth_type": "basic",
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:26:40",
"update_time": "2025-02-20 11:47:25",
"openapi_schema": null
},
{
"id": 13,
"name": "飞书消息",
"logo": "",
"description": "支持获取飞书单聊或群聊历史消息，向指定用户或者群聊发送消息",
"server_host": "",
"auth_method": 0,
"api_key": "",
"auth_type": "basic",
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:26:40",
"update_time": "2025-02-24 14:35:04",
"openapi_schema": null
},
{
"id": 14,
"name": "发送邮件",
"logo": "",
"description": "通过smtp协议发送电子邮件",
"server_host": "",
"auth_method": 0,
"api_key": "",
"auth_type": "basic",
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:26:40",
"update_time": "2025-03-03 22:10:20",
"openapi_schema": null
},
{
"id": 15,
"name": "企业微信",
"logo": "",
"description": "企业微信机器人发送群消息工具",
"server_host": "",
"auth_method": 0,
"api_key": "",
"auth_type": "basic",
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"create_time": "2024-04-30 10:26:40",
"update_time": "2025-01-10 21:22:37",
"openapi_schema": null
},
{
"id": 16,
"name": "联网搜索",
"logo": "",
"description": "搜索互联网信息，可配置使用不同的搜索引擎，目前支持 Bing、博查、Jina 深度搜索、SerpApi、Tavily。",
"server_host": "",
"auth_method": 0,
"api_key": "",
"auth_type": "basic",
"is_preset": 1,
"user_id": null,
"is_delete": 0,
"openapi_schema": null
}
]
================================================
FILE: src/backend/bisheng/database/data/template.json
================================================
[
{
"update_time": "2024-08-20 16:27:48",
"parameters": null,
"name": "⭐️大模型对话（支持多轮）",
"description": "直接与大模型对话",
"flow_id": "544b7e31a34e4bdcaf2fc02d6bd67f4a",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 16:27:48",
"id": 171,
"data": {
"edges": [
{
"id": "reactflow__edge-BishengLLM-2839cBishengLLM|BishengLLM-2839c|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-ConversationChain-A1J5dBaseLanguageModel|llm|ConversationChain-A1J5d",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-2839c",
"target": "ConversationChain-A1J5d",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-2839c|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|ConversationChain-A1J5d"
}
],
"nodes": [
{
"id": "ConversationChain-A1J5d",
"data": {
"id": "ConversationChain-A1J5d",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "ConversationChain",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"prompt": {
"l2": false,
"info": "",
"list": false,
"name": "prompt",
"show": false,
"type": "BasePromptTemplate",
"value": {
"_type": "prompt",
"template": "The following is a friendly conversation between a human and an AI. The AI is talkative and provides lots of specific details from its context. If the AI does not know the answer to a question, it truthfully says it does not know.\n\nCurrent conversation:\n{history}\nHuman: {input}\nAI:",
"output_parser": null,
"input_variables": [
"history",
"input"
],
"template_format": "f-string",
"partial_variables": {},
"validate_template": true
},
"l2_name": "prompt",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "input",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"llm_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "llm_kwargs",
"show": false,
"type": "code",
"l2_name": "llm_kwargs",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "response",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "BaseLLMOutputParser",
"value": {
"_type": "default"
},
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"return_final_only": {
"l2": false,
"info": "",
"list": false,
"name": "return_final_only",
"show": false,
"type": "bool",
"value": true,
"l2_name": "return_final_only",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain to have a conversation and load context from memory.",
"base_classes": [
"ConversationChain",
"LLMChain",
"Chain",
"function"
],
"display_name": "ConversationChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "ConversationChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 301,
"dragging": false,
"position": {
"x": 960.9557711159376,
"y": 247.42733841910135
},
"selected": false,
"positionAbsolute": {
"x": 960.9557711159376,
"y": 247.42733841910135
}
},
{
"id": "BishengLLM-2839c",
"data": {
"id": "BishengLLM-2839c",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 5,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": 387.2949740637953,
"y": 154.06227685122178
},
"selected": true,
"positionAbsolute": {
"x": 387.2949740637953,
"y": 154.06227685122178
}
}
],
"viewport": {
"x": 417.19304131976264,
"y": 206.78938474334075,
"zoom": 0.598528187050664
}
},
"order_num": 6094817,
"guide_word": null
},
{
"update_time": "2024-08-20 16:32:19",
"parameters": null,
"name": "⭐️ 输入网址进行问答",
"description": "输入网址，进行问答。该技能点击上线按钮以及创建会话时，都需要等待几秒钟，因为需要先执行网址数据爬取的流程。",
"flow_id": "dc4f9e141b8d4cf390ddc2c2c51efd8e",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 16:32:19",
"id": 172,
"data": {
"edges": [
{
"id": "reactflow__edge-CombineDocsChain-RCPNACombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function-RetrievalQA-x5napBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"source": "CombineDocsChain-RCPNA",
"target": "RetrievalQA-x5nap",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "CombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap"
},
{
"id": "reactflow__edge-Milvus-CWpaeMilvus|Milvus-CWpae|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever-RetrievalQA-x5napBaseRetriever|retriever|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"source": "Milvus-CWpae",
"target": "RetrievalQA-x5nap",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "Milvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-x5nap"
},
{
"id": "reactflow__edge-WebBaseLoader-af6dbWebBaseLoader|WebBaseLoader-af6db|Document-Milvus-CWpaeDocument|documents|Milvus-CWpae",
"style": {
"stroke": "#555"
},
"source": "WebBaseLoader-af6db",
"target": "Milvus-CWpae",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "WebBaseLoader|WebBaseLoader-af6db|Document",
"targetHandle": "Document|documents|Milvus-CWpae"
},
{
"id": "reactflow__edge-OpenAIProxyEmbedding-1771bOpenAIProxyEmbedding|OpenAIProxyEmbedding-1771b|Embeddings-Milvus-CWpaeEmbeddings|embedding|Milvus-CWpae",
"style": {
"stroke": "#555"
},
"source": "OpenAIProxyEmbedding-1771b",
"target": "Milvus-CWpae",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "OpenAIProxyEmbedding|OpenAIProxyEmbedding-1771b|Embeddings",
"targetHandle": "Embeddings|embedding|Milvus-CWpae"
},
{
"id": "reactflow__edge-BishengLLM-db49dBishengLLM|BishengLLM-db49d|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-CombineDocsChain-RCPNABaseLanguageModel|llm|CombineDocsChain-RCPNA",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-db49d",
"target": "CombineDocsChain-RCPNA",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-db49d|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-RCPNA"
}
],
"nodes": [
{
"id": "CombineDocsChain-RCPNA",
"data": {
"id": "CombineDocsChain-RCPNA",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"token_max": {
"l2": false,
"info": "当前只对stuff 生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"l2_name": "token_max",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 377,
"position": {
"x": 845.0164562381523,
"y": 124.74220334881426
},
"selected": false,
"positionAbsolute": {
"x": 845.0164562381523,
"y": 124.74220334881426
}
},
{
"id": "Milvus-CWpae",
"data": {
"id": "Milvus-CWpae",
"node": {
"l2_name": "知识库",
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"collection_name": {
"l2": false,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "选择知识库",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": ""
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"Milvus",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 347,
"dragging": false,
"position": {
"x": 857.7079649166786,
"y": 667.3224068392545
},
"selected": false,
"positionAbsolute": {
"x": 857.7079649166786,
"y": 667.3224068392545
}
},
{
"id": "RetrievalQA-x5nap",
"data": {
"id": "RetrievalQA-x5nap",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "RetrievalQA",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"combine_documents_chain": {
"l2": false,
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"l2_name": "combine_documents_chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"RetrievalQA",
"Chain",
"BaseRetrievalQA",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 317,
"position": {
"x": 1521.2877663953464,
"y": 287.6148502766154
},
"selected": false,
"positionAbsolute": {
"x": 1521.2877663953464,
"y": 287.6148502766154
}
},
{
"id": "WebBaseLoader-af6db",
"data": {
"id": "WebBaseLoader-af6db",
"node": {
"l2_name": "输入网址",
"template": {
"_type": "WebBaseLoader",
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"value": "{}",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Metadata"
},
"web_path": {
"l2": true,
"info": "",
"list": false,
"name": "web_path",
"show": true,
"type": "str",
"value": "https://dataelem.feishu.cn/wiki/ZxW6wZyAJicX4WkG0NqcWsbynde",
"l2_name": "网址",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Web Page"
}
},
"description": "Load HTML pages using `urllib` and parse them with `BeautifulSoup'.",
"base_classes": [
"Document"
],
"display_name": "WebBaseLoader",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/web_base"
},
"type": "WebBaseLoader",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 357,
"position": {
"x": 333.05874389504135,
"y": 651.7320290450776
},
"positionAbsolute": {
"x": 333.05874389504135,
"y": 651.7320290450776
}
},
{
"id": "OpenAIProxyEmbedding-1771b",
"data": {
"id": "OpenAIProxyEmbedding-1771b",
"node": {
"template": {
"_type": "proxy_embedding",
"proxy_url": {
"info": "",
"list": false,
"name": "proxy_url",
"show": false,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "http://43.133.35.137:8080"
}
},
"description": " 使用自建的embedding服务使用openai进行embed ",
"base_classes": [
"Embeddings"
],
"display_name": "OpenAIProxyEmbedding",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "OpenAIProxyEmbedding",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 173,
"position": {
"x": 345.75546354081786,
"y": 1048.504517975594
},
"positionAbsolute": {
"x": 345.75546354081786,
"y": 1048.504517975594
}
},
{
"id": "BishengLLM-db49d",
"data": {
"id": "BishengLLM-db49d",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 5,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": 338.00327145994027,
"y": 66.93400331593091
},
"selected": false,
"positionAbsolute": {
"x": 338.00327145994027,
"y": 66.93400331593091
}
}
],
"viewport": {
"x": 267.90795205311633,
"y": 35.63343348513274,
"zoom": 0.6777805639494686
}
},
"order_num": 6160352,
"guide_word": null
},
{
"update_time": "2024-08-20 16:33:49",
"parameters": null,
"name": "⭐️ 表格数据问答",
"description": "上传CSV进行问答，使用CSV专用解析模块，相比通用解析模块，效果更好。\nExcel文件可以转成CSV后上传体验。",
"flow_id": "b4a7482adcae44aa819f9ea5c7617ae5",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 16:33:49",
"id": 173,
"data": {
"edges": [
{
"id": "reactflow__edge-CombineDocsChain-RCPNACombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function-RetrievalQA-x5napBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"source": "CombineDocsChain-RCPNA",
"target": "RetrievalQA-x5nap",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "CombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap"
},
{
"id": "reactflow__edge-Milvus-CWpaeMilvus|Milvus-CWpae|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever-RetrievalQA-x5napBaseRetriever|retriever|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"source": "Milvus-CWpae",
"target": "RetrievalQA-x5nap",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "Milvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-x5nap"
},
{
"id": "reactflow__edge-OpenAIProxyEmbedding-1771bOpenAIProxyEmbedding|OpenAIProxyEmbedding-1771b|Embeddings-Milvus-CWpaeEmbeddings|embedding|Milvus-CWpae",
"style": {
"stroke": "#555"
},
"source": "OpenAIProxyEmbedding-1771b",
"target": "Milvus-CWpae",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "OpenAIProxyEmbedding|OpenAIProxyEmbedding-1771b|Embeddings",
"targetHandle": "Embeddings|embedding|Milvus-CWpae"
},
{
"id": "reactflow__edge-CSVLoader-854c0CSVLoader|CSVLoader-854c0|Document-Milvus-CWpaeDocument|documents|Milvus-CWpae",
"style": {
"stroke": "#555"
},
"source": "CSVLoader-854c0",
"target": "Milvus-CWpae",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "CSVLoader|CSVLoader-854c0|Document",
"targetHandle": "Document|documents|Milvus-CWpae"
},
{
"id": "reactflow__edge-InputFileNode-e4363InputFileNode|InputFileNode-e4363|fileNode-CSVLoader-854c0fileNode|file_path|CSVLoader-854c0",
"style": {
"stroke": "#555"
},
"source": "InputFileNode-e4363",
"target": "CSVLoader-854c0",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "InputFileNode|InputFileNode-e4363|fileNode",
"targetHandle": "fileNode|file_path|CSVLoader-854c0"
},
{
"id": "reactflow__edge-BishengLLM-83706BishengLLM|BishengLLM-83706|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-CombineDocsChain-RCPNABaseLanguageModel|llm|CombineDocsChain-RCPNA",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-83706",
"target": "CombineDocsChain-RCPNA",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-83706|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-RCPNA"
}
],
"nodes": [
{
"id": "CombineDocsChain-RCPNA",
"data": {
"id": "CombineDocsChain-RCPNA",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"token_max": {
"l2": false,
"info": "当前只对stuff 生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"l2_name": "token_max",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 377,
"position": {
"x": 845.0164562381523,
"y": 124.74220334881426
},
"selected": false,
"positionAbsolute": {
"x": 845.0164562381523,
"y": 124.74220334881426
}
},
{
"id": "Milvus-CWpae",
"data": {
"id": "Milvus-CWpae",
"node": {
"l2_name": "知识库",
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"collection_name": {
"l2": false,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "选择知识库",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": ""
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"Milvus",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 347,
"dragging": false,
"position": {
"x": 857.7079649166786,
"y": 667.3224068392545
},
"selected": false,
"positionAbsolute": {
"x": 857.7079649166786,
"y": 667.3224068392545
}
},
{
"id": "RetrievalQA-x5nap",
"data": {
"id": "RetrievalQA-x5nap",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "RetrievalQA",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"combine_documents_chain": {
"l2": false,
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"l2_name": "combine_documents_chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"RetrievalQA",
"Chain",
"BaseRetrievalQA",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 317,
"position": {
"x": 1521.2877663953464,
"y": 287.6148502766154
},
"selected": false,
"positionAbsolute": {
"x": 1521.2877663953464,
"y": 287.6148502766154
}
},
{
"id": "OpenAIProxyEmbedding-1771b",
"data": {
"id": "OpenAIProxyEmbedding-1771b",
"node": {
"template": {
"_type": "proxy_embedding",
"proxy_url": {
"info": "",
"list": false,
"name": "proxy_url",
"show": false,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "http://43.133.35.137:8080"
}
},
"description": " 使用自建的embedding服务使用openai进行embed ",
"base_classes": [
"Embeddings"
],
"display_name": "OpenAIProxyEmbedding",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "OpenAIProxyEmbedding",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 173,
"dragging": false,
"position": {
"x": 345.75546354081786,
"y": 1048.504517975594
},
"selected": false,
"positionAbsolute": {
"x": 345.75546354081786,
"y": 1048.504517975594
}
},
{
"id": "CSVLoader-854c0",
"data": {
"id": "CSVLoader-854c0",
"node": {
"template": {
"_type": "CSVLoader",
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"value": "{}",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Metadata"
},
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "fileNode",
"value": "",
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".csv"
],
"fileTypes": [
"csv"
],
"multiline": false,
"placeholder": ""
}
},
"description": "Load a `CSV` file into a list of Documents.",
"base_classes": [
"Document"
],
"display_name": "CSVLoader",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/csv"
},
"type": "CSVLoader",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 291,
"position": {
"x": 326.7103840721531,
"y": 672.2493274470351
},
"positionAbsolute": {
"x": 326.7103840721531,
"y": 672.2493274470351
}
},
{
"id": "InputFileNode-e4363",
"data": {
"id": "InputFileNode-e4363",
"node": {
"template": {
"_type": "InputFileNode",
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "file",
"value": "",
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".csv"
],
"fileTypes": [
"csv"
],
"file_path": null,
"multiline": false,
"placeholder": ""
},
"file_type": {
"info": "Tips for which file should upload",
"list": false,
"name": "file_type",
"show": true,
"type": "str",
"value": "CSV文件",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "提示上传文件类型",
"display_name": "Name"
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"fileNode"
],
"display_name": "InputFileNode",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "InputFileNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 337,
"dragging": false,
"position": {
"x": -176.39713189174194,
"y": 643.681708244038
},
"selected": false,
"positionAbsolute": {
"x": -176.39713189174194,
"y": 643.681708244038
}
},
{
"id": "BishengLLM-83706",
"data": {
"id": "BishengLLM-83706",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 5,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": 318.5459219608139,
"y": 81.68804102649821
},
"selected": true,
"positionAbsolute": {
"x": 318.5459219608139,
"y": 81.68804102649821
}
}
],
"viewport": {
"x": 270.09576536959617,
"y": 21.63343348513274,
"zoom": 0.6777805639494686
}
},
"order_num": 6225887,
"guide_word": null
},
{
"update_time": "2024-08-20 16:43:27",
"parameters": null,
"name": "☘️合同审核",
"description": "上传合同文件，自动触发预设的审查问题",
"flow_id": "cfca77afb829477190d11b9caccaa9c4",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 16:43:27",
"id": 175,
"data": {
"edges": [
{
"id": "reactflow__edge-CombineDocsChain-5EyWUCombineDocsChain|CombineDocsChain-5EyWU|BaseCombineDocumentsChain|function-RetrievalQA-wKjqvBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-wKjqv",
"style": {
"stroke": "#555"
},
"source": "CombineDocsChain-5EyWU",
"target": "RetrievalQA-wKjqv",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "CombineDocsChain|CombineDocsChain-5EyWU|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-wKjqv"
},
{
"id": "reactflow__edge-RetrievalQA-wKjqvRetrievalQA|RetrievalQA-wKjqv|Chain|RetrievalQA|BaseRetrievalQA|function-Tool-0dIGDfunction|func|Tool-0dIGD",
"style": {
"stroke": "#555"
},
"source": "RetrievalQA-wKjqv",
"target": "Tool-0dIGD",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "RetrievalQA|RetrievalQA-wKjqv|BaseRetrievalQA|Chain|RetrievalQA|function",
"targetHandle": "function|func|Tool-0dIGD"
},
{
"id": "reactflow__edge-Milvus-hs3JXMilvus|Milvus-hs3JX|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever-RetrievalQA-wKjqvBaseRetriever|retriever|RetrievalQA-wKjqv",
"style": {
"stroke": "#555"
},
"source": "Milvus-hs3JX",
"target": "RetrievalQA-wKjqv",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "Milvus|Milvus-hs3JX|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-wKjqv"
},
{
"id": "reactflow__edge-Tool-0dIGDTool|Tool-0dIGD|Tool|BaseTool-ZeroShotAgent-Vpn4kBaseTool|tools|ZeroShotAgent-Vpn4k",
"style": {
"stroke": "#555"
},
"source": "Tool-0dIGD",
"target": "ZeroShotAgent-Vpn4k",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "Tool|Tool-0dIGD|Tool|BaseTool",
"targetHandle": "BaseTool|tools|ZeroShotAgent-Vpn4k"
},
{
"id": "reactflow__edge-OpenAIProxyEmbedding-5TAiXOpenAIProxyEmbedding|OpenAIProxyEmbedding-5TAiX|Embeddings-Milvus-hs3JXEmbeddings|embedding|Milvus-hs3JX",
"style": {
"stroke": "#555"
},
"source": "OpenAIProxyEmbedding-5TAiX",
"target": "Milvus-hs3JX",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "OpenAIProxyEmbedding|OpenAIProxyEmbedding-5TAiX|Embeddings",
"targetHandle": "Embeddings|embedding|Milvus-hs3JX"
},
{
"id": "reactflow__edge-BingSearchAPIWrapper-9285QBingSearchAPIWrapper|BingSearchAPIWrapper-9285Q|BingSearchAPIWrapper-BingSearchRun-TAJzwBingSearchAPIWrapper|api_wrapper|BingSearchRun-TAJzw",
"style": {
"stroke": "#555"
},
"source": "BingSearchAPIWrapper-9285Q",
"target": "BingSearchRun-TAJzw",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "BingSearchAPIWrapper|BingSearchAPIWrapper-9285Q|BingSearchAPIWrapper",
"targetHandle": "BingSearchAPIWrapper|api_wrapper|BingSearchRun-TAJzw"
},
{
"id": "reactflow__edge-BingSearchRun-TAJzwBingSearchRun|BingSearchRun-TAJzw|Tool|BaseTool|BaseTool|BingSearchRun-ZeroShotAgent-Vpn4kBaseTool|tools|ZeroShotAgent-Vpn4k",
"style": {
"stroke": "#555"
},
"source": "BingSearchRun-TAJzw",
"target": "ZeroShotAgent-Vpn4k",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "BingSearchRun|BingSearchRun-TAJzw|Tool|BaseTool|BingSearchRun|BaseTool",
"targetHandle": "BaseTool|tools|ZeroShotAgent-Vpn4k"
},
{
"id": "reactflow__edge-Calculator-g6AEOCalculator|Calculator-g6AEO|Tool|BaseTool-ZeroShotAgent-Vpn4kBaseTool|tools|ZeroShotAgent-Vpn4k",
"style": {
"stroke": "#555"
},
"source": "Calculator-g6AEO",
"target": "ZeroShotAgent-Vpn4k",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "Calculator|Calculator-g6AEO|Tool|BaseTool",
"targetHandle": "BaseTool|tools|ZeroShotAgent-Vpn4k"
},
{
"id": "reactflow__edge-InputNode-LENRkInputNode|InputNode-LENRk|input-ZeroShotAgent-Vpn4kinput|input_node|ZeroShotAgent-Vpn4k",
"style": {
"stroke": "#555"
},
"source": "InputNode-LENRk",
"target": "ZeroShotAgent-Vpn4k",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "InputNode|InputNode-LENRk|input",
"targetHandle": "input|input_node|ZeroShotAgent-Vpn4k"
},
{
"id": "reactflow__edge-PyPDFLoader-nVufmPyPDFLoader|PyPDFLoader-nVufm|Document-RecursiveCharacterTextSplitter-ahXPxDocument|documents|RecursiveCharacterTextSplitter-ahXPx",
"style": {
"stroke": "#555"
},
"source": "PyPDFLoader-nVufm",
"target": "RecursiveCharacterTextSplitter-ahXPx",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "PyPDFLoader|PyPDFLoader-nVufm|Document",
"targetHandle": "Document|documents|RecursiveCharacterTextSplitter-ahXPx"
},
{
"id": "reactflow__edge-RecursiveCharacterTextSplitter-ahXPxRecursiveCharacterTextSplitter|RecursiveCharacterTextSplitter-ahXPx|Document-Milvus-hs3JXDocument|documents|Milvus-hs3JX",
"style": {
"stroke": "#555"
},
"source": "RecursiveCharacterTextSplitter-ahXPx",
"target": "Milvus-hs3JX",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "RecursiveCharacterTextSplitter|RecursiveCharacterTextSplitter-ahXPx|Document",
"targetHandle": "Document|documents|Milvus-hs3JX"
},
{
"id": "reactflow__edge-BishengLLM-a591dBishengLLM|BishengLLM-a591d|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-CombineDocsChain-5EyWUBaseLanguageModel|llm|CombineDocsChain-5EyWU",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-a591d",
"target": "CombineDocsChain-5EyWU",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-a591d|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-5EyWU"
},
{
"id": "reactflow__edge-BishengLLM-a591dBishengLLM|BishengLLM-a591d|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-ZeroShotAgent-Vpn4kBaseLanguageModel|llm|ZeroShotAgent-Vpn4k",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-a591d",
"target": "ZeroShotAgent-Vpn4k",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-a591d|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|ZeroShotAgent-Vpn4k"
},
{
"id": "reactflow__edge-BishengLLM-a591dBishengLLM|BishengLLM-a591d|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-Calculator-g6AEOBaseLanguageModel|llm|Calculator-g6AEO",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-a591d",
"target": "Calculator-g6AEO",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-a591d|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|Calculator-g6AEO"
},
{
"id": "reactflow__edge-InputFileNode-dd5fcInputFileNode|InputFileNode-dd5fc|fileNode-PyPDFLoader-nVufmfileNode|file_path|PyPDFLoader-nVufm",
"style": {
"stroke": "#555"
},
"source": "InputFileNode-dd5fc",
"target": "PyPDFLoader-nVufm",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "InputFileNode|InputFileNode-dd5fc|fileNode",
"targetHandle": "fileNode|file_path|PyPDFLoader-nVufm"
}
],
"nodes": [
{
"id": "CombineDocsChain-5EyWU",
"data": {
"id": "CombineDocsChain-5EyWU",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"prompt": {
"l2": false,
"info": "只对Stuff类型生效",
"list": false,
"name": "prompt",
"type": "BasePromptTemplate",
"l2_name": "prompt",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
},
"token_max": {
"l2": false,
"info": "只对Stuff类型生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"l2_name": "token_max",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 377,
"position": {
"x": 1600.092442163245,
"y": 227.94525786055965
},
"selected": false,
"positionAbsolute": {
"x": 1600.092442163245,
"y": 227.94525786055965
}
},
{
"id": "RetrievalQA-wKjqv",
"data": {
"id": "RetrievalQA-wKjqv",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "RetrievalQA",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"combine_documents_chain": {
"l2": false,
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"l2_name": "combine_documents_chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": false,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"BaseRetrievalQA",
"Chain",
"RetrievalQA",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 317,
"dragging": false,
"position": {
"x": 2214.008643017949,
"y": 410.0707219409365
},
"selected": false,
"positionAbsolute": {
"x": 2214.008643017949,
"y": 410.0707219409365
}
},
{
"id": "Tool-0dIGD",
"data": {
"id": "Tool-0dIGD",
"node": {
"template": {
"func": {
"l2": false,
"info": "",
"list": false,
"name": "func",
"show": true,
"type": "function",
"l2_name": "func",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"name": {
"l2": false,
"info": "",
"list": false,
"name": "name",
"show": true,
"type": "str",
"value": "Basic_information",
"l2_name": "name",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"_type": "Tool",
"description": {
"l2": false,
"info": "",
"list": false,
"name": "description",
"show": true,
"type": "str",
"value": "如果需要知道关于这份合同的各方面信息，可以使用该工具。如：\n- 合同的价款是多少？\n- 合同的甲方和乙方分别是谁？\n- 合同关于违约金是如何约定的？\n- 合同的免责条款是什么？\n",
"l2_name": "description",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"return_direct": {
"l2": false,
"info": "",
"list": false,
"name": "return_direct",
"show": true,
"type": "bool",
"value": false,
"l2_name": "return_direct",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Converts a chain, agent or function into a tool.",
"base_classes": [
"Tool",
"BaseTool"
],
"display_name": "Tool",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "Tool",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 445,
"position": {
"x": 2726.848698975115,
"y": 439.56467505828414
},
"selected": false,
"positionAbsolute": {
"x": 2726.848698975115,
"y": 439.56467505828414
}
},
{
"id": "Milvus-hs3JX",
"data": {
"id": "Milvus-hs3JX",
"node": {
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"collection_name": {
"l2": false,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "collection_name",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": ""
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"VectorStore",
"Milvus",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 347,
"dragging": false,
"position": {
"x": 1624.5397713340649,
"y": 643.8868403230642
},
"selected": false,
"positionAbsolute": {
"x": 1624.5397713340649,
"y": 643.8868403230642
}
},
{
"id": "ZeroShotAgent-Vpn4k",
"data": {
"id": "ZeroShotAgent-Vpn4k",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"_type": "ZeroShotAgent",
"tools": {
"l2": false,
"info": "",
"list": true,
"name": "tools",
"show": true,
"type": "BaseTool",
"l2_name": "tools",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"prefix": {
"l2": false,
"info": "",
"list": false,
"name": "prefix",
"show": true,
"type": "str",
"value": "Answer the following questions in Chinese as best you can. Think step by step. You have access to the following tools:",
"l2_name": "prefix",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
},
"suffix": {
"l2": false,
"info": "",
"list": false,
"name": "suffix",
"show": true,
"type": "str",
"value": "Begin!\n\nQuestion: {query}\nThought:{agent_scratchpad}",
"l2_name": "suffix",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "AgentOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": true,
"type": "str",
"value": [
"query",
"agent_scratchpad"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callback_manager": {
"l2": false,
"info": "",
"list": false,
"name": "callback_manager",
"show": false,
"type": "BaseCallbackManager",
"l2_name": "callback_manager",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"format_instructions": {
"l2": false,
"info": "",
"list": false,
"name": "format_instructions",
"show": true,
"type": "str",
"value": "- If questions are ralated to the tool [Basic_information], you should transmit related-questions on to [Basic_information] with no split.\n\n\nUse the following format:\n\nQuestion: 该合同是根据什么法律签订的？\n# the query question you must answer\nThought: 找到该合同的法律依据\n# you should always think about what to do\nAction: Basic_information\n# the action to take, should be one of [{tool_names}]\nAction Input: 该合同是根据什么法律签订的？\n# the input to the action\nObservation: \n# the result of the action\n... (this Thought/Action/Action Input/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: \n# the final answer to the original input question",
"l2_name": "format_instructions",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
}
},
"description": "Construct an agent from an LLM and tools.",
"base_classes": [
"ZeroShotAgent",
"Agent",
"BaseSingleActionAgent",
"function"
],
"display_name": "ZeroShotAgent",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/agents/how_to/custom_mrkl_agent"
},
"type": "ZeroShotAgent",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 659,
"dragging": false,
"position": {
"x": 3738.5526393054743,
"y": 722.4823701379853
},
"selected": false,
"positionAbsolute": {
"x": 3738.5526393054743,
"y": 722.4823701379853
}
},
{
"id": "OpenAIProxyEmbedding-5TAiX",
"data": {
"id": "OpenAIProxyEmbedding-5TAiX",
"node": {
"template": {
"_type": "proxy_embedding",
"proxy_url": {
"l2": false,
"info": "",
"list": false,
"name": "proxy_url",
"show": false,
"type": "str",
"value": "",
"l2_name": "proxy_url",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "http://43.133.35.137:8080"
}
},
"description": " 使用自建的embedding服务使用openai进行embed ",
"base_classes": [
"Embeddings"
],
"display_name": "OpenAIProxyEmbedding",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "OpenAIProxyEmbedding",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 173,
"position": {
"x": 901.2922965468372,
"y": 1334.1551610865345
},
"selected": false,
"positionAbsolute": {
"x": 901.2922965468372,
"y": 1334.1551610865345
}
},
{
"id": "BingSearchRun-TAJzw",
"data": {
"id": "BingSearchRun-TAJzw",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": false,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "BingSearchRun",
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": false,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"api_wrapper": {
"l2": false,
"info": "",
"list": false,
"name": "api_wrapper",
"show": true,
"type": "BingSearchAPIWrapper",
"l2_name": "api_wrapper",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": ""
},
"args_schema": {
"l2": false,
"info": "",
"list": false,
"name": "args_schema",
"show": false,
"type": "Type[pydantic.main.BaseModel]",
"l2_name": "args_schema",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"handle_tool_error": {
"l2": false,
"info": "",
"list": false,
"name": "handle_tool_error",
"show": false,
"type": "bool",
"value": false,
"l2_name": "handle_tool_error",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "A wrapper around Bing Search. Useful for when you need to answer questions about current events. Input should be a search query.",
"base_classes": [
"Tool",
"BaseTool",
"BingSearchRun",
"BaseTool"
],
"display_name": "BingSearchRun",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "BingSearchRun",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 249,
"dragging": false,
"position": {
"x": 2740.6814356916484,
"y": 1030.8252309669626
},
"selected": false,
"positionAbsolute": {
"x": 2740.6814356916484,
"y": 1030.8252309669626
}
},
{
"id": "BingSearchAPIWrapper-9285Q",
"data": {
"id": "BingSearchAPIWrapper-9285Q",
"node": {
"template": {
"k": {
"l2": false,
"info": "",
"list": false,
"name": "k",
"show": false,
"type": "int",
"value": 10,
"l2_name": "k",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "BingSearchAPIWrapper",
"bing_search_url": {
"l2": false,
"info": "",
"list": false,
"name": "bing_search_url",
"show": true,
"type": "str",
"value": "https://api.bing.microsoft.com/v7.0/search",
"l2_name": "bing_search_url",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"bing_subscription_key": {
"l2": false,
"info": "",
"list": false,
"name": "bing_subscription_key",
"show": true,
"type": "str",
"value": "137b003c89f04966ac3b1081ee5b10bc",
"l2_name": "bing_subscription_key",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Wrapper for Bing Search API.",
"base_classes": [
"BingSearchAPIWrapper"
],
"display_name": "BingSearchAPIWrapper",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "BingSearchAPIWrapper",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 337,
"dragging": false,
"position": {
"x": 2055.4764429926604,
"y": 1160.385181052582
},
"selected": false,
"positionAbsolute": {
"x": 2055.4764429926604,
"y": 1160.385181052582
}
},
{
"id": "Calculator-g6AEO",
"data": {
"id": "Calculator-g6AEO",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"_type": "Calculator"
},
"description": "Useful for when you need to answer questions about math.",
"base_classes": [
"Tool",
"BaseTool"
],
"display_name": "Calculator",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "Calculator",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 229,
"dragging": false,
"position": {
"x": 2732.119874152286,
"y": 131.7347928894963
},
"selected": false,
"positionAbsolute": {
"x": 2732.119874152286,
"y": 131.7347928894963
}
},
{
"id": "InputNode-LENRk",
"data": {
"id": "InputNode-LENRk",
"node": {
"template": {
"_type": "input",
"input": {
"l2": true,
"info": "",
"list": true,
"name": "input",
"show": true,
"type": "str",
"value": [
"这份合同中有约定价款的支付途径吗？",
"合同中是否约定了履行时间？",
"合同中约定了知识产权侵权责任吗？",
"合同中是否约定了甲方未按期支付款项的违约责任？",
"合同是否约定了解决纠纷的方法？如果是通过仲裁解决纠纷，是否有约定仲裁机构？",
"合同中提到了几次甲方名称？这几次的甲方名称都一样吗？",
"合同中提到了几次乙方名称？提到的乙方名称都一样吗？",
"合同全部费用的大写金额和小写金额是多少？这两个数字一样吗？",
"合同的发票项目是什么？搜索该发票项目对应的一般纳税人税率是多少",
"合同的违约金赔偿利率是多少？合同签订日期是哪年哪月？使用获得的答案，搜索合同签订日期当年当月的中国人民银行授权全国银行间同业拆借中心公布的一年期贷款市场报价利率（LPR）是多少？计算合同的违约金赔偿利率大于LPR的1.5倍吗？"
],
"l2_name": "审核问题",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "输入内容"
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"input"
],
"display_name": "InputNode",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "InputNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 705,
"dragging": false,
"position": {
"x": 3228.770319808037,
"y": -405.0472199949312
},
"selected": false,
"positionAbsolute": {
"x": 3228.770319808037,
"y": -405.0472199949312
}
},
{
"id": "PyPDFLoader-nVufm",
"data": {
"id": "PyPDFLoader-nVufm",
"node": {
"template": {
"_type": "PyPDFLoader",
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Metadata"
},
"file_path": {
"l2": false,
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "fileNode",
"value": "",
"l2_name": "file_path",
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".pdf"
],
"fileTypes": [
"pdf"
],
"multiline": false,
"placeholder": ""
}
},
"description": "Loads a PDF with pypdf and chunks at character level.",
"base_classes": [
"Document"
],
"display_name": "PyPDFLoader",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/data_connection/document_loaders/how_to/pdf"
},
"type": "PyPDFLoader",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 311,
"position": {
"x": 545.6794870854769,
"y": 576.4247493553407
},
"selected": false,
"positionAbsolute": {
"x": 545.6794870854769,
"y": 576.4247493553407
}
},
{
"id": "RecursiveCharacterTextSplitter-ahXPx",
"data": {
"id": "RecursiveCharacterTextSplitter-ahXPx",
"node": {
"template": {
"_type": "RecursiveCharacterTextSplitter",
"documents": {
"l2": false,
"info": "",
"list": false,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"chunk_size": {
"l2": false,
"info": "",
"list": false,
"name": "chunk_size",
"show": true,
"type": "int",
"value": 1000,
"l2_name": "chunk_size",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Chunk Size"
},
"separators": {
"l2": false,
"info": "",
"list": false,
"name": "separators",
"show": true,
"type": "str",
"value": "\\n",
"l2_name": "separators",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Separator"
},
"chunk_overlap": {
"l2": false,
"info": "",
"list": false,
"name": "chunk_overlap",
"show": true,
"type": "int",
"value": 200,
"l2_name": "chunk_overlap",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Chunk Overlap"
},
"separator_type": {
"l2": false,
"info": "",
"list": true,
"name": "separator_type",
"show": true,
"type": "str",
"value": "Text",
"l2_name": "separator_type",
"options": [
"Text",
"cpp",
"go",
"html",
"java",
"js",
"latex",
"markdown",
"php",
"proto",
"python",
"rst",
"ruby",
"rust",
"scala",
"sol",
"swift"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Separator Type"
}
},
"description": "Implementation of splitting text that looks at characters.",
"base_classes": [
"Document"
],
"display_name": "RecursiveCharacterTextSplitter",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/data_connection/document_transformers/text_splitters/recursive_text_splitter"
},
"type": "RecursiveCharacterTextSplitter",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 561,
"position": {
"x": 993.2830986988388,
"y": 672.9264454371207
},
"selected": false,
"positionAbsolute": {
"x": 993.2830986988388,
"y": 672.9264454371207
}
},
{
"id": "BishengLLM-a591d",
"data": {
"id": "BishengLLM-a591d",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 5,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": 1053.9828212208936,
"y": 59.894169470406155
},
"selected": false,
"positionAbsolute": {
"x": 1053.9828212208936,
"y": 59.894169470406155
}
},
{
"id": "InputFileNode-dd5fc",
"data": {
"id": "InputFileNode-dd5fc",
"node": {
"beta": false,
"template": {
"_type": "InputFileNode",
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "file",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".pdf"
],
"fileTypes": [
"pdf"
],
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "File Path",
"model_config": {}
},
"file_type": {
"info": "Tips for which file should upload",
"list": false,
"name": "file_type",
"show": true,
"type": "str",
"value": "请上传文件",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "提示上传文件类型",
"display_name": "Name",
"model_config": {}
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"fileNode"
],
"display_name": "InputFileNode",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "InputFileNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 337,
"dragging": false,
"position": {
"x": 66.88900058530322,
"y": 650.6996948374481
},
"selected": false,
"positionAbsolute": {
"x": 66.88900058530322,
"y": 650.6996948374481
}
}
],
"viewport": {
"x": -20.328962469404587,
"y": -153.06745791725848,
"zoom": 0.9467769276749988
}
},
"order_num": 6291422,
"guide_word": null
},
{
"update_time": "2024-08-20 16:45:43",
"parameters": null,
"name": "⭐️角色扮演",
"description": "模板是一个面试官角色，可以更换提示词变成别的角色。",
"flow_id": "523971dc79e84d9bbe4187ab0bdb2583",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 16:45:43",
"id": 176,
"data": {
"edges": [
{
"id": "reactflow__edge-PromptTemplate-ZoUVTPromptTemplate|PromptTemplate-ZoUVT|StringPromptTemplate|PromptTemplate|BasePromptTemplate-LLMChain-wRz0cBasePromptTemplate|prompt|LLMChain-wRz0c",
"style": {
"stroke": "#555"
},
"source": "PromptTemplate-ZoUVT",
"target": "LLMChain-wRz0c",
"animated": true,
"className": "",
"sourceHandle": "PromptTemplate|PromptTemplate-ZoUVT|PromptTemplate|StringPromptTemplate|BasePromptTemplate",
"targetHandle": "BasePromptTemplate|prompt|LLMChain-wRz0c"
},
{
"id": "reactflow__edge-ConversationBufferMemory-lI5PtConversationBufferMemory|ConversationBufferMemory-lI5Pt|ConversationBufferMemory|BaseMemory|BaseChatMemory-LLMChain-wRz0cBaseMemory|memory|LLMChain-wRz0c",
"style": {
"stroke": "#555"
},
"source": "ConversationBufferMemory-lI5Pt",
"target": "LLMChain-wRz0c",
"animated": true,
"className": "",
"sourceHandle": "ConversationBufferMemory|ConversationBufferMemory-lI5Pt|BaseChatMemory|ConversationBufferMemory|BaseMemory",
"targetHandle": "BaseMemory|memory|LLMChain-wRz0c"
},
{
"id": "reactflow__edge-BishengLLM-3bb75BishengLLM|BishengLLM-3bb75|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-LLMChain-wRz0cBaseLanguageModel|llm|LLMChain-wRz0c",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-3bb75",
"target": "LLMChain-wRz0c",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-3bb75|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|LLMChain-wRz0c"
}
],
"nodes": [
{
"id": "LLMChain-wRz0c",
"data": {
"id": "LLMChain-wRz0c",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "LLMChain",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"prompt": {
"l2": false,
"info": "",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"l2_name": "prompt",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"llm_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "llm_kwargs",
"show": false,
"type": "code",
"l2_name": "llm_kwargs",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "text",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "BaseLLMOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"return_final_only": {
"l2": false,
"info": "",
"list": false,
"name": "return_final_only",
"show": false,
"type": "bool",
"value": true,
"l2_name": "return_final_only",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain to run queries against LLMs.",
"base_classes": [
"Chain",
"LLMChain",
"function"
],
"display_name": "LLMChain",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/foundational/llm_chain"
},
"type": "LLMChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 317,
"dragging": false,
"position": {
"x": 1556.1679330787729,
"y": 1103.1743625556214
},
"selected": false,
"positionAbsolute": {
"x": 1556.1679330787729,
"y": 1103.1743625556214
}
},
{
"id": "PromptTemplate-ZoUVT",
"data": {
"id": "PromptTemplate-ZoUVT",
"node": {
"name": "",
"template": {
"_type": "PromptTemplate",
"input": {
"info": "",
"list": false,
"name": "input",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser"
],
"placeholder": "",
"display_name": "input"
},
"template": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "System：\n你将扮演一个科技公司的面试官，面试者是一名产品经理，提出 3 个犀利的问题。\n请注意：\n- 每次只问一个问题\n- 面试者回答问题后请简短点评一下，然后问下一个问题，不要试图纠正候选人的错误；\n- 如果你认为用户连续几次回答的都不对，就少问一点；\n- 问完最后一个问题后，你可以问这样一个问题：上一份工作为什么离职？用户回答该问题后，请表示理解与支持，并停止与面试者的对话。若面试者继续说话你可以礼貌拒绝。\n\n以下是你们的对话记录\n\n{chat_history}\n\n面试者：{input}\n\n你的回答：",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"chat_history": {
"info": "",
"list": false,
"name": "chat_history",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser"
],
"placeholder": "",
"display_name": "chat_history"
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"chat_history",
"input"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"template_format": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"partial_variables": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"validate_template": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Schema to represent a prompt for an LLM.",
"base_classes": [
"PromptTemplate",
"StringPromptTemplate",
"BasePromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"template": [
"chat_history",
"input"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {
"formatters": {
"openai_api_key": {}
},
"base_formatters": {
"dict": {},
"list": {},
"show": {},
"union": {},
"kwargs": {},
"default": {},
"headers": {},
"optional": {},
"password": {},
"multiline": {},
"model_fields": {
"MODEL_DICT": {
"OpenAI": [
"text-davinci-003",
"text-davinci-002",
"text-curie-001",
"text-babbage-001",
"text-ada-001"
],
"Anthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
],
"ChatOpenAI": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k"
],
"ChatAnthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
]
}
},
"dict_code_file": {}
}
}
},
"type": "PromptTemplate",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 419,
"dragging": false,
"position": {
"x": 782.4698380009005,
"y": 1522.4191451020877
},
"selected": false,
"positionAbsolute": {
"x": 782.4698380009005,
"y": 1522.4191451020877
}
},
{
"id": "ConversationBufferMemory-lI5Pt",
"data": {
"id": "ConversationBufferMemory-lI5Pt",
"node": {
"template": {
"_type": "ConversationBufferMemory",
"ai_prefix": {
"l2": false,
"info": "",
"list": false,
"name": "ai_prefix",
"show": false,
"type": "str",
"value": "AI",
"l2_name": "ai_prefix",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"l2": false,
"info": "The variable to be used as Chat Input when more than one variable is available.",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "",
"l2_name": "input_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"memory_key": {
"l2": false,
"info": "",
"list": false,
"name": "memory_key",
"show": true,
"type": "str",
"value": "chat_history",
"l2_name": "memory_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"output_key": {
"l2": false,
"info": "The variable to be used as Chat Output (e.g. answer in a ConversationalRetrievalChain)",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "",
"l2_name": "output_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"chat_memory": {
"l2": false,
"info": "",
"list": false,
"name": "chat_memory",
"show": true,
"type": "BaseChatMessageHistory",
"l2_name": "chat_memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"human_prefix": {
"l2": false,
"info": "",
"list": false,
"name": "human_prefix",
"show": false,
"type": "str",
"value": "Human",
"l2_name": "human_prefix",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"return_messages": {
"l2": false,
"info": "",
"list": false,
"name": "return_messages",
"show": true,
"type": "bool",
"l2_name": "return_messages",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Buffer for storing conversation memory.",
"base_classes": [
"BaseChatMemory",
"ConversationBufferMemory",
"BaseMemory"
],
"display_name": "ConversationBufferMemory",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/memory/how_to/buffer"
},
"type": "ConversationBufferMemory",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 527,
"dragging": false,
"position": {
"x": 378.4886538486901,
"y": 959.6786678431888
},
"selected": false,
"positionAbsolute": {
"x": 378.4886538486901,
"y": 959.6786678431888
}
},
{
"id": "BishengLLM-3bb75",
"data": {
"id": "BishengLLM-3bb75",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 5,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"position": {
"x": 892.0260725263943,
"y": 617.6812234671859
},
"positionAbsolute": {
"x": 892.0260725263943,
"y": 617.6812234671859
}
}
],
"viewport": {
"x": 147.39902426086758,
"y": -337.0826199129119,
"zoom": 0.6112168397288411
}
},
"order_num": 6356957,
"guide_word": null
},
{
"update_time": "2024-08-20 16:51:57",
"parameters": null,
"name": "⭐️文章摘要",
"description": "上传文件自动触发文档摘要，当前支持文件格式：png, jpeg, jpg,pdf",
"flow_id": "c642e5031c1244829ddc0a59c7a5e3fd",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 16:51:57",
"id": 178,
"data": {
"edges": [
{
"id": "reactflow__edge-CombineDocsChain-RCPNACombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function-RetrievalQA-x5napBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"source": "CombineDocsChain-RCPNA",
"target": "RetrievalQA-x5nap",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "CombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap"
},
{
"id": "reactflow__edge-Milvus-CWpaeMilvus|Milvus-CWpae|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever-RetrievalQA-x5napBaseRetriever|retriever|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"source": "Milvus-CWpae",
"target": "RetrievalQA-x5nap",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "Milvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-x5nap"
},
{
"id": "reactflow__edge-InputNode-EfneiInputNode|InputNode-Efnei|input-RetrievalQA-x5napinput|input_node|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"source": "InputNode-Efnei",
"target": "RetrievalQA-x5nap",
"animated": true,
"className": "",
"sourceHandle": "InputNode|InputNode-Efnei|input",
"targetHandle": "input|input_node|RetrievalQA-x5nap"
},
{
"id": "reactflow__edge-OpenAIProxyEmbedding-swx68OpenAIProxyEmbedding|OpenAIProxyEmbedding-swx68|Embeddings-Milvus-CWpaeEmbeddings|embedding|Milvus-CWpae",
"style": {
"stroke": "#555"
},
"source": "OpenAIProxyEmbedding-swx68",
"target": "Milvus-CWpae",
"animated": true,
"className": "",
"sourceHandle": "OpenAIProxyEmbedding|OpenAIProxyEmbedding-swx68|Embeddings",
"targetHandle": "Embeddings|embedding|Milvus-CWpae"
},
{
"id": "reactflow__edge-InputFileNode-rUBtAInputFileNode|InputFileNode-rUBtA|fileNode-ElemUnstructuredLoaderV0-2K7FCfileNode|file_path|ElemUnstructuredLoaderV0-2K7FC",
"style": {
"stroke": "#555"
},
"source": "InputFileNode-rUBtA",
"target": "ElemUnstructuredLoaderV0-2K7FC",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "InputFileNode|InputFileNode-rUBtA|fileNode",
"targetHandle": "fileNode|file_path|ElemUnstructuredLoaderV0-2K7FC"
},
{
"id": "reactflow__edge-ElemUnstructuredLoaderV0-2K7FCElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-2K7FC|Document-Milvus-CWpaeDocument|documents|Milvus-CWpae",
"style": {
"stroke": "#555"
},
"source": "ElemUnstructuredLoaderV0-2K7FC",
"target": "Milvus-CWpae",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "ElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-2K7FC|Document",
"targetHandle": "Document|documents|Milvus-CWpae"
},
{
"id": "reactflow__edge-BishengLLM-86c36BishengLLM|BishengLLM-86c36|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-CombineDocsChain-RCPNABaseLanguageModel|llm|CombineDocsChain-RCPNA",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-86c36",
"target": "CombineDocsChain-RCPNA",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-86c36|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-RCPNA"
}
],
"nodes": [
{
"id": "CombineDocsChain-RCPNA",
"data": {
"id": "CombineDocsChain-RCPNA",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"prompt": {
"l2": false,
"info": "",
"list": false,
"name": "prompt",
"type": "BasePromptTemplate",
"l2_name": "prompt",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
},
"token_max": {
"l2": false,
"info": "当前只对stuff 生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"l2_name": "token_max",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 377,
"dragging": false,
"position": {
"x": 845.0164562381523,
"y": 204.35580627436775
},
"selected": false,
"positionAbsolute": {
"x": 845.0164562381523,
"y": 204.35580627436775
}
},
{
"id": "Milvus-CWpae",
"data": {
"id": "Milvus-CWpae",
"node": {
"l2_name": "知识库",
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"collection_name": {
"l2": false,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "选择知识库",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": ""
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"Milvus",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 347,
"dragging": false,
"position": {
"x": 1084.3422855390577,
"y": 695.7475330381905
},
"selected": false,
"positionAbsolute": {
"x": 1084.3422855390577,
"y": 695.7475330381905
}
},
{
"id": "RetrievalQA-x5nap",
"data": {
"id": "RetrievalQA-x5nap",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "RetrievalQA",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"combine_documents_chain": {
"l2": false,
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"l2_name": "combine_documents_chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"BaseRetrievalQA",
"RetrievalQA",
"Chain",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 317,
"dragging": false,
"position": {
"x": 1516.3297428236956,
"y": 563.8820893628344
},
"selected": false,
"positionAbsolute": {
"x": 1516.3297428236956,
"y": 563.8820893628344
}
},
{
"id": "InputNode-Efnei",
"data": {
"id": "InputNode-Efnei",
"node": {
"template": {
"_type": "input",
"input": {
"l2": false,
"info": "",
"list": true,
"name": "input",
"show": true,
"type": "str",
"value": [
"这篇文章的主要内容是什么？请生成该文章的摘要"
],
"l2_name": "input",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "输入内容"
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"input"
],
"display_name": "InputNode",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "InputNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 255,
"dragging": false,
"position": {
"x": 1245.632226598747,
"y": 190.02499863969456
},
"selected": false,
"positionAbsolute": {
"x": 1245.632226598747,
"y": 190.02499863969456
}
},
{
"id": "InputFileNode-rUBtA",
"data": {
"id": "InputFileNode-rUBtA",
"node": {
"template": {
"_type": "InputFileNode",
"file_path": {
"l2": false,
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "file",
"value": "",
"l2_name": "file_path",
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".jpg",
".png",
".jpeg",
".csv",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
".xlsx",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"jpg",
"png",
"jpeg",
"csv",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
"tiff",
"xlsx"
],
"file_path": null,
"multiline": false,
"placeholder": "",
"display_name": "输入内容"
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"fileNode"
],
"display_name": "InputFileNode",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "InputFileNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 255,
"dragging": false,
"position": {
"x": 189.1572111197503,
"y": 644.2048195861047
},
"selected": false,
"positionAbsolute": {
"x": 189.1572111197503,
"y": 644.2048195861047
}
},
{
"id": "OpenAIProxyEmbedding-swx68",
"data": {
"id": "OpenAIProxyEmbedding-swx68",
"node": {
"template": {
"_type": "proxy_embedding",
"proxy_url": {
"l2": false,
"info": "",
"list": false,
"name": "proxy_url",
"show": false,
"type": "str",
"value": "",
"l2_name": "proxy_url",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "http://43.133.35.137:8080"
}
},
"description": " 使用自建的embedding服务使用openai进行embed ",
"base_classes": [
"Embeddings"
],
"display_name": "OpenAIProxyEmbedding",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "OpenAIProxyEmbedding",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 173,
"dragging": false,
"position": {
"x": 617.8305142283027,
"y": 1050.5852539090324
},
"selected": false,
"positionAbsolute": {
"x": 617.8305142283027,
"y": 1050.5852539090324
}
},
{
"id": "ElemUnstructuredLoaderV0-2K7FC",
"data": {
"id": "ElemUnstructuredLoaderV0-2K7FC",
"node": {
"template": {
"_type": "ElemUnstructuredLoaderV0",
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"value": "{}",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Metadata"
},
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "fileNode",
"value": "",
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".jpg",
".png",
".jpeg",
".csv",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
".xlsx",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"jpg",
"png",
"jpeg",
"csv",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
"tiff",
"xlsx"
],
"multiline": false,
"placeholder": ""
},
"unstructured_api_url": {
"info": "",
"list": false,
"name": "unstructured_api_url",
"show": true,
"type": "str",
"value": "https://bisheng.dataelem.com/api/v1/etl4llm/predict",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "unstructured_api_url"
}
},
"description": "Loads a PDF with pypdf and chunks at character level. dummy version",
"base_classes": [
"Document"
],
"display_name": "ElemUnstructuredLoaderV0",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": ""
},
"type": "ElemUnstructuredLoaderV0",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 393,
"dragging": false,
"position": {
"x": 619.7399756791771,
"y": 615.1613231146783
},
"selected": false,
"positionAbsolute": {
"x": 619.7399756791771,
"y": 615.1613231146783
}
},
{
"id": "BishengLLM-86c36",
"data": {
"id": "BishengLLM-86c36",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 5,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": 321.78279586604197,
"y": 75.49875367670856
},
"selected": true,
"positionAbsolute": {
"x": 321.78279586604197,
"y": 75.49875367670856
}
}
],
"viewport": {
"x": 235.78430160378775,
"y": 22.158441172777714,
"zoom": 0.6998997500474495
}
},
"order_num": 6488027,
"guide_word": null
},
{
"update_time": "2024-08-20 16:57:42",
"parameters": null,
"name": "☘️招股书分析",
"description": "上传招股书报告，自动触发预设的分析问题",
"flow_id": "159a4536c3834406b5e134c25ab6fb4b",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 16:57:42",
"id": 179,
"data": {
"edges": [
{
"id": "reactflow__edge-CombineDocsChain-rc2FkCombineDocsChain|CombineDocsChain-rc2Fk|BaseCombineDocumentsChain|function-RetrievalQA-u9DZuBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-u9DZu",
"style": {
"stroke": "#555"
},
"source": "CombineDocsChain-rc2Fk",
"target": "RetrievalQA-u9DZu",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "CombineDocsChain|CombineDocsChain-rc2Fk|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-u9DZu"
},
{
"id": "reactflow__edge-RetrievalQA-u9DZuRetrievalQA|RetrievalQA-u9DZu|Chain|BaseRetrievalQA|RetrievalQA|function-Tool-FbOTwfunction|func|Tool-FbOTw",
"style": {
"stroke": "#555"
},
"source": "RetrievalQA-u9DZu",
"target": "Tool-FbOTw",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "RetrievalQA|RetrievalQA-u9DZu|Chain|BaseRetrievalQA|RetrievalQA|function",
"targetHandle": "function|func|Tool-FbOTw"
},
{
"id": "reactflow__edge-Milvus-4LcfVMilvus|Milvus-4LcfV|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever-RetrievalQA-u9DZuBaseRetriever|retriever|RetrievalQA-u9DZu",
"style": {
"stroke": "#555"
},
"source": "Milvus-4LcfV",
"target": "RetrievalQA-u9DZu",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "Milvus|Milvus-4LcfV|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-u9DZu"
},
{
"id": "reactflow__edge-Tool-FbOTwTool|Tool-FbOTw|Tool|BaseTool-ZeroShotAgent-o94zNBaseTool|tools|ZeroShotAgent-o94zN",
"style": {
"stroke": "#555"
},
"source": "Tool-FbOTw",
"target": "ZeroShotAgent-o94zN",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "Tool|Tool-FbOTw|Tool|BaseTool",
"targetHandle": "BaseTool|tools|ZeroShotAgent-o94zN"
},
{
"id": "reactflow__edge-Calculator-CfVqiCalculator|Calculator-CfVqi|Tool|BaseTool-ZeroShotAgent-o94zNBaseTool|tools|ZeroShotAgent-o94zN",
"style": {
"stroke": "#555"
},
"source": "Calculator-CfVqi",
"target": "ZeroShotAgent-o94zN",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "Calculator|Calculator-CfVqi|Tool|BaseTool",
"targetHandle": "BaseTool|tools|ZeroShotAgent-o94zN"
},
{
"id": "reactflow__edge-OpenAIProxyEmbedding-NPuMIOpenAIProxyEmbedding|OpenAIProxyEmbedding-NPuMI|Embeddings-Milvus-4LcfVEmbeddings|embedding|Milvus-4LcfV",
"style": {
"stroke": "#555"
},
"source": "OpenAIProxyEmbedding-NPuMI",
"target": "Milvus-4LcfV",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "OpenAIProxyEmbedding|OpenAIProxyEmbedding-NPuMI|Embeddings",
"targetHandle": "Embeddings|embedding|Milvus-4LcfV"
},
{
"id": "reactflow__edge-RecursiveCharacterTextSplitter-sTTDTRecursiveCharacterTextSplitter|RecursiveCharacterTextSplitter-sTTDT|Document-Milvus-4LcfVDocument|documents|Milvus-4LcfV",
"style": {
"stroke": "#555"
},
"source": "RecursiveCharacterTextSplitter-sTTDT",
"target": "Milvus-4LcfV",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "RecursiveCharacterTextSplitter|RecursiveCharacterTextSplitter-sTTDT|Document",
"targetHandle": "Document|documents|Milvus-4LcfV"
},
{
"id": "reactflow__edge-InputNode-HLAg0InputNode|InputNode-HLAg0|input-ZeroShotAgent-o94zNinput|input_node|ZeroShotAgent-o94zN",
"style": {
"stroke": "#555"
},
"source": "InputNode-HLAg0",
"target": "ZeroShotAgent-o94zN",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "InputNode|InputNode-HLAg0|input",
"targetHandle": "input|input_node|ZeroShotAgent-o94zN"
},
{
"id": "reactflow__edge-BingSearchAPIWrapper-I86u8BingSearchAPIWrapper|BingSearchAPIWrapper-I86u8|BingSearchAPIWrapper-BingSearchRun-gmZjQBingSearchAPIWrapper|api_wrapper|BingSearchRun-gmZjQ",
"style": {
"stroke": "#555"
},
"source": "BingSearchAPIWrapper-I86u8",
"target": "BingSearchRun-gmZjQ",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "BingSearchAPIWrapper|BingSearchAPIWrapper-I86u8|BingSearchAPIWrapper",
"targetHandle": "BingSearchAPIWrapper|api_wrapper|BingSearchRun-gmZjQ"
},
{
"id": "reactflow__edge-BingSearchRun-gmZjQBingSearchRun|BingSearchRun-gmZjQ|Tool|BaseTool|BaseTool|BingSearchRun-ZeroShotAgent-o94zNBaseTool|tools|ZeroShotAgent-o94zN",
"style": {
"stroke": "#555"
},
"source": "BingSearchRun-gmZjQ",
"target": "ZeroShotAgent-o94zN",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "BingSearchRun|BingSearchRun-gmZjQ|Tool|BaseTool|BaseTool|BingSearchRun",
"targetHandle": "BaseTool|tools|ZeroShotAgent-o94zN"
},
{
"id": "reactflow__edge-InputFileNode-IR0Z1InputFileNode|InputFileNode-IR0Z1|fileNode-ElemUnstructuredLoaderV0-4rq3AfileNode|file_path|ElemUnstructuredLoaderV0-4rq3A",
"style": {
"stroke": "#555"
},
"source": "InputFileNode-IR0Z1",
"target": "ElemUnstructuredLoaderV0-4rq3A",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "InputFileNode|InputFileNode-IR0Z1|fileNode",
"targetHandle": "fileNode|file_path|ElemUnstructuredLoaderV0-4rq3A"
},
{
"id": "reactflow__edge-ElemUnstructuredLoaderV0-4rq3AElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-4rq3A|Document-RecursiveCharacterTextSplitter-sTTDTDocument|documents|RecursiveCharacterTextSplitter-sTTDT",
"style": {
"stroke": "#555"
},
"source": "ElemUnstructuredLoaderV0-4rq3A",
"target": "RecursiveCharacterTextSplitter-sTTDT",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "ElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-4rq3A|Document",
"targetHandle": "Document|documents|RecursiveCharacterTextSplitter-sTTDT"
},
{
"id": "reactflow__edge-BishengLLM-6dc86BishengLLM|BishengLLM-6dc86|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-CombineDocsChain-rc2FkBaseLanguageModel|llm|CombineDocsChain-rc2Fk",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-6dc86",
"target": "CombineDocsChain-rc2Fk",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-6dc86|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-rc2Fk"
},
{
"id": "reactflow__edge-BishengLLM-6dc86BishengLLM|BishengLLM-6dc86|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-Calculator-CfVqiBaseLanguageModel|llm|Calculator-CfVqi",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-6dc86",
"target": "Calculator-CfVqi",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-6dc86|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|Calculator-CfVqi"
},
{
"id": "reactflow__edge-BishengLLM-6dc86BishengLLM|BishengLLM-6dc86|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-ZeroShotAgent-o94zNBaseLanguageModel|llm|ZeroShotAgent-o94zN",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-6dc86",
"target": "ZeroShotAgent-o94zN",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-6dc86|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|ZeroShotAgent-o94zN"
}
],
"nodes": [
{
"id": "CombineDocsChain-rc2Fk",
"data": {
"id": "CombineDocsChain-rc2Fk",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"token_max": {
"l2": false,
"info": "当前只对stuff 生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"l2_name": "token_max",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "map_rerank",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 377,
"position": {
"x": 32.375529385543246,
"y": 759.3887339718593
},
"selected": false,
"positionAbsolute": {
"x": 32.375529385543246,
"y": 759.3887339718593
}
},
{
"id": "RetrievalQA-u9DZu",
"data": {
"id": "RetrievalQA-u9DZu",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "RetrievalQA",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"combine_documents_chain": {
"l2": false,
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"l2_name": "combine_documents_chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"Chain",
"BaseRetrievalQA",
"RetrievalQA",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 317,
"position": {
"x": 1055.9846105131896,
"y": 1319.580472587641
},
"selected": false,
"positionAbsolute": {
"x": 1055.9846105131896,
"y": 1319.580472587641
}
},
{
"id": "Tool-FbOTw",
"data": {
"id": "Tool-FbOTw",
"node": {
"template": {
"func": {
"l2": false,
"info": "",
"list": false,
"name": "func",
"show": true,
"type": "function",
"l2_name": "func",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"name": {
"l2": false,
"info": "",
"list": false,
"name": "name",
"show": true,
"type": "str",
"value": "Basic_information",
"l2_name": "name",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"_type": "Tool",
"description": {
"l2": false,
"info": "",
"list": false,
"name": "description",
"show": true,
"type": "str",
"value": "如果用户需要知道关于这家公司的各方面信息，可以使用该工具。如：\n- 某公司的盈利模式是什么\n- 某公司的保荐机构是哪家公司？\n- 某公司的审计机构是哪家公司？\n- 某公司的主要产品是什么？",
"l2_name": "如何使用用户上传的文档",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"return_direct": {
"l2": false,
"info": "",
"list": false,
"name": "return_direct",
"show": true,
"type": "bool",
"value": false,
"l2_name": "return_direct",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Converts a chain, agent or function into a tool.",
"base_classes": [
"Tool",
"BaseTool"
],
"display_name": "Tool",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "Tool",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 445,
"position": {
"x": 1621.5305224346598,
"y": 1282.841953911718
},
"selected": false,
"positionAbsolute": {
"x": 1621.5305224346598,
"y": 1282.841953911718
}
},
{
"id": "Milvus-4LcfV",
"data": {
"id": "Milvus-4LcfV",
"node": {
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"collection_name": {
"l2": false,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "招股书",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": ""
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"VectorStore",
"Milvus",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 347,
"position": {
"x": 522.9361398003147,
"y": 1462.3167297901632
},
"selected": false,
"positionAbsolute": {
"x": 522.9361398003147,
"y": 1462.3167297901632
}
},
{
"id": "ZeroShotAgent-o94zN",
"data": {
"id": "ZeroShotAgent-o94zN",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"_type": "ZeroShotAgent",
"tools": {
"l2": false,
"info": "",
"list": true,
"name": "tools",
"show": true,
"type": "BaseTool",
"l2_name": "tools",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"prefix": {
"l2": false,
"info": "",
"list": false,
"name": "prefix",
"show": true,
"type": "str",
"value": "Answer the following questions as best you can. Think step by step. You have access to the following tools:",
"l2_name": "prefix",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
},
"suffix": {
"l2": false,
"info": "",
"list": false,
"name": "suffix",
"show": true,
"type": "str",
"value": "Begin!\n\nQuestion: {query}\nThought:{agent_scratchpad}",
"l2_name": "suffix",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "AgentOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": true,
"type": "str",
"value": [
"query",
"agent_scratchpad"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callback_manager": {
"l2": false,
"info": "",
"list": false,
"name": "callback_manager",
"show": false,
"type": "BaseCallbackManager",
"l2_name": "callback_manager",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"format_instructions": {
"l2": false,
"info": "",
"list": false,
"name": "format_instructions",
"show": true,
"type": "str",
"value": "Use the following format:\n\nQuestion: 该公司的保荐机构是哪家公司？\n# the query question you must answer\nThought: 我需要从文档中查询达梦股份有限公司的保荐机构\n# you should always think about what to do\nAction: Basic_information\n# the action to take, should be one of [{tool_names}]\nAction Input: 该公司的保荐机构是哪家公司？\n# the input to the action\nObservation: 我发现文章中提到达梦股份有限公司的保荐机构是招商证券\n# the result of the action\n... (this Thought/Action/Action Input/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: #the final answer to the original input question",
"l2_name": "format_instructions",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
}
},
"description": "Construct an agent from an LLM and tools.",
"base_classes": [
"BaseSingleActionAgent",
"ZeroShotAgent",
"Agent",
"function"
],
"display_name": "ZeroShotAgent",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/agents/how_to/custom_mrkl_agent"
},
"type": "ZeroShotAgent",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 659,
"position": {
"x": 2724.755245757752,
"y": 792.4038456822296
},
"selected": false,
"positionAbsolute": {
"x": 2724.755245757752,
"y": 792.4038456822296
}
},
{
"id": "Calculator-CfVqi",
"data": {
"id": "Calculator-CfVqi",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"_type": "Calculator"
},
"description": "Useful for when you need to answer questions about math.",
"base_classes": [
"Tool",
"BaseTool"
],
"display_name": "Calculator",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "Calculator",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 229,
"position": {
"x": 1604.8480818394096,
"y": 973.4644452164116
},
"selected": false,
"positionAbsolute": {
"x": 1604.8480818394096,
"y": 973.4644452164116
}
},
{
"id": "OpenAIProxyEmbedding-NPuMI",
"data": {
"id": "OpenAIProxyEmbedding-NPuMI",
"node": {
"template": {
"_type": "proxy_embedding",
"proxy_url": {
"l2": false,
"info": "",
"list": false,
"name": "proxy_url",
"show": false,
"type": "str",
"value": "",
"l2_name": "proxy_url",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "http://43.133.35.137:8080"
}
},
"description": " 使用自建的embedding服务使用openai进行embed ",
"base_classes": [
"Embeddings"
],
"display_name": "OpenAIProxyEmbedding",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "OpenAIProxyEmbedding",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 173,
"position": {
"x": 27.86167751822711,
"y": 1796.72713464364
},
"selected": false,
"positionAbsolute": {
"x": 27.86167751822711,
"y": 1796.72713464364
}
},
{
"id": "InputFileNode-IR0Z1",
"data": {
"id": "InputFileNode-IR0Z1",
"node": {
"template": {
"_type": "InputFileNode",
"file_path": {
"l2": false,
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "file",
"value": "",
"l2_name": "file_path",
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".jpg",
".png",
".jpeg",
".csv",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
".xlsx",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"jpg",
"png",
"jpeg",
"csv",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
"tiff",
"xlsx"
],
"file_path": null,
"multiline": false,
"placeholder": "",
"display_name": "输入内容"
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"fileNode"
],
"display_name": "InputFileNode",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "InputFileNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 255,
"position": {
"x": -932.810583795316,
"y": 1190.272406644502
},
"selected": false,
"positionAbsolute": {
"x": -932.810583795316,
"y": 1190.272406644502
}
},
{
"id": "RecursiveCharacterTextSplitter-sTTDT",
"data": {
"id": "RecursiveCharacterTextSplitter-sTTDT",
"node": {
"template": {
"_type": "RecursiveCharacterTextSplitter",
"documents": {
"l2": false,
"info": "",
"list": false,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"chunk_size": {
"l2": false,
"info": "",
"list": false,
"name": "chunk_size",
"show": true,
"type": "int",
"value": 1000,
"l2_name": "chunk_size",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Chunk Size"
},
"separators": {
"l2": false,
"info": "",
"list": false,
"name": "separators",
"show": true,
"type": "str",
"value": "\\n",
"l2_name": "separators",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Separator"
},
"chunk_overlap": {
"l2": false,
"info": "",
"list": false,
"name": "chunk_overlap",
"show": true,
"type": "int",
"value": 200,
"l2_name": "chunk_overlap",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Chunk Overlap"
},
"separator_type": {
"l2": false,
"info": "",
"list": true,
"name": "separator_type",
"show": true,
"type": "str",
"value": "Text",
"l2_name": "separator_type",
"options": [
"Text",
"cpp",
"go",
"html",
"java",
"js",
"latex",
"markdown",
"php",
"proto",
"python",
"rst",
"ruby",
"rust",
"scala",
"sol",
"swift"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Separator Type"
}
},
"description": "Implementation of splitting text that looks at characters.",
"base_classes": [
"Document"
],
"display_name": "RecursiveCharacterTextSplitter",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/data_connection/document_transformers/text_splitters/recursive_text_splitter"
},
"type": "RecursiveCharacterTextSplitter",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 561,
"position": {
"x": 28.985765191920905,
"y": 1167.7466689744615
},
"selected": false,
"positionAbsolute": {
"x": 28.985765191920905,
"y": 1167.7466689744615
}
},
{
"id": "InputNode-HLAg0",
"data": {
"id": "InputNode-HLAg0",
"node": {
"template": {
"_type": "input",
"input": {
"l2": true,
"info": "",
"list": true,
"name": "input",
"show": true,
"type": "str",
"value": [
"该公司的主营业务包括哪些，请概述",
"该公司的盈利模式是什么？",
"该公司的销售模式有哪些？",
"该公司有多少专利和知识产权？",
"该公司处在哪个行业，查询一下该行业最近几年中国的市场规模分别为多少？",
"该公司的主要股东及其持股比例分别是多少",
"该公司是否面临法律诉讼或其他争议事项？如果是，请提供相关的细节和进展情况",
"该公司是否存在专利或其他知识产权方面的风险？请说明相关情况",
"该公司面临的主要市场竞争对手和他们在市场中的地位是什么？",
"该公司的主要产品或服务是什么？提供关键产品的详细描述",
"该公司的市场份额如何？是否有计划扩大市场份额？",
"该公司的盈利模式是什么？请解释公司如何获得收入",
"请详细说明公司的资金需求及其资金用途",
"该公司是否有债务或其他财务义务",
"请提供公司的人力资源情况，包括员工总数和关键管理团队成员的背景信息",
"请根据公司募集资金用途以及未来发展战略，推断公司是否计划在未来几年内进行研发和创新投资？",
"公司目标客户是谁？",
"从市场份额方面分析该公司的核心竞争力是什么？",
"从核心技术或核心产品方面分析该公司的核心竞争力是什么？",
"公司重要的合作伙伴与供应商有哪些？从他们那里能获取哪些核心资源?",
"该公司最重要的固定成本是什么?",
"结合财务报表，哪类费用占比高？这类费用主要用于哪些方面的支出？",
"公司2021年的应收账款周转率为多少，对比可比公司的情况如何？",
"公司的息税折旧摊销前利润相较于三年前增长了多少？",
"公司的息税折旧摊销前利润的增长趋势如何？",
"公司的归属于发行人股东的净利润的增长是否持续稳定？",
"公司归属于发行人股东扣除非经常性损益后的净利润有何变化？",
"公司在研发方面的投入与研发费用占营业收入的比例如何评价？",
"公司研发费用占营业收入的比例有何变化？",
"每股经营活动产生的现金流量反映了公司的经营效益如何？",
"公司在过去三年中，每股经营活动产生的现金流量有何变化？",
"公司每股净现金流量有何变化？",
"公司的流动比率水平如何评价？",
"公司的速动比率有何变化？",
"公司的资产负债率（母公司口径）有何变化？",
"公司的资产负债率（合并口径）水平如何评价？",
"公司的归属于发行人股东的每股净资产有何变化？",
""
],
"l2_name": "报告分析维度",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "输入内容"
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"input"
],
"display_name": "InputNode",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "InputNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 706,
"position": {
"x": 2176.519795047715,
"y": 1422.887731295922
},
"selected": false,
"positionAbsolute": {
"x": 2176.519795047715,
"y": 1422.887731295922
}
},
{
"id": "BingSearchAPIWrapper-I86u8",
"data": {
"id": "BingSearchAPIWrapper-I86u8",
"node": {
"template": {
"k": {
"l2": false,
"info": "",
"list": false,
"name": "k",
"show": false,
"type": "int",
"value": 10,
"l2_name": "k",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "BingSearchAPIWrapper",
"bing_search_url": {
"l2": false,
"info": "",
"list": false,
"name": "bing_search_url",
"show": true,
"type": "str",
"value": "https://api.bing.microsoft.com/v7.0/search",
"l2_name": "bing_search_url",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"bing_subscription_key": {
"l2": false,
"info": "",
"list": false,
"name": "bing_subscription_key",
"show": true,
"type": "str",
"value": "137b003c89f04966ac3b1081ee5b10bc",
"l2_name": "bing_subscription_key",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Wrapper for Bing Search API.",
"base_classes": [
"BingSearchAPIWrapper"
],
"display_name": "BingSearchAPIWrapper",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "BingSearchAPIWrapper",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 337,
"position": {
"x": 1084.3965767342606,
"y": 279.87865179934
},
"selected": false,
"positionAbsolute": {
"x": 1084.3965767342606,
"y": 279.87865179934
}
},
{
"id": "BingSearchRun-gmZjQ",
"data": {
"id": "BingSearchRun-gmZjQ",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": false,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "BingSearchRun",
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": false,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"api_wrapper": {
"l2": false,
"info": "",
"list": false,
"name": "api_wrapper",
"show": true,
"type": "BingSearchAPIWrapper",
"l2_name": "api_wrapper",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": ""
},
"args_schema": {
"l2": false,
"info": "",
"list": false,
"name": "args_schema",
"show": false,
"type": "Type[pydantic.main.BaseModel]",
"l2_name": "args_schema",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"handle_tool_error": {
"l2": false,
"info": "",
"list": false,
"name": "handle_tool_error",
"show": false,
"type": "bool",
"value": false,
"l2_name": "handle_tool_error",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "A wrapper around Bing Search. Useful for when you need to answer questions about current events. Input should be a search query.",
"base_classes": [
"Tool",
"BaseTool",
"BaseTool",
"BingSearchRun"
],
"display_name": "BingSearchRun",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "BingSearchRun",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 249,
"position": {
"x": 1587.0735944565677,
"y": 564.1271785797638
},
"selected": false,
"positionAbsolute": {
"x": 1587.0735944565677,
"y": 564.1271785797638
}
},
{
"id": "ElemUnstructuredLoaderV0-4rq3A",
"data": {
"id": "ElemUnstructuredLoaderV0-4rq3A",
"node": {
"template": {
"_type": "ElemUnstructuredLoaderV0",
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"value": "{}",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Metadata"
},
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "fileNode",
"value": "",
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".jpg",
".png",
".jpeg",
".csv",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
".xlsx",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"jpg",
"png",
"jpeg",
"csv",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
"tiff",
"xlsx"
],
"multiline": false,
"placeholder": ""
},
"unstructured_api_url": {
"info": "",
"list": false,
"name": "unstructured_api_url",
"show": true,
"type": "str",
"value": "https://bisheng.dataelem.com/api/v1/etl4llm/predict",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "unstructured_api_url"
}
},
"description": "Loads a PDF with pypdf and chunks at character level. dummy version",
"base_classes": [
"Document"
],
"display_name": "ElemUnstructuredLoaderV0",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": ""
},
"type": "ElemUnstructuredLoaderV0",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 393,
"dragging": false,
"position": {
"x": -468.3981411294128,
"y": 1150.9063802769972
},
"selected": false,
"positionAbsolute": {
"x": -468.3981411294128,
"y": 1150.9063802769972
}
},
{
"id": "BishengLLM-6dc86",
"data": {
"id": "BishengLLM-6dc86",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 5,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": -966.0777669425454,
"y": 569.5274332939994
},
"selected": false,
"positionAbsolute": {
"x": -966.0777669425454,
"y": 569.5274332939994
}
}
],
"viewport": {
"x": 446.8071199372391,
"y": -15.008593660173858,
"zoom": 0.3819453782554565
}
},
"order_num": 6553562,
"guide_word": null
},
{
"update_time": "2024-10-15 18:39:27",
"parameters": null,
"name": "⭐️知识库问答",
"description": "选择知识库，进行知识库问答",
"flow_id": "0c1649f7cd774248990bf7d961e8dcff",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 17:04:03",
"id": 180,
"data": {
"edges": [
{
"id": "reactflow__edge-Milvus-CWpaeMilvus|Milvus-CWpae|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever-RetrievalQA-x5napBaseRetriever|retriever|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"source": "Milvus-CWpae",
"target": "RetrievalQA-x5nap",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "Milvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-x5nap"
},
{
"id": "reactflow__edge-CombineDocsChain-71a85CombineDocsChain|CombineDocsChain-71a85|BaseCombineDocumentsChain|function-RetrievalQA-x5napBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"source": "CombineDocsChain-71a85",
"target": "RetrievalQA-x5nap",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "CombineDocsChain|CombineDocsChain-71a85|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap"
},
{
"id": "reactflow__edge-PromptTemplate-3c164PromptTemplate|PromptTemplate-3c164|Runnable|StringPromptTemplate|RunnableSerializable|PromptTemplate|Generic|BasePromptTemplate-CombineDocsChain-71a85BasePromptTemplate|document_prompt|CombineDocsChain-71a85",
"style": {
"stroke": "#555"
},
"source": "PromptTemplate-3c164",
"target": "CombineDocsChain-71a85",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "PromptTemplate|PromptTemplate-3c164|Runnable|StringPromptTemplate|RunnableSerializable|PromptTemplate|Generic|BasePromptTemplate",
"targetHandle": "BasePromptTemplate|document_prompt|CombineDocsChain-71a85"
},
{
"id": "reactflow__edge-PromptTemplate-ced29PromptTemplate|PromptTemplate-ced29|Runnable|StringPromptTemplate|RunnableSerializable|PromptTemplate|Generic|BasePromptTemplate-CombineDocsChain-71a85BasePromptTemplate|prompt|CombineDocsChain-71a85",
"style": {
"stroke": "#555"
},
"source": "PromptTemplate-ced29",
"target": "CombineDocsChain-71a85",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "PromptTemplate|PromptTemplate-ced29|Runnable|StringPromptTemplate|RunnableSerializable|PromptTemplate|Generic|BasePromptTemplate",
"targetHandle": "BasePromptTemplate|prompt|CombineDocsChain-71a85"
},
{
"id": "reactflow__edge-BishengLLM-02d50BishengLLM|BishengLLM-02d50|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-CombineDocsChain-71a85BaseLanguageModel|llm|CombineDocsChain-71a85",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-02d50",
"target": "CombineDocsChain-71a85",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-02d50|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-71a85"
}
],
"nodes": [
{
"id": "Milvus-CWpae",
"data": {
"id": "Milvus-CWpae",
"node": {
"l2_name": "知识库",
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": false,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"collection_name": {
"l2": true,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "毕昇产品文档20240421",
"l2_name": "选择知识库",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": 4597
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"Milvus",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 311,
"dragging": false,
"position": {
"x": 857.7079649166786,
"y": 666.0353925000366
},
"selected": false,
"positionAbsolute": {
"x": 857.7079649166786,
"y": 666.0353925000366
}
},
{
"id": "RetrievalQA-x5nap",
"data": {
"id": "RetrievalQA-x5nap",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "RetrievalQA",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"combine_documents_chain": {
"l2": false,
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"l2_name": "combine_documents_chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"RetrievalQA",
"Chain",
"BaseRetrievalQA",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 317,
"position": {
"x": 1521.2877663953464,
"y": 287.6148502766154
},
"selected": false,
"positionAbsolute": {
"x": 1521.2877663953464,
"y": 287.6148502766154
}
},
{
"id": "CombineDocsChain-71a85",
"data": {
"id": "CombineDocsChain-71a85",
"node": {
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"prompt": {
"info": "只对Stuff类型生效",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
},
"token_max": {
"info": "只对Stuff类型生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"document_prompt": {
"info": "",
"list": false,
"name": "document_prompt",
"show": true,
"type": "BasePromptTemplate",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 449,
"position": {
"x": 847.130082564399,
"y": 10.769465837008791
},
"selected": false,
"positionAbsolute": {
"x": 847.130082564399,
"y": 10.769465837008791
}
},
{
"id": "PromptTemplate-3c164",
"data": {
"id": "PromptTemplate-3c164",
"node": {
"name": "",
"l2_name": "召回段落拼接提示词",
"template": {
"_type": "PromptTemplate",
"source": {
"info": "",
"list": false,
"name": "source",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "source"
},
"template": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "title: {source}\n{page_content}\n----------------",
"l2_name": "提示词模板",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"input_types": {
"info": "",
"list": false,
"name": "input_types",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"page_content": {
"info": "",
"list": false,
"name": "page_content",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "page_content"
},
"output_parser": {
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"source",
"page_content"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"template_format": {
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "Union[Literal['f-string'], Literal['jinja2']]",
"value": "f-string",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"partial_variables": {
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"validate_template": {
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "A prompt template for a language model.",
"base_classes": [
"Runnable",
"StringPromptTemplate",
"RunnableSerializable",
"PromptTemplate",
"Generic",
"BasePromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"source",
"page_content"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {}
},
"type": "PromptTemplate",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 455,
"dragging": false,
"position": {
"x": 249.5669301132289,
"y": -472.7534674830942
},
"selected": false,
"positionAbsolute": {
"x": 249.5669301132289,
"y": -472.7534674830942
}
},
{
"id": "PromptTemplate-ced29",
"data": {
"id": "PromptTemplate-ced29",
"node": {
"name": "",
"l2_name": "大模型问答提示词",
"template": {
"_type": "PromptTemplate",
"context": {
"info": "",
"list": false,
"name": "context",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "context"
},
"question": {
"info": "",
"list": false,
"name": "question",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "question"
},
"template": {
"l2": true,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "参考文本：\n{context}\n\n----------------------\n# Role：企业客服\n\n# Background :\n你是一家名叫XX公司的客服人员。你的日常工作是回复公司内用户的各类咨询问题，所以当用户提到“公司”时，指的是XX公司。\n\n# Goals:\n- 以上【参考文本】中是从企业知识库中查找到的相关信息，只能结合以上信息进行回答，若以上内容为空或其中没有找到能回答【用户问题】的内容时，则回复“没有找到相关内容”，不能根据你自己的知识自己发挥。\n- 特别要注意区分【用户问题】与【参考文本】中不同的日期、人名、公司名这些关键信息。\n- 用中文回答问题，并且答案要严谨、专业、清晰、可读性好。\n- 拥有排版审美, 会利用序号, 缩进, 分隔线和换行符等等来美化信息排版。\n\n----------------------\n\n用户问题: {question}\n你的回答:",
"l2_name": "提示词模板",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"input_types": {
"info": "",
"list": false,
"name": "input_types",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"output_parser": {
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"context",
"question"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"template_format": {
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "Union[Literal['f-string'], Literal['jinja2']]",
"value": "f-string",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"partial_variables": {
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"validate_template": {
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "A prompt template for a language model.",
"base_classes": [
"Runnable",
"StringPromptTemplate",
"RunnableSerializable",
"PromptTemplate",
"Generic",
"BasePromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"context",
"question"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {}
},
"type": "PromptTemplate",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 455,
"dragging": false,
"position": {
"x": 239.9768809008542,
"y": 26.706867771652675
},
"selected": false,
"positionAbsolute": {
"x": 239.9768809008542,
"y": 26.706867771652675
}
},
{
"id": "BishengLLM-02d50",
"data": {
"id": "BishengLLM-02d50",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 5,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": 233.388284890662,
"y": 540.5502864443121
},
"selected": true,
"positionAbsolute": {
"x": 233.388284890662,
"y": 540.5502864443121
}
}
],
"viewport": {
"x": 282.57805303810756,
"y": 294.51040071091927,
"zoom": 0.5373960694755959
}
},
"order_num": null,
"guide_word": null
},
{
"update_time": "2024-08-20 17:08:01",
"parameters": null,
"name": "⭐️ Word 格式专业报告生成【需配置】",
"description": "【word 模板默认是空的，需要自己进行配置】\n将复杂的专业报告拆分成独立的具体章节任务，然后将大模型生成内容自动填入到预设 word 模板中，从而完成专业报告的生成。",
"flow_id": "212df93d0f8445c690a9aeebe684db84",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 17:08:01",
"id": 181,
"data": {
"edges": [
{
"id": "reactflow__edge-CombineDocsChain-tv0S2CombineDocsChain|CombineDocsChain-tv0S2|BaseCombineDocumentsChain|function-RetrievalQA-F6quWBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-F6quW",
"style": {
"stroke": "#555"
},
"source": "CombineDocsChain-tv0S2",
"target": "RetrievalQA-F6quW",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "CombineDocsChain|CombineDocsChain-tv0S2|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-F6quW"
},
{
"id": "reactflow__edge-RetrievalQA-F6quWRetrievalQA|RetrievalQA-F6quW|BaseRetrievalQA|Chain|RetrievalQA|function-Report-PRcYLfunction|chains|Report-PRcYL",
"style": {
"stroke": "#555"
},
"source": "RetrievalQA-F6quW",
"target": "Report-PRcYL",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "RetrievalQA|RetrievalQA-F6quW|BaseRetrievalQA|Chain|RetrievalQA|function",
"targetHandle": "function|chains|Report-PRcYL"
},
{
"id": "reactflow__edge-InputNode-TdX3BInputNode|InputNode-TdX3B|input-RetrievalQA-F6quWinput|input_node|RetrievalQA-F6quW",
"style": {
"stroke": "#555"
},
"source": "InputNode-TdX3B",
"target": "RetrievalQA-F6quW",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "InputNode|InputNode-TdX3B|input",
"targetHandle": "input|input_node|RetrievalQA-F6quW"
},
{
"id": "reactflow__edge-ElemUnstructuredLoaderV0-nYygCElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-nYygC|Document-RecursiveCharacterTextSplitter-6g2FADocument|documents|RecursiveCharacterTextSplitter-6g2FA",
"style": {
"stroke": "#555"
},
"source": "ElemUnstructuredLoaderV0-nYygC",
"target": "RecursiveCharacterTextSplitter-6g2FA",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "ElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-nYygC|Document",
"targetHandle": "Document|documents|RecursiveCharacterTextSplitter-6g2FA"
},
{
"id": "reactflow__edge-VariableNode-UB934VariableNode|VariableNode-UB934|VariableNode-Report-PRcYLVariableNode|variables|Report-PRcYL",
"style": {
"stroke": "#555"
},
"source": "VariableNode-UB934",
"target": "Report-PRcYL",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "VariableNode|VariableNode-UB934|VariableNode",
"targetHandle": "VariableNode|variables|Report-PRcYL"
},
{
"id": "reactflow__edge-Milvus-fSZ8oMilvus|Milvus-fSZ8o|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever-RetrievalQA-F6quWBaseRetriever|retriever|RetrievalQA-F6quW",
"style": {
"stroke": "#555"
},
"source": "Milvus-fSZ8o",
"target": "RetrievalQA-F6quW",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "Milvus|Milvus-fSZ8o|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-F6quW"
},
{
"id": "reactflow__edge-RecursiveCharacterTextSplitter-6g2FARecursiveCharacterTextSplitter|RecursiveCharacterTextSplitter-6g2FA|Document-Milvus-fSZ8oDocument|documents|Milvus-fSZ8o",
"style": {
"stroke": "#555"
},
"source": "RecursiveCharacterTextSplitter-6g2FA",
"target": "Milvus-fSZ8o",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "RecursiveCharacterTextSplitter|RecursiveCharacterTextSplitter-6g2FA|Document",
"targetHandle": "Document|documents|Milvus-fSZ8o"
},
{
"id": "reactflow__edge-OpenAIProxyEmbedding-0OoJhOpenAIProxyEmbedding|OpenAIProxyEmbedding-0OoJh|Embeddings-Milvus-fSZ8oEmbeddings|embedding|Milvus-fSZ8o",
"style": {
"stroke": "#555"
},
"source": "OpenAIProxyEmbedding-0OoJh",
"target": "Milvus-fSZ8o",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "OpenAIProxyEmbedding|OpenAIProxyEmbedding-0OoJh|Embeddings",
"targetHandle": "Embeddings|embedding|Milvus-fSZ8o"
},
{
"id": "reactflow__edge-InputFileNode-a4516InputFileNode|InputFileNode-a4516|fileNode-ElemUnstructuredLoaderV0-nYygCfileNode|file_path|ElemUnstructuredLoaderV0-nYygC",
"style": {
"stroke": "#555"
},
"source": "InputFileNode-a4516",
"target": "ElemUnstructuredLoaderV0-nYygC",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "InputFileNode|InputFileNode-a4516|fileNode",
"targetHandle": "fileNode|file_path|ElemUnstructuredLoaderV0-nYygC"
},
{
"id": "reactflow__edge-BishengLLM-a972dBishengLLM|BishengLLM-a972d|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-CombineDocsChain-tv0S2BaseLanguageModel|llm|CombineDocsChain-tv0S2",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-a972d",
"target": "CombineDocsChain-tv0S2",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-a972d|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-tv0S2"
}
],
"nodes": [
{
"id": "Report-PRcYL",
"data": {
"id": "Report-PRcYL",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "Report",
"chains": {
"l2": false,
"info": "",
"list": true,
"name": "chains",
"show": true,
"type": "function",
"l2_name": "chains",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "functions"
},
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": false,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"l2_name": "verbose",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": false,
"type": "str",
"value": "report_name",
"l2_name": "input_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"stop_flag": {
"l2": false,
"info": "",
"list": false,
"name": "stop_flag",
"show": false,
"type": "bool",
"value": false,
"l2_name": "stop_flag",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"variables": {
"l2": false,
"info": "",
"list": true,
"name": "variables",
"show": true,
"type": "VariableNode",
"l2_name": "variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": false,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": false,
"type": "str",
"value": "text",
"l2_name": "output_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"report_name": {
"l2": false,
"info": "the file name we generate",
"list": false,
"name": "report_name",
"show": true,
"type": "str",
"value": "尽调报告",
"l2_name": "report_name",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Report Name"
}
},
"description": "",
"base_classes": [
"Chain",
"Report"
],
"display_name": "Report",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "Report",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 359,
"dragging": false,
"position": {
"x": 2893.098387884978,
"y": 854.5001552884537
},
"selected": false,
"positionAbsolute": {
"x": 2893.098387884978,
"y": 854.5001552884537
}
},
{
"id": "InputNode-TdX3B",
"data": {
"id": "InputNode-TdX3B",
"node": {
"template": {
"_type": "input",
"input": {
"l2": false,
"info": "",
"list": true,
"name": "input",
"show": true,
"type": "str",
"value": [
"公司主营业务的市场定位是什么？",
"公司的主要供应商情况是什么？",
"公司营业成本是多少？各项成本占营业成本的比例是多少？并结合行业特点分析合理性",
"对公司进行核心竞争力、盈利模式及可持续性分析"
],
"l2_name": "input",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "输入内容"
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"input"
],
"display_name": "InputNode",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "InputNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 405,
"position": {
"x": 1882.0029133695327,
"y": 20
},
"selected": false,
"positionAbsolute": {
"x": 1882.0029133695327,
"y": 20
}
},
{
"id": "RetrievalQA-F6quW",
"data": {
"id": "RetrievalQA-F6quW",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "RetrievalQA",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"combine_documents_chain": {
"l2": false,
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"l2_name": "combine_documents_chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"BaseRetrievalQA",
"Chain",
"RetrievalQA",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 317,
"position": {
"x": 2365.4233944189255,
"y": 484.70381798330607
},
"selected": false,
"positionAbsolute": {
"x": 2365.4233944189255,
"y": 484.70381798330607
}
},
{
"id": "CombineDocsChain-tv0S2",
"data": {
"id": "CombineDocsChain-tv0S2",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"prompt": {
"l2": false,
"info": "只对Stuff类型生效",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"l2_name": "prompt",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
},
"token_max": {
"l2": false,
"info": "只对Stuff类型生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"l2_name": "token_max",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 413,
"position": {
"x": 1430.7645538078127,
"y": 262.02647664648924
},
"selected": false,
"positionAbsolute": {
"x": 1430.7645538078127,
"y": 262.02647664648924
}
},
{
"id": "ElemUnstructuredLoaderV0-nYygC",
"data": {
"id": "ElemUnstructuredLoaderV0-nYygC",
"node": {
"template": {
"_type": "ElemUnstructuredLoaderV0",
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Metadata"
},
"file_path": {
"l2": false,
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "fileNode",
"value": "",
"l2_name": "file_path",
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".bmp",
".jpg",
".png",
".jpeg",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
".xls",
".xlsx",
".tsv",
".csv",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"bmp",
"jpg",
"png",
"jpeg",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
".xls",
"xlsx",
"tsv",
"csv",
"tiff"
],
"multiline": false,
"placeholder": ""
},
"unstructured_api_url": {
"l2": false,
"info": "",
"list": false,
"name": "unstructured_api_url",
"show": true,
"type": "str",
"value": "https://bisheng.dataelem.com/api/v1/etl4llm/predict",
"l2_name": "unstructured_api_url",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "unstructured_api_url"
}
},
"description": "The appropriate parser is automatically selected based on the file format and OCR is supported",
"base_classes": [
"Document"
],
"display_name": "ElemUnstructuredLoaderV0",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": ""
},
"type": "ElemUnstructuredLoaderV0",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 393,
"position": {
"x": 687.5802734859502,
"y": 887.9738676014504
},
"selected": false,
"positionAbsolute": {
"x": 687.5802734859502,
"y": 887.9738676014504
}
},
{
"id": "RecursiveCharacterTextSplitter-6g2FA",
"data": {
"id": "RecursiveCharacterTextSplitter-6g2FA",
"node": {
"template": {
"_type": "RecursiveCharacterTextSplitter",
"documents": {
"l2": false,
"info": "",
"list": false,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"chunk_size": {
"l2": false,
"info": "",
"list": false,
"name": "chunk_size",
"show": true,
"type": "int",
"value": 1000,
"l2_name": "chunk_size",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Chunk Size"
},
"separators": {
"l2": false,
"info": "",
"list": false,
"name": "separators",
"show": true,
"type": "str",
"value": "\\n",
"l2_name": "separators",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Separator"
},
"chunk_overlap": {
"l2": false,
"info": "",
"list": false,
"name": "chunk_overlap",
"show": true,
"type": "int",
"value": 200,
"l2_name": "chunk_overlap",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Chunk Overlap"
},
"separator_type": {
"l2": false,
"info": "",
"list": true,
"name": "separator_type",
"show": true,
"type": "str",
"value": "Text",
"l2_name": "separator_type",
"options": [
"Text",
"cpp",
"go",
"html",
"java",
"js",
"latex",
"markdown",
"php",
"proto",
"python",
"rst",
"ruby",
"rust",
"scala",
"sol",
"swift"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Separator Type"
}
},
"description": "Implementation of splitting text that looks at characters.",
"base_classes": [
"Document"
],
"display_name": "RecursiveCharacterTextSplitter",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/data_connection/document_transformers/text_splitters/recursive_text_splitter"
},
"type": "RecursiveCharacterTextSplitter",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 561,
"position": {
"x": 1126.323112777516,
"y": 958.025245303465
},
"selected": false,
"positionAbsolute": {
"x": 1126.323112777516,
"y": 958.025245303465
}
},
{
"id": "VariableNode-UB934",
"data": {
"id": "VariableNode-UB934",
"node": {
"template": {
"_type": "VariableNode",
"variables": {
"l2": false,
"info": "",
"list": true,
"name": "variables",
"show": true,
"type": "variable",
"value": [
"客户名称",
"客户规模"
],
"l2_name": "variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"variable_value": {
"l2": false,
"info": "",
"list": true,
"name": "variable_value",
"show": false,
"type": "str",
"l2_name": "variable_value",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "用来设置变量",
"base_classes": [
"VariableNode"
],
"display_name": "VariableNode",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "VariableNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 233,
"dragging": false,
"position": {
"x": 2311.569198584317,
"y": 991.6304728944774
},
"selected": false,
"positionAbsolute": {
"x": 2311.569198584317,
"y": 991.6304728944774
}
},
{
"id": "Milvus-fSZ8o",
"data": {
"id": "Milvus-fSZ8o",
"node": {
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"collection_name": {
"l2": false,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "collection_name",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": ""
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"Milvus",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 429,
"position": {
"x": 1715.041130224926,
"y": 967.8237662817272
},
"selected": false,
"positionAbsolute": {
"x": 1715.041130224926,
"y": 967.8237662817272
}
},
{
"id": "OpenAIProxyEmbedding-0OoJh",
"data": {
"id": "OpenAIProxyEmbedding-0OoJh",
"node": {
"template": {
"_type": "proxy_embedding",
"proxy_url": {
"l2": false,
"info": "",
"list": false,
"name": "proxy_url",
"show": false,
"type": "str",
"value": "",
"l2_name": "proxy_url",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "http://43.133.35.137:8080"
}
},
"description": " 使用自建的embedding服务使用openai进行embed ",
"base_classes": [
"Embeddings"
],
"display_name": "OpenAIProxyEmbedding",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "OpenAIProxyEmbedding",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 173,
"position": {
"x": 1138.4741947789414,
"y": 704.7914052206025
},
"selected": false,
"positionAbsolute": {
"x": 1138.4741947789414,
"y": 704.7914052206025
}
},
{
"id": "InputFileNode-a4516",
"data": {
"id": "InputFileNode-a4516",
"node": {
"beta": false,
"template": {
"_type": "InputFileNode",
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "file",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".bmp",
".jpg",
".png",
".jpeg",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
"..xls",
".xlsx",
".tsv",
".csv",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"bmp",
"jpg",
"png",
"jpeg",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
".xls",
"xlsx",
"tsv",
"csv",
"tiff"
],
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "File Path",
"model_config": {}
},
"file_type": {
"info": "Tips for which file should upload",
"list": false,
"name": "file_type",
"show": true,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "提示上传文件类型",
"display_name": "Name",
"model_config": {}
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"fileNode"
],
"display_name": "InputFileNode",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "InputFileNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 337,
"dragging": false,
"position": {
"x": 232.94631949418465,
"y": 876.7981526493783
},
"selected": false,
"positionAbsolute": {
"x": 232.94631949418465,
"y": 876.7981526493783
}
},
{
"id": "BishengLLM-a972d",
"data": {
"id": "BishengLLM-a972d",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 5,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": 526.6927690842307,
"y": 137.74438088815248
},
"selected": true,
"positionAbsolute": {
"x": 526.6927690842307,
"y": 137.74438088815248
}
}
],
"viewport": {
"x": -41.278753137878425,
"y": 51.57633498215188,
"zoom": 0.5112634327714005
}
},
"order_num": 6684632,
"guide_word": null
},
{
"update_time": "2024-10-15 18:39:29",
"parameters": null,
"name": "⭐️知识库问答【增强版】",
"description": "选择知识库，进行知识库问答。特点：\n1.知识多轮问答；\n2.使用关键词检索、向量检索两种方式，提高召回率；\n3.优化过的提示词，降低胡说的出现；",
"flow_id": "9c964a6cabba4d6ea76fde38e72e559d",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 17:24:17",
"id": 182,
"data": {
"edges": [
{
"id": "reactflow__edge-Milvus-cyR5WMilvus|Milvus-cyR5W|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever-MixEsVectorRetriever-J35CZBaseRetriever|vector_retriever|MixEsVectorRetriever-J35CZ",
"style": {
"stroke": "#555"
},
"source": "Milvus-cyR5W",
"target": "MixEsVectorRetriever-J35CZ",
"animated": true,
"className": "",
"sourceHandle": "Milvus|Milvus-cyR5W|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|vector_retriever|MixEsVectorRetriever-J35CZ"
},
{
"id": "reactflow__edge-MixEsVectorRetriever-J35CZMixEsVectorRetriever|MixEsVectorRetriever-J35CZ|MixEsVectorRetriever|BaseRetriever-ConversationalRetrievalChain-SmY3wBaseRetriever|retriever|ConversationalRetrievalChain-SmY3w",
"style": {
"stroke": "#555"
},
"source": "MixEsVectorRetriever-J35CZ",
"target": "ConversationalRetrievalChain-SmY3w",
"animated": true,
"className": "",
"sourceHandle": "MixEsVectorRetriever|MixEsVectorRetriever-J35CZ|BaseRetriever|MixEsVectorRetriever",
"targetHandle": "BaseRetriever|retriever|ConversationalRetrievalChain-SmY3w"
},
{
"id": "reactflow__edge-ConversationBufferMemory-K1tssConversationBufferMemory|ConversationBufferMemory-K1tss|BaseChatMemory|BaseMemory|ConversationBufferMemory-ConversationalRetrievalChain-SmY3wBaseChatMemory|memory|ConversationalRetrievalChain-SmY3w",
"style": {
"stroke": "#555"
},
"source": "ConversationBufferMemory-K1tss",
"target": "ConversationalRetrievalChain-SmY3w",
"animated": true,
"className": "",
"sourceHandle": "ConversationBufferMemory|ConversationBufferMemory-K1tss|ConversationBufferMemory|BaseChatMemory|BaseMemory",
"targetHandle": "BaseChatMemory|memory|ConversationalRetrievalChain-SmY3w"
},
{
"id": "reactflow__edge-PromptTemplate-bs0vjPromptTemplate|PromptTemplate-bs0vj|StringPromptTemplate|BasePromptTemplate|PromptTemplate-ConversationalRetrievalChain-SmY3wBasePromptTemplate|combine_docs_chain_kwargs|ConversationalRetrievalChain-SmY3w",
"style": {
"stroke": "#555"
},
"source": "PromptTemplate-bs0vj",
"target": "ConversationalRetrievalChain-SmY3w",
"animated": true,
"className": "",
"sourceHandle": "PromptTemplate|PromptTemplate-bs0vj|StringPromptTemplate|PromptTemplate|BasePromptTemplate",
"targetHandle": "BasePromptTemplate|combine_docs_chain_kwargs|ConversationalRetrievalChain-SmY3w"
},
{
"id": "reactflow__edge-BishengLLM-768acBishengLLM|BishengLLM-768ac|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-ConversationalRetrievalChain-SmY3wBaseLanguageModel|llm|ConversationalRetrievalChain-SmY3w",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-768ac",
"target": "ConversationalRetrievalChain-SmY3w",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-768ac|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|ConversationalRetrievalChain-SmY3w"
},
{
"id": "reactflow__edge-ElasticKeywordsSearch-1c80eElasticKeywordsSearch|ElasticKeywordsSearch-1c80e|VectorStore|ElasticKeywordsSearch|BaseRetriever|VectorStoreRetriever-MixEsVectorRetriever-J35CZBaseRetriever|keyword_retriever|MixEsVectorRetriever-J35CZ",
"style": {
"stroke": "#555"
},
"source": "ElasticKeywordsSearch-1c80e",
"target": "MixEsVectorRetriever-J35CZ",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "ElasticKeywordsSearch|ElasticKeywordsSearch-1c80e|VectorStore|ElasticKeywordsSearch|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|keyword_retriever|MixEsVectorRetriever-J35CZ"
}
],
"nodes": [
{
"id": "MixEsVectorRetriever-J35CZ",
"data": {
"id": "MixEsVectorRetriever-J35CZ",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": true,
"type": "str",
"l2_name": "tags",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "MixEsVectorRetriever",
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"l2_name": "metadata",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"combine_strategy": {
"l2": false,
"info": "",
"list": true,
"name": "combine_strategy",
"show": true,
"type": "str",
"value": "mix",
"l2_name": "combine_strategy",
"options": [
"keyword_front",
"vector_front",
"mix"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"vector_retriever": {
"l2": false,
"info": "",
"list": false,
"name": "vector_retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "vector_retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"keyword_retriever": {
"l2": false,
"info": "",
"list": false,
"name": "keyword_retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "keyword_retriever",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "This class ensemble the results of es retriever and vector retriever.",
"base_classes": [
"BaseRetriever",
"MixEsVectorRetriever"
],
"display_name": "MixEsVectorRetriever",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "MixEsVectorRetriever",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 351,
"dragging": false,
"position": {
"x": 855.7285654150652,
"y": 657.020339011865
},
"selected": false,
"positionAbsolute": {
"x": 855.7285654150652,
"y": 657.020339011865
}
},
{
"id": "Milvus-cyR5W",
"data": {
"id": "Milvus-cyR5W",
"node": {
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": false,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"collection_name": {
"l2": true,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "毕昇产品文档20240421",
"l2_name": "collection_name",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": 4597
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"VectorStore",
"Milvus",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 393,
"dragging": false,
"position": {
"x": 252.82794993869493,
"y": 1067.6627689948364
},
"selected": false,
"positionAbsolute": {
"x": 252.82794993869493,
"y": 1067.6627689948364
}
},
{
"id": "ConversationalRetrievalChain-SmY3w",
"data": {
"id": "ConversationalRetrievalChain-SmY3w",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"_type": "ConversationalRetrievalChain",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseChatMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": false,
"name": "callbacks",
"show": false,
"type": "Callbacks",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"condense_question_llm": {
"l2": false,
"info": "",
"list": false,
"name": "condense_question_llm",
"show": false,
"type": "BaseLanguageModel",
"l2_name": "condense_question_llm",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Return source documents"
},
"condense_question_prompt": {
"l2": false,
"info": "",
"list": false,
"name": "condense_question_prompt",
"show": true,
"type": "BasePromptTemplate",
"value": {
"_type": "prompt",
"template": "Given the following conversation and a follow up question, rephrase the follow up question to be a standalone question, in its original language.\n\nChat History:\n{chat_history}\nFollow Up Input: {question}\nStandalone question:",
"output_parser": null,
"input_variables": [
"chat_history",
"question"
],
"template_format": "f-string",
"partial_variables": {},
"validate_template": true
},
"l2_name": "condense_question_prompt",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"combine_docs_chain_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "combine_docs_chain_kwargs",
"show": true,
"type": "BasePromptTemplate",
"l2_name": "combine_docs_chain_kwargs",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
}
},
"description": "Convenience method to load chain from LLM and retriever.",
"base_classes": [
"ConversationalRetrievalChain",
"Chain",
"BaseConversationalRetrievalChain",
"function"
],
"display_name": "ConversationalRetrievalChain",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/chat_vector_db"
},
"type": "ConversationalRetrievalChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 495,
"dragging": false,
"position": {
"x": 1716.0776457768882,
"y": 243.73689122583036
},
"selected": false,
"positionAbsolute": {
"x": 1716.0776457768882,
"y": 243.73689122583036
}
},
{
"id": "ConversationBufferMemory-K1tss",
"data": {
"id": "ConversationBufferMemory-K1tss",
"node": {
"template": {
"_type": "ConversationBufferMemory",
"ai_prefix": {
"l2": false,
"info": "",
"list": false,
"name": "ai_prefix",
"show": false,
"type": "str",
"value": "AI",
"l2_name": "ai_prefix",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"l2": false,
"info": "The variable to be used as Chat Input when more than one variable is available.",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "",
"l2_name": "input_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"memory_key": {
"l2": false,
"info": "",
"list": false,
"name": "memory_key",
"show": true,
"type": "str",
"value": "chat_history",
"l2_name": "memory_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"output_key": {
"l2": false,
"info": "The variable to be used as Chat Output (e.g. answer in a ConversationalRetrievalChain)",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "answer",
"l2_name": "output_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"chat_memory": {
"l2": false,
"info": "",
"list": false,
"name": "chat_memory",
"show": true,
"type": "BaseChatMessageHistory",
"l2_name": "chat_memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"human_prefix": {
"l2": false,
"info": "",
"list": false,
"name": "human_prefix",
"show": false,
"type": "str",
"value": "Human",
"l2_name": "human_prefix",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"return_messages": {
"l2": false,
"info": "",
"list": false,
"name": "return_messages",
"show": true,
"type": "bool",
"l2_name": "return_messages",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Buffer for storing conversation memory.",
"base_classes": [
"ConversationBufferMemory",
"BaseChatMemory",
"BaseMemory"
],
"display_name": "ConversationBufferMemory",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/memory/how_to/buffer"
},
"type": "ConversationBufferMemory",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 527,
"dragging": false,
"position": {
"x": 843.4894981704888,
"y": 77.9938911218216
},
"selected": false,
"positionAbsolute": {
"x": 843.4894981704888,
"y": 77.9938911218216
}
},
{
"id": "PromptTemplate-bs0vj",
"data": {
"id": "PromptTemplate-bs0vj",
"node": {
"name": "",
"template": {
"_type": "PromptTemplate",
"template": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "参考文本：\n{context}\n\n----------------------\n\n1.用中文回答用户问题，并且答案要严谨专业。\n2.你需要依据以上【参考文本】中的内容来回答，当【参考文本】中有明确与用户问题相关的内容时才进行回答，不可根据自己的知识来回答。\n3.由于【参考文本】可能包含多个来自不同信息源的信息，所以根据这些不同的信息源可能得出有差异甚至冲突的答案，当发现这种情况时，这些答案都列举出来；如果没有冲突或差异，则只需要给出一个最终结果。\n4.若【参考文本】中内容与用户问题不相关则回复“没有找到相关内容”。\n\n----------------------\n\n用户问题: {question}\n你的回答:",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"context",
"question"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"template_format": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"partial_variables": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"validate_template": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Schema to represent a prompt for an LLM.",
"base_classes": [
"StringPromptTemplate",
"PromptTemplate",
"BasePromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"context",
"question"
],
"template": [
"context",
"question"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {
"formatters": {
"openai_api_key": {}
},
"base_formatters": {
"dict": {},
"list": {},
"show": {},
"union": {},
"kwargs": {},
"default": {},
"headers": {},
"optional": {},
"password": {},
"multiline": {},
"model_fields": {
"MODEL_DICT": {
"OpenAI": [
"text-davinci-003",
"text-davinci-002",
"text-curie-001",
"text-babbage-001",
"text-ada-001"
],
"Anthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
],
"ChatOpenAI": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k"
],
"ChatAnthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
]
}
},
"dict_code_file": {}
}
}
},
"type": "PromptTemplate",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 291,
"dragging": false,
"position": {
"x": 838.6147040497829,
"y": -450.1088052880104
},
"selected": false,
"positionAbsolute": {
"x": 838.6147040497829,
"y": -450.1088052880104
}
},
{
"id": "BishengLLM-768ac",
"data": {
"id": "BishengLLM-768ac",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 5,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": 185.91731964952712,
"y": -45.1192362802976
},
"selected": true,
"positionAbsolute": {
"x": 185.91731964952712,
"y": -45.1192362802976
}
},
{
"id": "ElasticKeywordsSearch-1c80e",
"data": {
"id": "ElasticKeywordsSearch-1c80e",
"node": {
"beta": false,
"template": {
"ids": {
"info": "",
"list": true,
"name": "ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Ids",
"model_config": {}
},
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLLM",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "LLM",
"model_config": {}
},
"_type": "ElasticKeywordsSearch",
"prompt": {
"info": "",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "prompt",
"model_config": {}
},
"drop_old": {
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Drop Old",
"model_config": {}
},
"documents": {
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Documents",
"model_config": {}
},
"embedding": {
"info": "",
"list": false,
"name": "embedding",
"show": false,
"type": "Embeddings",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Embedding",
"model_config": {}
},
"metadatas": {
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadatas",
"model_config": {}
},
"index_name": {
"info": "",
"list": false,
"name": "index_name",
"show": true,
"type": "knowledge_one",
"value": "毕昇产品文档20240421",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Index Name",
"model_config": {},
"collection_id": 4597
},
"ssl_verify": {
"info": "",
"list": false,
"name": "ssl_verify",
"show": true,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "ssl_verify",
"model_config": {}
},
"search_type": {
"info": "",
"list": false,
"name": "search_type",
"show": true,
"type": "str",
"value": "similarity",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Search Type",
"model_config": {}
},
"search_kwargs": {
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Search Kwargs",
"model_config": {}
},
"refresh_indices": {
"info": "",
"list": false,
"name": "refresh_indices",
"show": false,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Refresh Indices",
"model_config": {}
},
"elasticsearch_url": {
"info": "",
"list": false,
"name": "elasticsearch_url",
"show": true,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "ES_connection_url",
"model_config": {}
}
},
"description": "Construct ElasticKeywordsSearch wrapper from raw documents.",
"base_classes": [
"VectorStore",
"ElasticKeywordsSearch",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "ElasticKeywordsSearch",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030",
"field_formatters": {}
},
"type": "ElasticKeywordsSearch",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 547,
"dragging": false,
"position": {
"x": 216.8898375498219,
"y": 461.0085204207429
},
"selected": false,
"positionAbsolute": {
"x": 216.8898375498219,
"y": 461.0085204207429
}
}
],
"viewport": {
"x": 365.45540726853375,
"y": 231.0471559799472,
"zoom": 0.42343675192812
}
},
"order_num": null,
"guide_word": null
},
{
"update_time": "2024-08-20 17:25:43",
"parameters": null,
"name": "⭐️SQLAgent",
"description": "通过自然语言进行数据库查询与分析。当前共享key只支持GPT3.5模型，效果可能不如4。\n\n暂不支持多轮问答。当前提供了上市公司相关的示例数据库，具体数据库表见：https://dataelem.feishu.cn/wiki/XaIzwkZNbisIoRkdPv2cuOqwnQf?sheet=d9b32d",
"flow_id": "2623add6f93f40cf9c707de670e44b59",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 17:25:43",
"id": 183,
"data": {
"edges": [
{
"id": "reactflow__edge-BishengLLM-268bdBishengLLM|BishengLLM-268bd|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM-SQLAgent-coFffBaseLanguageModel|llm|SQLAgent-coFff",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-268bd",
"target": "SQLAgent-coFff",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-268bd|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|SQLAgent-coFff"
}
],
"nodes": [
{
"id": "SQLAgent-coFff",
"data": {
"id": "SQLAgent-coFff",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"_type": "SQLAgent",
"top_k": {
"l2": false,
"info": "",
"list": false,
"name": "top_k",
"show": false,
"type": "int",
"value": 10,
"l2_name": "top_k",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"prefix": {
"l2": false,
"info": "",
"list": false,
"name": "prefix",
"show": true,
"type": "str",
"value": "You are an agent designed to interact with a SQL database.\nGiven an input question, create a syntactically correct {dialect} query to run, then look at the results of the query and return the answer.\nUnless the user specifies a specific number of examples they wish to obtain, always limit your query to at most {top_k} results.\nYou can order the results by a relevant column to return the most interesting examples in the database.\nNever query for all the columns from a specific table, only ask for the relevant columns given the question.\nYou have access to tools for interacting with the database.\nOnly use the below tools. Only use the information returned by the below tools to construct your final answer.\nYou MUST double check your query before executing it. If you get an error while executing a query, rewrite the query and try again.\n\nDO NOT make any DML statements (INSERT, UPDATE, DELETE, DROP etc.) to the database.\n\nIf the question does not seem related to the database, just return \"I don't know\" as the answer.\n",
"l2_name": "prefix",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
},
"suffix": {
"l2": false,
"info": "",
"list": false,
"name": "suffix",
"show": true,
"type": "str",
"value": "Begin!\n\nQuestion: {input}\nThought: I should look at the tables in the database to see what I can query. Then I should query the schema of the most relevant tables.\n{agent_scratchpad}",
"l2_name": "suffix",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"database_uri": {
"l2": true,
"info": "",
"list": false,
"name": "database_uri",
"show": true,
"type": "str",
"value": "mysql+pymysql://root:E1SkG0PaDMEPTAxY@192.168.106.116:3306/company_stock",
"l2_name": "database_uri",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": true,
"type": "str",
"value": [
"input",
"agent_scratchpad"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"format_instructions": {
"l2": false,
"info": "",
"list": false,
"name": "format_instructions",
"show": true,
"type": "str",
"value": "Use the following format:\n\nQuestion: the input question you must answer\nThought: you should always think about what to do\nAction: the action to take, should be one of [{tool_names}]\nAction Input: the input to the action\nObservation: the result of the action\n... (this Thought/Action/Action Input/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question",
"l2_name": "format_instructions",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
}
},
"description": "Construct an SQL agent from an LLM and tools.",
"base_classes": [
"CustomAgentExecutor",
"SQLAgent",
"Chain",
"AgentExecutor",
"function"
],
"display_name": "SQLAgent",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "SQLAgent",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 705,
"position": {
"x": 560.4986060286186,
"y": 43.27482866991457
},
"selected": false,
"positionAbsolute": {
"x": 560.4986060286186,
"y": 43.27482866991457
}
},
{
"id": "BishengLLM-268bd",
"data": {
"id": "BishengLLM-268bd",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 4,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BishengLLM",
"BaseLanguageModel",
"BaseChatModel",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": -234.44315040341712,
"y": 160.48338217607173
},
"selected": true,
"positionAbsolute": {
"x": -234.44315040341712,
"y": 160.48338217607173
}
}
],
"viewport": {
"x": 448.55360702357893,
"y": -9.209667977077856,
"zoom": 1.147646679561573
}
},
"order_num": 6815702,
"guide_word": null
},
{
"update_time": "2024-08-20 17:30:30",
"parameters": null,
"name": "⭐️ QA生成",
"description": "上传文档自动生成QA对，上传文档后，发送“开始”触发",
"flow_id": "23966d5d7f234ef2827e6128ae6bd8bf",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 17:30:30",
"id": 185,
"data": {
"edges": [
{
"id": "reactflow__edge-ElemUnstructuredLoaderV0-fcc08ElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-fcc08|Document-QAGenerationChainV2-c3ec3Document|documents|QAGenerationChainV2-c3ec3",
"style": {
"stroke": "#555"
},
"source": "ElemUnstructuredLoaderV0-fcc08",
"target": "QAGenerationChainV2-c3ec3",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "ElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-fcc08|Document",
"targetHandle": "Document|documents|QAGenerationChainV2-c3ec3"
},
{
"id": "reactflow__edge-InputFileNode-09d8bInputFileNode|InputFileNode-09d8b|fileNode-ElemUnstructuredLoaderV0-fcc08fileNode|file_path|ElemUnstructuredLoaderV0-fcc08",
"style": {
"stroke": "#555"
},
"source": "InputFileNode-09d8b",
"target": "ElemUnstructuredLoaderV0-fcc08",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "InputFileNode|InputFileNode-09d8b|fileNode",
"targetHandle": "fileNode|file_path|ElemUnstructuredLoaderV0-fcc08"
},
{
"id": "reactflow__edge-BishengLLM-f47edBishengLLM|BishengLLM-f47ed|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM-QAGenerationChainV2-c3ec3BaseLanguageModel|llm|QAGenerationChainV2-c3ec3",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-f47ed",
"target": "QAGenerationChainV2-c3ec3",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-f47ed|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|QAGenerationChainV2-c3ec3"
}
],
"nodes": [
{
"id": "QAGenerationChainV2-c3ec3",
"data": {
"id": "QAGenerationChainV2-c3ec3",
"node": {
"beta": false,
"template": {
"k": {
"info": "",
"list": false,
"name": "k",
"show": true,
"type": "int",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "K",
"model_config": {}
},
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"_type": "QAGenerationChainV2",
"prompt": {
"info": "",
"list": false,
"name": "prompt",
"show": false,
"type": "ChatPromptTemplate",
"value": {
"name": null,
"tags": null,
"_type": "chat",
"messages": [
{
"prompt": {
"name": null,
"tags": null,
"_type": "prompt",
"metadata": null,
"template": "您的任务是遵循以下规则从给定的上下文中提出一个问题，规则如下：\n\n 1. 即使在没有给定上下文的情况下，问题也应该对人类有意义。\n 2. 应该可以从给定上下文中完全回答问题。\n 3. 问题应该来自包含重要信息的上下文部分。它也可以来自表格、段落、或者代码等。\n 4. 回答问题时不应包含任何链接。\n 5. 问题的难度应该是中等的。\n 6. 问题必须是合理的，并且必须能被人理解和回答。\n 7. 不要在问题中使用“提供的上下文”等短语。\n 8. 避免使用可以分解成多个问题的“和”字样来构建问题。\n 9. 如果上下文是中文，那么问题也应该是中文的。\n\nExamples:\ncontext:武汉达梦数据库股份有限公司 招股说明书 （申报稿） 1-1-226 表中作出恰当列报。 2、研发费用 2021年度、 2020年度、 2019 年度，达梦数据 研发费用金额分别 为11,786.99 万元、 9,660.26 万元、 6,255.86万元， 各年度研发费用占营 业收入的比例分别为 15.86 % 、 21.46 %、20.74 %。 由于研发投入金额及其占当期 营业收入的比例是 达梦数据 的关键 指标之一，可能存在因为核算不准 确而导致的错报风险。因此， 中天 运会计师 将研发费用的归集和核算 确定为关键审计事项。 针对研发费用的真实性与准确性，会计师执行的 重要审计程序主要包括： （1）了解与研发费用相关的关键内部控制，评价 这些控制的设计，确定其是否得到执行，并对相关内 部控制的运行有效性进行测试； （2）获取研发项目立项、审批资料，抽查重要研 发项目的过程文档，判断研发项目的真实性； （3）获取研发费用按项目、性质分类明细表，分\nquestion:达梦2021年的研发费用占营业收入的比例是多少？\n\ncontext:武汉达梦数据库股份有限公司 招股说明书 （申报稿） 1-1-329 （2）存货周转率 公司与同行业可比公司存货周转率对比情况如下： 公司简称 2021年度 2020年度 2019年度 中望软件 6.93 5.62 10.66 星环科技 3.38 3.21 2.24 金山办公 212.60 175.46 162.91 平均值 74.30 61.43 58.60 本公司 1.13 0.57 0.87 数据来源：可比公司招股说明书、定期报告。 报告期各期， 公司存货周转率显著低于同行业可比公司存货周转率平均水平， 主要是因为公司将未验收的数据及行业解决方案项目所发生的累 计成本均作为 存货核算。报告期各期末，公司存在 “湖北省司法行政数据中心项目 ”、“政法云 大数据中心基础设施服务及大数据中心软件采购 项目”等金额较大且实施周期较 长的数据及行业解决方案项目，导致年末存货金额较大。\nquestion:达梦2021年的存货周转率相较于前一年有何变化？\n",
"input_types": {},
"output_parser": null,
"input_variables": [],
"template_format": "f-string",
"partial_variables": {},
"validate_template": false
},
"additional_kwargs": {}
},
{
"prompt": {
"name": null,
"tags": null,
"_type": "prompt",
"metadata": null,
"template": "\ncontext:{context}\nquestion:\n",
"input_types": {},
"output_parser": null,
"input_variables": [
"context"
],
"template_format": "f-string",
"partial_variables": {},
"validate_template": false
},
"additional_kwargs": {}
}
],
"metadata": null,
"input_types": {},
"output_parser": null,
"input_variables": [
"context"
],
"partial_variables": {},
"validate_template": false
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Prompt",
"model_config": {}
},
"documents": {
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Documents",
"model_config": {}
},
"chunk_size": {
"info": "",
"list": false,
"name": "chunk_size",
"show": true,
"type": "int",
"value": 512,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Chunk Size",
"model_config": {}
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": false,
"type": "input",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Preset Question",
"model_config": {}
}
},
"description": "Create a QAGenerationChain from a language model.",
"base_classes": [
"QAGenerationChainV2",
"Chain",
"function"
],
"display_name": "QAGenerationChainV2",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "QAGenerationChainV2",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 429,
"dragging": false,
"position": {
"x": 1052.5270781868935,
"y": 242.3629681024322
},
"selected": false,
"positionAbsolute": {
"x": 1052.5270781868935,
"y": 242.3629681024322
}
},
{
"id": "ElemUnstructuredLoaderV0-fcc08",
"data": {
"id": "ElemUnstructuredLoaderV0-fcc08",
"node": {
"beta": false,
"template": {
"_type": "ElemUnstructuredLoaderV0",
"kwargs": {
"info": "",
"list": false,
"name": "kwargs",
"show": true,
"type": "dict",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "kwargs",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"value": "{}",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "fileNode",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"suffixes": [
".html",
".md",
".txt",
".bmp",
".jpg",
".png",
".jpeg",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
".xls",
".xlsx",
".tsv",
".csv",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"bmp",
"jpg",
"png",
"jpeg",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
".xls",
"xlsx",
"tsv",
"csv",
"tiff"
],
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "File Path",
"model_config": {}
},
"unstructured_api_url": {
"info": "",
"list": false,
"name": "unstructured_api_url",
"show": true,
"type": "str",
"value": "https://bisheng.dataelem.com/api/v1/etl4llm/predict",
"dynamic": false,
"advanced": false,
"password": true,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "unstructured_api_url",
"model_config": {}
}
},
"description": "The appropriate parser is automatically selected based on the file format and OCR is supported",
"base_classes": [
"Document"
],
"display_name": "ElemUnstructuredLoaderV0",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "ElemUnstructuredLoaderV0",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 477,
"dragging": false,
"position": {
"x": 311.70357393794484,
"y": -85.01385230407082
},
"selected": false,
"positionAbsolute": {
"x": 311.70357393794484,
"y": -85.01385230407082
}
},
{
"id": "InputFileNode-09d8b",
"data": {
"id": "InputFileNode-09d8b",
"node": {
"beta": false,
"template": {
"_type": "InputFileNode",
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "file",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".bmp",
".jpg",
".png",
".jpeg",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
"..xls",
".xlsx",
".tsv",
".csv",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"bmp",
"jpg",
"png",
"jpeg",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
".xls",
"xlsx",
"tsv",
"csv",
"tiff"
],
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "File Path",
"model_config": {}
},
"file_type": {
"info": "Tips for which file should upload",
"list": false,
"name": "file_type",
"show": true,
"type": "str",
"value": "请上传文件",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "提示上传文件类型",
"display_name": "Name",
"model_config": {}
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"fileNode"
],
"display_name": "InputFileNode",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "InputFileNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 337,
"dragging": false,
"position": {
"x": -353.26090599027486,
"y": -72.66649282333755
},
"selected": false,
"positionAbsolute": {
"x": -353.26090599027486,
"y": -72.66649282333755
}
},
{
"id": "BishengLLM-f47ed",
"data": {
"id": "BishengLLM-f47ed",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 5,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BishengLLM",
"BaseLanguageModel",
"BaseChatModel",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": 289.08494703205145,
"y": 479.77562233501465
},
"selected": true,
"positionAbsolute": {
"x": 289.08494703205145,
"y": 479.77562233501465
}
}
],
"viewport": {
"x": 387.6997915647513,
"y": 113.95818800464951,
"zoom": 0.8646078322295284
}
},
"order_num": 6946772,
"guide_word": null
},
{
"update_time": "2024-08-20 17:39:24",
"parameters": null,
"name": "⭐️ API调用（使用APIChain）",
"description": "可以参考示例复制完整的接口文档，即可开始进行接口调用",
"flow_id": "44b8448120ba4d65b5d16e55984a583f",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 17:39:24",
"id": 186,
"data": {
"edges": [
{
"id": "reactflow__edge-Tool-OL3JiTool|Tool-OL3Ji|Tool|BaseTool-ZeroShotAgent-aFNWBBaseTool|tools|ZeroShotAgent-aFNWB",
"style": {
"stroke": "#555"
},
"source": "Tool-OL3Ji",
"target": "ZeroShotAgent-aFNWB",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "Tool|Tool-OL3Ji|Tool|BaseTool",
"targetHandle": "BaseTool|tools|ZeroShotAgent-aFNWB"
},
{
"id": "reactflow__edge-APIChain-613fcAPIChain|APIChain-613fc|APIChain|Chain|function-Tool-OL3Jifunction|func|Tool-OL3Ji",
"style": {
"stroke": "#555"
},
"source": "APIChain-613fc",
"target": "Tool-OL3Ji",
"animated": true,
"className": "",
"sourceHandle": "APIChain|APIChain-613fc|APIChain|Chain|function",
"targetHandle": "function|func|Tool-OL3Ji"
},
{
"id": "reactflow__edge-PromptTemplate-48659PromptTemplate|PromptTemplate-48659|BasePromptTemplate|StringPromptTemplate|PromptTemplate-APIChain-613fcBasePromptTemplate|api_url_prompt|APIChain-613fc",
"style": {
"stroke": "#555"
},
"source": "PromptTemplate-48659",
"target": "APIChain-613fc",
"animated": true,
"className": "",
"sourceHandle": "PromptTemplate|PromptTemplate-48659|BasePromptTemplate|StringPromptTemplate|PromptTemplate",
"targetHandle": "BasePromptTemplate|api_url_prompt|APIChain-613fc"
},
{
"id": "reactflow__edge-BishengLLM-687d0BishengLLM|BishengLLM-687d0|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM-APIChain-613fcBaseLanguageModel|llm|APIChain-613fc",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-687d0",
"target": "APIChain-613fc",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-687d0|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|APIChain-613fc"
},
{
"id": "reactflow__edge-BishengLLM-687d0BishengLLM|BishengLLM-687d0|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM-ZeroShotAgent-aFNWBBaseLanguageModel|llm|ZeroShotAgent-aFNWB",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-687d0",
"target": "ZeroShotAgent-aFNWB",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-687d0|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|ZeroShotAgent-aFNWB"
}
],
"nodes": [
{
"id": "ZeroShotAgent-aFNWB",
"data": {
"id": "ZeroShotAgent-aFNWB",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"_type": "ZeroShotAgent",
"tools": {
"l2": false,
"info": "",
"list": true,
"name": "tools",
"show": true,
"type": "BaseTool",
"l2_name": "tools",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"prefix": {
"l2": false,
"info": "",
"list": false,
"name": "prefix",
"show": true,
"type": "str",
"value": "Answer the following questions as best you can. You have access to the following tools:",
"l2_name": "prefix",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
},
"suffix": {
"l2": false,
"info": "",
"list": false,
"name": "suffix",
"show": true,
"type": "str",
"value": "Begin!\n\nQuestion: {input}\nThought:{agent_scratchpad}",
"l2_name": "suffix",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "AgentOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": true,
"type": "str",
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callback_manager": {
"l2": false,
"info": "",
"list": false,
"name": "callback_manager",
"show": false,
"type": "BaseCallbackManager",
"l2_name": "callback_manager",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"format_instructions": {
"l2": false,
"info": "",
"list": false,
"name": "format_instructions",
"show": true,
"type": "str",
"value": "Use the following format:\n\nQuestion: the input question you must answer\nThought: you should always think about what to do\nAction: the action to take, should be one of [{tool_names}]\nAction Input: the input to the action\nObservation: the result of the action\n... (this Thought/Action/Action Input/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question",
"l2_name": "format_instructions",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
}
},
"description": "Construct an agent from an LLM and tools.",
"base_classes": [
"BaseSingleActionAgent",
"ZeroShotAgent",
"Agent",
"function"
],
"display_name": "ZeroShotAgent",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/agents/how_to/custom_mrkl_agent"
},
"type": "ZeroShotAgent",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 609,
"position": {
"x": 1504.8003657712438,
"y": 142.75763054618068
},
"selected": false,
"positionAbsolute": {
"x": 1504.8003657712438,
"y": 142.75763054618068
}
},
{
"id": "Tool-OL3Ji",
"data": {
"id": "Tool-OL3Ji",
"node": {
"template": {
"func": {
"l2": false,
"info": "",
"list": false,
"name": "func",
"show": true,
"type": "function",
"l2_name": "func",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"name": {
"l2": false,
"info": "",
"list": false,
"name": "name",
"show": true,
"type": "str",
"value": "天气查询",
"l2_name": "name",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"_type": "Tool",
"description": {
"l2": false,
"info": "",
"list": false,
"name": "description",
"show": true,
"type": "str",
"value": "提供地址名称，查询该地址天气信息",
"l2_name": "description",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"return_direct": {
"l2": false,
"info": "",
"list": false,
"name": "return_direct",
"show": true,
"type": "bool",
"value": false,
"l2_name": "return_direct",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Converts a chain, agent or function into a tool.",
"base_classes": [
"Tool",
"BaseTool"
],
"display_name": "Tool",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "Tool",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 445,
"position": {
"x": 918.0602598736352,
"y": 513.6611974885975
},
"selected": false,
"positionAbsolute": {
"x": 918.0602598736352,
"y": 513.6611974885975
}
},
{
"id": "APIChain-613fc",
"data": {
"id": "APIChain-613fc",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"_type": "APIChain",
"headers": {
"info": "",
"list": false,
"name": "headers",
"show": true,
"type": "code",
"value": "{'Authorization': 'Bearer '}",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": true,
"title_case": true,
"placeholder": "",
"display_name": "Headers",
"model_config": {}
},
"api_docs": {
"info": "",
"list": false,
"name": "api_docs",
"show": true,
"type": "str",
"value": "key SGVquxtnZYK9WHJVD 接口地址 https://api.seniverse.com/v3/weather/now.json?key=your_api_key&location=beijing&language=zh-Hans&unit=c 请求参数说明 参数名称 类型 默认值 必填 备注 key String 无 是 你的 API 密钥 location Location 无 是 所查询的位置 language Language zh-Hans 否 语言 unit Unit c 否 单位 返回结果示例 JSON 复制代码 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 { \"results\": [ { \"location\": { \"id\": \"C23NB62W20TF\", \"name\": \"西雅图\", \"country\": \"US\", \"path\": \"西雅图,华盛顿州,美国\", \"timezone\": \"America/Los_Angeles\", \"timezone_offset\": \"-07:00\" }, \"now\": { \"text\": \"多云\", //天气现象文字 \"code\": \"4\", //天气现象代码 \"temperature\": \"14\", //温度，单位为c摄氏度或f华氏度 \"feels_like\": \"14\", //体感温度，单位为c摄氏度或f华氏度 \"pressure\": \"1018\", //气压，单位为mb百帕或in英寸 \"humidity\": \"76\", //相对湿度，0~100，单位为百分比 \"visibility\": \"16.09\", //能见度，单位为km公里或mi英里 \"wind_direction\": \"西北\", //风向文字 \"wind_direction_degree\": \"340\", //风向角度，范围0~360，0为正北，90为正东，180为正南，270为正西 \"wind_speed\": \"8.05\", //风速，单位为km/h公里每小时或mph英里每小时 \"wind_scale\": \"2\", //风力等级，请参考：http://baike.baidu.com/view/465076.htm \"clouds\": \"90\", //云量，单位%，范围0~100，天空被云覆盖的百分比 #目前不支持中国城市# \"dew_point\": \"-12\" //露点温度，请参考：http://baike.baidu.com/view/118348.htm #目前不支持中国城市# }, \"last_update\": \"2015-09-25T22:45:00-07:00\" //数据更新时间（该城市的本地时间） } ] }",
"dynamic": false,
"advanced": false,
"password": true,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Api Docs",
"model_config": {}
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Preset Question",
"model_config": {}
},
"api_url_prompt": {
"info": "",
"list": false,
"name": "api_url_prompt",
"show": true,
"type": "BasePromptTemplate",
"value": {
"name": null,
"tags": null,
"_type": "prompt",
"metadata": null,
"template": "You are given the below API Documentation:\n{api_docs}\nUsing this documentation, generate the full API url to call for answering the user question.\nYou should build the API url in order to get a response that is as short as possible, while still getting the necessary information to answer the question. Pay attention to deliberately exclude any unnecessary pieces of data in the API call.\n\nQuestion:{question}\nAPI url:",
"input_types": {},
"output_parser": null,
"input_variables": [
"api_docs",
"question"
],
"template_format": "f-string",
"partial_variables": {},
"validate_template": false
},
"dynamic": false,
"advanced": false,
"password": true,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Api Url Prompt",
"model_config": {}
},
"limit_to_domains": {
"info": "",
"list": true,
"name": "limit_to_domains",
"show": true,
"type": "str",
"value": [
"https://api.seniverse.com"
],
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Limit To Domains",
"model_config": {}
},
"api_response_prompt": {
"info": "",
"list": false,
"name": "api_response_prompt",
"show": true,
"type": "BasePromptTemplate",
"value": {
"name": null,
"tags": null,
"_type": "prompt",
"metadata": null,
"template": "You are given the below API Documentation:\n{api_docs}\nUsing this documentation, generate the full API url to call for answering the user question.\nYou should build the API url in order to get a response that is as short as possible, while still getting the necessary information to answer the question. Pay attention to deliberately exclude any unnecessary pieces of data in the API call.\n\nQuestion:{question}\nAPI url: {api_url}\n\nHere is the response from the API:\n\n{api_response}\n\nSummarize this response to answer the original question.\n\nSummary:",
"input_types": {},
"output_parser": null,
"input_variables": [
"api_docs",
"api_response",
"api_url",
"question"
],
"template_format": "f-string",
"partial_variables": {},
"validate_template": false
},
"dynamic": false,
"advanced": false,
"password": true,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Api Response Prompt",
"model_config": {}
}
},
"description": "Load chain from just an LLM and the api docs.",
"base_classes": [
"APIChain",
"Chain",
"function"
],
"display_name": "APIChain",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "APIChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 563,
"dragging": false,
"position": {
"x": 477.6277694871776,
"y": 380.571320729037
},
"selected": false,
"positionAbsolute": {
"x": 477.6277694871776,
"y": 380.571320729037
}
},
{
"id": "PromptTemplate-48659",
"data": {
"id": "PromptTemplate-48659",
"node": {
"beta": false,
"icon": null,
"name": "",
"error": null,
"template": {
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "PromptTemplate",
"api_docs": {
"info": "",
"list": false,
"name": "api_docs",
"show": true,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": true,
"title_case": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "api_docs",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"question": {
"info": "",
"list": false,
"name": "question",
"show": true,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": true,
"title_case": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "question",
"model_config": {}
},
"template": {
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "You are given the below API Documentation:\n{api_docs}\n\nUsing this documentation, generate the full API url to call for answering the user question.\nYou should build the API url in order to get a response that is as short as possible, while still getting the necessary information to answer the question. \n\nPlease do not return any other words except the full API url.\n\nQuestion:{question}\nAPI url:",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": true,
"title_case": true,
"placeholder": "",
"display_name": "Template",
"model_config": {}
},
"input_types": {
"info": "",
"list": false,
"name": "input_types",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Input Types",
"model_config": {}
},
"output_parser": {
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Output Parser",
"model_config": {}
},
"input_variables": {
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"api_docs",
"question"
],
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Input Variables",
"model_config": {}
},
"template_format": {
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "Literal['f-string', 'mustache', 'jinja2']",
"value": "f-string",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Template Format",
"model_config": {}
},
"partial_variables": {
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Partial Variables",
"model_config": {}
},
"validate_template": {
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": false,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Validate Template",
"model_config": {}
}
},
"full_path": null,
"description": "Prompt template for a language model.",
"base_classes": [
"BasePromptTemplate",
"StringPromptTemplate",
"PromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"api_docs",
"question"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {}
},
"type": "PromptTemplate",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 455,
"dragging": false,
"position": {
"x": -52.092556516963214,
"y": 621.8035505128619
},
"selected": false,
"positionAbsolute": {
"x": -52.092556516963214,
"y": 621.8035505128619
}
},
{
"id": "BishengLLM-687d0",
"data": {
"id": "BishengLLM-687d0",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 4,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BishengLLM",
"BaseLanguageModel",
"BaseChatModel",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": -85.44442588200178,
"y": 56.91781929370879
},
"selected": false,
"positionAbsolute": {
"x": -85.44442588200178,
"y": 56.91781929370879
}
}
],
"viewport": {
"x": 254.0096889201243,
"y": 143.6897914083358,
"zoom": 0.6555101557762626
}
},
"order_num": 7012307,
"guide_word": null
},
{
"update_time": "2024-08-20 17:41:21",
"parameters": null,
"name": "⭐️知识库问答【严谨版】",
"description": "选择知识库，进行知识库问答。通过ES搜索控制召回内容强相关（当然还有许多细节优化，详见：https://dataelem.feishu.cn/wiki/R7OYwlN9Ii8jRFkmOuKcMKtEnZc）",
"flow_id": "95c28c47291641d680a380cf8b03d104",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 17:41:21",
"id": 187,
"data": {
"edges": [
{
"id": "reactflow__edge-CombineDocsChain-bMaXkCombineDocsChain|CombineDocsChain-bMaXk|BaseCombineDocumentsChain|function-RetrievalQA-ypS4mBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-ypS4m",
"style": {
"stroke": "#555"
},
"source": "CombineDocsChain-bMaXk",
"target": "RetrievalQA-ypS4m",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "CombineDocsChain|CombineDocsChain-bMaXk|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-ypS4m"
},
{
"id": "reactflow__edge-PromptTemplate-gM5zgPromptTemplate|PromptTemplate-gM5zg|PromptTemplate|BasePromptTemplate|StringPromptTemplate-CombineDocsChain-bMaXkBasePromptTemplate|prompt|CombineDocsChain-bMaXk",
"style": {
"stroke": "#555"
},
"source": "PromptTemplate-gM5zg",
"target": "CombineDocsChain-bMaXk",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "PromptTemplate|PromptTemplate-gM5zg|StringPromptTemplate|BasePromptTemplate|PromptTemplate",
"targetHandle": "BasePromptTemplate|prompt|CombineDocsChain-bMaXk"
},
{
"id": "reactflow__edge-ElasticKeywordsSearch-f27a9ElasticKeywordsSearch|ElasticKeywordsSearch-f27a9|ElasticKeywordsSearch|VectorStore|BaseRetriever|VectorStoreRetriever-RetrievalQA-ypS4mBaseRetriever|retriever|RetrievalQA-ypS4m",
"style": {
"stroke": "#555"
},
"source": "ElasticKeywordsSearch-f27a9",
"target": "RetrievalQA-ypS4m",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "ElasticKeywordsSearch|ElasticKeywordsSearch-f27a9|ElasticKeywordsSearch|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-ypS4m"
},
{
"id": "reactflow__edge-PromptTemplate-4iBFjPromptTemplate|PromptTemplate-4iBFj|StringPromptTemplate|BasePromptTemplate|PromptTemplate-ElasticKeywordsSearch-f27a9BasePromptTemplate|prompt|ElasticKeywordsSearch-f27a9",
"style": {
"stroke": "#555"
},
"source": "PromptTemplate-4iBFj",
"target": "ElasticKeywordsSearch-f27a9",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "PromptTemplate|PromptTemplate-4iBFj|StringPromptTemplate|BasePromptTemplate|PromptTemplate",
"targetHandle": "BasePromptTemplate|prompt|ElasticKeywordsSearch-f27a9"
},
{
"id": "reactflow__edge-BishengLLM-bc77fBishengLLM|BishengLLM-bc77f|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM-CombineDocsChain-bMaXkBaseLanguageModel|llm|CombineDocsChain-bMaXk",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-bc77f",
"target": "CombineDocsChain-bMaXk",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-bc77f|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-bMaXk"
},
{
"id": "reactflow__edge-BishengLLM-bc77fBishengLLM|BishengLLM-bc77f|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM-ElasticKeywordsSearch-f27a9BaseLLM|llm|ElasticKeywordsSearch-f27a9",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-bc77f",
"target": "ElasticKeywordsSearch-f27a9",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-bc77f|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM",
"targetHandle": "BaseLLM|llm|ElasticKeywordsSearch-f27a9"
}
],
"nodes": [
{
"id": "RetrievalQA-ypS4m",
"data": {
"id": "RetrievalQA-ypS4m",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "RetrievalQA",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"combine_documents_chain": {
"l2": false,
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"l2_name": "combine_documents_chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"RetrievalQA",
"BaseRetrievalQA",
"Chain",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 317,
"position": {
"x": 1941.838444212454,
"y": 902.7207397625788
},
"selected": false,
"positionAbsolute": {
"x": 1941.838444212454,
"y": 902.7207397625788
}
},
{
"id": "CombineDocsChain-bMaXk",
"data": {
"id": "CombineDocsChain-bMaXk",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"prompt": {
"l2": false,
"info": "只对Stuff类型生效",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"l2_name": "prompt",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
},
"token_max": {
"l2": false,
"info": "只对Stuff类型生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": 7000,
"l2_name": "token_max",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"document_prompt": {
"info": "",
"list": false,
"name": "document_prompt",
"type": "BasePromptTemplate",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 413,
"dragging": false,
"position": {
"x": 1334.4528869077485,
"y": 404.7844512964858
},
"selected": false,
"positionAbsolute": {
"x": 1334.4528869077485,
"y": 404.7844512964858
}
},
{
"id": "PromptTemplate-gM5zg",
"data": {
"id": "PromptTemplate-gM5zg",
"node": {
"name": "",
"template": {
"_type": "PromptTemplate",
"context": {
"info": "",
"list": false,
"name": "context",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "context"
},
"question": {
"info": "",
"list": false,
"name": "question",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "question"
},
"template": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "参考文本：\n{context}\n\n----------------------\n# Role：客服\n\n# Background :\n- 你是一名客服人员。你的日常工作是根据【参考文本】的内容回复用户的各类咨询问题。\n\n# Goals:\n- 以上【参考文本】中是从企业知识库中查找到的相关信息，只能结合以上信息进行回答，若以上内容为空或其中没有找到能回答【用户问题】的内容时，则回复“没有找到相关内容”，不能根据你自己的知识自己发挥。\n- 特别要注意区分【用户问题】与【参考文本】中不同的日期、人名、公司名这些关键信息。\n- 用中文回答问题，并且答案要严谨、专业、清晰、可读性好。\n- 拥有排版审美, 会利用序号, 缩进, 分隔线和换行符等等来美化信息排版。\n\n----------------------\n\n用户问题: {question}\n你的回答:",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"context",
"question"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"template_format": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"partial_variables": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"validate_template": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Schema to represent a prompt for an LLM.",
"base_classes": [
"StringPromptTemplate",
"BasePromptTemplate",
"PromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"context",
"question"
],
"template": [
"context",
"question"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {
"formatters": {
"openai_api_key": {}
},
"base_formatters": {
"dict": {},
"list": {},
"show": {},
"union": {},
"kwargs": {},
"default": {},
"headers": {},
"optional": {},
"password": {},
"multiline": {},
"model_fields": {
"MODEL_DICT": {
"OpenAI": [
"text-davinci-003",
"text-davinci-002",
"text-curie-001",
"text-babbage-001",
"text-ada-001"
],
"Anthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
],
"ChatOpenAI": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k"
],
"ChatAnthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
]
}
},
"dict_code_file": {}
}
}
},
"type": "PromptTemplate",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 455,
"dragging": false,
"position": {
"x": 832.6495717437198,
"y": -8.979469210830944
},
"selected": false,
"positionAbsolute": {
"x": 832.6495717437198,
"y": -8.979469210830944
}
},
{
"id": "PromptTemplate-4iBFj",
"data": {
"id": "PromptTemplate-4iBFj",
"node": {
"name": "",
"template": {
"_type": "PromptTemplate",
"question": {
"info": "",
"list": false,
"name": "question",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "question"
},
"template": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "分析给定Question，提取Question中包含的KeyWords，输出列表形式\n\nExamples:\nQuestion: 数据项素2022年营收是多少？\nKeyWords: ['数据项素', '2022', '营收, '营业收入']\nQuestion: 深圳出差住宿标准是什么？\nKeyWords: ['深圳', '出差', '住宿标准', '报销标准']\n\n\n----------------\nQuestion: {question}\nKeyWords: ",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"question"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"template_format": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"partial_variables": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"validate_template": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Schema to represent a prompt for an LLM.",
"base_classes": [
"StringPromptTemplate",
"BasePromptTemplate",
"PromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"question"
],
"template": [
"question"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {
"formatters": {
"openai_api_key": {}
},
"base_formatters": {
"dict": {},
"list": {},
"show": {},
"union": {},
"kwargs": {},
"default": {},
"headers": {},
"optional": {},
"password": {},
"multiline": {},
"model_fields": {
"MODEL_DICT": {
"OpenAI": [
"text-davinci-003",
"text-davinci-002",
"text-curie-001",
"text-babbage-001",
"text-ada-001"
],
"Anthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
],
"ChatOpenAI": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k"
],
"ChatAnthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
]
}
},
"dict_code_file": {}
}
}
},
"type": "PromptTemplate",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 373,
"dragging": false,
"position": {
"x": 162.28620398428382,
"y": 658.7665439891539
},
"selected": false,
"positionAbsolute": {
"x": 162.28620398428382,
"y": 658.7665439891539
}
},
{
"id": "ElasticKeywordsSearch-f27a9",
"data": {
"id": "ElasticKeywordsSearch-f27a9",
"node": {
"beta": false,
"template": {
"ids": {
"info": "",
"list": true,
"name": "ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Ids",
"model_config": {}
},
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLLM",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "LLM",
"model_config": {}
},
"_type": "ElasticKeywordsSearch",
"prompt": {
"info": "",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "prompt",
"model_config": {}
},
"drop_old": {
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Drop Old",
"model_config": {}
},
"documents": {
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Documents",
"model_config": {}
},
"embedding": {
"info": "",
"list": false,
"name": "embedding",
"show": false,
"type": "Embeddings",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Embedding",
"model_config": {}
},
"metadatas": {
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadatas",
"model_config": {}
},
"index_name": {
"info": "",
"list": false,
"name": "index_name",
"show": true,
"type": "knowledge_one",
"value": "毕昇产品文档20240421",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Index Name",
"model_config": {},
"collection_id": 4597
},
"ssl_verify": {
"info": "",
"list": false,
"name": "ssl_verify",
"show": true,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "ssl_verify",
"model_config": {}
},
"search_type": {
"info": "",
"list": false,
"name": "search_type",
"show": true,
"type": "str",
"value": "similarity",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Search Type",
"model_config": {}
},
"search_kwargs": {
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{\"k\":6}",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Search Kwargs",
"model_config": {}
},
"refresh_indices": {
"info": "",
"list": false,
"name": "refresh_indices",
"show": false,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Refresh Indices",
"model_config": {}
},
"elasticsearch_url": {
"info": "",
"list": false,
"name": "elasticsearch_url",
"show": true,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "ES_connection_url",
"model_config": {}
}
},
"description": "Construct ElasticKeywordsSearch wrapper from raw documents.",
"base_classes": [
"ElasticKeywordsSearch",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "ElasticKeywordsSearch",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030",
"field_formatters": {}
},
"type": "ElasticKeywordsSearch",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 711,
"dragging": false,
"position": {
"x": 812.1149007520644,
"y": 664.6494588266206
},
"selected": false,
"positionAbsolute": {
"x": 812.1149007520644,
"y": 664.6494588266206
}
},
{
"id": "BishengLLM-bc77f",
"data": {
"id": "BishengLLM-bc77f",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 5,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BishengLLM",
"BaseLanguageModel",
"BaseChatModel",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": 121.42385839968928,
"y": 86.07764034320081
},
"selected": false,
"positionAbsolute": {
"x": 121.42385839968928,
"y": 86.07764034320081
}
}
],
"viewport": {
"x": 129.04745958386252,
"y": 45.70158800856939,
"zoom": 0.5843377187256229
}
},
"order_num": 7077842,
"guide_word": null
},
{
"update_time": "2024-08-20 18:13:33",
"parameters": null,
"name": "⭐️联网搜索对话",
"description": "支持通过搜索引擎获取信息并回答问题",
"flow_id": "8374d36f23af4c438c4ce4e6e080e6cf",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-20 18:13:33",
"id": 189,
"data": {
"edges": [
{
"id": "reactflow__edge-BingSearchAPIWrapper-Rikh1BingSearchAPIWrapper|BingSearchAPIWrapper-Rikh1|BingSearchAPIWrapper-BingSearchRun-S3aq7BingSearchAPIWrapper|api_wrapper|BingSearchRun-S3aq7",
"style": {
"stroke": "#555"
},
"source": "BingSearchAPIWrapper-Rikh1",
"target": "BingSearchRun-S3aq7",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "BingSearchAPIWrapper|BingSearchAPIWrapper-Rikh1|BingSearchAPIWrapper",
"targetHandle": "BingSearchAPIWrapper|api_wrapper|BingSearchRun-S3aq7"
},
{
"id": "reactflow__edge-BingSearchRun-S3aq7BingSearchRun|BingSearchRun-S3aq7|Tool|BaseTool|BingSearchRun|BaseTool-ZeroShotAgent-6vSFwBaseTool|tools|ZeroShotAgent-6vSFw",
"style": {
"stroke": "#555"
},
"source": "BingSearchRun-S3aq7",
"target": "ZeroShotAgent-6vSFw",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "BingSearchRun|BingSearchRun-S3aq7|Tool|BaseTool|BingSearchRun|BaseTool",
"targetHandle": "BaseTool|tools|ZeroShotAgent-6vSFw"
},
{
"id": "reactflow__edge-BishengLLM-bc485BishengLLM|BishengLLM-bc485|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-ZeroShotAgent-6vSFwBaseLanguageModel|llm|ZeroShotAgent-6vSFw",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-bc485",
"target": "ZeroShotAgent-6vSFw",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-bc485|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|ZeroShotAgent-6vSFw"
}
],
"nodes": [
{
"id": "BingSearchRun-S3aq7",
"data": {
"id": "BingSearchRun-S3aq7",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": false,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "BingSearchRun",
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": false,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"api_wrapper": {
"l2": false,
"info": "",
"list": false,
"name": "api_wrapper",
"show": true,
"type": "BingSearchAPIWrapper",
"l2_name": "api_wrapper",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": ""
},
"args_schema": {
"l2": false,
"info": "",
"list": false,
"name": "args_schema",
"show": false,
"type": "Type[pydantic.main.BaseModel]",
"l2_name": "args_schema",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"handle_tool_error": {
"l2": false,
"info": "",
"list": false,
"name": "handle_tool_error",
"show": false,
"type": "bool",
"value": false,
"l2_name": "handle_tool_error",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "A wrapper around Bing Search. Useful for when you need to answer questions about current events. Input should be a search query.",
"base_classes": [
"Tool",
"BaseTool",
"BingSearchRun",
"BaseTool"
],
"display_name": "BingSearchRun",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "BingSearchRun",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 249,
"dragging": false,
"position": {
"x": 430.3491112346748,
"y": 526.7220206834627
},
"selected": true,
"positionAbsolute": {
"x": 430.3491112346748,
"y": 526.7220206834627
}
},
{
"id": "BingSearchAPIWrapper-Rikh1",
"data": {
"id": "BingSearchAPIWrapper-Rikh1",
"node": {
"l2_name": "BING搜索",
"template": {
"k": {
"l2": false,
"info": "",
"list": false,
"name": "k",
"show": false,
"type": "int",
"value": 10,
"l2_name": "k",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "BingSearchAPIWrapper",
"bing_search_url": {
"l2": false,
"info": "",
"list": false,
"name": "bing_search_url",
"show": true,
"type": "str",
"value": "https://api.bing.microsoft.com/v7.0/search",
"l2_name": "bing_search_url",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"bing_subscription_key": {
"l2": true,
"info": "",
"list": false,
"name": "bing_subscription_key",
"show": true,
"type": "str",
"value": "137b003c89f04966ac3b1081ee5b10bc",
"l2_name": "bing_subscription_key",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Wrapper for Bing Search API.",
"base_classes": [
"BingSearchAPIWrapper"
],
"display_name": "BingSearchAPIWrapper",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "BingSearchAPIWrapper",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 337,
"position": {
"x": -17.635040947620837,
"y": 468.8157898410929
},
"selected": false,
"positionAbsolute": {
"x": -17.635040947620837,
"y": 468.8157898410929
}
},
{
"id": "ZeroShotAgent-6vSFw",
"data": {
"id": "ZeroShotAgent-6vSFw",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"_type": "ZeroShotAgent",
"tools": {
"l2": false,
"info": "",
"list": true,
"name": "tools",
"show": true,
"type": "BaseTool",
"l2_name": "tools",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"prefix": {
"l2": false,
"info": "",
"list": false,
"name": "prefix",
"show": true,
"type": "str",
"value": "Answer the following questions as best you can. You have access to the following tools:",
"l2_name": "prefix",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
},
"suffix": {
"l2": false,
"info": "",
"list": false,
"name": "suffix",
"show": true,
"type": "str",
"value": "answer in Chinese.\nBegin!Question: {query}Thought:{agent_scratchpad}",
"l2_name": "suffix",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "AgentOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": true,
"type": "str",
"value": [
"query",
"agent_scratchpad"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callback_manager": {
"l2": false,
"info": "",
"list": false,
"name": "callback_manager",
"show": false,
"type": "BaseCallbackManager",
"l2_name": "callback_manager",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"format_instructions": {
"l2": false,
"info": "",
"list": false,
"name": "format_instructions",
"show": true,
"type": "str",
"value": "Use the following format:\n\nQuestion: the input question you must answer\nThought: you should always think about what to do\nAction: the action to take, should be one of [{tool_names}]\nAction Input: the input to the action\nObservation: the result of the action\n... (this Thought/Action/Action Input/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question",
"l2_name": "format_instructions",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
}
},
"description": "Construct an agent from an LLM and tools.",
"base_classes": [
"ZeroShotAgent",
"Agent",
"BaseSingleActionAgent",
"function"
],
"display_name": "ZeroShotAgent",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/agents/how_to/custom_mrkl_agent"
},
"type": "ZeroShotAgent",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 659,
"position": {
"x": 937.4050050593946,
"y": 62.48387006449201
},
"selected": false,
"positionAbsolute": {
"x": 937.4050050593946,
"y": 62.48387006449201
}
},
{
"id": "BishengLLM-bc485",
"data": {
"id": "BishengLLM-bc485",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 4,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"position": {
"x": 31.941398539174173,
"y": -96.4545131206016
},
"positionAbsolute": {
"x": 31.941398539174173,
"y": -96.4545131206016
}
}
],
"viewport": {
"x": 271.43658145124573,
"y": 126.9479936490626,
"zoom": 0.8967278502185819
}
},
"order_num": 7208912,
"guide_word": null
},
{
"update_time": "2024-08-23 17:41:28",
"parameters": null,
"name": "⭐️ 通用信息提取",
"description": "上传文件，触发自动提取，提取文件中的所有keyvalue信息。",
"flow_id": "947c03db78784932a344314844e32940",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-08-23 17:41:28",
"id": 190,
"data": {
"edges": [
{
"id": "reactflow__edge-PromptTemplate-5UHOIPromptTemplate|PromptTemplate-5UHOI|StringPromptTemplate|PromptTemplate|BasePromptTemplate-LLMChain-aSC8HBasePromptTemplate|prompt|LLMChain-aSC8H",
"style": {
"stroke": "#555"
},
"source": "PromptTemplate-5UHOI",
"target": "LLMChain-aSC8H",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "PromptTemplate|PromptTemplate-5UHOI|StringPromptTemplate|PromptTemplate|BasePromptTemplate",
"targetHandle": "BasePromptTemplate|prompt|LLMChain-aSC8H"
},
{
"id": "reactflow__edge-BishengLLM-2763fBishengLLM|BishengLLM-2763f|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-LLMChain-aSC8HBaseLanguageModel|llm|LLMChain-aSC8H",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-2763f",
"target": "LLMChain-aSC8H",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-2763f|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|LLMChain-aSC8H"
},
{
"id": "reactflow__edge-ElemUnstructuredLoaderV0-293c3ElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-293c3|Document-PromptTemplate-5UHOIDocument;BaseOutputParser|context|PromptTemplate-5UHOI",
"style": {
"stroke": "#555"
},
"source": "ElemUnstructuredLoaderV0-293c3",
"target": "PromptTemplate-5UHOI",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "ElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-293c3|Document",
"targetHandle": "Document;BaseOutputParser|context|PromptTemplate-5UHOI"
},
{
"id": "reactflow__edge-InputNode-3ccbfInputNode|InputNode-3ccbf|input-LLMChain-aSC8Hinput|input_node|LLMChain-aSC8H",
"style": {
"stroke": "#555"
},
"source": "InputNode-3ccbf",
"target": "LLMChain-aSC8H",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "InputNode|InputNode-3ccbf|input",
"targetHandle": "input|input_node|LLMChain-aSC8H"
},
{
"id": "reactflow__edge-InputFileNode-fc806InputFileNode|InputFileNode-fc806|fileNode-ElemUnstructuredLoaderV0-293c3fileNode|file_path|ElemUnstructuredLoaderV0-293c3",
"style": {
"stroke": "#555"
},
"source": "InputFileNode-fc806",
"target": "ElemUnstructuredLoaderV0-293c3",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "InputFileNode|InputFileNode-fc806|fileNode",
"targetHandle": "fileNode|file_path|ElemUnstructuredLoaderV0-293c3"
}
],
"nodes": [
{
"id": "LLMChain-aSC8H",
"data": {
"id": "LLMChain-aSC8H",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "LLMChain",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"prompt": {
"l2": false,
"info": "",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"l2_name": "prompt",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"llm_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "llm_kwargs",
"show": false,
"type": "code",
"l2_name": "llm_kwargs",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "text",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "BaseLLMOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"return_final_only": {
"l2": false,
"info": "",
"list": false,
"name": "return_final_only",
"show": false,
"type": "bool",
"value": true,
"l2_name": "return_final_only",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain to run queries against LLMs.",
"base_classes": [
"Chain",
"LLMChain",
"function"
],
"display_name": "LLMChain",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/foundational/llm_chain"
},
"type": "LLMChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 315,
"position": {
"x": 1313.6980454632887,
"y": 900.921715712506
},
"selected": false,
"positionAbsolute": {
"x": 1313.6980454632887,
"y": 900.921715712506
}
},
{
"id": "PromptTemplate-5UHOI",
"data": {
"id": "PromptTemplate-5UHOI",
"node": {
"name": "",
"template": {
"_type": "PromptTemplate",
"input": {
"info": "",
"list": false,
"name": "input",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser"
],
"placeholder": "",
"display_name": "input"
},
"context": {
"info": "",
"list": false,
"name": "context",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser"
],
"placeholder": "",
"display_name": "context"
},
"template": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "提取以下文本内容中的所有keyvalue信息，符合如下格式定义：\n\nThe output should be a Markdown code snippet formatted in the following schema, including the leading and trailing \"```json\" and \"```\":\n\n```json\n\"Key1\": Value\n\"Key2\": Value\n```\n\n文本内容\n{context}\n\n{input}",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"context",
"input"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"template_format": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"partial_variables": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"validate_template": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Schema to represent a prompt for an LLM.",
"base_classes": [
"StringPromptTemplate",
"PromptTemplate",
"BasePromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"context"
],
"template": [
"context",
"input"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {
"formatters": {
"openai_api_key": {}
},
"base_formatters": {
"dict": {},
"list": {},
"show": {},
"union": {},
"kwargs": {},
"default": {},
"headers": {},
"optional": {},
"password": {},
"multiline": {},
"model_fields": {
"MODEL_DICT": {
"OpenAI": [
"text-davinci-003",
"text-davinci-002",
"text-curie-001",
"text-babbage-001",
"text-ada-001"
],
"Anthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
],
"ChatOpenAI": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k"
],
"ChatAnthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
]
}
},
"dict_code_file": {}
}
}
},
"type": "PromptTemplate",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 416,
"dragging": false,
"position": {
"x": 809.51004530817,
"y": 1237.7696518068285
},
"selected": false,
"positionAbsolute": {
"x": 809.51004530817,
"y": 1237.7696518068285
}
},
{
"id": "BishengLLM-2763f",
"data": {
"id": "BishengLLM-2763f",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 5,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 406,
"position": {
"x": -68.63553056857684,
"y": 696.600000304993
},
"positionAbsolute": {
"x": -68.63553056857684,
"y": 696.600000304993
}
},
{
"id": "ElemUnstructuredLoaderV0-293c3",
"data": {
"id": "ElemUnstructuredLoaderV0-293c3",
"node": {
"beta": false,
"template": {
"_type": "ElemUnstructuredLoaderV0",
"kwargs": {
"info": "",
"list": false,
"name": "kwargs",
"show": true,
"type": "dict",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "kwargs",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"value": "{}",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "fileNode",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"suffixes": [
".html",
".md",
".txt",
".bmp",
".jpg",
".png",
".jpeg",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
".xls",
".xlsx",
".tsv",
".csv",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"bmp",
"jpg",
"png",
"jpeg",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
".xls",
"xlsx",
"tsv",
"csv",
"tiff"
],
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "File Path",
"model_config": {}
},
"unstructured_api_url": {
"info": "",
"list": false,
"name": "unstructured_api_url",
"show": true,
"type": "str",
"value": "https://bisheng.dataelem.com/api/v1/etl4llm/predict",
"dynamic": false,
"advanced": false,
"password": true,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "unstructured_api_url",
"model_config": {}
}
},
"description": "The appropriate parser is automatically selected based on the file format and OCR is supported",
"base_classes": [
"Document"
],
"display_name": "ElemUnstructuredLoaderV0",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "ElemUnstructuredLoaderV0",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 475,
"position": {
"x": 237.30334166517488,
"y": 1166.7647849898744
},
"positionAbsolute": {
"x": 237.30334166517488,
"y": 1166.7647849898744
}
},
{
"id": "InputNode-3ccbf",
"data": {
"id": "InputNode-3ccbf",
"node": {
"beta": false,
"template": {
"_type": "input",
"input": {
"info": "",
"list": true,
"name": "input",
"show": true,
"type": "str",
"value": [
"开始"
],
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": true,
"title_case": true,
"placeholder": "",
"display_name": "输入内容",
"model_config": {}
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"input"
],
"display_name": "InputNode",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "InputNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 253,
"position": {
"x": 809.0064819287146,
"y": 663.3484087022575
},
"positionAbsolute": {
"x": 809.0064819287146,
"y": 663.3484087022575
}
},
{
"id": "InputFileNode-fc806",
"data": {
"id": "InputFileNode-fc806",
"node": {
"beta": false,
"template": {
"_type": "InputFileNode",
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "file",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".bmp",
".jpg",
".png",
".jpeg",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
"..xls",
".xlsx",
".tsv",
".csv",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"bmp",
"jpg",
"png",
"jpeg",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
".xls",
"xlsx",
"tsv",
"csv",
"tiff"
],
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "File Path",
"model_config": {}
},
"file_type": {
"info": "Tips for which file should upload",
"list": false,
"name": "file_type",
"show": true,
"type": "str",
"value": "请上传文件",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "提示上传文件类型",
"display_name": "Name",
"model_config": {}
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"fileNode"
],
"display_name": "InputFileNode",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "InputFileNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 335,
"dragging": false,
"position": {
"x": -283.58174168605024,
"y": 1208.0544562311302
},
"selected": true,
"positionAbsolute": {
"x": -283.58174168605024,
"y": 1208.0544562311302
}
}
],
"viewport": {
"x": 249.5759133596914,
"y": -408.5108685516501,
"zoom": 0.6296974332414016
}
},
"order_num": 7274447,
"guide_word": null
},
{
"update_time": "2024-09-03 21:23:10",
"parameters": null,
"name": "⭐️单文档问答",
"description": "上传文件，然后可以查询文件中相关的问题",
"flow_id": "9f59249acf76448899f8918b2aa964af",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-09-03 21:23:10",
"id": 191,
"data": {
"edges": [
{
"id": "reactflow__edge-ElemUnstructuredLoaderV0-LAGXMElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-LAGXM|Document-RecursiveCharacterTextSplitter-a7fc9Document|documents|RecursiveCharacterTextSplitter-a7fc9",
"style": {
"stroke": "#555"
},
"source": "ElemUnstructuredLoaderV0-LAGXM",
"target": "RecursiveCharacterTextSplitter-a7fc9",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "ElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-LAGXM|Document",
"targetHandle": "Document|documents|RecursiveCharacterTextSplitter-a7fc9"
},
{
"id": "reactflow__edge-RecursiveCharacterTextSplitter-a7fc9RecursiveCharacterTextSplitter|RecursiveCharacterTextSplitter-a7fc9|Document-Milvus-T3kRHDocument|documents|Milvus-T3kRH",
"style": {
"stroke": "#555"
},
"source": "RecursiveCharacterTextSplitter-a7fc9",
"target": "Milvus-T3kRH",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "RecursiveCharacterTextSplitter|RecursiveCharacterTextSplitter-a7fc9|Document",
"targetHandle": "Document|documents|Milvus-T3kRH"
},
{
"id": "reactflow__edge-Milvus-T3kRHMilvus|Milvus-T3kRH|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever-RetrievalQA-37d31BaseRetriever|retriever|RetrievalQA-37d31",
"style": {
"stroke": "#555"
},
"source": "Milvus-T3kRH",
"target": "RetrievalQA-37d31",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "Milvus|Milvus-T3kRH|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-37d31"
},
{
"id": "reactflow__edge-CombineDocsChain-Pud2pCombineDocsChain|CombineDocsChain-Pud2p|BaseCombineDocumentsChain|function-RetrievalQA-37d31BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-37d31",
"style": {
"stroke": "#555"
},
"source": "CombineDocsChain-Pud2p",
"target": "RetrievalQA-37d31",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "CombineDocsChain|CombineDocsChain-Pud2p|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-37d31"
},
{
"id": "reactflow__edge-InputFileNode-eda80InputFileNode|InputFileNode-eda80|fileNode-ElemUnstructuredLoaderV0-LAGXMfileNode|file_path|ElemUnstructuredLoaderV0-LAGXM",
"style": {
"stroke": "#555"
},
"source": "InputFileNode-eda80",
"target": "ElemUnstructuredLoaderV0-LAGXM",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "InputFileNode|InputFileNode-eda80|fileNode",
"targetHandle": "fileNode|file_path|ElemUnstructuredLoaderV0-LAGXM"
},
{
"id": "reactflow__edge-BishengLLM-1bcf8BishengLLM|BishengLLM-1bcf8|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM-CombineDocsChain-Pud2pBaseLanguageModel|llm|CombineDocsChain-Pud2p",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-1bcf8",
"target": "CombineDocsChain-Pud2p",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-1bcf8|BishengLLM|BaseLanguageModel|BaseChatModel|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-Pud2p"
},
{
"id": "reactflow__edge-BishengEmbedding-eda33BishengEmbedding|BishengEmbedding-eda33|Embeddings|BishengEmbedding-Milvus-T3kRHEmbeddings|embedding|Milvus-T3kRH",
"style": {
"stroke": "#555"
},
"source": "BishengEmbedding-eda33",
"target": "Milvus-T3kRH",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengEmbedding|BishengEmbedding-eda33|Embeddings|BishengEmbedding",
"targetHandle": "Embeddings|embedding|Milvus-T3kRH"
}
],
"nodes": [
{
"id": "CombineDocsChain-Pud2p",
"data": {
"id": "CombineDocsChain-Pud2p",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"token_max": {
"l2": false,
"info": "当前只对stuff 生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"l2_name": "token_max",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 377,
"position": {
"x": 762.7783327725605,
"y": 31.94438969676179
},
"selected": false,
"positionAbsolute": {
"x": 762.7783327725605,
"y": 31.94438969676179
}
},
{
"id": "Milvus-T3kRH",
"data": {
"id": "Milvus-T3kRH",
"node": {
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"collection_name": {
"l2": false,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "collection_name",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": ""
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"VectorStore",
"Milvus",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 347,
"dragging": false,
"position": {
"x": 885.364722441176,
"y": 638.0514447977121
},
"selected": false,
"positionAbsolute": {
"x": 885.364722441176,
"y": 638.0514447977121
}
},
{
"id": "ElemUnstructuredLoaderV0-LAGXM",
"data": {
"id": "ElemUnstructuredLoaderV0-LAGXM",
"node": {
"template": {
"_type": "ElemUnstructuredLoaderV0",
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"value": "{}",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Metadata"
},
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "fileNode",
"value": "",
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".jpg",
".png",
".jpeg",
".csv",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
".xlsx",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"jpg",
"png",
"jpeg",
"csv",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
"tiff",
"xlsx"
],
"multiline": false,
"placeholder": ""
},
"unstructured_api_url": {
"info": "",
"list": false,
"name": "unstructured_api_url",
"show": true,
"type": "str",
"value": "https://bisheng.dataelem.com/api/v1/etl4llm/predict",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "unstructured_api_url"
}
},
"description": "Loads a PDF with pypdf and chunks at character level. dummy version",
"base_classes": [
"Document"
],
"display_name": "ElemUnstructuredLoaderV0",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": ""
},
"type": "ElemUnstructuredLoaderV0",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 393,
"dragging": false,
"position": {
"x": -174.48653868947127,
"y": 475.0990452635182
},
"selected": false,
"positionAbsolute": {
"x": -174.48653868947127,
"y": 475.0990452635182
}
},
{
"id": "RecursiveCharacterTextSplitter-a7fc9",
"data": {
"id": "RecursiveCharacterTextSplitter-a7fc9",
"node": {
"template": {
"_type": "RecursiveCharacterTextSplitter",
"documents": {
"info": "",
"list": false,
"name": "documents",
"show": true,
"type": "Document",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"chunk_size": {
"info": "",
"list": false,
"name": "chunk_size",
"show": true,
"type": "int",
"value": 1000,
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Chunk Size"
},
"separators": {
"info": "",
"list": false,
"name": "separators",
"show": true,
"type": "str",
"value": "\\n\\n",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Separator"
},
"chunk_overlap": {
"info": "",
"list": false,
"name": "chunk_overlap",
"show": true,
"type": "int",
"value": 200,
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Chunk Overlap"
},
"separator_type": {
"info": "",
"list": true,
"name": "separator_type",
"show": true,
"type": "str",
"value": "Text",
"options": [
"Text",
"cobol",
"cpp",
"csharp",
"go",
"html",
"java",
"js",
"kotlin",
"latex",
"markdown",
"php",
"proto",
"python",
"rst",
"ruby",
"rust",
"scala",
"sol",
"swift",
"ts"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Separator Type"
}
},
"description": "Splitting text by recursively look at characters.",
"base_classes": [
"Document"
],
"display_name": "RecursiveCharacterTextSplitter",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/data_connection/document_transformers/text_splitters/recursive_text_splitter"
},
"type": "RecursiveCharacterTextSplitter",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 541,
"dragging": false,
"position": {
"x": 380.122171588396,
"y": 483.5257541316808
},
"selected": false,
"positionAbsolute": {
"x": 380.122171588396,
"y": 483.5257541316808
}
},
{
"id": "RetrievalQA-37d31",
"data": {
"id": "RetrievalQA-37d31",
"node": {
"beta": false,
"template": {
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "RetrievalQA",
"memory": {
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Memory",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"input_key": {
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Input Key",
"model_config": {}
},
"retriever": {
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Retriever",
"model_config": {}
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Preset Question",
"model_config": {}
},
"output_key": {
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Output Key",
"model_config": {}
},
"combine_documents_chain": {
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Combine Documents Chain",
"model_config": {}
},
"return_source_documents": {
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Return Source Documents",
"model_config": {}
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"Chain",
"BaseRetrievalQA",
"RetrievalQA",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa",
"field_formatters": {}
},
"type": "RetrievalQA",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 481,
"dragging": false,
"position": {
"x": 1497.3865616690784,
"y": 525.8440613798767
},
"selected": false,
"positionAbsolute": {
"x": 1497.3865616690784,
"y": 525.8440613798767
}
},
{
"id": "InputFileNode-eda80",
"data": {
"id": "InputFileNode-eda80",
"node": {
"beta": false,
"template": {
"_type": "InputFileNode",
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "file",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".jpg",
".png",
".jpeg",
".csv",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
".tiff",
".xlsx"
],
"fileTypes": [
"html",
"md",
"txt",
"jpg",
"png",
"jpeg",
"csv",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
"tiff",
"xlsx"
],
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "File Path",
"model_config": {}
},
"file_type": {
"info": "Tips for which file should upload",
"list": false,
"name": "file_type",
"show": true,
"type": "str",
"value": "请上传文件",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "提示上传文件类型",
"display_name": "Name",
"model_config": {}
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"fileNode"
],
"display_name": "InputFileNode",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "InputFileNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 337,
"dragging": false,
"position": {
"x": -728.0965344734079,
"y": 480.2837762791313
},
"selected": false,
"positionAbsolute": {
"x": -728.0965344734079,
"y": 480.2837762791313
}
},
{
"id": "BishengLLM-1bcf8",
"data": {
"id": "BishengLLM-1bcf8",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 9,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BishengLLM",
"BaseLanguageModel",
"BaseChatModel",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 407,
"dragging": false,
"position": {
"x": 220.52841712247573,
"y": -113.58850984307922
},
"selected": false,
"positionAbsolute": {
"x": 220.52841712247573,
"y": -113.58850984307922
}
},
{
"id": "BishengEmbedding-eda33",
"data": {
"id": "BishengEmbedding-eda33",
"node": {
"beta": false,
"template": {
"_type": "BishengEmbedding",
"model": {
"info": "",
"list": false,
"name": "model",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_embedding",
"value": 1,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"embeddings": {
"info": "",
"list": false,
"name": "embeddings",
"show": true,
"type": "Embeddings",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Embeddings",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": true,
"type": "LLMModel",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"max_retries": {
"info": "",
"list": false,
"name": "max_retries",
"show": true,
"type": "int",
"value": 6,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Max Retries",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": true,
"type": "LLMServer",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"model_kwargs": {
"info": "",
"list": false,
"name": "model_kwargs",
"show": true,
"type": "code",
"value": {},
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Kwargs",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "DashScopeEmbeddings",
"vllm": "OpenAIEmbeddings",
"zhipu": "OpenAIEmbeddings",
"ollama": "OllamaEmbeddings",
"openai": "OpenAIEmbeddings",
"minimax": "OpenAIEmbeddings",
"qianfan": "QianfanEmbeddingsEndpoint",
"llamacpp": "OpenAIEmbeddings",
"bisheng_rt": "HostEmbeddings",
"xinference": "OpenAIEmbeddings",
"azure_openai": "AzureOpenAIEmbeddings"
},
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"request_timeout": {
"info": "",
"list": false,
"name": "request_timeout",
"show": true,
"type": "int",
"value": 200,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Request Timeout",
"model_config": {}
},
"embedding_ctx_length": {
"info": "",
"list": false,
"name": "embedding_ctx_length",
"show": true,
"type": "int",
"value": 8192,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Embedding Ctx Length",
"model_config": {}
}
},
"description": "Use the embedding model that has been launched in model management ",
"base_classes": [
"Embeddings",
"BishengEmbedding"
],
"display_name": "BishengEmbedding",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengEmbedding",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 273,
"dragging": false,
"position": {
"x": 390.035554735752,
"y": 1105.577580188958
},
"selected": true,
"positionAbsolute": {
"x": 390.035554735752,
"y": 1105.577580188958
}
}
],
"viewport": {
"x": 543.6795293445036,
"y": 101.93050532990264,
"zoom": 0.5416165288707682
}
},
"order_num": 7339982,
"guide_word": null
},
{
"update_time": "2024-09-20 18:40:48",
"parameters": null,
"name": "演示-券商研报生成",
"description": "生成关于某家公司的研究报告",
"flow_id": "7bbdcd64287343f98effc90a70119f0a",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-09-20 18:40:48",
"id": 192,
"data": {
"edges": [
{
"id": "reactflow__edge-CombineDocsChain-foMvuCombineDocsChain|CombineDocsChain-foMvu|BaseCombineDocumentsChain|function-RetrievalQA-1fCXFBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-1fCXF",
"style": {
"stroke": "#555"
},
"source": "CombineDocsChain-foMvu",
"target": "RetrievalQA-1fCXF",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "CombineDocsChain|CombineDocsChain-foMvu|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-1fCXF"
},
{
"id": "reactflow__edge-RetrievalQA-1fCXFRetrievalQA|RetrievalQA-1fCXF|Chain|BaseRetrievalQA|RetrievalQA|function-Report-fSFx4function|chains|Report-fSFx4",
"style": {
"stroke": "#555"
},
"source": "RetrievalQA-1fCXF",
"target": "Report-fSFx4",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "RetrievalQA|RetrievalQA-1fCXF|Generic|Chain|Runnable|RetrievalQA|RunnableSerializable|BaseRetrievalQA|function",
"targetHandle": "function|chains|Report-fSFx4"
},
{
"id": "reactflow__edge-Milvus-E2BnsMilvus|Milvus-E2Bns|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever-RetrievalQA-1fCXFBaseRetriever|retriever|RetrievalQA-1fCXF",
"style": {
"stroke": "#555"
},
"source": "Milvus-E2Bns",
"target": "RetrievalQA-1fCXF",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "Milvus|Milvus-E2Bns|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-1fCXF"
},
{
"id": "reactflow__edge-ProxyChatLLM-CzZfAProxyChatLLM|ProxyChatLLM-CzZfA|BaseLanguageModel|BaseChatModel|ProxyChatLLM|BaseLLM-CombineDocsChain-foMvuBaseLanguageModel|llm|CombineDocsChain-foMvu",
"style": {
"stroke": "#555"
},
"source": "ProxyChatLLM-CzZfA",
"target": "CombineDocsChain-foMvu",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "ProxyChatLLM|ProxyChatLLM-CzZfA|Generic|BaseLanguageModel|Runnable|ProxyChatLLM|RunnableSerializable|BaseChatModel|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-foMvu"
},
{
"id": "reactflow__edge-RetrievalQA-QWfeURetrievalQA|RetrievalQA-QWfeU|Chain|BaseRetrievalQA|RetrievalQA|function-Report-fSFx4function|chains|Report-fSFx4",
"style": {
"stroke": "#555"
},
"source": "RetrievalQA-QWfeU",
"target": "Report-fSFx4",
"animated": true,
"className": "",
"sourceHandle": "RetrievalQA|RetrievalQA-QWfeU|Generic|Chain|Runnable|RetrievalQA|RunnableSerializable|BaseRetrievalQA|function",
"targetHandle": "function|chains|Report-fSFx4"
},
{
"id": "reactflow__edge-CombineDocsChain-48oi2CombineDocsChain|CombineDocsChain-48oi2|BaseCombineDocumentsChain|function-RetrievalQA-QWfeUBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-QWfeU",
"style": {
"stroke": "#555"
},
"source": "CombineDocsChain-48oi2",
"target": "RetrievalQA-QWfeU",
"animated": true,
"className": "",
"sourceHandle": "CombineDocsChain|CombineDocsChain-48oi2|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-QWfeU"
},
{
"id": "reactflow__edge-ProxyChatLLM-3oL2CProxyChatLLM|ProxyChatLLM-3oL2C|BaseLanguageModel|BaseChatModel|ProxyChatLLM|BaseLLM-CombineDocsChain-48oi2BaseLanguageModel|llm|CombineDocsChain-48oi2",
"style": {
"stroke": "#555"
},
"source": "ProxyChatLLM-3oL2C",
"target": "CombineDocsChain-48oi2",
"animated": true,
"className": "",
"sourceHandle": "ProxyChatLLM|ProxyChatLLM-3oL2C|Generic|BaseLanguageModel|Runnable|ProxyChatLLM|RunnableSerializable|BaseChatModel|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-48oi2"
},
{
"id": "reactflow__edge-Milvus-YOosCMilvus|Milvus-YOosC|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever-RetrievalQA-QWfeUBaseRetriever|retriever|RetrievalQA-QWfeU",
"style": {
"stroke": "#555"
},
"source": "Milvus-YOosC",
"target": "RetrievalQA-QWfeU",
"animated": true,
"className": "",
"sourceHandle": "Milvus|Milvus-YOosC|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-QWfeU"
},
{
"id": "reactflow__edge-InputNode-IZUU3InputNode|InputNode-IZUU3|input-RetrievalQA-QWfeUinput|input_node|RetrievalQA-QWfeU",
"style": {
"stroke": "#555"
},
"source": "InputNode-IZUU3",
"target": "RetrievalQA-QWfeU",
"animated": true,
"className": "",
"sourceHandle": "InputNode|InputNode-IZUU3|input",
"targetHandle": "input|input_node|RetrievalQA-QWfeU"
},
{
"id": "reactflow__edge-InputNode-NXC32InputNode|InputNode-NXC32|input-RetrievalQA-1fCXFinput|input_node|RetrievalQA-1fCXF",
"style": {
"stroke": "#555"
},
"source": "InputNode-NXC32",
"target": "RetrievalQA-1fCXF",
"animated": true,
"className": "",
"sourceHandle": "InputNode|InputNode-NXC32|input",
"targetHandle": "input|input_node|RetrievalQA-1fCXF"
},
{
"id": "reactflow__edge-PromptTemplate-wxEKtPromptTemplate|PromptTemplate-wxEKt|BasePromptTemplate|PromptTemplate|StringPromptTemplate-CombineDocsChain-48oi2BasePromptTemplate|prompt|CombineDocsChain-48oi2",
"style": {
"stroke": "#555"
},
"source": "PromptTemplate-wxEKt",
"target": "CombineDocsChain-48oi2",
"animated": true,
"className": "",
"sourceHandle": "PromptTemplate|PromptTemplate-wxEKt|Generic|PromptTemplate|StringPromptTemplate|Runnable|RunnableSerializable|BasePromptTemplate",
"targetHandle": "BasePromptTemplate|prompt|CombineDocsChain-48oi2"
},
{
"id": "reactflow__edge-PromptTemplate-ot33yPromptTemplate|PromptTemplate-ot33y|BasePromptTemplate|PromptTemplate|StringPromptTemplate-CombineDocsChain-foMvuBasePromptTemplate|prompt|CombineDocsChain-foMvu",
"style": {
"stroke": "#555"
},
"source": "PromptTemplate-ot33y",
"target": "CombineDocsChain-foMvu",
"animated": true,
"className": "",
"sourceHandle": "PromptTemplate|PromptTemplate-ot33y|Generic|PromptTemplate|StringPromptTemplate|Runnable|RunnableSerializable|BasePromptTemplate",
"targetHandle": "BasePromptTemplate|prompt|CombineDocsChain-foMvu"
},
{
"id": "reactflow__edge-VariableNode-d14a6VariableNode|VariableNode-d14a6|VariableNode-Report-fSFx4VariableNode|variables|Report-fSFx4",
"style": {
"stroke": "#555"
},
"source": "VariableNode-d14a6",
"target": "Report-fSFx4",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "VariableNode|VariableNode-d14a6|VariableNode",
"targetHandle": "VariableNode|variables|Report-fSFx4"
}
],
"nodes": [
{
"id": "Report-fSFx4",
"data": {
"id": "Report-fSFx4",
"node": {
"template": {
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "Report",
"chains": {
"info": "",
"list": true,
"name": "chains",
"show": true,
"type": "function",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "functions"
},
"memory": {
"info": "",
"list": false,
"name": "memory",
"show": false,
"type": "BaseMemory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"info": "",
"list": false,
"name": "input_key",
"show": false,
"type": "str",
"value": "report_name",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"stop_flag": {
"info": "",
"list": false,
"name": "stop_flag",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"variables": {
"info": "",
"list": true,
"name": "variables",
"show": true,
"type": "VariableNode",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": false,
"type": "input",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"info": "",
"list": false,
"name": "output_key",
"show": false,
"type": "str",
"value": "text",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"report_name": {
"info": "the file name we generate",
"list": false,
"name": "report_name",
"show": true,
"type": "str",
"value": "行业研报",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Report Name"
}
},
"description": "",
"base_classes": [
"Generic",
"Report",
"Chain",
"Runnable",
"RunnableSerializable"
],
"display_name": "Report",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "Report",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 359,
"dragging": false,
"position": {
"x": 3290.5383305655187,
"y": 409.3538527610235
},
"selected": true,
"positionAbsolute": {
"x": 3290.5383305655187,
"y": 409.3538527610235
}
},
{
"id": "InputNode-IZUU3",
"data": {
"id": "InputNode-IZUU3",
"node": {
"template": {
"_type": "input",
"input": {
"info": "",
"list": true,
"name": "input",
"show": true,
"type": "str",
"value": [
"公司的核心业务和主要产品和服务",
"公司在行业中的地位和市场份额",
"公司的核心竞争力，如专利技术、品牌影响力等",
"对公司进行核心竞争力、盈利模式及可持续性分析",
"影响公司业务的风险，包括市场风险、财务风险、法规风险",
"公司的创新能力和技术水平",
"公司未来展望和战略规划"
],
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "输入内容"
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"input"
],
"display_name": "InputNode",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "InputNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 555,
"dragging": false,
"position": {
"x": 2071.9473113130907,
"y": -480.7563001986203
},
"selected": false,
"positionAbsolute": {
"x": 2071.9473113130907,
"y": -480.7563001986203
}
},
{
"id": "RetrievalQA-1fCXF",
"data": {
"id": "RetrievalQA-1fCXF",
"node": {
"template": {
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "RetrievalQA",
"memory": {
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"retriever": {
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"combine_documents_chain": {
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"Generic",
"Chain",
"Runnable",
"RetrievalQA",
"RunnableSerializable",
"BaseRetrievalQA",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 317,
"dragging": false,
"position": {
"x": 2237.06298695131,
"y": 1483.4909209372413
},
"selected": false,
"positionAbsolute": {
"x": 2237.06298695131,
"y": 1483.4909209372413
}
},
{
"id": "CombineDocsChain-foMvu",
"data": {
"id": "CombineDocsChain-foMvu",
"node": {
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"prompt": {
"info": "只对Stuff类型生效",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
},
"token_max": {
"info": "只对Stuff类型生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"document_prompt": {
"info": "",
"list": false,
"name": "document_prompt",
"show": true,
"type": "BasePromptTemplate",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 449,
"dragging": false,
"position": {
"x": 1552.0278708162225,
"y": 1323.4452741641828
},
"selected": false,
"positionAbsolute": {
"x": 1552.0278708162225,
"y": 1323.4452741641828
}
},
{
"id": "Milvus-E2Bns",
"data": {
"id": "Milvus-E2Bns",
"node": {
"template": {
"_type": "Milvus",
"drop_old": {
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"documents": {
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"info": "",
"list": false,
"name": "embedding",
"show": false,
"type": "Embeddings",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_type": {
"info": "",
"list": false,
"name": "search_type",
"show": true,
"type": "str",
"value": "similarity",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"index_params": {
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_kwargs": {
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_params": {
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"collection_name": {
"l2": true,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "AI行业研报",
"l2_name": "行业库",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": 1855
},
"connection_args": {
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"consistency_level": {
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"Milvus",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 393,
"dragging": false,
"position": {
"x": 1559.8817737276297,
"y": 1858.4770607573223
},
"selected": false,
"positionAbsolute": {
"x": 1559.8817737276297,
"y": 1858.4770607573223
}
},
{
"id": "ProxyChatLLM-CzZfA",
"data": {
"id": "ProxyChatLLM-CzZfA",
"node": {
"template": {
"n": {
"info": "",
"list": false,
"name": "n",
"show": false,
"type": "int",
"value": 1,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "ProxyChatLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": false,
"type": "bool",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 0.9,
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"client": {
"info": "",
"list": false,
"name": "client",
"show": false,
"type": "Any",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"headers": {
"info": "",
"list": false,
"name": "headers",
"show": false,
"type": "code",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"max_tokens": {
"info": "",
"list": false,
"name": "max_tokens",
"show": true,
"type": "int",
"value": 2048,
"advanced": true,
"password": true,
"required": false,
"multiline": false,
"placeholder": ""
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": true,
"type": "str",
"value": "gpt-3.5-turbo-16k-0613",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"max_retries": {
"info": "",
"list": false,
"name": "max_retries",
"show": false,
"type": "int",
"value": 0,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.7,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"model_kwargs": {
"info": "",
"list": false,
"name": "model_kwargs",
"show": true,
"type": "code",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"elemai_api_key": {
"info": "",
"list": false,
"name": "elemai_api_key",
"show": true,
"type": "str",
"value": "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJiaXNoZW5nX29wZW42IiwiZXhwIjoxNzEzNzU0MjUyfQ.ww1l-GTBYJiHV3-U1JcacvWOqYPd-QMpuJIeuO9_OM8",
"advanced": false,
"password": true,
"required": false,
"multiline": false,
"placeholder": ""
},
"elemai_base_url": {
"info": "",
"list": false,
"name": "elemai_base_url",
"show": true,
"type": "str",
"value": "http://192.168.106.12:6001/v1/chatcompletion",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"request_timeout": {
"info": "",
"list": false,
"name": "request_timeout",
"show": false,
"type": "float",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"tiktoken_model_name": {
"info": "",
"list": false,
"name": "tiktoken_model_name",
"show": false,
"type": "str",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Wrapper around proxy Chat large language models.",
"base_classes": [
"Generic",
"BaseLanguageModel",
"Runnable",
"ProxyChatLLM",
"RunnableSerializable",
"BaseChatModel",
"BaseLLM"
],
"display_name": "ProxyChatLLM",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "ProxyChatLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 501,
"dragging": false,
"position": {
"x": 1021.6972307239736,
"y": 950.9394648638606
},
"selected": false,
"positionAbsolute": {
"x": 1021.6972307239736,
"y": 950.9394648638606
}
},
{
"id": "RetrievalQA-QWfeU",
"data": {
"id": "RetrievalQA-QWfeU",
"node": {
"template": {
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "RetrievalQA",
"memory": {
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"retriever": {
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"combine_documents_chain": {
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"Generic",
"Chain",
"Runnable",
"RetrievalQA",
"RunnableSerializable",
"BaseRetrievalQA",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 317,
"dragging": false,
"position": {
"x": 2621.435785834929,
"y": 5.613751609646442
},
"selected": false,
"positionAbsolute": {
"x": 2621.435785834929,
"y": 5.613751609646442
}
},
{
"id": "CombineDocsChain-48oi2",
"data": {
"id": "CombineDocsChain-48oi2",
"node": {
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"prompt": {
"info": "只对Stuff类型生效",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
},
"token_max": {
"info": "只对Stuff类型生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"document_prompt": {
"info": "",
"list": false,
"name": "document_prompt",
"show": true,
"type": "BasePromptTemplate",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 449,
"dragging": false,
"position": {
"x": 1557.52279817175,
"y": -432.4678010752729
},
"selected": false,
"positionAbsolute": {
"x": 1557.52279817175,
"y": -432.4678010752729
}
},
{
"id": "ProxyChatLLM-3oL2C",
"data": {
"id": "ProxyChatLLM-3oL2C",
"node": {
"template": {
"n": {
"info": "",
"list": false,
"name": "n",
"show": false,
"type": "int",
"value": 1,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "ProxyChatLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": false,
"type": "bool",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 0.9,
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"client": {
"info": "",
"list": false,
"name": "client",
"show": false,
"type": "Any",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"headers": {
"info": "",
"list": false,
"name": "headers",
"show": false,
"type": "code",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"placeholder": ""
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"max_tokens": {
"info": "",
"list": false,
"name": "max_tokens",
"show": true,
"type": "int",
"value": 2048,
"advanced": true,
"password": true,
"required": false,
"multiline": false,
"placeholder": ""
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": true,
"type": "str",
"value": "gpt-3.5-turbo-16k-0613",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"max_retries": {
"info": "",
"list": false,
"name": "max_retries",
"show": false,
"type": "int",
"value": 0,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.7,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"model_kwargs": {
"info": "",
"list": false,
"name": "model_kwargs",
"show": true,
"type": "code",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"elemai_api_key": {
"info": "",
"list": false,
"name": "elemai_api_key",
"show": true,
"type": "str",
"value": "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJiaXNoZW5nX29wZW42IiwiZXhwIjoxNzEzNzU0MjUyfQ.ww1l-GTBYJiHV3-U1JcacvWOqYPd-QMpuJIeuO9_OM8",
"advanced": false,
"password": true,
"required": false,
"multiline": false,
"placeholder": ""
},
"elemai_base_url": {
"info": "",
"list": false,
"name": "elemai_base_url",
"show": true,
"type": "str",
"value": "http://192.168.106.12:6001/v1/chatcompletion",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"request_timeout": {
"info": "",
"list": false,
"name": "request_timeout",
"show": false,
"type": "float",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"tiktoken_model_name": {
"info": "",
"list": false,
"name": "tiktoken_model_name",
"show": false,
"type": "str",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Wrapper around proxy Chat large language models.",
"base_classes": [
"Generic",
"BaseLanguageModel",
"Runnable",
"ProxyChatLLM",
"RunnableSerializable",
"BaseChatModel",
"BaseLLM"
],
"display_name": "ProxyChatLLM",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "ProxyChatLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 501,
"dragging": false,
"position": {
"x": 1024.1163217007638,
"y": -520.5274231154001
},
"selected": false,
"positionAbsolute": {
"x": 1024.1163217007638,
"y": -520.5274231154001
}
},
{
"id": "InputNode-NXC32",
"data": {
"id": "InputNode-NXC32",
"node": {
"template": {
"_type": "input",
"input": {
"info": "",
"list": true,
"name": "input",
"show": true,
"type": "str",
"value": [
"人工智能行业的市场概况，如市场规模等",
"“AI人工智能”行业的趋势、动态和主要推动因素，例如技术创新、市场需求变化等",
"“AI人工智能”涉及哪些领域，分析各个领域的特点、规模和增长前景",
"”AI人工智能“行业的优势、劣势、机会和威胁进行评估",
"“AI人工智能”行业影响较大的法规、政策和法律环境",
"“AI人工智能”行业中的投资机会，可能影响行业表现的风险因素",
"“AI人工智能”行业未来发展趋势和预测的观点，考虑到技术发展、市场需求变化等因素",
"写一个“AI人工智能“行业的现状的总结题目，15个字以内，示例：AI+产品矩阵完善，利润端短期承压"
],
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "输入内容"
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"input"
],
"display_name": "InputNode",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "InputNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 595,
"dragging": false,
"position": {
"x": 1618.88491687723,
"y": 660.651152090995
},
"selected": false,
"positionAbsolute": {
"x": 1618.88491687723,
"y": 660.651152090995
}
},
{
"id": "Milvus-YOosC",
"data": {
"id": "Milvus-YOosC",
"node": {
"template": {
"_type": "Milvus",
"drop_old": {
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"documents": {
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"info": "",
"list": false,
"name": "embedding",
"show": false,
"type": "Embeddings",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_type": {
"info": "",
"list": false,
"name": "search_type",
"show": true,
"type": "str",
"value": "similarity",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"index_params": {
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_kwargs": {
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_params": {
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"collection_name": {
"l2": true,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "昆仑万维年报",
"l2_name": "公司库",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": 1557
},
"connection_args": {
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"consistency_level": {
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"Milvus",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 393,
"dragging": false,
"position": {
"x": 2093.7332965707155,
"y": 167.43148034535486
},
"selected": false,
"positionAbsolute": {
"x": 2093.7332965707155,
"y": 167.43148034535486
}
},
{
"id": "PromptTemplate-wxEKt",
"data": {
"id": "PromptTemplate-wxEKt",
"node": {
"name": "",
"template": {
"_type": "PromptTemplate",
"template": {
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "参考文本：\n{context}\n\n----------------------\n# Role：行业分析师\n\n# Background :\n- 你是一名行业分析师。你的日常工作是根据【参考文本】的内容总结问题。\n\n# Goals:\n- 以上【参考文本】中是从企业知识库中查找到的相关信息，\n- 尽可能体现你的专业性，将答案尽可能详细阐述。\n- 用中文回答问题，并且答案要严谨、专业、清晰、可读性好。\n- 拥有排版审美, 会利用序号, 缩进, 分隔线和换行符等等来美化信息排版。\n- 只回复内容，在段落的开头和结尾，不要有“根据文本内容”和“希望对你有帮助”等过渡语句。\n\n\n----------------------\n\n用户问题: {question}\n你的回答:",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"input_types": {
"info": "",
"list": false,
"name": "input_types",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"output_parser": {
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"context",
"question"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"template_format": {
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "Union[Literal['f-string'], Literal['jinja2']]",
"value": "f-string",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"partial_variables": {
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"validate_template": {
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "A prompt template for a language model.",
"base_classes": [
"Generic",
"PromptTemplate",
"StringPromptTemplate",
"Runnable",
"RunnableSerializable",
"BasePromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"context",
"question"
],
"template": [
"context",
"question"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {}
},
"type": "PromptTemplate",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 291,
"dragging": false,
"position": {
"x": 1023.4514262700376,
"y": 11.02496084544926
},
"selected": false,
"positionAbsolute": {
"x": 1023.4514262700376,
"y": 11.02496084544926
}
},
{
"id": "PromptTemplate-ot33y",
"data": {
"id": "PromptTemplate-ot33y",
"node": {
"name": "",
"template": {
"_type": "PromptTemplate",
"template": {
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "参考文本：\n{context}\n\n----------------------\n# Role：行业分析师\n\n# Background :\n- 你是一名行业分析师。你的日常工作是根据【参考文本】的内容总结问题。\n\n# Goals:\n- 以上【参考文本】中是从企业知识库中查找到的相关信息，\n- 尽可能体现你的专业性，将答案尽可能详细阐述。\n- 用中文回答问题，并且答案要严谨、专业、清晰、可读性好。\n- 拥有排版审美, 会利用序号, 缩进, 分隔线和换行符等等来美化信息排版。\n- 只回复内容，在段落的开头和结尾，不要有“根据文本内容”和“希望对你有帮助”等过渡语句。\n\n\n----------------------\n\n用户问题: {question}\n你的回答:",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"input_types": {
"info": "",
"list": false,
"name": "input_types",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"output_parser": {
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"context",
"question"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"template_format": {
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "Union[Literal['f-string'], Literal['jinja2']]",
"value": "f-string",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"partial_variables": {
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"validate_template": {
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "A prompt template for a language model.",
"base_classes": [
"Generic",
"PromptTemplate",
"StringPromptTemplate",
"Runnable",
"RunnableSerializable",
"BasePromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"context",
"question"
],
"template": [
"context",
"question"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {}
},
"type": "PromptTemplate",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 291,
"dragging": false,
"position": {
"x": 858.2666097750891,
"y": 1523.0705323867192
},
"selected": false,
"positionAbsolute": {
"x": 858.2666097750891,
"y": 1523.0705323867192
}
},
{
"id": "VariableNode-d14a6",
"data": {
"id": "VariableNode-d14a6",
"node": {
"template": {
"_type": "VariableNode",
"variables": {
"info": "",
"list": true,
"name": "variables",
"show": true,
"type": "variable",
"value": [
"公司名称"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"variable_value": {
"info": "",
"list": true,
"name": "variable_value",
"show": false,
"type": "str",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "用来设置变量",
"base_classes": [
"VariableNode"
],
"display_name": "VariableNode",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "VariableNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 283,
"position": {
"x": 2566.3869873706526,
"y": 567.5902839403417
},
"positionAbsolute": {
"x": 2566.3869873706526,
"y": 567.5902839403417
}
}
],
"viewport": {
"x": 87.48881258369431,
"y": 160.39352066140611,
"zoom": 0.243342122449621
}
},
"order_num": 7405517,
"guide_word": null
},
{
"update_time": "2024-10-15 18:43:05",
"parameters": null,
"name": "⭐️先检索 QA 后检索文档知识库问答",
"description": "优先检索 QA 知识库，有则直接返回答案，无则继续检索文档知识库，基于 RAG 生成回复",
"flow_id": "c7fa82f69d4e495e87e4e44a6ab44f30",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-10-15 18:40:16",
"id": 194,
"data": {
"edges": [
{
"id": "reactflow__edge-PythonFunction-wBU1iPythonFunction|PythonFunction-wBU1i|function-RuleBasedRouter-apZIqfunction|rule_function|RuleBasedRouter-apZIq",
"style": {
"stroke": "#555"
},
"source": "PythonFunction-wBU1i",
"target": "RuleBasedRouter-apZIq",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "PythonFunction|PythonFunction-wBU1i|function",
"targetHandle": "function|rule_function|RuleBasedRouter-apZIq"
},
{
"id": "reactflow__edge-PythonFunction-ChvDxPythonFunction|PythonFunction-ChvDx|function-TransformChain-jBTgXfunction|transform|TransformChain-jBTgX",
"style": {
"stroke": "#555"
},
"source": "PythonFunction-ChvDx",
"target": "TransformChain-jBTgX",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "PythonFunction|PythonFunction-ChvDx|function",
"targetHandle": "function|transform|TransformChain-jBTgX"
},
{
"id": "reactflow__edge-RetrievalChain-ryEdkRetrievalChain|RetrievalChain-ryEdk|Chain|RetrievalChain|function-SequentialChain-ZcqZ1Chain|chains|SequentialChain-ZcqZ1",
"style": {
"stroke": "#555"
},
"source": "RetrievalChain-ryEdk",
"target": "SequentialChain-ZcqZ1",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "RetrievalChain|RetrievalChain-ryEdk|Chain|RetrievalChain|function",
"targetHandle": "Chain|chains|SequentialChain-ZcqZ1"
},
{
"id": "reactflow__edge-Milvus-8pyknMilvus|Milvus-8pykn|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever-RetrievalChain-ryEdkBaseRetriever|retriever|RetrievalChain-ryEdk",
"style": {
"stroke": "#555"
},
"source": "Milvus-8pykn",
"target": "RetrievalChain-ryEdk",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "Milvus|Milvus-8pykn|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalChain-ryEdk"
},
{
"id": "reactflow__edge-ConversationBufferMemory-i1PuBConversationBufferMemory|ConversationBufferMemory-i1PuB|BaseMemory|ConversationBufferMemory|BaseChatMemory-ConversationalRetrievalChain-aS5HYBaseChatMemory|memory|ConversationalRetrievalChain-aS5HY",
"style": {
"stroke": "#555"
},
"source": "ConversationBufferMemory-i1PuB",
"target": "ConversationalRetrievalChain-aS5HY",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "ConversationBufferMemory|ConversationBufferMemory-i1PuB|BaseMemory|ConversationBufferMemory|BaseChatMemory",
"targetHandle": "BaseChatMemory|memory|ConversationalRetrievalChain-aS5HY"
},
{
"id": "reactflow__edge-Milvus-u5L3KMilvus|Milvus-u5L3K|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever-ConversationalRetrievalChain-aS5HYBaseRetriever|retriever|ConversationalRetrievalChain-aS5HY",
"style": {
"stroke": "#555"
},
"source": "Milvus-u5L3K",
"target": "ConversationalRetrievalChain-aS5HY",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "Milvus|Milvus-u5L3K|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|ConversationalRetrievalChain-aS5HY"
},
{
"id": "reactflow__edge-ConversationalRetrievalChain-aS5HYConversationalRetrievalChain|ConversationalRetrievalChain-aS5HY|BaseConversationalRetrievalChain|ConversationalRetrievalChain|Chain|function-MultiRuleChain-s5i5CChain|default_chain|MultiRuleChain-s5i5C",
"style": {
"stroke": "#555"
},
"source": "ConversationalRetrievalChain-aS5HY",
"target": "MultiRuleChain-s5i5C",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "ConversationalRetrievalChain|ConversationalRetrievalChain-aS5HY|BaseConversationalRetrievalChain|ConversationalRetrievalChain|Chain|function",
"targetHandle": "Chain|default_chain|MultiRuleChain-s5i5C"
},
{
"id": "reactflow__edge-RuleBasedRouter-apZIqRuleBasedRouter|RuleBasedRouter-apZIq|Chain|RuleBasedRouter|RouterChain|function-MultiRuleChain-s5i5CRouterChain|router_chain|MultiRuleChain-s5i5C",
"style": {
"stroke": "#555"
},
"source": "RuleBasedRouter-apZIq",
"target": "MultiRuleChain-s5i5C",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "RuleBasedRouter|RuleBasedRouter-apZIq|Chain|RuleBasedRouter|RouterChain|function",
"targetHandle": "RouterChain|router_chain|MultiRuleChain-s5i5C"
},
{
"id": "reactflow__edge-TransformChain-jBTgXTransformChain|TransformChain-jBTgX|TransformChain|Chain|function-MultiRuleChain-s5i5CChain|LLMChains|MultiRuleChain-s5i5C",
"style": {
"stroke": "#555"
},
"source": "TransformChain-jBTgX",
"target": "MultiRuleChain-s5i5C",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "TransformChain|TransformChain-jBTgX|TransformChain|Chain|function",
"targetHandle": "Chain|LLMChains|MultiRuleChain-s5i5C"
},
{
"id": "reactflow__edge-MultiRuleChain-s5i5CMultiRuleChain|MultiRuleChain-s5i5C|MultiRouteChain|Chain|MultiRuleChain|function-SequentialChain-ZcqZ1Chain|chains|SequentialChain-ZcqZ1",
"style": {
"stroke": "#555"
},
"source": "MultiRuleChain-s5i5C",
"target": "SequentialChain-ZcqZ1",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "MultiRuleChain|MultiRuleChain-s5i5C|MultiRouteChain|Chain|MultiRuleChain|function",
"targetHandle": "Chain|chains|SequentialChain-ZcqZ1"
},
{
"id": "reactflow__edge-BishengLLM-90fb5BishengLLM|BishengLLM-90fb5|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-ConversationalRetrievalChain-aS5HYBaseLanguageModel|llm|ConversationalRetrievalChain-aS5HY",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-90fb5",
"target": "ConversationalRetrievalChain-aS5HY",
"animated": false,
"selected": false,
"className": "stroke-gray-900 ",
"sourceHandle": "BishengLLM|BishengLLM-90fb5|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|ConversationalRetrievalChain-aS5HY"
}
],
"nodes": [
{
"id": "RuleBasedRouter-apZIq",
"data": {
"id": "RuleBasedRouter-apZIq",
"node": {
"template": {
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "RuleBasedRouter",
"memory": {
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"rule_function": {
"info": "",
"list": false,
"name": "rule_function",
"show": true,
"type": "function",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"info": "",
"list": true,
"name": "input_variables",
"show": true,
"type": "str",
"value": [
"result"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "",
"base_classes": [
"Chain",
"RuleBasedRouter",
"RouterChain",
"function"
],
"display_name": "RuleBasedRouter",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "RuleBasedRouter",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 343,
"position": {
"x": 1080.3103197789733,
"y": 1337.4878712865757
},
"selected": false,
"positionAbsolute": {
"x": 1080.3103197789733,
"y": 1337.4878712865757
}
},
{
"id": "PythonFunction-wBU1i",
"data": {
"id": "PythonFunction-wBU1i",
"node": {
"template": {
"code": {
"info": "",
"list": false,
"name": "code",
"show": true,
"type": "code",
"value": "def rule_fuc(inputs: dict) -> dict:\n if not len(inputs['result']):\n return {'destination': None, 'next_inputs': {\"question\": inputs['query']}}\n else:\n return {'destination': 'norm', 'next_inputs': inputs}",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"_type": "PythonFunction"
},
"description": "Python function to be executed.",
"base_classes": [
"function"
],
"display_name": "PythonFunction",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "PythonFunction",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 255,
"position": {
"x": 522.7838419708218,
"y": 1571.2197796319172
},
"selected": false,
"positionAbsolute": {
"x": 522.7838419708218,
"y": 1571.2197796319172
}
},
{
"id": "PythonFunction-ChvDx",
"data": {
"id": "PythonFunction-ChvDx",
"node": {
"template": {
"code": {
"info": "",
"list": false,
"name": "code",
"show": true,
"type": "code",
"value": "def func(inputs: dict) -> dict:\n return {'answer': inputs['result'][0], 'source_documents':''} # 仅取第一个",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"_type": "PythonFunction"
},
"description": "Python function to be executed.",
"base_classes": [
"function"
],
"display_name": "PythonFunction",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "PythonFunction",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 255,
"position": {
"x": 549.6671717852333,
"y": 187.7892888367361
},
"selected": false,
"positionAbsolute": {
"x": 549.6671717852333,
"y": 187.7892888367361
}
},
{
"id": "TransformChain-jBTgX",
"data": {
"id": "TransformChain-jBTgX",
"node": {
"template": {
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "TransformChain",
"memory": {
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"transform": {
"info": "",
"list": false,
"name": "transform",
"show": true,
"type": "function",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"input_variables": {
"info": "",
"list": true,
"name": "input_variables",
"show": true,
"type": "str",
"value": [
"result"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"output_variables": {
"info": "",
"list": true,
"name": "output_variables",
"show": true,
"type": "str",
"value": [
"answer",
"source_documents"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain transform chain output.",
"base_classes": [
"TransformChain",
"Chain",
"function"
],
"display_name": "TransformChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "TransformChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 495,
"position": {
"x": 1069.678074540037,
"y": 153.32667222210148
},
"selected": false,
"positionAbsolute": {
"x": 1069.678074540037,
"y": 153.32667222210148
}
},
{
"id": "SequentialChain-ZcqZ1",
"data": {
"id": "SequentialChain-ZcqZ1",
"node": {
"template": {
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "SequentialChain",
"chains": {
"info": "",
"list": true,
"name": "chains",
"show": true,
"type": "Chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"memory": {
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"return_all": {
"info": "",
"list": false,
"name": "return_all",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"chain_order": {
"info": "",
"list": false,
"name": "chain_order",
"show": true,
"type": "str",
"value": "[\"RetrievalChain-ryEdk\", \"MultiRuleChain-s5i5C\"]",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"info": "",
"list": true,
"name": "input_variables",
"show": true,
"type": "str",
"value": [
"query"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"output_variables": {
"info": "",
"list": true,
"name": "output_variables",
"show": true,
"type": "str",
"value": [
"answer",
"source_documents"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain where the outputs of one chain feed directly into next.",
"base_classes": [
"SequentialChain",
"Chain",
"function"
],
"display_name": "SequentialChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "SequentialChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 597,
"position": {
"x": 2554.0955901640964,
"y": 323.87489548587064
},
"selected": false,
"positionAbsolute": {
"x": 2554.0955901640964,
"y": 323.87489548587064
}
},
{
"id": "RetrievalChain-ryEdk",
"data": {
"id": "RetrievalChain-ryEdk",
"node": {
"template": {
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "RetrievalChain",
"memory": {
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"retriever": {
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": false,
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain to use to combine the documents.",
"base_classes": [
"Chain",
"RetrievalChain",
"function"
],
"display_name": "RetrievalChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "RetrievalChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 281,
"position": {
"x": 2047.2642182152017,
"y": 258.5734077753632
},
"selected": false,
"positionAbsolute": {
"x": 2047.2642182152017,
"y": 258.5734077753632
}
},
{
"id": "Milvus-u5L3K",
"data": {
"id": "Milvus-u5L3K",
"node": {
"template": {
"_type": "Milvus",
"drop_old": {
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"documents": {
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"info": "",
"list": false,
"name": "embedding",
"show": false,
"type": "Embeddings",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_type": {
"info": "",
"list": false,
"name": "search_type",
"type": "str",
"value": "similarity",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"index_params": {
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_kwargs": {
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_params": {
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"collection_name": {
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "BISHENG产品文档20240529版本",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": 6642
},
"connection_args": {
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"consistency_level": {
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"Milvus",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 391,
"position": {
"x": 520.0097972326654,
"y": 1125.7035041561169
},
"selected": false,
"positionAbsolute": {
"x": 520.0097972326654,
"y": 1125.7035041561169
}
},
{
"id": "Milvus-8pykn",
"data": {
"id": "Milvus-8pykn",
"node": {
"template": {
"_type": "Milvus",
"drop_old": {
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"documents": {
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"info": "",
"list": false,
"name": "embedding",
"show": false,
"type": "Embeddings",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_type": {
"info": "",
"list": false,
"name": "search_type",
"show": true,
"type": "str",
"value": "similarity_score_threshold",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"index_params": {
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_kwargs": {
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{\"k\": 1, \"score_threshold\": 0.8}",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"search_params": {
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"collection_name": {
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "毕昇 QA 知识库",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": 6938
},
"connection_args": {
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"consistency_level": {
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"Milvus",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 555,
"position": {
"x": 1570.1640482171665,
"y": 20
},
"selected": false,
"positionAbsolute": {
"x": 1570.1640482171665,
"y": 20
}
},
{
"id": "ConversationalRetrievalChain-aS5HY",
"data": {
"id": "ConversationalRetrievalChain-aS5HY",
"node": {
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"_type": "ConversationalRetrievalChain",
"memory": {
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseChatMemory",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"info": "",
"list": false,
"name": "callbacks",
"show": false,
"type": "Callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"retriever": {
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"chain_type": {
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"document_prompt": {
"info": "",
"list": false,
"name": "document_prompt",
"show": true,
"type": "BasePromptTemplate",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"condense_question_llm": {
"info": "",
"list": false,
"name": "condense_question_llm",
"show": false,
"type": "BaseLanguageModel",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Return source documents"
},
"condense_question_prompt": {
"info": "",
"list": false,
"name": "condense_question_prompt",
"show": true,
"type": "BasePromptTemplate",
"value": {
"_type": "prompt",
"template": "Given the following conversation and a follow up question, rephrase the follow up question to be a standalone question, in its original language.\n\nChat History:\n{chat_history}\nFollow Up Input: {question}\nStandalone question:",
"output_parser": null,
"input_variables": [
"chat_history",
"question"
],
"template_format": "f-string",
"partial_variables": {},
"validate_template": true
},
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"combine_docs_chain_kwargs": {
"info": "",
"list": false,
"name": "combine_docs_chain_kwargs",
"show": true,
"type": "BasePromptTemplate",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
}
},
"description": "Convenience method to load chain from LLM and retriever.",
"base_classes": [
"BaseConversationalRetrievalChain",
"ConversationalRetrievalChain",
"Chain",
"function"
],
"display_name": "ConversationalRetrievalChain",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/chat_vector_db"
},
"type": "ConversationalRetrievalChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 531,
"position": {
"x": 1076.7431790118962,
"y": 720.4023383381091
},
"selected": false,
"positionAbsolute": {
"x": 1076.7431790118962,
"y": 720.4023383381091
}
},
{
"id": "ConversationBufferMemory-i1PuB",
"data": {
"id": "ConversationBufferMemory-i1PuB",
"node": {
"template": {
"_type": "ConversationBufferMemory",
"ai_prefix": {
"info": "",
"list": false,
"name": "ai_prefix",
"show": false,
"type": "str",
"value": "AI",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_key": {
"info": "The variable to be used as Chat Input when more than one variable is available.",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"memory_key": {
"info": "",
"list": false,
"name": "memory_key",
"show": true,
"type": "str",
"value": "chat_history",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"output_key": {
"info": "The variable to be used as Chat Output (e.g. answer in a ConversationalRetrievalChain)",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "answer",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"chat_memory": {
"info": "",
"list": false,
"name": "chat_memory",
"show": true,
"type": "BaseChatMessageHistory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"human_prefix": {
"info": "",
"list": false,
"name": "human_prefix",
"show": false,
"type": "str",
"value": "Human",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"return_messages": {
"info": "",
"list": false,
"name": "return_messages",
"show": true,
"type": "bool",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Buffer for storing conversation memory.",
"base_classes": [
"BaseMemory",
"ConversationBufferMemory",
"BaseChatMemory"
],
"display_name": "ConversationBufferMemory",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/memory/how_to/buffer"
},
"type": "ConversationBufferMemory",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 527,
"position": {
"x": 519.6689261937935,
"y": 539.0200298494985
},
"selected": false,
"positionAbsolute": {
"x": 519.6689261937935,
"y": 539.0200298494985
}
},
{
"id": "MultiRuleChain-s5i5C",
"data": {
"id": "MultiRuleChain-s5i5C",
"node": {
"template": {
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "MultiRuleChain",
"memory": {
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"LLMChains": {
"info": "",
"list": true,
"name": "LLMChains",
"show": true,
"type": "Chain",
"value": "[]",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"router_chain": {
"info": "",
"list": false,
"name": "router_chain",
"show": true,
"type": "RouterChain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"default_chain": {
"info": "",
"list": false,
"name": "default_chain",
"show": true,
"type": "Chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"silent_errors": {
"info": "",
"list": false,
"name": "silent_errors",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"output_variables": {
"info": "",
"list": true,
"name": "output_variables",
"show": true,
"type": "str",
"value": [
"source_documents",
"answer"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"destination_chains": {
"info": "",
"list": false,
"name": "destination_chains",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"destination_chain_name": {
"info": "{chain_id: name}",
"list": true,
"name": "destination_chain_name",
"show": true,
"type": "NestedDict",
"value": "{\"TransformChain-jBTgX\": \"norm\"}",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
}
},
"description": "",
"base_classes": [
"MultiRouteChain",
"Chain",
"MultiRuleChain",
"function"
],
"display_name": "MultiRuleChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "MultiRuleChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 549,
"position": {
"x": 1787.231955024217,
"y": 753.8514799486979
},
"selected": false,
"positionAbsolute": {
"x": 1787.231955024217,
"y": 753.8514799486979
}
},
{
"id": "BishengLLM-90fb5",
"data": {
"id": "BishengLLM-90fb5",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 4,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "ChatOpenAI",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "Use the llm model that has been launched in model management",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 427,
"position": {
"x": 18.000000000000004,
"y": 309.62500515920107
},
"selected": false,
"positionAbsolute": {
"x": 18.000000000000004,
"y": 309.62500515920107
}
}
],
"viewport": {
"x": 194.6566973414132,
"y": 31.460215170984327,
"zoom": 0.4474437869053313
}
},
"order_num": null,
"guide_word": null
},
{
"update_time": "2024-10-17 16:44:57",
"parameters": null,
"name": "⭐️ 查询数据库并画图",
"description": "大模型根据用户请求可查询SQL数据库并画图（若用户没有要求画图则不会画图），画图是通过调用代码运行环境绘制图表并最终保存为图片存储到Minio，然后以Markdown格式返回图片地址在前端展示。\n\n我们提供了模板技能，运行该应用对模型能力有一定要求，测试Qwen2.5-72B可以运行成功，但会有一些不稳定，可以多尝试换几种问法。",
"flow_id": "387757240d3e4f3a9b5cea781bdb4d48",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-10-17 16:44:57",
"id": 196,
"data": {
"edges": [
{
"id": "reactflow__edge-BishengLLM-bc7dcBishengLLM|BishengLLM-bc7dc|BaseChatModel|BaseLanguageModel|BishengLLM|BaseLLM-SQLDatabaseChain-01203BaseLanguageModel|llm|SQLDatabaseChain-01203",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-bc7dc",
"target": "SQLDatabaseChain-01203",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "BishengLLM|BishengLLM-bc7dc|BaseChatModel|BaseLanguageModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|SQLDatabaseChain-01203"
},
{
"id": "reactflow__edge-SQLDatabase-14a1bSQLDatabase|SQLDatabase-14a1b|SQLDatabase|function-SQLDatabaseChain-01203SQLDatabase|db|SQLDatabaseChain-01203",
"style": {
"stroke": "#555"
},
"source": "SQLDatabase-14a1b",
"target": "SQLDatabaseChain-01203",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "SQLDatabase|SQLDatabase-14a1b|SQLDatabase|function",
"targetHandle": "SQLDatabase|db|SQLDatabaseChain-01203"
},
{
"id": "reactflow__edge-PromptTemplate-06f11PromptTemplate|PromptTemplate-06f11|BasePromptTemplate|StringPromptTemplate|PromptTemplate-SQLDatabaseChain-01203BasePromptTemplate|prompt|SQLDatabaseChain-01203",
"style": {
"stroke": "#555"
},
"source": "PromptTemplate-06f11",
"target": "SQLDatabaseChain-01203",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "PromptTemplate|PromptTemplate-06f11|BasePromptTemplate|StringPromptTemplate|PromptTemplate",
"targetHandle": "BasePromptTemplate|prompt|SQLDatabaseChain-01203"
},
{
"id": "reactflow__edge-BishengLLM-bc7dcBishengLLM|BishengLLM-bc7dc|BaseChatModel|BaseLanguageModel|BishengLLM|BaseLLM-ZeroShotAgent-50e40BaseLanguageModel|llm|ZeroShotAgent-50e40",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-bc7dc",
"target": "ZeroShotAgent-50e40",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "BishengLLM|BishengLLM-bc7dc|BaseChatModel|BaseLanguageModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|ZeroShotAgent-50e40"
},
{
"id": "reactflow__edge-SQLDatabaseChain-01203SQLDatabaseChain|SQLDatabaseChain-01203|SQLDatabaseChain|Chain|function-Tool-f4cb0function|func|Tool-f4cb0",
"style": {
"stroke": "#555"
},
"source": "SQLDatabaseChain-01203",
"target": "Tool-f4cb0",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "SQLDatabaseChain|SQLDatabaseChain-01203|SQLDatabaseChain|Chain|function",
"targetHandle": "function|func|Tool-f4cb0"
},
{
"id": "reactflow__edge-Tool-f4cb0Tool|Tool-f4cb0|Tool|BaseTool-ZeroShotAgent-50e40BaseTool|tools|ZeroShotAgent-50e40",
"style": {
"stroke": "#555"
},
"source": "Tool-f4cb0",
"target": "ZeroShotAgent-50e40",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "Tool|Tool-f4cb0|Tool|BaseTool",
"targetHandle": "BaseTool|tools|ZeroShotAgent-50e40"
},
{
"id": "reactflow__edge-Data-2e84bData|Data-2e84b|BaseTool|Tool-ZeroShotAgent-50e40BaseTool|tools|ZeroShotAgent-50e40",
"style": {
"stroke": "#555"
},
"source": "Data-2e84b",
"target": "ZeroShotAgent-50e40",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "Data|Data-2e84b|BaseTool|Tool",
"targetHandle": "BaseTool|tools|ZeroShotAgent-50e40"
}
],
"nodes": [
{
"id": "SQLDatabaseChain-01203",
"data": {
"id": "SQLDatabaseChain-01203",
"node": {
"beta": false,
"template": {
"db": {
"info": "",
"list": false,
"name": "db",
"show": true,
"type": "SQLDatabase",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Db",
"model_config": {}
},
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"_type": "SQLDatabaseChain",
"prompt": {
"info": "",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Prompt",
"model_config": {}
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Preset Question",
"model_config": {}
}
},
"description": "Create a SQLDatabaseChain from an LLM and a database connection.",
"base_classes": [
"SQLDatabaseChain",
"Chain",
"function"
],
"display_name": "SQLDatabaseChain",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "SQLDatabaseChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 337,
"dragging": false,
"position": {
"x": 676.0138039780265,
"y": 556.0498254993254
},
"selected": false,
"positionAbsolute": {
"x": 676.0138039780265,
"y": 556.0498254993254
}
},
{
"id": "BishengLLM-bc7dc",
"data": {
"id": "BishengLLM-bc7dc",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": false,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 27,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "ChatOpenAI",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "Use the llm model that has been launched in model management",
"base_classes": [
"BaseChatModel",
"BaseLanguageModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 427,
"position": {
"x": 18,
"y": 20
},
"selected": false,
"positionAbsolute": {
"x": 18,
"y": 20
}
},
{
"id": "SQLDatabase-14a1b",
"data": {
"id": "SQLDatabase-14a1b",
"node": {
"beta": false,
"template": {
"_type": "SQLDatabase",
"engine_args": {
"info": "",
"list": false,
"name": "engine_args",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Engine Args",
"model_config": {}
},
"database_uri": {
"info": "",
"list": false,
"name": "database_uri",
"show": true,
"type": "str",
"value": "mysql+pymysql://root:E1SkG0PaDMEPTAxY@192.168.106.116:3306/company_stock",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Database Uri",
"model_config": {}
}
},
"description": "Construct a SQLAlchemy engine from URI.",
"base_classes": [
"SQLDatabase",
"function"
],
"display_name": "SQLDatabase",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "SQLDatabase",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 255,
"position": {
"x": 18.61695520886218,
"y": 577.4905684719494
},
"selected": false,
"positionAbsolute": {
"x": 18.61695520886218,
"y": 577.4905684719494
}
},
{
"id": "PromptTemplate-06f11",
"data": {
"id": "PromptTemplate-06f11",
"node": {
"beta": false,
"icon": null,
"name": "",
"error": null,
"template": {
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "PromptTemplate",
"input": {
"info": "",
"list": false,
"name": "input",
"show": true,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": true,
"title_case": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "input",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"template": {
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "你是一个与 SQL 数据库交互的智能代理。\n\n## 规则\n1. 给定一个输入问题，输出一个语法正确的 mysql 查询来运行。\n2. 除非用户指定了希望获得的示例的具体数量，否则始终将查询限制为最多5个结果。\n3. 切勿查询特定表中的所有列，只查询问题中的相关列。\n4. 一定要严格根据查询结果来回答，如果查询结果为空，表示未找到相关记录。\n5. 请用中文回答用户的问题。\n6. 当前系统时间是2024年x月xx日。\n7. 如果SQL查询为空，不要输出错误，则回答为空。\n8. 如果出现SQL相关报错，请调整后重试。\n9. 文件链接前不需要添加额外的https或者http前缀\n\n## 供参考的信息\n`stock_price` 为股价表。字段及含义如下：\nid: 主键，自动递增。\nsotck_no: 股票代码。\nname: 股票名称。\ncomment: 千股千评，可能是指对该股票的评论或分析。\nnew_price: 最新价。\nchange_amount: 涨跌额。\namplitude: 涨跌幅。\nprev_close_price: 昨日收盘价。\nopen_price: 今日开盘价。\nhighest: 最高价。\nlowest: 最低价。\ntrade_volume: 成交量（万股）。\ntrade_money: 成交额（万元）。\n\n#####\n用户问题：\n{input}\n\n问题对应sql（注意开头不要包含”sql“标识符，直接输入sql语句即可）：",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": true,
"title_case": true,
"placeholder": "",
"display_name": "Template",
"model_config": {}
},
"input_types": {
"info": "",
"list": false,
"name": "input_types",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Input Types",
"model_config": {}
},
"output_parser": {
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Output Parser",
"model_config": {}
},
"input_variables": {
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"input"
],
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Input Variables",
"model_config": {}
},
"template_format": {
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "Literal['f-string', 'mustache', 'jinja2']",
"value": "f-string",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Template Format",
"model_config": {}
},
"partial_variables": {
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Partial Variables",
"model_config": {}
},
"validate_template": {
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": false,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Validate Template",
"model_config": {}
}
},
"full_path": null,
"description": "Prompt template for a language model.",
"base_classes": [
"BasePromptTemplate",
"StringPromptTemplate",
"PromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"input"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {}
},
"type": "PromptTemplate",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 373,
"position": {
"x": 25.822320988711738,
"y": 874.460009094774
},
"selected": false,
"positionAbsolute": {
"x": 25.822320988711738,
"y": 874.460009094774
}
},
{
"id": "ZeroShotAgent-50e40",
"data": {
"id": "ZeroShotAgent-50e40",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"_type": "ZeroShotAgent",
"tools": {
"info": "",
"list": true,
"name": "tools",
"show": true,
"type": "BaseTool",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tools",
"model_config": {}
},
"prefix": {
"info": "",
"list": false,
"name": "prefix",
"show": true,
"type": "str",
"value": "Answer the following questions as best you can. \n一般来讲需要先根据用户的问题使用”数据库查询“tool来查询数据，然后使用”Python REPL Tool“tool来绘制图表，”Python REPL Tool“tool绘制完成的图片会存储到对象存储并返回存储的链接，最终请用Markdown的语法来输出链接\nYou have access to the following tools:",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": true,
"title_case": true,
"placeholder": "",
"display_name": "Prefix",
"model_config": {}
},
"suffix": {
"info": "",
"list": false,
"name": "suffix",
"show": true,
"type": "str",
"value": "Begin!\n\nQuestion: {input}\nThought:{agent_scratchpad}",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": true,
"title_case": true,
"placeholder": "",
"display_name": "Suffix",
"model_config": {}
},
"input_node": {
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Preset Question",
"model_config": {}
},
"output_parser": {
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "AgentOutputParser",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Output Parser",
"model_config": {}
},
"input_variables": {
"info": "",
"list": true,
"name": "input_variables",
"show": true,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Input Variables",
"model_config": {}
},
"callback_manager": {
"info": "",
"list": false,
"name": "callback_manager",
"show": false,
"type": "BaseCallbackManager",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callback Manager",
"model_config": {}
},
"format_instructions": {
"info": "",
"list": false,
"name": "format_instructions",
"show": true,
"type": "str",
"value": "Use the following format:\n\nQuestion: the input question you must answer\nThought: you should always think about what to do\nAction: the action to take, should be one of [{tool_names}]\nAction Input: the input to the action\nObservation: the result of the action\n... (this Thought/Action/Action Input/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": true,
"title_case": true,
"placeholder": "",
"display_name": "Format Instructions",
"model_config": {}
}
},
"description": "Construct an agent from an LLM and tools.",
"base_classes": [
"BaseSingleActionAgent",
"Agent",
"ZeroShotAgent",
"function"
],
"display_name": "ZeroShotAgent",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/agents/how_to/custom_mrkl_agent",
"field_formatters": {}
},
"type": "ZeroShotAgent",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 609,
"position": {
"x": 1881.5603475961273,
"y": 749.2188069126355
},
"selected": false,
"positionAbsolute": {
"x": 1881.5603475961273,
"y": 749.2188069126355
}
},
{
"id": "Tool-f4cb0",
"data": {
"id": "Tool-f4cb0",
"node": {
"beta": false,
"template": {
"func": {
"info": "",
"list": false,
"name": "func",
"show": true,
"type": "function",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": true,
"title_case": true,
"placeholder": "",
"display_name": "Func",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": true,
"type": "str",
"value": "数据库查询",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": true,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"_type": "Tool",
"args_schema": {
"info": "",
"list": false,
"name": "args_schema",
"show": true,
"type": "NestedDict",
"value": "{\"arg1\": {\"type\": \"string\"}}",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": true,
"title_case": true,
"placeholder": "",
"display_name": "Args Schema",
"model_config": {}
},
"description": {
"info": "",
"list": false,
"name": "description",
"show": true,
"type": "str",
"value": "用户问题先从数据库中查询结果",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": true,
"title_case": true,
"placeholder": "",
"display_name": "Description",
"model_config": {}
},
"return_direct": {
"info": "",
"list": false,
"name": "return_direct",
"show": true,
"type": "bool",
"value": false,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Return Direct",
"model_config": {}
}
},
"description": "Converts a chain, agent or function into a tool.",
"base_classes": [
"Tool",
"BaseTool"
],
"display_name": "Tool",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "Tool",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 529,
"dragging": false,
"position": {
"x": 1207.7539633864767,
"y": 636.1580608046685
},
"selected": false,
"positionAbsolute": {
"x": 1207.7539633864767,
"y": 636.1580608046685
}
},
{
"id": "Data-2e84b",
"data": {
"id": "Data-2e84b",
"node": {
"beta": true,
"template": {
"code": {
"info": "",
"list": false,
"name": "code",
"show": true,
"type": "code",
"value": "import importlib\r\nimport time\r\nimport re\r\nfrom typing import Any\r\n\r\nfrom langchain_experimental.utilities.python import PythonREPL\r\nfrom langchain.callbacks.manager import (\r\n AsyncCallbackManagerForToolRun,\r\n CallbackManagerForToolRun,\r\n)\r\nfrom bisheng import CustomComponent\r\nfrom bisheng.utils.minio_client import MinioClient\r\nfrom langchain_core.tools import Tool\r\n\r\n\r\nclass PythonREPLToolComponent(CustomComponent):\r\n display_name = \"Python REPL Tool\"\r\n description = \"A Python shell. Use this to execute python commands. Input should be a valid python command. If you want to see the output of a value, you should print it out with `print(...)`.\"\r\n name = \"PythonREPLTool\"\r\n\r\n def build_config(self):\r\n return {\r\n \"name\": {\"display_name\": \"Name\", \"info\": \"The name of the tool.\"},\r\n \"description\": {\"display_name\": \"Description\", \"info\": \"A description of the tool.\"},\r\n \"global_imports\": {\r\n \"display_name\": \"Global Imports\",\r\n \"info\": \"A list of modules to import globally, e.g. ['math', 'numpy'].\",\r\n },\r\n }\r\n\r\n def get_globals(self, globals: list[str]) -> dict:\r\n \"\"\"\r\n Retrieves the global variables from the specified modules.\r\n\r\n Args:\r\n globals (list[str]): A list of module names.\r\n\r\n Returns:\r\n dict: A dictionary containing the global variables from the specified modules.\r\n \"\"\"\r\n global_dict = {}\r\n for module in globals:\r\n try:\r\n imported_module = importlib.import_module(module)\r\n global_dict[imported_module.__name__] = imported_module\r\n except ImportError:\r\n raise ImportError(f\"Could not import module {module}\")\r\n return global_dict\r\n\r\n def build(\r\n self,\r\n name: str = \"python_repl\",\r\n description: str = \"A Python shell. Use this to execute python commands. Input should be a valid python command. If you want to see the output of a value, you should print it out with `print(...)`.\",\r\n global_imports: list[str] = [\"math\"],\r\n ) -> Tool:\r\n \"\"\"\r\n Builds a Python REPL tool.\r\n\r\n Args:\r\n name (str, optional): The name of the tool. Defaults to \"python_repl\".\r\n description (str, optional): The description of the tool. Defaults to \"A Python shell. Use this to execute python commands. Input should be a valid python command. If you want to see the output of a value, you should print it out with `print(...)`. \".\r\n global_imports (list[str], optional): A list of global imports to be available in the Python REPL. Defaults to [\"math\"].\r\n\r\n Returns:\r\n Tool: The built Python REPL tool.\r\n \"\"\"\r\n _globals = self.get_globals(global_imports)\r\n python_repl = PythonREPL(_globals=_globals)\r\n \r\n class PersonalTool(Tool):\r\n def _run(\r\n self,\r\n query: str,\r\n run_manager: Optional[CallbackManagerForToolRun] = None,\r\n ) -> Any:\r\n \"\"\"Use the tool.\"\"\"\r\n print(\"-----start---\")\r\n \r\n minio_client = MinioClient()\r\n query = re.compile('plt.savefig(.*)').sub(\"\", query)\r\n \r\n # 把图片存到本地\r\n file_path = f\"/tmp/{int(time.time())}.png\"\r\n query = query.replace(\"plt.show()\", f\"plt.savefig('{file_path}')\")\r\n query = 'import matplotlib\\nmatplotlib.rc(\"font\", family=\"WenQuanYi Zen Hei\")\\n' + query\r\n print(f\"-----exec code: {query}\")\r\n \r\n result = super(PersonalTool, self)._run(query, run_manager=run_manager)\r\n print(\"-----code exec over-----\")\r\n # 上传文件到minio\r\n \r\n minio_client.upload_minio(file_path, file_path)\r\n file_url = minio_client.get_share_link(file_path)\r\n file_url = minio_client.clear_minio_share_host(file_url)\r\n print(f\"-----over---\", file_url)\r\n return file_url\r\n \r\n async def _arun(\r\n self,\r\n query: str,\r\n run_manager: Optional[AsyncCallbackManagerForToolRun] = None,\r\n ) -> Any:\r\n \"\"\"Use the tool asynchronously.\"\"\"\r\n print(\"-----astart---\")\r\n result = await super(PersonalTool, self)._arun(query, run_manager=run_manager)\r\n print(\"---aover---\")\r\n return result\r\n \r\n tool = PersonalTool(\r\n name=name,\r\n description=description,\r\n func=python_repl.run,\r\n )\r\n return tool\r\n\r\n",
"dynamic": true,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": true,
"title_case": true,
"placeholder": "",
"display_name": "Code",
"model_config": {}
},
"name": {
"info": "The name of the tool.",
"list": false,
"name": "name",
"show": true,
"type": "str",
"value": "python_repl",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"_type": "CustomComponent",
"description": {
"info": "A description of the tool.",
"list": false,
"name": "description",
"show": true,
"type": "str",
"value": "A Python shell. Use this to execute python commands. Input should be a valid python command. If you want to see the output of a value, you should print it out with `print(...)`.",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Description",
"model_config": {}
},
"global_imports": {
"info": "A list of modules to import globally, e.g. ['math', 'numpy'].",
"list": true,
"name": "global_imports",
"show": true,
"type": "str",
"value": [
"math"
],
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Global Imports",
"model_config": {}
}
},
"description": "A Python shell. Use this to execute python commands. Input should be a valid python command. If you want to see the output of a value, you should print it out with `print(...)`.",
"base_classes": [
"BaseTool",
"Tool"
],
"display_name": "Python REPL Tool",
"output_types": [
"Tool"
],
"custom_fields": {
"name": null,
"description": null,
"global_imports": null
},
"documentation": "",
"field_formatters": {}
},
"type": "Data",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 561,
"dragging": false,
"position": {
"x": 778.9175511051337,
"y": 1149.9902599062198
},
"selected": true,
"positionAbsolute": {
"x": 778.9175511051337,
"y": 1149.9902599062198
}
}
],
"viewport": {
"x": 339.1544304743195,
"y": 29.21939842035414,
"zoom": 0.45266644261865624
}
},
"order_num": 7471052,
"guide_word": null
},
{
"update_time": "2024-10-28 15:51:35",
"parameters": null,
"name": "⭐️ 指定Keyvalue信息提取",
"description": "上传文件，指定需要提取的Keyvalue，分号分隔，点击【开始】按钮",
"flow_id": "caae4fa4c0b6405ba3ebfda20895849e",
"api_parameters": null,
"flow_type": 1,
"create_time": "2024-10-28 15:51:35",
"id": 197,
"data": {
"edges": [
{
"id": "reactflow__edge-PromptTemplate-5UHOIPromptTemplate|PromptTemplate-5UHOI|StringPromptTemplate|PromptTemplate|BasePromptTemplate-LLMChain-aSC8HBasePromptTemplate|prompt|LLMChain-aSC8H",
"style": {
"stroke": "#555"
},
"source": "PromptTemplate-5UHOI",
"target": "LLMChain-aSC8H",
"animated": false,
"selected": false,
"className": "",
"sourceHandle": "PromptTemplate|PromptTemplate-5UHOI|StringPromptTemplate|PromptTemplate|BasePromptTemplate",
"targetHandle": "BasePromptTemplate|prompt|LLMChain-aSC8H"
},
{
"id": "reactflow__edge-BishengLLM-2763fBishengLLM|BishengLLM-2763f|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM-LLMChain-aSC8HBaseLanguageModel|llm|LLMChain-aSC8H",
"style": {
"stroke": "#555"
},
"source": "BishengLLM-2763f",
"target": "LLMChain-aSC8H",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "BishengLLM|BishengLLM-2763f|BaseLanguageModel|BaseChatModel|BishengLLM|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|LLMChain-aSC8H"
},
{
"id": "reactflow__edge-InputNode-3ccbfInputNode|InputNode-3ccbf|input-LLMChain-aSC8Hinput|input_node|LLMChain-aSC8H",
"style": {
"stroke": "#555"
},
"source": "InputNode-3ccbf",
"target": "LLMChain-aSC8H",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "InputNode|InputNode-3ccbf|input",
"targetHandle": "input|input_node|LLMChain-aSC8H"
},
{
"id": "reactflow__edge-InputFileNode-fc806InputFileNode|InputFileNode-fc806|fileNode-ElemUnstructuredLoaderV0-293c3fileNode|file_path|ElemUnstructuredLoaderV0-293c3",
"style": {
"stroke": "#555"
},
"source": "InputFileNode-fc806",
"target": "ElemUnstructuredLoaderV0-293c3",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "InputFileNode|InputFileNode-fc806|fileNode",
"targetHandle": "fileNode|file_path|ElemUnstructuredLoaderV0-293c3"
},
{
"id": "reactflow__edge-ElemUnstructuredLoaderV0-293c3ElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-293c3|Document-PromptTemplate-5UHOIDocument;BaseOutputParser;VariableNode|context|PromptTemplate-5UHOI",
"style": {
"stroke": "#555"
},
"source": "ElemUnstructuredLoaderV0-293c3",
"target": "PromptTemplate-5UHOI",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "ElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-293c3|Document",
"targetHandle": "Document;BaseOutputParser;VariableNode|context|PromptTemplate-5UHOI"
},
{
"id": "reactflow__edge-VariableNode-e314cVariableNode|VariableNode-e314c|VariableNode-PromptTemplate-5UHOIDocument;BaseOutputParser;VariableNode|keyvalue|PromptTemplate-5UHOI",
"style": {
"stroke": "#555"
},
"source": "VariableNode-e314c",
"target": "PromptTemplate-5UHOI",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "VariableNode|VariableNode-e314c|VariableNode",
"targetHandle": "Document;BaseOutputParser;VariableNode|keyvalue|PromptTemplate-5UHOI"
}
],
"nodes": [
{
"id": "LLMChain-aSC8H",
"data": {
"id": "LLMChain-aSC8H",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"_type": "LLMChain",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"prompt": {
"l2": false,
"info": "",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"l2_name": "prompt",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"llm_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "llm_kwargs",
"show": false,
"type": "code",
"l2_name": "llm_kwargs",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "text",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "BaseLLMOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"return_final_only": {
"l2": false,
"info": "",
"list": false,
"name": "return_final_only",
"show": false,
"type": "bool",
"value": true,
"l2_name": "return_final_only",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Chain to run queries against LLMs.",
"base_classes": [
"Chain",
"LLMChain",
"function"
],
"display_name": "LLMChain",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/foundational/llm_chain"
},
"type": "LLMChain",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 317,
"position": {
"x": 1313.6980454632887,
"y": 900.921715712506
},
"selected": false,
"positionAbsolute": {
"x": 1313.6980454632887,
"y": 900.921715712506
}
},
{
"id": "PromptTemplate-5UHOI",
"data": {
"id": "PromptTemplate-5UHOI",
"node": {
"beta": false,
"icon": null,
"name": "",
"error": null,
"template": {
"_type": "PromptTemplate",
"input": {
"info": "",
"list": false,
"name": "input",
"show": true,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": true,
"title_case": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "input",
"model_config": {}
},
"context": {
"info": "",
"list": false,
"name": "context",
"show": true,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": true,
"title_case": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "context",
"model_config": {}
},
"keyvalue": {
"info": "",
"list": false,
"name": "keyvalue",
"show": true,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": true,
"title_case": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "keyvalue",
"model_config": {}
},
"template": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "提取以下文本内容中的keyvalue信息，需要抽取的key包括：\n\n{keyvalue}\n\nThe output should be a Markdown code snippet formatted in the following schema, including the leading and trailing \"```json\" and \"```\":\n\n```json\n\"Key1\": Value\n\"Key2\": Value\n```\n\n文本内容\n{context}\n\n{input}",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": ""
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"keyvalue",
"context",
"input"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"template_format": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"partial_variables": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"validate_template": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"full_path": null,
"description": "Schema to represent a prompt for an LLM.",
"base_classes": [
"StringPromptTemplate",
"PromptTemplate",
"BasePromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"keyvalue",
"context",
"input"
],
"template": [
"context",
"input"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {}
},
"type": "PromptTemplate",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 501,
"dragging": false,
"position": {
"x": 809.51004530817,
"y": 1237.7696518068285
},
"selected": false,
"positionAbsolute": {
"x": 809.51004530817,
"y": 1237.7696518068285
}
},
{
"id": "BishengLLM-2763f",
"data": {
"id": "BishengLLM-2763f",
"node": {
"beta": false,
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": false,
"type": "BaseChatModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "BishengLLM",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"top_p": {
"info": "",
"list": false,
"name": "top_p",
"show": true,
"type": "float",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Top P",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"model_id": {
"info": "",
"list": false,
"name": "model_id",
"show": true,
"type": "bisheng_model",
"value": 31,
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": true,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"model_info": {
"info": "",
"list": false,
"name": "model_info",
"show": false,
"type": "LLMModel",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Info",
"model_config": {}
},
"model_name": {
"info": "",
"list": false,
"name": "model_name",
"show": false,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"server_info": {
"info": "",
"list": false,
"name": "server_info",
"show": false,
"type": "LLMServer",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Server Info",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.3,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"llm_node_type": {
"info": "",
"list": false,
"name": "llm_node_type",
"show": false,
"type": "code",
"value": {
"qwen": "ChatOpenAI",
"vllm": "ChatOpenAI",
"spark": "ChatOpenAI",
"zhipu": "ChatOpenAI",
"ollama": "ChatOllama",
"openai": "ChatOpenAI",
"minimax": "MiniMaxChat",
"qianfan": "ChatWenxin",
"deepseek": "ChatOpenAI",
"llamacpp": "ChatOpenAI",
"anthropic": "ChatAnthropic",
"bisheng_rt": "HostChatGLM",
"xinference": "ChatOpenAI",
"azure_openai": "AzureChatOpenAI"
},
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Llm Node Type",
"model_config": {}
},
"custom_get_token_ids": {
"info": "",
"list": true,
"name": "custom_get_token_ids",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Custom Get Token Ids",
"model_config": {}
}
},
"description": "依赖bisheng后端服务的llm组件",
"base_classes": [
"BaseLanguageModel",
"BaseChatModel",
"BishengLLM",
"BaseLLM"
],
"display_name": "BishengLLM",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "BishengLLM",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 408,
"position": {
"x": -68.63553056857684,
"y": 696.600000304993
},
"positionAbsolute": {
"x": -68.63553056857684,
"y": 696.600000304993
}
},
{
"id": "ElemUnstructuredLoaderV0-293c3",
"data": {
"id": "ElemUnstructuredLoaderV0-293c3",
"node": {
"beta": false,
"template": {
"_type": "ElemUnstructuredLoaderV0",
"kwargs": {
"info": "",
"list": false,
"name": "kwargs",
"show": true,
"type": "dict",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "kwargs",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"value": "{}",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "fileNode",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"suffixes": [
".html",
".md",
".txt",
".bmp",
".jpg",
".png",
".jpeg",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
".xls",
".xlsx",
".tsv",
".csv",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"bmp",
"jpg",
"png",
"jpeg",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
".xls",
"xlsx",
"tsv",
"csv",
"tiff"
],
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "File Path",
"model_config": {}
},
"unstructured_api_url": {
"info": "",
"list": false,
"name": "unstructured_api_url",
"show": true,
"type": "str",
"value": "https://bisheng.dataelem.com/api/v1/etl4llm/predict",
"dynamic": false,
"advanced": false,
"password": true,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "unstructured_api_url",
"model_config": {}
}
},
"description": "The appropriate parser is automatically selected based on the file format and OCR is supported",
"base_classes": [
"Document"
],
"display_name": "ElemUnstructuredLoaderV0",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "ElemUnstructuredLoaderV0",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 477,
"position": {
"x": 237.30334166517488,
"y": 1166.7647849898744
},
"positionAbsolute": {
"x": 237.30334166517488,
"y": 1166.7647849898744
}
},
{
"id": "InputNode-3ccbf",
"data": {
"id": "InputNode-3ccbf",
"node": {
"beta": false,
"template": {
"_type": "input",
"input": {
"info": "",
"list": true,
"name": "input",
"show": true,
"type": "str",
"value": [
"开始"
],
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": true,
"title_case": true,
"placeholder": "",
"display_name": "输入内容",
"model_config": {}
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"input"
],
"display_name": "InputNode",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "InputNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 255,
"position": {
"x": 809.0064819287146,
"y": 663.3484087022575
},
"positionAbsolute": {
"x": 809.0064819287146,
"y": 663.3484087022575
}
},
{
"id": "InputFileNode-fc806",
"data": {
"id": "InputFileNode-fc806",
"node": {
"beta": false,
"template": {
"_type": "InputFileNode",
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "file",
"value": "",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".bmp",
".jpg",
".png",
".jpeg",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
"..xls",
".xlsx",
".tsv",
".csv",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"bmp",
"jpg",
"png",
"jpeg",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
".xls",
"xlsx",
"tsv",
"csv",
"tiff"
],
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "File Path",
"model_config": {}
},
"file_type": {
"info": "Tips for which file should upload",
"list": false,
"name": "file_type",
"show": true,
"type": "str",
"value": "请上传文件",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "提示上传文件类型",
"display_name": "Name",
"model_config": {}
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"fileNode"
],
"display_name": "InputFileNode",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "InputFileNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 337,
"dragging": false,
"position": {
"x": -283.58174168605024,
"y": 1208.0544562311302
},
"selected": false,
"positionAbsolute": {
"x": -283.58174168605024,
"y": 1208.0544562311302
}
},
{
"id": "VariableNode-e314c",
"data": {
"id": "VariableNode-e314c",
"node": {
"beta": false,
"template": {
"_type": "VariableNode",
"variables": {
"info": "",
"list": true,
"name": "variables",
"show": true,
"type": "variable",
"value": [
"要素"
],
"dynamic": false,
"advanced": false,
"password": false,
"required": true,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Variables",
"model_config": {}
},
"variable_value": {
"info": "",
"list": true,
"name": "variable_value",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Variable Value",
"model_config": {}
}
},
"description": "用来设置变量",
"base_classes": [
"VariableNode"
],
"display_name": "VariableNode",
"output_types": [],
"custom_fields": {},
"documentation": "",
"field_formatters": {}
},
"type": "VariableNode",
"value": null
},
"type": "genericNode",
"width": 384,
"height": 283,
"position": {
"x": 191.2097358237865,
"y": 1669.692405161968
},
"positionAbsolute": {
"x": 191.2097358237865,
"y": 1669.692405161968
}
}
],
"viewport": {
"x": 186.80586866476656,
"y": -288.498161293062,
"zoom": 0.4797259326255045
}
},
"order_num": 7536587,
"guide_word": null
},
{
"update_time": "2026-02-05 19:54:04",
"parameters": null,
"name": "文档知识库问答",
"description": "功能简介：这是文档知识库问答流程，通过 “开始” 引导用户提问，“输入” 模块支持对话或表单录入文档内容，“文档知识库问答” 模块利用配置参数，基于文档知识解答用户问题 。\n使用说明：在 “开始” 填引导语与示例问题；“输入” 选录入方式，设置文档相关参数；“文档知识库问答” 配置模型、问题等参数，保存后点击 “运行” 测试，确认无误可 “上线” 使用 。",
"flow_id": "0da365f9709842e29a5b1476d94ba8d9",
"api_parameters": null,
"create_time": "2026-02-05 19:54:04",
"id": 358,
"data": {
"edges": [
{
"id": "xy-edge__start_fa9afright_handle-input_cc36cleft_handle",
"type": "customEdge",
"source": "start_fa9af",
"target": "input_cc36c",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_cc36cright_handle-rag_896caleft_handle",
"type": "customEdge",
"source": "input_cc36c",
"target": "rag_896ca",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__rag_896caright_handle-input_cc36cleft_handle",
"type": "customEdge",
"source": "rag_896ca",
"target": "input_cc36c",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
}
],
"nodes": [
{
"id": "start_fa9af",
"data": {
"v": 3,
"id": "start_fa9af",
"name": "开始",
"type": "start",
"description": "工作流运行的起始节点。",
"group_params": [
{
"name": "开场引导",
"params": [
{
"key": "guide_word",
"type": "textarea",
"label": "开场白",
"value": "",
"placeholder": "每次工作流开始执行时向用户发送此消息，支持 Markdown 格式，为空时不发送。"
},
{
"key": "guide_question",
"help": "为用户提供推荐问题，引导用户输入，超过3个时将随机选取3个。",
"type": "input_list",
"label": "引导问题",
"value": [
""
],
"placeholder": "请输入引导问题"
}
]
},
{
"name": "全局变量",
"params": [
{
"key": "user_info",
"type": "var",
"label": "true",
"value": "",
"global": "key"
},
{
"key": "current_time",
"type": "var",
"label": "当前时间",
"value": "",
"global": "key"
},
{
"key": "chat_history",
"type": "chat_history_num",
"value": 10,
"global": "key"
},
{
"key": "preset_question",
"help": "适合文档审核、报告生成等场景，利用提前预置的问题批量进行 RAG 问答。",
"type": "input_list",
"label": "预置问题列表",
"value": [
{
"key": "e6c745",
"value": ""
}
],
"global": "item:input_list",
"placeholder": "输入批量预置问题"
},
{
"key": "custom_variables",
"help": "true",
"type": "global_var",
"label": "true",
"value": [],
"global": "item:input_list"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 711
},
"position": {
"x": 469,
"y": 150
},
"selected": false
},
{
"id": "rag_896ca",
"data": {
"v": 3,
"id": "rag_896ca",
"name": "文档知识库问答",
"type": "rag",
"expand": true,
"description": "根据用户问题从知识库中检索相关内容，结合检索结果调用大模型生成最终结果，支持多个问题并行执行。",
"group_params": [
{
"name": "知识库检索设置",
"params": [
{
"key": "user_question",
"help": "当选择多个问题时，将会多次运行本节点，每次运行时从批量问题中取一项进行处理。",
"test": "var",
"type": "user_question",
"label": "用户问题",
"value": [
"input_cc36c.user_input"
],
"varZh": {
"input_cc36c.user_input": "输入/user_input",
"input_e1758.user_input": "输入/user_input",
"start_fa9af.current_time": "开始/current_time"
},
"global": "self=user_prompt",
"linkage": "output_user_input",
"required": true,
"placeholder": "请选择用户问题"
},
{
"key": "knowledge",
"type": "knowledge_select_multi",
"label": "检索范围",
"value": {
"type": "knowledge",
"value": [
{
"key": 1283,
"label": "毕昇文档知识库"
}
]
},
"required": true,
"placeholder": "请选择知识库"
},
{
"key": "metadata_filter",
"type": "metadata_filter",
"label": "true",
"value": {
"enabled": false
}
},
{
"key": "advanced_retrieval_switch",
"type": "search_switch",
"label": "true",
"value": {
"user_auth": false,
"rerank_flag": false,
"rerank_model": "",
"search_switch": true,
"vector_weight": 0.5,
"keyword_weight": 0.5,
"max_chunk_size": 15000
}
},
{
"key": "retrieved_result",
"type": "var",
"label": "检索结果",
"global": "self=user_prompt"
}
]
},
{
"name": "AI回复生成设置",
"params": [
{
"key": "system_prompt",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是一个知识库问答助手： \n1.用中文回答用户问题，并且答案要严谨专业。\n2.你需要依据以下【参考文本】中的内容来回答，当【参考文本】中有明确与用户问题相关的内容时才进行回答，不可根据自己的知识来回答。\n3.由于【参考文本】可能包含多个来自不同信息源的信息，所以根据这些不同的信息源可能得出有差异甚至冲突的答案，当发现这种情况时，这些答案都列举出来；如果没有冲突或差异，则只需要给出一个最终结果。\n4.若【参考文本】中内容与用户问题不相关则回复“没有找到相关内容”。",
"required": true
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "用户问题：{{#rag_896ca.user_question#}}\n参考文本：{{#rag_896ca.retrieved_result#}}\n你的回答：",
"varZh": {
"rag_896ca.user_question": "user_question",
"rag_896ca.retrieved_result": "retrieved_result"
},
"required": true
},
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output_user_input",
"type": "var",
"label": "输出变量",
"value": [
{
"key": "output_user_input",
"label": "output_user_input"
}
],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 1275
},
"position": {
"x": 1338.9283536315477,
"y": 152.54665865258175
},
"selected": true
},
{
"id": "input_cc36c",
"data": {
"v": "3",
"id": "input_cc36c",
"tab": {
"value": "dialog_input",
"options": [
{
"key": "dialog_input",
"help": "true",
"label": "true"
},
{
"key": "form_input",
"help": "true",
"label": "true"
}
]
},
"name": "输入",
"type": "input",
"description": "接收用户在会话页面的输入，支持 2 种形式：对话框输入，表单输入。",
"group_params": [
{
"name": "接收文本",
"params": [
{
"key": "user_input",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
}
]
},
{
"name": "",
"params": [
{
"key": "user_input_file",
"tab": "dialog_input",
"value": true,
"groupTitle": true
},
{
"key": "file_parse_mode",
"tab": "dialog_input",
"type": "select_parsemode",
"value": "extract_text"
},
{
"key": "dialog_files_content",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_files_content_size",
"min": 0,
"tab": "dialog_input",
"type": "char_number",
"label": "true",
"value": 15000
},
{
"key": "dialog_file_accept",
"tab": "dialog_input",
"type": "select_fileaccept",
"label": "true",
"value": "all"
},
{
"key": "dialog_image_files",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_file_paths",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
}
],
"groupKey": "inputfile"
},
{
"name": "",
"params": [
{
"key": "recommended_questions_flag",
"tab": "dialog_input",
"help": "true",
"label": "true",
"value": false,
"hidden": "true",
"groupTitle": true
},
{
"key": "recommended_llm",
"tab": "dialog_input",
"type": "bisheng_model",
"label": "true",
"value": 261,
"required": true,
"placeholder": "true"
},
{
"key": "recommended_system_prompt",
"tab": "dialog_input",
"type": "var_textarea",
"label": "true",
"value": "# Role\n你是一个极具洞察力的\"对话延续预测专家\"。你的任务是根据当前的对话历史，预测用户接下来最有可能输入的 3 条短语或问题。\n\n# Context Awareness\n在生成预测前，请执行以下深度思考步骤：\n1. **意图分析**：用户当前是在寻求事实、表达情绪、请求建议，还是在尝试结束对话？\n2. **逻辑推演**：根据 AI 上一条回复的内容，用户最自然产生的联想或疑问是什么？\n3. **风格对齐**：观察用户的用词习惯（专业、随性、简短还是详细），预测结果需保持语气一致。\n\n# Goals\n- **多样性**：提供不同方向的后续（例如：一个深入探究的问题，一个相关的新话题，一个反馈性评价）。\n- **简洁性**：预测内容应短小精悍，通常不超过 15 个字，方便用户点击。\n- **前瞻性**：能够预判用户的\"痛点\"，直接提供解决方案的入口。\n\n# Output Format\n请严格遵守 JSON 格式输出。\n返回一个包含单一键名 \"suggestions\" 的 JSON 对象。\n- 所有字符串必须使用**双引号**。\n- 不要输出 Markdown 代码块标记（如 ```json），直接输出 JSON 字符串。\n**示例：**\n{\"suggestions\": [\"你能举个例子吗？\", \"这个方案的成本是多少？\", \"听起来不错，怎么开始？\"]}",
"required": true
},
{
"key": "recommended_history_num",
"tab": "dialog_input",
"help": "true",
"step": 1,
"type": "slide",
"label": "true",
"scope": [
1,
10
],
"value": 2
}
],
"groupKey": "custom"
},
{
"name": "",
"params": [
{
"key": "form_input",
"tab": "form_input",
"type": "form",
"label": "true",
"value": [],
"global": "item:form_input"
}
]
}
]
},
"type": "flowNode",
"measured": {
"width": 334,
"height": 570
},
"position": {
"x": 935.6608232521362,
"y": 163.53933298957216
}
}
],
"viewport": {
"x": -365.03852145949816,
"y": -107.15501189516708,
"zoom": 0.6295427860409281
}
},
"order_num": 12238723,
"guide_word": null,
"flow_type": 10
},
{
"update_time": "2026-02-05 20:05:11",
"parameters": null,
"name": "优先知识库问答+搜索引擎兜底",
"description": "功能简介：这是 “优先知识库问答 + 搜索引擎兜底” 工作流，先借 “开始” 引导对话，“输入” 模块获取用户提问与文件，经 “文档知识库问答” 检索知识；若未匹配内容，“条件分支” 触发 “助手” 调用搜索引擎补充，最终 “输出” 回答，保障问答覆盖度。\n使用说明：“开始” 填引导语；“输入” 选录入方式，设文件参数。\n运行流程，“文档知识库问答” 优先检索知识；若output_user含 “没有找到相关内容”，触发 “助手” 用搜索引擎兜底。",
"flow_id": "2853ac0e3e3e483985a414911e409463",
"api_parameters": null,
"create_time": "2026-02-05 20:05:11",
"id": 360,
"data": {
"edges": [
{
"id": "xy-edge__condition_448edright_handle-output_2a3beleft_handle",
"type": "customEdge",
"source": "condition_448ed",
"target": "output_2a3be",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__rag_3e24bright_handle-condition_448edleft_handle",
"type": "customEdge",
"source": "rag_3e24b",
"target": "condition_448ed",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__condition_448ed78028559-agent_49ec7left_handle",
"type": "customEdge",
"source": "condition_448ed",
"target": "agent_49ec7",
"animated": true,
"sourceHandle": "78028559",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__start_9f092right_handle-input_c284fleft_handle",
"type": "customEdge",
"source": "start_9f092",
"target": "input_c284f",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_c284fright_handle-rag_3e24bleft_handle",
"type": "customEdge",
"source": "input_c284f",
"target": "rag_3e24b",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__output_2a3beright_handle-input_c284fleft_handle",
"type": "customEdge",
"source": "output_2a3be",
"target": "input_c284f",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__agent_49ec7right_handle-input_c284fleft_handle",
"type": "customEdge",
"source": "agent_49ec7",
"target": "input_c284f",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
}
],
"nodes": [
{
"id": "start_9f092",
"data": {
"v": 3,
"id": "start_9f092",
"name": "开始",
"type": "start",
"description": "工作流运行的起始节点。",
"group_params": [
{
"name": "开场引导",
"params": [
{
"key": "guide_word",
"type": "textarea",
"label": "开场白",
"value": "你好，我是问答助手",
"placeholder": "每次工作流开始执行时向用户发送此消息，支持 Markdown 格式，为空时不发送。"
},
{
"key": "guide_question",
"help": "为用户提供推荐问题，引导用户输入，超过3个时将随机选取3个。",
"type": "input_list",
"label": "引导问题",
"value": [
""
],
"placeholder": "请输入引导问题"
}
]
},
{
"name": "全局变量",
"params": [
{
"key": "user_info",
"type": "var",
"label": "true",
"value": "",
"global": "key"
},
{
"key": "current_time",
"type": "var",
"label": "当前时间",
"value": "",
"global": "key"
},
{
"key": "chat_history",
"type": "chat_history_num",
"value": 10,
"global": "key"
},
{
"key": "preset_question",
"help": "适合文档审核、报告生成等场景，利用提前预置的问题批量进行 RAG 问答。",
"type": "input_list",
"label": "预置问题列表",
"value": [
{
"key": "a3a02e",
"value": ""
}
],
"global": "item:input_list",
"placeholder": "输入批量预置问题"
},
{
"key": "custom_variables",
"help": "true",
"type": "global_var",
"label": "true",
"value": [],
"global": "item:input_list"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 711
},
"position": {
"x": 237.20000000000005,
"y": -161
},
"selected": false
},
{
"id": "condition_448ed",
"data": {
"id": "condition_448ed",
"name": "条件分支",
"type": "condition",
"description": "根据条件表达式执行不同的分支。",
"group_params": [
{
"params": [
{
"key": "condition",
"type": "condition",
"label": "",
"value": [
{
"id": "78028559",
"operator": "and",
"conditions": [
{
"id": "59451337",
"left_var": "rag_3e24b.output_user_input",
"left_label": "output_user_input",
"right_label": "",
"right_value": "没有找到相关内容",
"right_value_type": "input",
"comparison_operation": "contains"
}
]
}
]
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 562,
"height": 347
},
"position": {
"x": 1614.1888377214964,
"y": -254.09438817125263
},
"selected": false
},
{
"id": "output_2a3be",
"data": {
"v": 2,
"id": "output_2a3be",
"name": "输出",
"type": "output",
"description": "可向用户发送消息，并且支持进行更丰富的交互，例如请求用户批准进行某项敏感操作、允许用户在模型输出内容的基础上直接修改并提交。",
"group_params": [
{
"params": [
{
"key": "message",
"type": "var_textarea_file",
"label": "消息内容",
"value": {
"msg": "{{#rag_3e24b.output_user_input#}}",
"files": []
},
"varZh": {
"agent_140fe.output": "助手/output",
"rag_3e24b.output_user_input": "文档知识库问答/output_user_input"
},
"global": "key",
"required": true,
"placeholder": "输入需要发送给用户的消息，例如“接下来我将执行 XX 操作，请您确认”，“以下是我的初版草稿，您可以在其基础上进行修改”"
},
{
"key": "output_result",
"type": "output_form",
"label": "用户输入内容",
"value": {
"type": "",
"value": ""
},
"global": "value.type=input",
"options": [],
"required": false
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 463
},
"position": {
"x": 2260.2238409262695,
"y": 26.60625066348959
},
"selected": false
},
{
"id": "rag_3e24b",
"data": {
"v": 3,
"id": "rag_3e24b",
"name": "文档知识库问答",
"type": "rag",
"expand": false,
"description": "根据用户问题从知识库中检索相关内容，结合检索结果调用大模型生成最终结果，支持多个问题并行执行。",
"group_params": [
{
"name": "知识库检索设置",
"params": [
{
"key": "user_question",
"help": "当选择多个问题时，将会多次运行本节点，每次运行时从批量问题中取一项进行处理。",
"test": "var",
"type": "user_question",
"label": "用户问题",
"value": [
"input_c284f.user_input"
],
"varZh": {
"input_1b610.user_input": "输入/user_input",
"input_2d0d4.user_input": "输入/user_input",
"input_c284f.user_input": "输入/user_input"
},
"global": "self=user_prompt",
"linkage": "output_user_input",
"required": true,
"placeholder": "请选择用户问题"
},
{
"key": "knowledge",
"type": "knowledge_select_multi",
"label": "检索范围",
"value": {
"type": "knowledge",
"value": [
{
"key": 1301,
"label": "毕昇产品文档"
}
]
},
"required": true,
"placeholder": "请选择知识库"
},
{
"key": "metadata_filter",
"type": "metadata_filter",
"label": "true",
"value": {
"enabled": false
}
},
{
"key": "advanced_retrieval_switch",
"type": "search_switch",
"label": "true",
"value": {
"user_auth": false,
"rerank_flag": false,
"rerank_model": "",
"search_switch": true,
"vector_weight": 0.5,
"keyword_weight": 0.5,
"max_chunk_size": 15000
}
},
{
"key": "retrieved_result",
"type": "var",
"label": "检索结果",
"global": "self=user_prompt"
}
]
},
{
"name": "AI回复生成设置",
"params": [
{
"key": "system_prompt",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是一个知识库问答助手： \n1.用中文回答用户问题，并且答案要严谨专业。\n2.你需要依据以上【参考文本】中的内容来回答，当【参考文本】中有明确与用户问题相关的内容时才进行回答，不可根据自己的知识来回答。\n3.由于【参考文本】可能包含多个来自不同信息源的信息，所以根据这些不同的信息源可能得出有差异甚至冲突的答案，当发现这种情况时，这些答案都列举出来；如果没有冲突或差异，则只需要给出一个最终结果。\n4.若【参考文本】中内容与用户问题不相关则回复“没有找到相关内容”。",
"required": true
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "用户问题：{{#rag_3e24b.user_question#}}\n参考文本：{{#rag_3e24b.retrieved_result#}}\n你的回答：",
"varZh": {
"rag_3e24b.user_question": "user_question",
"rag_3e24b.retrieved_result": "retrieved_result"
},
"required": true
},
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": false
},
{
"key": "output_user_input",
"type": "var",
"label": "输出变量",
"value": [
{
"key": "output_user_input",
"label": "output_user_input"
}
],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 150
},
"position": {
"x": 1145.3311585293775,
"y": -159.21938594285209
},
"selected": false
},
{
"id": "agent_49ec7",
"data": {
"v": 2,
"id": "agent_49ec7",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "助手",
"type": "agent",
"expand": false,
"description": "AI 自主进行任务规划，选择合适的知识库或工具进行调用。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "agent_model",
"label": "模型",
"value": 274,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "# role \n\n你是一个善于根据互联网搜索引擎获取信息并回答用户问题的 AI 助理# skill\n\n根据用户问题，调用 Bing Web 工具检索需要的信息，并结合检索结果回答用户问题。",
"required": true,
"placeholder": "助手画像"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "{{#input_c284f.user_input#}}",
"varZh": {
"input_c284f.user_input": "输入/user_input"
},
"required": true,
"placeholder": "用户消息内容"
},
{
"key": "chat_history_flag",
"help": "带入模型上下文的历史消息条数，为 0 时代表不包含上下文信息。",
"step": 1,
"type": "slide_switch",
"label": "历史聊天记录",
"scope": [
0,
100
],
"value": {
"flag": true,
"value": 50
}
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": ""
}
]
},
{
"name": "知识库",
"params": [
{
"key": "knowledge_id",
"type": "knowledge_select_multi",
"label": "检索知识库范围",
"value": {
"type": "knowledge",
"value": []
},
"placeholder": "请选择知识库"
}
]
},
{
"name": "数据库",
"params": [
{
"key": "sql_agent",
"type": "sql_config",
"value": {
"open": false,
"db_name": "",
"db_address": "",
"db_password": "",
"db_username": "",
"database_engine": "mysql"
}
}
]
},
{
"name": "工具",
"params": [
{
"key": "tool_list",
"type": "add_tool",
"label": "+ 添加工具",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 150
},
"position": {
"x": 2265.0840022667044,
"y": -456.86140838548886
},
"selected": true
},
{
"id": "input_c284f",
"data": {
"v": "3",
"id": "input_c284f",
"tab": {
"value": "dialog_input",
"options": [
{
"key": "dialog_input",
"help": "true",
"label": "true"
},
{
"key": "form_input",
"help": "true",
"label": "true"
}
]
},
"name": "输入",
"type": "input",
"description": "接收用户在会话页面的输入，支持 2 种形式：对话框输入，表单输入。",
"group_params": [
{
"name": "接收文本",
"params": [
{
"key": "user_input",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
}
]
},
{
"name": "",
"params": [
{
"key": "user_input_file",
"tab": "dialog_input",
"value": true,
"groupTitle": true
},
{
"key": "file_parse_mode",
"tab": "dialog_input",
"type": "select_parsemode",
"value": "extract_text"
},
{
"key": "dialog_files_content",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_files_content_size",
"min": 0,
"tab": "dialog_input",
"type": "char_number",
"label": "true",
"value": 15000
},
{
"key": "dialog_file_accept",
"tab": "dialog_input",
"type": "select_fileaccept",
"label": "true",
"value": "all"
},
{
"key": "dialog_image_files",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_file_paths",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
}
],
"groupKey": "inputfile"
},
{
"name": "",
"params": [
{
"key": "recommended_questions_flag",
"tab": "dialog_input",
"help": "true",
"label": "true",
"value": false,
"hidden": "true",
"groupTitle": true
},
{
"key": "recommended_llm",
"tab": "dialog_input",
"type": "bisheng_model",
"label": "true",
"value": 261,
"required": true,
"placeholder": "true"
},
{
"key": "recommended_system_prompt",
"tab": "dialog_input",
"type": "var_textarea",
"label": "true",
"value": "# Role\n你是一个极具洞察力的\"对话延续预测专家\"。你的任务是根据当前的对话历史，预测用户接下来最有可能输入的 3 条短语或问题。\n\n# Context Awareness\n在生成预测前，请执行以下深度思考步骤：\n1. **意图分析**：用户当前是在寻求事实、表达情绪、请求建议，还是在尝试结束对话？\n2. **逻辑推演**：根据 AI 上一条回复的内容，用户最自然产生的联想或疑问是什么？\n3. **风格对齐**：观察用户的用词习惯（专业、随性、简短还是详细），预测结果需保持语气一致。\n\n# Goals\n- **多样性**：提供不同方向的后续（例如：一个深入探究的问题，一个相关的新话题，一个反馈性评价）。\n- **简洁性**：预测内容应短小精悍，通常不超过 15 个字，方便用户点击。\n- **前瞻性**：能够预判用户的\"痛点\"，直接提供解决方案的入口。\n\n# Output Format\n请严格遵守 JSON 格式输出。\n返回一个包含单一键名 \"suggestions\" 的 JSON 对象。\n- 所有字符串必须使用**双引号**。\n- 不要输出 Markdown 代码块标记（如 ```json），直接输出 JSON 字符串。\n**示例：**\n{\"suggestions\": [\"你能举个例子吗？\", \"这个方案的成本是多少？\", \"听起来不错，怎么开始？\"]}",
"required": true
},
{
"key": "recommended_history_num",
"tab": "dialog_input",
"help": "true",
"step": 1,
"type": "slide",
"label": "true",
"scope": [
1,
10
],
"value": 2
}
],
"groupKey": "custom"
},
{
"name": "",
"params": [
{
"key": "form_input",
"tab": "form_input",
"type": "form",
"label": "true",
"value": [],
"global": "item:form_input"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 570
},
"position": {
"x": 735.2652144940947,
"y": -165.84643370976673
},
"selected": false
}
],
"viewport": {
"x": -69.1807775874338,
"y": 311.2637313881481,
"zoom": 0.5741179493568034
}
},
"order_num": 12369793,
"guide_word": null,
"flow_type": 10
},
{
"update_time": "2026-02-05 20:11:36",
"parameters": null,
"name": "五步翻译法",
"description": "功能简介\n这是翻译工作流，通过 “输入” 环节录入文本，经 “英文要点直译”“中文意图反译” 等多轮模型处理，利用大模型分步骤优化翻译内容，最终输出高质量译文，适用于需精准翻译的场景。\n使用说明\n在 “输入” 节点填入待翻译文本，依次运行 “英文要点直译”“中文意图反译” 等后续节点，各环节依托大模型自动处理，流程结束后在 “结束” 节点查看最终翻译结果，可按需调整模型参数优化效果。",
"flow_id": "a2fb8daa1af942d58c306c164d2e7c9e",
"api_parameters": null,
"create_time": "2026-02-05 20:11:36",
"id": 361,
"data": {
"edges": [
{
"id": "xy-edge__llm_ebfb8right_handle-llm_3fa0dleft_handle",
"type": "customEdge",
"source": "llm_ebfb8",
"target": "llm_3fa0d",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_3fa0dright_handle-llm_834d3left_handle",
"type": "customEdge",
"source": "llm_3fa0d",
"target": "llm_834d3",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_834d3right_handle-llm_9ae96left_handle",
"type": "customEdge",
"source": "llm_834d3",
"target": "llm_9ae96",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_9ae96right_handle-llm_414d9left_handle",
"type": "customEdge",
"source": "llm_9ae96",
"target": "llm_414d9",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_414d9right_handle-end_6ee61left_handle",
"type": "customEdge",
"source": "llm_414d9",
"target": "end_6ee61",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__start_faec1right_handle-input_c5b73left_handle",
"type": "customEdge",
"source": "start_faec1",
"target": "input_c5b73",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_c5b73right_handle-llm_ebfb8left_handle",
"type": "customEdge",
"source": "input_c5b73",
"target": "llm_ebfb8",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
}
],
"nodes": [
{
"id": "start_faec1",
"data": {
"v": 3,
"id": "start_faec1",
"name": "开始",
"type": "start",
"expand": true,
"description": "工作流运行的起始节点。",
"group_params": [
{
"name": "开场引导",
"params": [
{
"key": "guide_word",
"type": "textarea",
"label": "开场白",
"value": "五步翻译法，适合论文等对于翻译质量有较高要求的场景，包含 5 个环节： 1）直译初稿 2）意译初稿 3）回译 4）编辑校对 5）意译终稿 。\n\n请发送您要翻译的英文材料：",
"placeholder": "每次工作流开始执行时向用户发送此消息，支持 Markdown 格式，为空时不发送。"
},
{
"key": "guide_question",
"help": "为用户提供推荐问题，引导用户输入，超过3个时将随机选取3个。",
"type": "input_list",
"label": "引导问题",
"value": [
""
],
"placeholder": "请输入引导问题"
}
]
},
{
"name": "全局变量",
"params": [
{
"key": "user_info",
"type": "var",
"label": "true",
"value": "",
"global": "key"
},
{
"key": "current_time",
"type": "var",
"label": "当前时间",
"value": "",
"global": "key"
},
{
"key": "chat_history",
"type": "chat_history_num",
"value": 10,
"global": "key"
},
{
"key": "preset_question",
"help": "适合文档审核、报告生成等场景，利用提前预置的问题批量进行 RAG 问答。",
"type": "input_list",
"label": "预置问题列表",
"value": [
{
"key": "63e87c",
"value": ""
}
],
"global": "item:input_list",
"placeholder": "输入批量预置问题"
},
{
"key": "custom_variables",
"help": "true",
"type": "global_var",
"label": "true",
"value": [],
"global": "item:input_list"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 711
},
"position": {
"x": 138.9223951864526,
"y": 81.99792129963708
},
"selected": false
},
{
"id": "end_6ee61",
"data": {
"id": "end_6ee61",
"name": "结束",
"type": "end",
"description": "工作流运行到此结束。",
"group_params": []
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2946.3789207819623,
"y": 403.8501940420881
},
"selected": false
},
{
"id": "llm_ebfb8",
"data": {
"v": 2,
"id": "llm_ebfb8",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "英文教授直译",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "# role \n\n你是一位资深英语教授，能精确的理解英文并用中文表达，同时也是一名熟悉神经科学、machine learning、deep learning、NLP和CV领域的教授，掌握这些领域的专业术语。 \n\n# task \n\n翻译解释一篇专业的技术文章成简体中文给本科生阅读，务必遵守原意，翻译时保持原始英文的段落结构，不要合并分段。\n\n# Rules\n\n- 翻译时要准确传达学术论文的事实和背景，同时风格上保持为通俗易懂并且严谨的科普文风格。\n\n- 保留特定的英文术语、数字或名字，并在其前后加上空格，例如：\"中 UN 文\"，\"不超过 10 秒\"。\n\n- 即使意译也要保留术语，例如 token、transformer、LLM、zero-shot 等。保留公司缩写，例如 Microsoft, Amazon 等。\n\n- 保留引用的论文，例如 [20] 这样的引用；同时也要保留针对图例的引用，例如保留 Figure 1 并翻译为图 1。\n\n- 输入格式为 Markdown 格式，输出格式也必须保留原始 Markdown 格式。- 翻译时全角括号换成半角括号，并在左括号前面加半角空格，右括号后面加半角空格。\n\n"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "{{#input_c5b73.user_input#}}",
"varZh": {
"input_c5b73.user_input": "输入/user_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 1107.194280093113,
"y": 84.7835136533582
},
"selected": false
},
{
"id": "llm_3fa0d",
"data": {
"v": 2,
"id": "llm_3fa0d",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "中文教授意译初稿",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "# role\n\n你是一位资深中文教授，擅长按照中文使用习惯，撰写通俗易懂的科普文。 \n\n# task \n\n对英语教授翻译的内容重新意译，遵守原意的前提下让内容更通俗易懂，符合中文表达习惯，但不要增加和删减内容，保持原始分段。\n\n"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "# 英文原文\n{{#input_c5b73.user_input#}}\n\n\n# 英文教授直译文 \n\n{{#llm_ebfb8.output#}}\n",
"varZh": {
"llm_ebfb8.output": "英文教授直译/output",
"input_c5b73.user_input": "输入/user_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 1574.4947033403798,
"y": 85.49355402754313
},
"selected": false
},
{
"id": "llm_834d3",
"data": {
"v": 2,
"id": "llm_834d3",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "英文教授回译",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "# role \n\n你是一位资深英语教授，能精确的理解英文并用中文表达，同时也是一名熟悉神经科学、machine learning、deep learning、NLP和CV领域的教授，掌握这些领域的专业术语。 \n\n# task \n\n将中文教授的中文意译稿反向翻译成英文回译稿。"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "# 中文教授中文意译稿 \n{{#llm_3fa0d.output#}}\n",
"varZh": {
"llm_3fa0d.output": "中文教授意译初稿/output"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2030.102513363393,
"y": 89.35591616008031
},
"selected": false
},
{
"id": "llm_9ae96",
"data": {
"v": 2,
"id": "llm_9ae96",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "审查编辑校对",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "# role \n\n你是一位资深出版编辑，精通中文和英文，擅长校对审查。\n\n你现在手上有3份文稿：\n\n- 英文原稿。 \n\n- 中文译稿。 \n\n- 英文回译稿（原稿翻译成中文后，重新回译成英文的回译稿） \n\n# task \n\n1. 通过对比【英文原稿】和【英文回译稿】的出入，便于找到中文译稿中不恰当的地方（重复以下列表，直到列出所有有出入并且需要修改译文的内容）： \n\n- 回译稿对比原文缺失或表达歧义部分[1...n]: \n\n- 原文：“{English}” \n\n- 回译文：“{English}”\n\n- 中文译文：“{译文}” \n\n- 建议：{译文修改建议} \n\n2. 指出中文翻译表达不符合中文习惯的部分（重复以下列表，直到列出所有需要修改的内容）： \n\n- 修改{1...n}: \n\n- 原文：“{English}” \n\n- 译文：“{译文}” \n\n- 建议：{译文修改建议}\n\n\n\n"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "{{#input_c5b73.user_input#}}# 英文原稿\n``````\n\n# 英文回译稿 \n```{{#llm_834d3.output#}}```\n# 中文译稿```{{#llm_3fa0d.output#}}```",
"varZh": {
"llm_3fa0d.output": "中文教授意译初稿/output",
"llm_834d3.output": "英文教授回译/output",
"input_c5b73.user_input": "输入/user_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2464.072445850047,
"y": 90.30383209830156
},
"selected": true
},
{
"id": "llm_414d9",
"data": {
"v": 2,
"id": "llm_414d9",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "中文教授意译终稿",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "# role\n你是一位资深中文教授，擅长按照中文使用习惯，撰写通俗易懂的科普文。\n\n\n# task\n对照英文原文材料，在意译初稿的基础上，针对出版编辑的修改建议，翻译出意译终稿。\n\n\n\n\n\n\n"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "# 英文原文\n```{{#input_c5b73.user_input#}}```\n# 意译初稿\n```{{#llm_3fa0d.output#}}```\n# 出版编辑修改意见 \n```{{#llm_9ae96.output#}}```",
"varZh": {
"llm_3fa0d.output": "中文教授意译初稿/output",
"llm_9ae96.output": "审查编辑校对/output",
"input_c5b73.user_input": "输入/user_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2916.394693106114,
"y": 97.6941646914677
},
"selected": false
},
{
"id": "input_c5b73",
"data": {
"v": "3",
"id": "input_c5b73",
"tab": {
"value": "dialog_input",
"options": [
{
"key": "dialog_input",
"help": "true",
"label": "true"
},
{
"key": "form_input",
"help": "true",
"label": "true"
}
]
},
"name": "输入",
"type": "input",
"expand": false,
"description": "接收用户在会话页面的输入，支持 2 种形式：对话框输入，表单输入。",
"group_params": [
{
"name": "接收文本",
"params": [
{
"key": "user_input",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
}
]
},
{
"name": "",
"params": [
{
"key": "user_input_file",
"tab": "dialog_input",
"value": true,
"groupTitle": true
},
{
"key": "file_parse_mode",
"tab": "dialog_input",
"type": "select_parsemode",
"value": "extract_text"
},
{
"key": "dialog_files_content",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_files_content_size",
"min": 0,
"tab": "dialog_input",
"type": "char_number",
"label": "true",
"value": 15000
},
{
"key": "dialog_file_accept",
"tab": "dialog_input",
"type": "select_fileaccept",
"label": "true",
"value": "all"
},
{
"key": "dialog_image_files",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_file_paths",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
}
],
"groupKey": "inputfile"
},
{
"name": "",
"params": [
{
"key": "recommended_questions_flag",
"tab": "dialog_input",
"help": "true",
"label": "true",
"value": false,
"hidden": "true",
"groupTitle": true
},
{
"key": "recommended_llm",
"tab": "dialog_input",
"type": "bisheng_model",
"label": "true",
"value": 261,
"required": true,
"placeholder": "true"
},
{
"key": "recommended_system_prompt",
"tab": "dialog_input",
"type": "var_textarea",
"label": "true",
"value": "# Role\n你是一个极具洞察力的\"对话延续预测专家\"。你的任务是根据当前的对话历史，预测用户接下来最有可能输入的 3 条短语或问题。\n\n# Context Awareness\n在生成预测前，请执行以下深度思考步骤：\n1. **意图分析**：用户当前是在寻求事实、表达情绪、请求建议，还是在尝试结束对话？\n2. **逻辑推演**：根据 AI 上一条回复的内容，用户最自然产生的联想或疑问是什么？\n3. **风格对齐**：观察用户的用词习惯（专业、随性、简短还是详细），预测结果需保持语气一致。\n\n# Goals\n- **多样性**：提供不同方向的后续（例如：一个深入探究的问题，一个相关的新话题，一个反馈性评价）。\n- **简洁性**：预测内容应短小精悍，通常不超过 15 个字，方便用户点击。\n- **前瞻性**：能够预判用户的\"痛点\"，直接提供解决方案的入口。\n\n# Output Format\n请严格遵守 JSON 格式输出。\n返回一个包含单一键名 \"suggestions\" 的 JSON 对象。\n- 所有字符串必须使用**双引号**。\n- 不要输出 Markdown 代码块标记（如 ```json），直接输出 JSON 字符串。\n**示例：**\n{\"suggestions\": [\"你能举个例子吗？\", \"这个方案的成本是多少？\", \"听起来不错，怎么开始？\"]}",
"required": true
},
{
"key": "recommended_history_num",
"tab": "dialog_input",
"help": "true",
"step": 1,
"type": "slide",
"label": "true",
"scope": [
1,
10
],
"value": 2
}
],
"groupKey": "custom"
},
{
"name": "",
"params": [
{
"key": "form_input",
"tab": "form_input",
"type": "form",
"label": "true",
"value": [],
"global": "item:form_input"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 150
},
"position": {
"x": 634.423952924182,
"y": 85.4401521708546
},
"selected": false
}
],
"viewport": {
"x": -48.47172678094421,
"y": 123.7367237461774,
"zoom": 0.4280931573424512
}
},
"order_num": 12435328,
"guide_word": null,
"flow_type": 10
},
{
"update_time": "2026-02-05 20:27:30",
"parameters": null,
"name": "基于产品知识库的 AI 营销文案生成 + 人类意见迭代",
"description": "功能简介\n基于产品知识库，通过输入用户需求，经文本识别、拆分等处理，调用大模型生成营销文案，辅助高效产出贴合产品知识的推广内容 。\n使用说明\n在 “输入” 节点填用户需求，依次运行 “文本识别”“拆分” 等节点，配置各环节参数（如文本长度、模型选择），最后经大模型节点生成营销文案，在 “输出” 查看结果 。",
"flow_id": "60338b009bb44d6d80086c2edeffcad9",
"api_parameters": null,
"create_time": "2026-02-05 20:27:30",
"id": 362,
"data": {
"edges": [
{
"id": "xy-edge__output_204469212bf1c-end_ed5e5left_handle",
"type": "customEdge",
"source": "output_20446",
"target": "end_ed5e5",
"animated": true,
"sourceHandle": "9212bf1c",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__output_2044664f4c390-output_84247left_handle",
"type": "customEdge",
"source": "output_20446",
"target": "output_84247",
"animated": true,
"sourceHandle": "64f4c390",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__rag_20bedright_handle-output_7d6b6left_handle",
"type": "customEdge",
"source": "rag_20bed",
"target": "output_7d6b6",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__output_7d6b6right_handle-rag_2b974left_handle",
"type": "customEdge",
"source": "output_7d6b6",
"target": "rag_2b974",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__rag_2b974right_handle-output_20446left_handle",
"type": "customEdge",
"source": "rag_2b974",
"target": "output_20446",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__output_84247right_handle-llm_c66feleft_handle",
"type": "customEdge",
"source": "output_84247",
"target": "llm_c66fe",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_c66feright_handle-output_20446left_handle",
"type": "customEdge",
"source": "llm_c66fe",
"target": "output_20446",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__start_d602dright_handle-input_60a99left_handle",
"type": "customEdge",
"source": "start_d602d",
"target": "input_60a99",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_60a99right_handle-rag_20bedleft_handle",
"type": "customEdge",
"source": "input_60a99",
"target": "rag_20bed",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
}
],
"nodes": [
{
"id": "start_d602d",
"data": {
"v": 3,
"id": "start_d602d",
"name": "开始",
"type": "start",
"description": "工作流运行的起始节点。",
"group_params": [
{
"name": "开场引导",
"params": [
{
"key": "guide_word",
"type": "textarea",
"label": "开场白",
"value": "您好，我是 AI 营销文案助手，请输入您想要生成文案的主题。",
"placeholder": "每次工作流开始执行时向用户发送此消息，支持 Markdown 格式，为空时不发送。"
},
{
"key": "guide_question",
"help": "为用户提供推荐问题，引导用户输入，超过3个时将随机选取3个。",
"type": "input_list",
"label": "引导问题",
"value": [
""
],
"placeholder": "请输入引导问题"
}
]
},
{
"name": "全局变量",
"params": [
{
"key": "user_info",
"type": "var",
"label": "true",
"value": "",
"global": "key"
},
{
"key": "current_time",
"type": "var",
"label": "当前时间",
"value": "",
"global": "key"
},
{
"key": "chat_history",
"type": "chat_history_num",
"value": 10,
"global": "key"
},
{
"key": "preset_question",
"help": "适合文档审核、报告生成等场景，利用提前预置的问题批量进行 RAG 问答。",
"type": "input_list",
"label": "预置问题列表",
"value": [
{
"key": "452b35",
"value": ""
}
],
"global": "item:input_list",
"placeholder": "输入批量预置问题"
},
{
"key": "custom_variables",
"help": "true",
"type": "global_var",
"label": "true",
"value": [],
"global": "item:input_list"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 711
},
"position": {
"x": -42.66122923778269,
"y": 241.06330242522972
},
"selected": false
},
{
"id": "output_20446",
"data": {
"v": 2,
"id": "output_20446",
"name": "输出",
"type": "output",
"description": "可向用户发送消息，并且支持进行更丰富的交互，例如请求用户批准进行某项敏感操作、允许用户在模型输出内容的基础上直接修改并提交。",
"group_params": [
{
"params": [
{
"key": "message",
"type": "var_textarea_file",
"label": "消息内容",
"value": {
"msg": "是否需要对以上营销文案进一步修改润色？",
"files": []
},
"global": "key",
"required": true,
"placeholder": "输入需要发送给用户的消息，例如“接下来我将执行 XX 操作，请您确认”，“以下是我的初版草稿，您可以在其基础上进行修改”"
},
{
"key": "output_result",
"type": "output_form",
"label": "用户输入内容",
"value": {
"type": "choose",
"value": ""
},
"global": "value.type=input",
"options": [
{
"id": "64f4c390",
"label": "是",
"value": ""
},
{
"id": "9212bf1c",
"label": "否",
"value": ""
}
],
"required": false
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 622
},
"position": {
"x": 2311.53466926792,
"y": 197.65713275159808
},
"selected": false
},
{
"id": "output_7d6b6",
"data": {
"v": 2,
"id": "output_7d6b6",
"name": "输出2",
"type": "output",
"description": "可向用户发送消息，并且支持进行更丰富的交互，例如请求用户批准进行某项敏感操作、允许用户在模型输出内容的基础上直接修改并提交。",
"group_params": [
{
"params": [
{
"key": "message",
"type": "var_textarea_file",
"label": "消息内容",
"value": {
"msg": "根据产品知识库的相关信息，我总结产品功能亮点如下，你可以直接在其基础上进行修改，接下来我会根据这些亮点生成营销文案。",
"files": []
},
"global": "key",
"required": true,
"placeholder": "输入需要发送给用户的消息，例如“接下来我将执行 XX 操作，请您确认”，“以下是我的初版草稿，您可以在其基础上进行修改”"
},
{
"key": "output_result",
"type": "output_form",
"label": "用户输入内容",
"value": {
"type": "input",
"value": "{{#rag_20bed.output_user_input#}}"
},
"varZh": {
"llm_07892.output": "大模型2/output",
"rag_20bed.output_user_input": "文档知识库问答/output_user_input"
},
"global": "value.type=input",
"options": [],
"required": false
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 615
},
"position": {
"x": 1268.3128337196854,
"y": 235.8165936047889
},
"selected": false
},
{
"id": "end_ed5e5",
"data": {
"id": "end_ed5e5",
"name": "结束",
"type": "end",
"description": "工作流运行到此结束。",
"group_params": []
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2912.857796074366,
"y": 1292.741328658892
},
"selected": false
},
{
"id": "output_84247",
"data": {
"v": 2,
"id": "output_84247",
"name": "输出3",
"type": "output",
"description": "可向用户发送消息，并且支持进行更丰富的交互，例如请求用户批准进行某项敏感操作、允许用户在模型输出内容的基础上直接修改并提交。",
"group_params": [
{
"params": [
{
"key": "message",
"type": "var_textarea_file",
"label": "消息内容",
"value": {
"msg": "请输入您的润色建议：",
"files": []
},
"global": "key",
"required": true,
"placeholder": "输入需要发送给用户的消息，例如“接下来我将执行 XX 操作，请您确认”，“以下是我的初版草稿，您可以在其基础上进行修改”"
},
{
"key": "output_result",
"type": "output_form",
"label": "用户输入内容",
"value": {
"type": "input",
"value": ""
},
"global": "value.type=input",
"options": [],
"required": false
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 615
},
"position": {
"x": 2792.23765505673,
"y": 179.75492008835002
},
"selected": false
},
{
"id": "rag_20bed",
"data": {
"v": 3,
"id": "rag_20bed",
"name": "文档知识库问答",
"type": "rag",
"expand": true,
"description": "根据用户问题从知识库中检索相关内容，结合检索结果调用大模型生成最终结果，支持多个问题并行执行。",
"group_params": [
{
"name": "知识库检索设置",
"params": [
{
"key": "user_question",
"help": "当选择多个问题时，将会多次运行本节点，每次运行时从批量问题中取一项进行处理。",
"test": "var",
"type": "user_question",
"label": "用户问题",
"value": [
"input_60a99.user_input"
],
"varZh": {
"input_1936c.user_input": "输入/user_input",
"input_60a99.user_input": "输入/user_input"
},
"global": "self=user_prompt",
"linkage": "output_user_input",
"required": true,
"placeholder": "请选择用户问题"
},
{
"key": "knowledge",
"type": "knowledge_select_multi",
"label": "检索范围",
"value": {
"type": "knowledge",
"value": [
{
"key": 1283,
"label": "毕昇文档知识库"
}
]
},
"required": true,
"placeholder": "请选择知识库"
},
{
"key": "metadata_filter",
"type": "metadata_filter",
"label": "true",
"value": {
"enabled": false
}
},
{
"key": "advanced_retrieval_switch",
"type": "search_switch",
"label": "true",
"value": {
"user_auth": false,
"rerank_flag": false,
"rerank_model": "",
"search_switch": true,
"vector_weight": 0.5,
"keyword_weight": 0.5,
"max_chunk_size": 15000
}
},
{
"key": "retrieved_result",
"type": "var",
"label": "检索结果",
"global": "self=user_prompt"
}
]
},
{
"name": "AI回复生成设置",
"params": [
{
"key": "system_prompt",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是一个营销助手，专注于根据用户指令结合相关参考资料，提取关键信息和数据，帮助用户找到相关产品的功能和亮点，以便后续生成具有吸引力的营销文案。 ",
"required": true
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "# 用户需求\n{{#input_60a99.user_input#}}\n# 参考资料：\n{{#rag_20bed.retrieved_result#}}\n----\n\n产品功能亮点：",
"varZh": {
"input_60a99.user_input": "输入/user_input",
"rag_20bed.user_question": "user_question",
"rag_20bed.retrieved_result": "retrieved_result"
},
"required": true
},
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output_user_input",
"type": "var",
"label": "输出变量",
"value": [
{
"key": "output_user_input",
"label": "output_user_input"
}
],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 1275
},
"position": {
"x": 826.2000049926346,
"y": 237.0998165893505
},
"selected": true
},
{
"id": "rag_2b974",
"data": {
"v": 3,
"id": "rag_2b974",
"name": "文档知识库问答2",
"type": "rag",
"expand": false,
"description": "根据用户问题从知识库中检索相关内容，结合检索结果调用大模型生成最终结果，支持多个问题并行执行。",
"group_params": [
{
"name": "知识库检索设置",
"params": [
{
"key": "user_question",
"help": "当选择多个问题时，将会多次运行本节点，每次运行时从批量问题中取一项进行处理。",
"test": "var",
"type": "user_question",
"label": "用户问题",
"value": [
"input_60a99.user_input"
],
"varZh": {
"input_1936c.user_input": "输入/user_input",
"input_60a99.user_input": "输入/user_input"
},
"global": "self=user_prompt",
"linkage": "output_user_input",
"required": true,
"placeholder": "请选择用户问题"
},
{
"key": "knowledge",
"type": "knowledge_select_multi",
"label": "检索范围",
"value": {
"type": "knowledge",
"value": [
{
"key": 1301,
"label": "毕昇产品文档"
}
]
},
"required": true,
"placeholder": "请选择知识库"
},
{
"key": "metadata_filter",
"type": "metadata_filter",
"label": "true",
"value": {
"enabled": false
}
},
{
"key": "advanced_retrieval_switch",
"type": "search_switch",
"label": "true",
"value": {
"user_auth": false,
"rerank_flag": false,
"rerank_model": "",
"search_switch": true,
"vector_weight": 0.5,
"keyword_weight": 0.5,
"max_chunk_size": 15000
}
},
{
"key": "retrieved_result",
"type": "var",
"label": "检索结果",
"global": "self=user_prompt"
}
]
},
{
"name": "AI回复生成设置",
"params": [
{
"key": "system_prompt",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是一个具有多年资深经验的产品营销大师，专注于根据用户指令，结合相关参考资料，突出产品功能和亮点，生成富有创意、情感共鸣、具有吸引力的营销文案。",
"required": true
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "# 用户需求\n{{#input_60a99.user_input#}}\n# 参考资料\n{{#rag_2b974.retrieved_result#}}\n\n# 产品功能亮点\n{{#output_7d6b6.output_result#}}\n\n\n\n\n",
"varZh": {
"input_60a99.user_input": "输入/user_input",
"rag_2b974.user_question": "user_question",
"output_7d6b6.output_result": "输出2/output_result",
"rag_2b974.retrieved_result": "文档知识库问答2/retrieved_result"
},
"required": true
},
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output_user_input",
"type": "var",
"label": "输出变量",
"value": [
{
"key": "output_user_input",
"label": "output_user_input"
}
],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 150
},
"position": {
"x": 1733.012638288989,
"y": 241.24634805976703
},
"selected": false
},
{
"id": "llm_c66fe",
"data": {
"v": 2,
"id": "llm_c66fe",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型",
"type": "llm",
"expand": true,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是一个具有多年资深经验的产品营销大师，专注于根据用户部门修改建议，对原始营销文案进行修改润色，以帮助企业提升品牌知名度和销售业绩。 "
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "# 原文\n{{#rag_2b974.output_user_input#}}\n\n\n\n# 用户修改建议\n\n{{#output_84247.output_result#}}\n\n# 润色修改后文案\n\n\n\n\n\n\n",
"varZh": {
"output_84247.output_result": "输出3/output_result",
"rag_2b974.output_user_input": "文档知识库问答2/output_user_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 833
},
"position": {
"x": 3243.668406668528,
"y": 182.05670813263373
},
"selected": false
},
{
"id": "input_60a99",
"data": {
"v": "3",
"id": "input_60a99",
"tab": {
"value": "dialog_input",
"options": [
{
"key": "dialog_input",
"help": "true",
"label": "true"
},
{
"key": "form_input",
"help": "true",
"label": "true"
}
]
},
"name": "输入",
"type": "input",
"description": "接收用户在会话页面的输入，支持 2 种形式：对话框输入，表单输入。",
"group_params": [
{
"name": "接收文本",
"params": [
{
"key": "user_input",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
}
]
},
{
"name": "",
"params": [
{
"key": "user_input_file",
"tab": "dialog_input",
"value": true,
"groupTitle": true
},
{
"key": "file_parse_mode",
"tab": "dialog_input",
"type": "select_parsemode",
"value": "extract_text"
},
{
"key": "dialog_files_content",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_files_content_size",
"min": 0,
"tab": "dialog_input",
"type": "char_number",
"label": "true",
"value": 15000
},
{
"key": "dialog_file_accept",
"tab": "dialog_input",
"type": "select_fileaccept",
"label": "true",
"value": "all"
},
{
"key": "dialog_image_files",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_file_paths",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
}
],
"groupKey": "inputfile"
},
{
"name": "",
"params": [
{
"key": "recommended_questions_flag",
"tab": "dialog_input",
"help": "true",
"label": "true",
"value": false,
"hidden": "true",
"groupTitle": true
},
{
"key": "recommended_llm",
"tab": "dialog_input",
"type": "bisheng_model",
"label": "true",
"value": 261,
"required": true,
"placeholder": "true"
},
{
"key": "recommended_system_prompt",
"tab": "dialog_input",
"type": "var_textarea",
"label": "true",
"value": "# Role\n你是一个极具洞察力的\"对话延续预测专家\"。你的任务是根据当前的对话历史，预测用户接下来最有可能输入的 3 条短语或问题。\n\n# Context Awareness\n在生成预测前，请执行以下深度思考步骤：\n1. **意图分析**：用户当前是在寻求事实、表达情绪、请求建议，还是在尝试结束对话？\n2. **逻辑推演**：根据 AI 上一条回复的内容，用户最自然产生的联想或疑问是什么？\n3. **风格对齐**：观察用户的用词习惯（专业、随性、简短还是详细），预测结果需保持语气一致。\n\n# Goals\n- **多样性**：提供不同方向的后续（例如：一个深入探究的问题，一个相关的新话题，一个反馈性评价）。\n- **简洁性**：预测内容应短小精悍，通常不超过 15 个字，方便用户点击。\n- **前瞻性**：能够预判用户的\"痛点\"，直接提供解决方案的入口。\n\n# Output Format\n请严格遵守 JSON 格式输出。\n返回一个包含单一键名 \"suggestions\" 的 JSON 对象。\n- 所有字符串必须使用**双引号**。\n- 不要输出 Markdown 代码块标记（如 ```json），直接输出 JSON 字符串。\n**示例：**\n{\"suggestions\": [\"你能举个例子吗？\", \"这个方案的成本是多少？\", \"听起来不错，怎么开始？\"]}",
"required": true
},
{
"key": "recommended_history_num",
"tab": "dialog_input",
"help": "true",
"step": 1,
"type": "slide",
"label": "true",
"scope": [
1,
10
],
"value": 2
}
],
"groupKey": "custom"
},
{
"name": "",
"params": [
{
"key": "form_input",
"tab": "form_input",
"type": "form",
"label": "true",
"value": [],
"global": "item:form_input"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 570
},
"position": {
"x": 411.31336047282423,
"y": 234.7627660417411
},
"selected": false
}
],
"viewport": {
"x": -933.842567126535,
"y": -504.2152476262525,
"zoom": 1.4199819732946106
}
},
"order_num": 12500863,
"guide_word": null,
"flow_type": 10
},
{
"update_time": "2026-02-05 20:45:48",
"parameters": null,
"name": "分诊场景",
"description": "该流程用于分诊场景，用户输入内容后，经大模型判断是否完整，通过条件分支决策，再由大模型生成科室，输出分诊结果，辅助高效分诊 。\n使用说明\n在 “输入” 节点填内容，选 “大模型 - 判断是否完整” 配置参数运行，根据分支结果，经 “大模型 - 推荐科室” 处理，最后在 “输出” 查看分诊结论 。",
"flow_id": "7353cc101f434234bbee3a61eb17959d",
"api_parameters": null,
"create_time": "2026-02-05 20:45:48",
"id": 364,
"data": {
"edges": [
{
"id": "xy-edge__condition_63775right_handle-output_15b82left_handle",
"type": "customEdge",
"source": "condition_63775",
"target": "output_15b82",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_8bbeeright_handle-condition_63775left_handle",
"type": "customEdge",
"source": "llm_8bbee",
"target": "condition_63775",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__condition_637750eda50f7-llm_2cf38left_handle",
"type": "customEdge",
"source": "condition_63775",
"target": "llm_2cf38",
"animated": true,
"sourceHandle": "0eda50f7",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_2cf38right_handle-end_1341eleft_handle",
"type": "customEdge",
"source": "llm_2cf38",
"target": "end_1341e",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__start_68c76right_handle-input_85ed8left_handle",
"type": "customEdge",
"source": "start_68c76",
"target": "input_85ed8",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_85ed8right_handle-llm_8bbeeleft_handle",
"type": "customEdge",
"source": "input_85ed8",
"target": "llm_8bbee",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__output_15b82right_handle-input_85ed8left_handle",
"type": "customEdge",
"source": "output_15b82",
"target": "input_85ed8",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
}
],
"nodes": [
{
"id": "start_68c76",
"data": {
"v": 3,
"id": "start_68c76",
"name": "开始",
"type": "start",
"description": "工作流运行的起始节点",
"group_params": [
{
"name": "开场引导",
"params": [
{
"key": "guide_word",
"type": "textarea",
"label": "开场白",
"value": "您好，请描述您的症状，并且提供您的性别、年龄信息。"
},
{
"key": "guide_question",
"help": "为用户提供推荐问题，引导用户输入，超过3个时将随机选取3个",
"type": "input_list",
"label": "引导问题",
"value": [
""
]
}
]
},
{
"name": "全局变量",
"params": [
{
"key": "user_info",
"type": "var",
"label": "true",
"value": "",
"global": "key"
},
{
"key": "current_time",
"type": "var",
"label": "当前时间",
"value": "",
"global": "key"
},
{
"key": "chat_history",
"type": "chat_history_num",
"value": 10,
"global": "key"
},
{
"key": "preset_question",
"type": "input_list",
"label": "预置问题列表",
"value": [
{
"key": "366504",
"value": ""
}
],
"global": "item:input_list"
},
{
"key": "custom_variables",
"help": "true",
"type": "global_var",
"label": "true",
"value": [],
"global": "item:input_list"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 711
},
"position": {
"x": -283.54706809376773,
"y": -86.40776212614759
},
"selected": false
},
{
"id": "condition_63775",
"data": {
"id": "condition_63775",
"name": "条件分支",
"type": "condition",
"description": "根据条件表达式执行不同的分支",
"group_params": [
{
"params": [
{
"key": "condition",
"type": "condition",
"label": "",
"value": [
{
"id": "0eda50f7",
"operator": "and",
"conditions": [
{
"id": "b78beaf9",
"left_var": "llm_8bbee.output",
"left_label": "output",
"right_label": "",
"right_value": "完整",
"right_value_type": "input",
"comparison_operation": "contains"
}
]
}
]
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 545,
"height": 347
},
"position": {
"x": 1031.063648112229,
"y": -84.34903331683131
},
"selected": false
},
{
"id": "end_1341e",
"data": {
"id": "end_1341e",
"name": "结束",
"type": "end",
"description": "工作流运行到此结束",
"group_params": []
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2099.286704970077,
"y": -113.81885566422974
},
"selected": false
},
{
"id": "output_15b82",
"data": {
"v": 2,
"id": "output_15b82",
"name": "输出",
"type": "output",
"description": "可向用户发送消息，并且支持进行更丰富的交互，例如请求用户批准进行某项敏感操作、允许用户在模型输出内容的基础上直接修改并提交。",
"group_params": [
{
"params": [
{
"key": "message",
"type": "var_textarea_file",
"label": "消息内容",
"value": {
"msg": "{{#llm_8bbee.output#}}",
"files": []
},
"varZh": {
"llm_8bbee.output": "大模型-判断信息是否完整/output",
"llm_a46d8.output": "大模型/output"
},
"global": "key",
"required": true,
"placeholder": "输入需要发送给用户的消息，例如“接下来我将执行 XX 操作，请您确认”，“以下是我的初版草稿，您可以在其基础上进行修改”"
},
{
"key": "output_result",
"type": "output_form",
"label": "用户输入内容",
"value": {
"type": "",
"value": ""
},
"global": "value.type=input",
"options": [],
"required": false
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 463
},
"position": {
"x": 1681.5053754933697,
"y": 244.61489030268024
},
"selected": false
},
{
"id": "llm_8bbee",
"data": {
"v": 2,
"id": "llm_8bbee",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型-判断信息是否完整",
"type": "llm",
"expand": true,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "您是一名分诊医生。请判断以下用户提供的信息是否完整，需要包含性别、年龄、症状三项信息。若用户提供了完整信息则回复“信息完整”，若不完整则反问用户，让对方补充相应缺失的信息。"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "你们的对话记录：\n\n```\n\n{{#start_68c76.chat_history#}}\n\n```\n\n你的回复：\n\n\n\n",
"varZh": {
"start_68c76.chat_history": "开始/chat_history"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": false
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 833
},
"position": {
"x": 622.8451256124496,
"y": -74.78472706554999
},
"selected": false
},
{
"id": "llm_2cf38",
"data": {
"v": 2,
"id": "llm_2cf38",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型-推荐科室",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "1. 根据提供的性别、年龄和症状，判断患者应该去医院的哪个科室就诊。\n\n2. 首先，分析患者的性别和年龄，这可能会影响某些症状的严重性和可能的科室选择。\n\n3. 然后，仔细阅读并理解患者的症状，确定这些症状可能与哪些科室相关。\n\n4. 根据常见的医疗科室和其对应的症状，匹配患者的症状到最合适的科室。\n\n5. 输出结果时，只需提供科室名称，不要包含任何XML标签。"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "完整的用户信息包含在以下聊天记录中：\n\n```\n\n{{#start_68c76.chat_history#}}\n\n```\n\n推荐科室：\n\n\n\n",
"varZh": {
"start_68c76.chat_history": "开始/chat_history"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 1686.1706694671357,
"y": -81.72638198724074
},
"selected": false
},
{
"id": "input_85ed8",
"data": {
"v": "3",
"id": "input_85ed8",
"tab": {
"value": "dialog_input",
"options": [
{
"key": "dialog_input",
"help": "true",
"label": "true"
},
{
"key": "form_input",
"help": "true",
"label": "true"
}
]
},
"name": "输入",
"type": "input",
"description": "接收用户在会话页面的输入，支持 2 种形式：对话框输入，表单输入。",
"group_params": [
{
"name": "接收文本",
"params": [
{
"key": "user_input",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
}
]
},
{
"name": "",
"params": [
{
"key": "user_input_file",
"tab": "dialog_input",
"value": true,
"groupTitle": true
},
{
"key": "file_parse_mode",
"tab": "dialog_input",
"type": "select_parsemode",
"value": "extract_text"
},
{
"key": "dialog_files_content",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_files_content_size",
"min": 0,
"tab": "dialog_input",
"type": "char_number",
"label": "true",
"value": 15000
},
{
"key": "dialog_file_accept",
"tab": "dialog_input",
"type": "select_fileaccept",
"label": "true",
"value": "all"
},
{
"key": "dialog_image_files",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_file_paths",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
}
],
"groupKey": "inputfile"
},
{
"name": "",
"params": [
{
"key": "recommended_questions_flag",
"tab": "dialog_input",
"help": "true",
"label": "true",
"value": false,
"hidden": "true",
"groupTitle": true
},
{
"key": "recommended_llm",
"tab": "dialog_input",
"type": "bisheng_model",
"label": "true",
"value": 261,
"required": true,
"placeholder": "true"
},
{
"key": "recommended_system_prompt",
"tab": "dialog_input",
"type": "var_textarea",
"label": "true",
"value": "# Role\n你是一个极具洞察力的\"对话延续预测专家\"。你的任务是根据当前的对话历史，预测用户接下来最有可能输入的 3 条短语或问题。\n\n# Context Awareness\n在生成预测前，请执行以下深度思考步骤：\n1. **意图分析**：用户当前是在寻求事实、表达情绪、请求建议，还是在尝试结束对话？\n2. **逻辑推演**：根据 AI 上一条回复的内容，用户最自然产生的联想或疑问是什么？\n3. **风格对齐**：观察用户的用词习惯（专业、随性、简短还是详细），预测结果需保持语气一致。\n\n# Goals\n- **多样性**：提供不同方向的后续（例如：一个深入探究的问题，一个相关的新话题，一个反馈性评价）。\n- **简洁性**：预测内容应短小精悍，通常不超过 15 个字，方便用户点击。\n- **前瞻性**：能够预判用户的\"痛点\"，直接提供解决方案的入口。\n\n# Output Format\n请严格遵守 JSON 格式输出。\n返回一个包含单一键名 \"suggestions\" 的 JSON 对象。\n- 所有字符串必须使用**双引号**。\n- 不要输出 Markdown 代码块标记（如 ```json），直接输出 JSON 字符串。\n**示例：**\n{\"suggestions\": [\"你能举个例子吗？\", \"这个方案的成本是多少？\", \"听起来不错，怎么开始？\"]}",
"required": true
},
{
"key": "recommended_history_num",
"tab": "dialog_input",
"help": "true",
"step": 1,
"type": "slide",
"label": "true",
"scope": [
1,
10
],
"value": 2
}
],
"groupKey": "custom"
},
{
"name": "",
"params": [
{
"key": "form_input",
"tab": "form_input",
"type": "form",
"label": "true",
"value": [],
"global": "item:form_input"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 570
},
"position": {
"x": 151.22647818562317,
"y": -68.89763571499358
},
"selected": false
}
],
"viewport": {
"x": 208.5212914927622,
"y": 177.1875221661041,
"zoom": 0.499110403236339
}
},
"order_num": 12631933,
"guide_word": null,
"flow_type": 10
},
{
"update_time": "2026-02-05 20:51:03",
"parameters": null,
"name": "工具调用",
"description": "功能简介\n该流程用于内容处理，经输入、大模型处理、PMI 分析等环节，利用大模型完成任务，输出处理后的内容，支持单 / 批量运行，实现文本内容的智能化加工 。\n使用说明\n选择输入方式（对话 / 文件），配置大模型参数，设置 PMI 工具参数，确认各节点连接，点击 “运行”，流程自动执行，输出结果可查看调用大模型处理后的内容 。\n",
"flow_id": "7d820dd5dbce46b4b38502af34d8fccb",
"api_parameters": null,
"create_time": "2026-02-05 20:51:03",
"id": 366,
"data": {
"edges": [
{
"id": "xy-edge__tool_7a316right_handle-output_45121left_handle",
"type": "customEdge",
"source": "tool_7a316",
"target": "output_45121",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_0c3fbright_handle-tool_7a316left_handle",
"type": "customEdge",
"source": "llm_0c3fb",
"target": "tool_7a316",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_148e4right_handle-tool_7a316left_handle",
"type": "customEdge",
"source": "llm_148e4",
"target": "tool_7a316",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__output_45121right_handle-llm_0fa85left_handle",
"type": "customEdge",
"source": "output_45121",
"target": "llm_0fa85",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__start_3d920right_handle-input_4c88bleft_handle",
"type": "customEdge",
"source": "start_3d920",
"target": "input_4c88b",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_4c88bright_handle-llm_0c3fbleft_handle",
"type": "customEdge",
"source": "input_4c88b",
"target": "llm_0c3fb",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_4c88bright_handle-llm_148e4left_handle",
"type": "customEdge",
"source": "input_4c88b",
"target": "llm_148e4",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_0fa85right_handle-input_4c88bleft_handle",
"type": "customEdge",
"source": "llm_0fa85",
"target": "input_4c88b",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
}
],
"nodes": [
{
"id": "start_3d920",
"data": {
"v": 3,
"id": "start_3d920",
"name": "开始",
"type": "start",
"expand": false,
"description": "工作流运行的起始节点。",
"group_params": [
{
"name": "开场引导",
"params": [
{
"key": "guide_word",
"type": "textarea",
"label": "开场白",
"value": "",
"placeholder": "每次工作流开始执行时向用户发送此消息，支持 Markdown 格式，为空时不发送。"
},
{
"key": "guide_question",
"help": "为用户提供推荐问题，引导用户输入，超过3个时将随机选取3个。",
"type": "input_list",
"label": "引导问题",
"value": [
"2024年上半年PMI变化情况",
""
],
"placeholder": "请输入引导问题"
}
]
},
{
"name": "全局变量",
"params": [
{
"key": "user_info",
"type": "var",
"label": "true",
"value": "",
"global": "key"
},
{
"key": "current_time",
"type": "var",
"label": "当前时间",
"value": "",
"global": "key"
},
{
"key": "chat_history",
"type": "chat_history_num",
"value": 10,
"global": "key"
},
{
"key": "preset_question",
"help": "适合文档审核、报告生成等场景，利用提前预置的问题批量进行 RAG 问答。",
"type": "input_list",
"label": "预置问题列表",
"value": [
{
"key": "4b91bc",
"value": ""
}
],
"global": "item:input_list",
"placeholder": "输入批量预置问题"
},
{
"key": "custom_variables",
"help": "true",
"type": "global_var",
"label": "true",
"value": [],
"global": "item:input_list"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 293.65992364972067,
"y": 74.81260064793756
},
"selected": false
},
{
"id": "tool_7a316",
"data": {
"id": "tool_7a316",
"name": "PMI",
"type": "tool",
"expand": true,
"tool_key": "macro_china_pmi",
"description": "中国 PMI （采购经理人指数）月度统计数据。返回数据包括：月份制造业 PMI，制造业 PMI 同比增长，非制造业 PMI，非制造业 PMI 同比增长。",
"group_params": [
{
"name": "工具参数",
"params": [
{
"key": "start_date",
"desc": "开始月份, 使用YYYY-MM-DD 方式表示",
"test": "input",
"type": "var_textarea",
"label": "start_date",
"value": "{{#llm_0c3fb.output#}}",
"varZh": {
"llm_0c3fb.output": "大模型-提取开始时间/output"
},
"required": true
},
{
"key": "end_date",
"desc": "结束月份, 使用YYYY-MM-DD 方式表示",
"test": "input",
"type": "var_textarea",
"label": "end_date",
"value": "{{#llm_148e4.output#}}",
"varZh": {
"llm_148e4.output": "大模型-提取结束时间/output"
},
"required": true
}
]
},
{
"name": "输出",
"params": [
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": "",
"global": "key"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 602
},
"position": {
"x": 1614.4895326144945,
"y": 100
},
"selected": false
},
{
"id": "output_45121",
"data": {
"v": 2,
"id": "output_45121",
"name": "输出",
"type": "output",
"description": "可向用户发送消息，并且支持进行更丰富的交互，例如请求用户批准进行某项敏感操作、允许用户在模型输出内容的基础上直接修改并提交。",
"group_params": [
{
"params": [
{
"key": "message",
"type": "var_textarea_file",
"label": "消息内容",
"value": {
"msg": "{{#tool_7a316.output#}}",
"files": []
},
"varZh": {
"tool_7a316.output": "PMI/output",
"start_3d920.current_time": "开始/current_time"
},
"global": "key",
"required": true,
"placeholder": "输入需要发送给用户的消息，例如“接下来我将执行 XX 操作，请您确认”，“以下是我的初版草稿，您可以在其基础上进行修改”"
},
{
"key": "output_result",
"type": "output_form",
"label": "用户输入内容",
"value": {
"type": "",
"value": ""
},
"global": "value.type=input",
"options": [],
"required": false
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 463
},
"position": {
"x": 2032.2233877193396,
"y": 115.19445604716492
},
"selected": false
},
{
"id": "llm_0c3fb",
"data": {
"v": 2,
"id": "llm_0c3fb",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型-提取开始时间",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "从用户的输入中抽取出开始时间，格式：YYYY-MM-DD，仅输出时间，不要包含其他内容"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "{{#input_4c88b.user_input#}}",
"varZh": {
"input_4c88b.user_input": "输入/user_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "true",
"type": "image_prompt",
"label": "true",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 1171.405788440929,
"y": 76.59789148675313
},
"selected": false
},
{
"id": "llm_148e4",
"data": {
"v": 2,
"id": "llm_148e4",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型-提取结束时间",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "从用户的输入中抽取出查询的结束时间，格式：YYYY-MM-DD，仅输出时间，不要包含其他内容"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "{{#input_4c88b.user_input#}}",
"varZh": {
"input_4c88b.user_input": "输入/user_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "true",
"type": "image_prompt",
"label": "true",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 1175.5581100672648,
"y": 267.4853845047534
},
"selected": true
},
{
"id": "llm_0fa85",
"data": {
"v": 2,
"id": "llm_0fa85",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型",
"type": "llm",
"expand": true,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是一名宏观经济学家，请点评以下经济数据。"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "{{#tool_7a316.output#}}",
"varZh": {
"tool_7a316.output": "PMI/output"
},
"required": true
},
{
"key": "image_prompt",
"help": "true",
"type": "image_prompt",
"label": "true",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 833
},
"position": {
"x": 2457.846463360472,
"y": 114.79561323887066
},
"selected": false
},
{
"id": "input_4c88b",
"data": {
"v": "3",
"id": "input_4c88b",
"tab": {
"value": "dialog_input",
"options": [
{
"key": "dialog_input",
"help": "true",
"label": "true"
},
{
"key": "form_input",
"help": "true",
"label": "true"
}
]
},
"name": "输入",
"type": "input",
"description": "接收用户在会话页面的输入，支持 2 种形式：对话框输入，表单输入。",
"group_params": [
{
"name": "接收文本",
"params": [
{
"key": "user_input",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
}
]
},
{
"name": "",
"params": [
{
"key": "user_input_file",
"tab": "dialog_input",
"value": true,
"groupTitle": true
},
{
"key": "file_parse_mode",
"tab": "dialog_input",
"type": "select_parsemode",
"value": "extract_text"
},
{
"key": "dialog_files_content",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_files_content_size",
"min": 0,
"tab": "dialog_input",
"type": "char_number",
"label": "true",
"value": 15000
},
{
"key": "dialog_file_accept",
"tab": "dialog_input",
"type": "select_fileaccept",
"label": "true",
"value": "all"
},
{
"key": "dialog_image_files",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_file_paths",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
}
],
"groupKey": "inputfile"
},
{
"name": "",
"params": [
{
"key": "recommended_questions_flag",
"tab": "dialog_input",
"help": "true",
"label": "true",
"value": false,
"hidden": "true",
"groupTitle": true
},
{
"key": "recommended_llm",
"tab": "dialog_input",
"type": "bisheng_model",
"label": "true",
"value": 261,
"required": true,
"placeholder": "true"
},
{
"key": "recommended_system_prompt",
"tab": "dialog_input",
"type": "var_textarea",
"label": "true",
"value": "# Role\n你是一个极具洞察力的\"对话延续预测专家\"。你的任务是根据当前的对话历史，预测用户接下来最有可能输入的 3 条短语或问题。\n\n# Context Awareness\n在生成预测前，请执行以下深度思考步骤：\n1. **意图分析**：用户当前是在寻求事实、表达情绪、请求建议，还是在尝试结束对话？\n2. **逻辑推演**：根据 AI 上一条回复的内容，用户最自然产生的联想或疑问是什么？\n3. **风格对齐**：观察用户的用词习惯（专业、随性、简短还是详细），预测结果需保持语气一致。\n\n# Goals\n- **多样性**：提供不同方向的后续（例如：一个深入探究的问题，一个相关的新话题，一个反馈性评价）。\n- **简洁性**：预测内容应短小精悍，通常不超过 15 个字，方便用户点击。\n- **前瞻性**：能够预判用户的\"痛点\"，直接提供解决方案的入口。\n\n# Output Format\n请严格遵守 JSON 格式输出。\n返回一个包含单一键名 \"suggestions\" 的 JSON 对象。\n- 所有字符串必须使用**双引号**。\n- 不要输出 Markdown 代码块标记（如 ```json），直接输出 JSON 字符串。\n**示例：**\n{\"suggestions\": [\"你能举个例子吗？\", \"这个方案的成本是多少？\", \"听起来不错，怎么开始？\"]}",
"required": true
},
{
"key": "recommended_history_num",
"tab": "dialog_input",
"help": "true",
"step": 1,
"type": "slide",
"label": "true",
"scope": [
1,
10
],
"value": 2
}
],
"groupKey": "custom"
},
{
"name": "",
"params": [
{
"key": "form_input",
"tab": "form_input",
"type": "form",
"label": "true",
"value": [],
"global": "item:form_input"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 570
},
"position": {
"x": 754.2400379031777,
"y": 74.30568610625028
},
"selected": false
}
],
"viewport": {
"x": -111.39676647009968,
"y": -49.532665393166326,
"zoom": 0.5427937339527105
}
},
"order_num": 12763003,
"guide_word": null,
"flow_type": 10
},
{
"update_time": "2026-02-05 20:57:04",
"parameters": null,
"name": "智能简历筛选",
"description": "上传简历文件触发筛选",
"flow_id": "7576761af1ae4ea2b6f401b1e0953a57",
"api_parameters": null,
"create_time": "2026-02-05 20:57:04",
"id": 367,
"data": {
"edges": [
{
"id": "xy-edge__condition_9afcb28a37878-llm_60dfcleft_handle",
"type": "customEdge",
"source": "condition_9afcb",
"target": "llm_60dfc",
"animated": true,
"sourceHandle": "28a37878",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_60dfcright_handle-output_08e3dleft_handle",
"type": "customEdge",
"source": "llm_60dfc",
"target": "output_08e3d",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__output_08e3dright_handle-end_ed827left_handle",
"type": "customEdge",
"source": "output_08e3d",
"target": "end_ed827",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_025afright_handle-output_dc624left_handle",
"type": "customEdge",
"source": "llm_025af",
"target": "output_dc624",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__output_dc624right_handle-end_ed827left_handle",
"type": "customEdge",
"source": "output_dc624",
"target": "end_ed827",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__rag_743b7right_handle-condition_9afcbleft_handle",
"type": "customEdge",
"source": "rag_743b7",
"target": "condition_9afcb",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__condition_9afcbright_handle-rag_6c446left_handle",
"type": "customEdge",
"source": "condition_9afcb",
"target": "rag_6c446",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__rag_6c446right_handle-llm_025afleft_handle",
"type": "customEdge",
"source": "rag_6c446",
"target": "llm_025af",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__start_d0027right_handle-input_51222left_handle",
"type": "customEdge",
"source": "start_d0027",
"target": "input_51222",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_51222right_handle-rag_743b7left_handle",
"type": "customEdge",
"source": "input_51222",
"target": "rag_743b7",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
}
],
"nodes": [
{
"id": "start_d0027",
"data": {
"v": 3,
"id": "start_d0027",
"name": "开始",
"type": "start",
"description": "工作流运行的起始节点。",
"group_params": [
{
"name": "开场引导",
"params": [
{
"key": "guide_word",
"type": "textarea",
"label": "开场白",
"value": "",
"placeholder": "每次工作流开始执行时向用户发送此消息，支持 Markdown 格式，为空时不发送。"
},
{
"key": "guide_question",
"help": "为用户提供推荐问题，引导用户输入，超过3个时将随机选取3个。",
"type": "input_list",
"label": "引导问题",
"value": [
""
],
"placeholder": "请输入引导问题"
}
]
},
{
"name": "全局变量",
"params": [
{
"key": "user_info",
"type": "var",
"label": "true",
"value": "",
"global": "key"
},
{
"key": "current_time",
"type": "var",
"label": "当前时间",
"value": "",
"global": "key"
},
{
"key": "chat_history",
"type": "chat_history_num",
"value": 10,
"global": "key"
},
{
"key": "preset_question",
"help": "适合文档审核、报告生成等场景，利用提前预置的问题批量进行 RAG 问答。",
"type": "input_list",
"label": "预置问题列表",
"value": [
{
"key": "6e5db5",
"value": "需有AI领域从业经历或为计算机相关专业"
},
{
"key": "33572c",
"value": "该候选人总工作时长需要3年以上"
},
{
"key": "677ad0",
"value": "工作经验中不能有多段工作经历持续时长低于3个月"
},
{
"key": "622ff1",
"value": "会Python"
},
{
"key": "ce9333",
"value": "有大模型相关项目经验"
},
{
"key": "2b83e5",
"value": "30岁以下"
},
{
"key": "2a3382",
"value": ""
}
],
"global": "item:input_list",
"placeholder": "输入批量预置问题"
},
{
"key": "custom_variables",
"help": "true",
"type": "global_var",
"label": "true",
"value": [],
"global": "item:input_list"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 879
},
"position": {
"x": 183.82157147745,
"y": 152.22958521319305
},
"selected": false
},
{
"id": "condition_9afcb",
"data": {
"id": "condition_9afcb",
"name": "条件分支",
"type": "condition",
"description": "根据条件表达式执行不同的分支。",
"group_params": [
{
"params": [
{
"key": "condition",
"type": "condition",
"label": "",
"value": [
{
"id": "28a37878",
"operator": "or",
"conditions": [
{
"id": "475500db",
"left_var": "rag_743b7.output_preset_question_6e5db5",
"left_label": "判断候选人是否满足必要要求/output_需有AI领域从业经历或为计算机相关专业",
"right_label": "",
"right_value": "不满足",
"right_value_type": "input",
"comparison_operation": "contains"
},
{
"id": "65e2732f",
"left_var": "rag_743b7.output_preset_question_33572c",
"left_label": "判断候选人是否满足必要要求/output_该候选人总工作时长需要3年以上",
"right_label": "",
"right_value": "不满足",
"right_value_type": "input",
"comparison_operation": "contains"
},
{
"id": "316abb7b",
"left_var": "rag_743b7.output_preset_question_677ad0",
"left_label": "判断候选人是否满足必要要求/output_工作经验中不能有多段工作经历持续时长低于3个月",
"right_label": "",
"right_value": "不满足",
"right_value_type": "input",
"comparison_operation": "contains"
}
]
}
]
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 562,
"height": 419
},
"position": {
"x": 1488.6523613323277,
"y": 145.17597465560576
},
"selected": false
},
{
"id": "llm_60dfc",
"data": {
"v": 2,
"id": "llm_60dfc",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "总结未通过简历筛选原因",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "input",
"type": "var_textarea",
"label": "系统提示词",
"value": "以下是某个候选人简历评估的结果。由于但凡有一项“不满足”则无法进入面试环节。请简要（50字以内）总结一下该候选人简历未通过筛选的原因。",
"varZh": {
"rag_49431.output_preset_question_0": "判断候选人是否满足必要条件/output_需有AI领域从业经历或为计算机相关专业"
}
},
{
"key": "user_prompt",
"test": "input",
"type": "var_textarea",
"label": "用户提示词",
"value": "要求1满足情况：{{#rag_743b7.output_preset_question_6e5db5#}}要求2满足情况：{{#rag_743b7.output_preset_question_33572c#}}\n要求3满足情况：{{#rag_743b7.output_preset_question_677ad0#}}",
"varZh": {
"rag_49431.output_preset_question_1": "判断候选人是否满足必要要求/output_需要有5年及以上工作经验",
"rag_98e0c.output_preset_question_0": "判断候选人是否满足必要条件/output_需有AI领域从业经历或为计算机相关专业",
"rag_98e0c.output_preset_question_1": "判断候选人是否满足必要条件/output_需要有5年及以上工作经验",
"rag_98e0c.output_preset_question_2": "判断候选人是否满足必要条件/output_工作经验中不能有多段工作经历持续时长低于3个月",
"rag_743b7.output_preset_question_33572c": "判断候选人是否满足必要要求/output_该候选人总工作时长需要3年以上",
"rag_743b7.output_preset_question_677ad0": "判断候选人是否满足必要要求/output_工作经验中不能有多段工作经历持续时长低于3个月",
"rag_743b7.output_preset_question_6e5db5": "判断候选人是否满足必要要求/output_需有AI领域从业经历或为计算机相关专业"
},
"required": true
},
{
"key": "image_prompt",
"help": "true",
"type": "image_prompt",
"label": "true",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": false
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2224.0533822371517,
"y": 147.86184733909764
},
"selected": true
},
{
"id": "output_08e3d",
"data": {
"v": 2,
"id": "output_08e3d",
"name": "输出",
"type": "output",
"description": "可向用户发送消息，并且支持进行更丰富的交互，例如请求用户批准进行某项敏感操作、允许用户在模型输出内容的基础上直接修改并提交。",
"group_params": [
{
"params": [
{
"key": "message",
"type": "var_textarea_file",
"label": "消息内容",
"value": {
"msg": "简历筛选未通过。\n\n{{#llm_60dfc.output#}}",
"files": []
},
"varZh": {
"llm_60dfc.output": "总结未通过简历筛选原因/output"
},
"global": "key",
"required": true,
"placeholder": "输入需要发送给用户的消息，例如“接下来我将执行 XX 操作，请您确认”，“以下是我的初版草稿，您可以在其基础上进行修改”"
},
{
"key": "output_result",
"type": "output_form",
"label": "用户输入内容",
"value": {
"type": "",
"value": ""
},
"global": "value.type=input",
"options": [],
"required": false
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 463
},
"position": {
"x": 2625.826899106754,
"y": 151.65379821350757
},
"selected": false
},
{
"id": "end_ed827",
"data": {
"id": "end_ed827",
"name": "结束",
"type": "end",
"description": "工作流运行到此结束。",
"group_params": []
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 3462.9772003335,
"y": 159.78541998491698
},
"selected": false
},
{
"id": "llm_025af",
"data": {
"v": 2,
"id": "llm_025af",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "加分项算分",
"type": "llm",
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "input",
"type": "var_textarea",
"label": "系统提示词",
"value": "以下是某个候选人简历评估中加分项的评估结果。每个“满足”加2分，每个“不满足”扣一分，“不确定”不加不减。输出最后的分数，仅输出数字即可。"
},
{
"key": "user_prompt",
"test": "input",
"type": "var_textarea",
"label": "用户提示词",
"value": "要求1满足情况：{{#rag_6c446.output_preset_question_622ff1#}}要求2满足情况：{{#rag_6c446.output_preset_question_ce9333#}}要求3满足情况：{{#rag_6c446.output_preset_question_2b83e5#}}",
"varZh": {
"rag_882f4.output_preset_question_3": "判断加分项/output_会Python",
"rag_882f4.output_preset_question_4": "判断加分项/output_有大模型相关项目经验",
"rag_882f4.output_preset_question_5": "判断加分项/output_30岁以下",
"rag_6c446.output_preset_question_2b83e5": "判断加分情况/output_30岁以下",
"rag_6c446.output_preset_question_622ff1": "判断加分情况/output_会Python",
"rag_6c446.output_preset_question_ce9333": "判断加分情况/output_有大模型相关项目经验"
},
"required": true
},
{
"key": "image_prompt",
"help": "true",
"type": "image_prompt",
"label": "true",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": false
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 833
},
"position": {
"x": 2630.260369780459,
"y": 630.0457897653762
},
"selected": false
},
{
"id": "output_dc624",
"data": {
"v": 2,
"id": "output_dc624",
"name": "输出2",
"type": "output",
"description": "可向用户发送消息，并且支持进行更丰富的交互，例如请求用户批准进行某项敏感操作、允许用户在模型输出内容的基础上直接修改并提交。",
"group_params": [
{
"params": [
{
"key": "message",
"type": "var_textarea_file",
"label": "消息内容",
"value": {
"msg": "该候选人得分：{{#llm_025af.output#}}",
"files": []
},
"varZh": {
"llm_025af.output": "加分项算分/output"
},
"global": "key",
"required": true,
"placeholder": "输入需要发送给用户的消息，例如“接下来我将执行 XX 操作，请您确认”，“以下是我的初版草稿，您可以在其基础上进行修改”"
},
{
"key": "output_result",
"type": "output_form",
"label": "用户输入内容",
"value": {
"type": "",
"value": ""
},
"global": "value.type=input",
"options": [],
"required": false
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 463
},
"position": {
"x": 3030,
"y": 631.2632435428188
},
"selected": false
},
{
"id": "rag_743b7",
"data": {
"v": 3,
"id": "rag_743b7",
"name": "判断候选人是否满足必要要求",
"type": "rag",
"description": "根据用户问题从知识库中检索相关内容，结合检索结果调用大模型生成最终结果，支持多个问题并行执行。",
"group_params": [
{
"name": "知识库检索设置",
"params": [
{
"key": "user_question",
"help": "当选择多个问题时，将会多次运行本节点，每次运行时从批量问题中取一项进行处理。",
"test": "var",
"type": "user_question",
"label": "用户问题",
"value": [
"start_d0027.preset_question#6e5db5",
"start_d0027.preset_question#33572c",
"start_d0027.preset_question#677ad0"
],
"varZh": {
"start_d0027.preset_question#33572c": "开始/该候选人总工作时长需要3年以上",
"start_d0027.preset_question#677ad0": "开始/工作经验中不能有多段工作经历持续时长低于3个月",
"start_d0027.preset_question#6e5db5": "开始/需有AI领域从业经历或为计算机相关专业"
},
"global": "self=user_prompt",
"linkage": "output_user_input",
"required": true,
"placeholder": "请选择用户问题"
},
{
"key": "knowledge",
"type": "knowledge_select_multi",
"label": "检索范围",
"value": {
"type": "tmp",
"value": [
{
"key": "input_51222.file",
"label": "file(简历)"
}
]
},
"required": true,
"placeholder": "请选择知识库"
},
{
"key": "metadata_filter",
"type": "metadata_filter",
"label": "true",
"value": {
"enabled": false
}
},
{
"key": "advanced_retrieval_switch",
"type": "search_switch",
"label": "true",
"value": {
"user_auth": false,
"rerank_flag": false,
"rerank_model": "",
"search_switch": true,
"vector_weight": 0.5,
"keyword_weight": 0.5,
"max_chunk_size": 15000
}
},
{
"key": "retrieved_result",
"type": "var",
"label": "检索结果",
"global": "self=user_prompt"
}
]
},
{
"name": "AI回复生成设置",
"params": [
{
"key": "system_prompt",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是一名HR，需要你根据审核项对简历进行审核。审核结论输出符合以下要求：1、包含“本项审核结论”、“结论依据”两项内容。2、“本项审核结论”只输出“满足”、“不满足”或“不确定”3、“结论依据”即对输出该结论的解释，解释中不要包含“满足”、“不满足”或“不确定”字眼。",
"required": true
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "审核项：{{#rag_743b7.user_question#}}\n参考文本：{{#rag_743b7.retrieved_result#}}\n审核结果：",
"varZh": {
"rag_743b7.user_question": "user_question",
"rag_743b7.retrieved_result": "retrieved_result"
},
"required": true
},
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output_user_input",
"help": "模型输出内容将会存储在该变量中。",
"type": "var",
"label": "输出变量",
"value": [
{
"key": "output_preset_question_6e5db5",
"label": "output_需有AI领域从业经历或为计算机相关专业"
},
{
"key": "output_preset_question_33572c",
"label": "output_该候选人总工作时长需要3年以上"
},
{
"key": "output_preset_question_677ad0",
"label": "output_工作经验中不能有多段工作经历持续时长低于3个月"
}
],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 1421
},
"position": {
"x": 1046.492501246415,
"y": 150.24255315187906
},
"selected": false
},
{
"id": "rag_6c446",
"data": {
"v": 3,
"id": "rag_6c446",
"name": "判断加分情况",
"type": "rag",
"description": "根据用户问题从知识库中检索相关内容，结合检索结果调用大模型生成最终结果，支持多个问题并行执行。",
"group_params": [
{
"name": "知识库检索设置",
"params": [
{
"key": "user_question",
"help": "当选择多个问题时，将会多次运行本节点，每次运行时从批量问题中取一项进行处理。",
"test": "var",
"type": "user_question",
"label": "用户问题",
"value": [
"start_d0027.preset_question#622ff1",
"start_d0027.preset_question#ce9333",
"start_d0027.preset_question#2b83e5"
],
"varZh": {
"start_d0027.preset_question#2b83e5": "开始/30岁以下",
"start_d0027.preset_question#622ff1": "开始/会Python",
"start_d0027.preset_question#ce9333": "开始/有大模型相关项目经验"
},
"global": "self=user_prompt",
"linkage": "output_user_input",
"required": true,
"placeholder": "请选择用户问题"
},
{
"key": "knowledge",
"type": "knowledge_select_multi",
"label": "检索范围",
"value": {
"type": "tmp",
"value": [
{
"key": "input_51222.file",
"label": "file(简历)"
}
]
},
"required": true,
"placeholder": "请选择知识库"
},
{
"key": "metadata_filter",
"type": "metadata_filter",
"label": "true",
"value": {
"enabled": false
}
},
{
"key": "advanced_retrieval_switch",
"type": "search_switch",
"label": "true",
"value": {
"user_auth": false,
"rerank_flag": false,
"rerank_model": "",
"search_switch": true,
"vector_weight": 0.5,
"keyword_weight": 0.5,
"max_chunk_size": 15000
}
},
{
"key": "retrieved_result",
"type": "var",
"label": "检索结果",
"global": "self=user_prompt"
}
]
},
{
"name": "AI回复生成设置",
"params": [
{
"key": "system_prompt",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是一名HR，需要你根据审核项对简历进行审核。审核结论输出符合以下要求：1、包含“本项审核结论”、“结论依据”两项内容。2、“本项审核结论”只输出“满足”、“不满足”或“不确定”3、“结论依据”即对输出该结论的解释，解释中不要包含“满足”、“不满足”或“不确定”字眼。",
"required": true
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "用户问题：{{#rag_6c446.user_question#}}\n参考文本：{{#rag_6c446.retrieved_result#}}\n你的回答：",
"varZh": {
"rag_6c446.user_question": "user_question",
"rag_6c446.retrieved_result": "retrieved_result"
},
"required": true
},
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output_user_input",
"help": "模型输出内容将会存储在该变量中。",
"type": "var",
"label": "输出变量",
"value": [
{
"key": "output_preset_question_622ff1",
"label": "output_会Python"
},
{
"key": "output_preset_question_ce9333",
"label": "output_有大模型相关项目经验"
},
{
"key": "output_preset_question_2b83e5",
"label": "output_30岁以下"
}
],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 1389
},
"position": {
"x": 2212.3843046432144,
"y": 548.5147332426619
},
"selected": false
},
{
"id": "input_51222",
"data": {
"v": "3",
"id": "input_51222",
"tab": {
"value": "form_input",
"options": [
{
"key": "dialog_input",
"help": "true",
"label": "true"
},
{
"key": "form_input",
"help": "true",
"label": "true"
}
]
},
"name": "输入",
"type": "input",
"description": "接收用户在会话页面的输入，支持 2 种形式：对话框输入，表单输入。",
"group_params": [
{
"name": "接收文本",
"params": [
{
"key": "user_input",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
}
]
},
{
"name": "",
"params": [
{
"key": "user_input_file",
"tab": "dialog_input",
"value": true,
"groupTitle": true
},
{
"key": "file_parse_mode",
"tab": "dialog_input",
"type": "select_parsemode",
"value": "extract_text"
},
{
"key": "dialog_files_content",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_files_content_size",
"min": 0,
"tab": "dialog_input",
"type": "char_number",
"label": "true",
"value": 15000
},
{
"key": "dialog_file_accept",
"tab": "dialog_input",
"type": "select_fileaccept",
"label": "true",
"value": "all"
},
{
"key": "dialog_image_files",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"value": {},
"global": "key"
},
{
"key": "dialog_file_paths",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"value": {},
"global": "key"
}
],
"groupKey": "inputfile"
},
{
"name": "",
"params": [
{
"key": "recommended_questions_flag",
"tab": "dialog_input",
"help": "true",
"label": "true",
"value": false,
"hidden": "true",
"groupTitle": true
},
{
"key": "recommended_llm",
"tab": "dialog_input",
"type": "bisheng_model",
"label": "true",
"value": 261,
"required": true,
"placeholder": "true"
},
{
"key": "recommended_system_prompt",
"tab": "dialog_input",
"type": "var_textarea",
"label": "true",
"value": "# Role\n你是一个极具洞察力的\"对话延续预测专家\"。你的任务是根据当前的对话历史，预测用户接下来最有可能输入的 3 条短语或问题。\n\n# Context Awareness\n在生成预测前，请执行以下深度思考步骤：\n1. **意图分析**：用户当前是在寻求事实、表达情绪、请求建议，还是在尝试结束对话？\n2. **逻辑推演**：根据 AI 上一条回复的内容，用户最自然产生的联想或疑问是什么？\n3. **风格对齐**：观察用户的用词习惯（专业、随性、简短还是详细），预测结果需保持语气一致。\n\n# Goals\n- **多样性**：提供不同方向的后续（例如：一个深入探究的问题，一个相关的新话题，一个反馈性评价）。\n- **简洁性**：预测内容应短小精悍，通常不超过 15 个字，方便用户点击。\n- **前瞻性**：能够预判用户的\"痛点\"，直接提供解决方案的入口。\n\n# Output Format\n请严格遵守 JSON 格式输出。\n返回一个包含单一键名 \"suggestions\" 的 JSON 对象。\n- 所有字符串必须使用**双引号**。\n- 不要输出 Markdown 代码块标记（如 ```json），直接输出 JSON 字符串。\n**示例：**\n{\"suggestions\": [\"你能举个例子吗？\", \"这个方案的成本是多少？\", \"听起来不错，怎么开始？\"]}",
"required": true
},
{
"key": "recommended_history_num",
"tab": "dialog_input",
"help": "true",
"step": 1,
"type": "slide",
"label": "true",
"scope": [
1,
10
],
"value": 2
}
],
"groupKey": "custom"
},
{
"name": "",
"params": [
{
"key": "form_input",
"tab": "form_input",
"type": "form",
"label": "true",
"value": [
{
"key": "file",
"type": "file",
"value": "简历",
"options": [],
"multiple": true,
"required": true,
"file_path": "",
"file_type": "all",
"image_file": "",
"file_content": "",
"file_parse_mode": "ingest_to_temp_kb",
"file_content_size": 15000
}
],
"global": "item:form_input"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 311
},
"position": {
"x": 591.9687179846699,
"y": 149.54936212710822
},
"selected": false
}
],
"viewport": {
"x": -638.4131557718965,
"y": -49.3950076425578,
"zoom": 0.5209290936198073
}
},
"order_num": 12828538,
"guide_word": null,
"flow_type": 10
},
{
"update_time": "2026-02-05 21:24:46",
"parameters": null,
"name": "文章助手",
"description": "功能简介\n这是 “文章助手” 工作流，用于文章创作优化。先通过输入明确创作需求，经条件分支判断是否需优化、是知识错误还是缺漏、哪种优化方向，调用大模型完成对应改写、纠错、补漏等操作，辅助产出高质量文章 。\n使用说明\n在 “输入” 节点填创作意图、内容方向，依次运行各节点。按条件分支逻辑，根据需求走 “改写优化”“检查错误 / 缺漏” 等分支，调用对应大模型工具，最终输出优化后的文章内容 。",
"flow_id": "b483af499e4c49498adf963fdefa692a",
"api_parameters": null,
"create_time": "2026-02-05 21:24:46",
"id": 373,
"data": {
"edges": [
{
"id": "xy-edge__condition_086c0f0af1112-condition_e4299left_handle",
"type": "customEdge",
"source": "condition_086c0",
"target": "condition_e4299",
"animated": true,
"sourceHandle": "f0af1112",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__condition_e4299172c7f82-llm_04f45left_handle",
"type": "customEdge",
"source": "condition_e4299",
"target": "llm_04f45",
"animated": true,
"sourceHandle": "172c7f82",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__condition_e4299right_handle-llm_0d11dleft_handle",
"type": "customEdge",
"source": "condition_e4299",
"target": "llm_0d11d",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__flowNode_4325b172c7f82-llm_319a5left_handle",
"type": "customEdge",
"source": "flowNode_4325b",
"target": "llm_319a5",
"animated": true,
"sourceHandle": "172c7f82",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__flowNode_4325b341a403c-llm_fcfd0left_handle",
"type": "customEdge",
"source": "flowNode_4325b",
"target": "llm_fcfd0",
"animated": true,
"sourceHandle": "341a403c",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__flowNode_4325bright_handle-llm_9b7ableft_handle",
"type": "customEdge",
"source": "flowNode_4325b",
"target": "llm_9b7ab",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_04f45right_handle-end_c54b8left_handle",
"type": "customEdge",
"source": "llm_04f45",
"target": "end_c54b8",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_0d11dright_handle-end_d3a29left_handle",
"type": "customEdge",
"source": "llm_0d11d",
"target": "end_d3a29",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_319a5right_handle-end_95a69left_handle",
"type": "customEdge",
"source": "llm_319a5",
"target": "end_95a69",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_fcfd0right_handle-end_b740cleft_handle",
"type": "customEdge",
"source": "llm_fcfd0",
"target": "end_b740c",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_9b7abright_handle-end_7a9d0left_handle",
"type": "customEdge",
"source": "llm_9b7ab",
"target": "end_7a9d0",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_c0564right_handle-condition_086c0left_handle",
"type": "customEdge",
"source": "input_c0564",
"target": "condition_086c0",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__start_68c76right_handle-input_c0564left_handle",
"type": "customEdge",
"source": "start_68c76",
"target": "input_c0564",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__condition_086c0right_handle-input_93a57left_handle",
"type": "customEdge",
"source": "condition_086c0",
"target": "input_93a57",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_93a57right_handle-flowNode_4325bleft_handle",
"type": "customEdge",
"source": "input_93a57",
"target": "flowNode_4325b",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
}
],
"nodes": [
{
"id": "start_68c76",
"data": {
"v": 3,
"id": "start_68c76",
"name": "开始",
"type": "start",
"description": "工作流运行的起始节点",
"group_params": [
{
"name": "开场引导",
"params": [
{
"key": "guide_word",
"type": "textarea",
"label": "开场白",
"value": "文章撰写助手。\n"
},
{
"key": "guide_question",
"help": "为用户提供推荐问题，引导用户输入，超过3个时将随机选取3个",
"type": "input_list",
"label": "引导问题",
"value": [
""
]
}
]
},
{
"name": "全局变量",
"params": [
{
"key": "user_info",
"type": "var",
"label": "true",
"value": "",
"global": "key"
},
{
"key": "current_time",
"type": "var",
"label": "当前时间",
"value": "",
"global": "key"
},
{
"key": "chat_history",
"type": "chat_history_num",
"value": 10,
"global": "key"
},
{
"key": "preset_question",
"type": "input_list",
"label": "预置问题列表",
"value": [
{
"key": "fdf2f6",
"value": ""
}
],
"global": "item:input_list"
},
{
"key": "custom_variables",
"help": "true",
"type": "global_var",
"label": "true",
"value": [],
"global": "item:input_list"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 711
},
"position": {
"x": -254.5074365009976,
"y": -54.95034320067019
},
"selected": false
},
{
"id": "condition_086c0",
"data": {
"id": "condition_086c0",
"name": "条件分支-是否为优化改写",
"type": "condition",
"description": "根据条件表达式执行不同的分支",
"group_params": [
{
"params": [
{
"key": "condition",
"type": "condition",
"label": "",
"value": [
{
"id": "f0af1112",
"operator": "and",
"conditions": [
{
"id": "b5744201",
"left_var": "input_c0564.category",
"left_label": "输入/category",
"right_label": "",
"right_value": "改写优化",
"right_value_type": "input",
"comparison_operation": "not_contains"
}
]
}
]
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 562,
"height": 347
},
"position": {
"x": 652.6065106681506,
"y": -56.557347234989656
},
"selected": false
},
{
"id": "condition_e4299",
"data": {
"id": "condition_e4299",
"name": "条件分支-检查错误还是起标题",
"type": "condition",
"description": "根据条件表达式执行不同的分支",
"group_params": [
{
"params": [
{
"key": "condition",
"type": "condition",
"label": "",
"value": [
{
"id": "172c7f82",
"operator": "and",
"conditions": [
{
"id": "1a0126bd",
"left_var": "input_c0564.category",
"left_label": "输入/category",
"right_label": "",
"right_value": "检查错误",
"right_value_type": "input",
"comparison_operation": "contains"
}
]
}
]
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 562,
"height": 347
},
"position": {
"x": 1403.6380617911316,
"y": -53.541044883239984
},
"selected": false
},
{
"id": "flowNode_4325b",
"data": {
"id": "flowNode_4325b",
"name": "条件分支-哪种优化",
"type": "condition",
"description": "根据条件表达式执行不同的分支",
"group_params": [
{
"params": [
{
"key": "condition",
"type": "condition",
"label": "",
"value": [
{
"id": "172c7f82",
"operator": "and",
"conditions": [
{
"id": "1a0126bd",
"left_var": "input_93a57.category2",
"left_label": "选择优化方向/category2",
"right_label": "",
"right_value": "变长",
"right_value_type": "input",
"comparison_operation": "contains"
}
]
},
{
"id": "341a403c",
"operator": "and",
"conditions": [
{
"id": "887e9105",
"left_var": "input_93a57.category2",
"left_label": "选择优化方向/category2",
"right_label": "",
"right_value": "变短",
"right_value_type": "input",
"comparison_operation": "contains"
}
]
}
]
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 562,
"height": 463
},
"position": {
"x": 1710.9835717741546,
"y": 352.22981329527255
},
"selected": false
},
{
"id": "llm_04f45",
"data": {
"v": 2,
"id": "llm_04f45",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型-检查错误",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是文章撰写专家，请检查以下文本中的错误，并一一列举出来"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "{{#input_c0564.text_input#}}",
"varZh": {
"input_c0564.a4089e": "输入/a4089e",
"input_c0564.text_input": "输入/text_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2357.376350158347,
"y": -99.5882796942481
},
"selected": false
},
{
"id": "llm_0d11d",
"data": {
"v": 2,
"id": "llm_0d11d",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型-起标题",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是文章撰写专家，请基于以下文本内容起一个标题"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "{{#input_c0564.text_input#}}",
"varZh": {
"input_93a57.category2": "选择优化方向/category2",
"input_c0564.text_input": "输入/text_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2366.37878639579,
"y": 135.57804927695716
},
"selected": false
},
{
"id": "llm_319a5",
"data": {
"v": 2,
"id": "llm_319a5",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型-变长",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是文章撰写专家，请将以下内容改写得更长。"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "{{#input_c0564.text_input#}}",
"varZh": {
"input_c0564.text_input": "输入/text_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2364.2222119477274,
"y": 311.77105431340004
},
"selected": false
},
{
"id": "llm_fcfd0",
"data": {
"v": 2,
"id": "llm_fcfd0",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型-变短",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是文章撰写专家，请将以下内容改写得更短。"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "{{#input_c0564.text_input#}}",
"varZh": {
"input_c0564.text_input": "输入/text_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2380.2587980739713,
"y": 523.9445747167125
},
"selected": false
},
{
"id": "llm_9b7ab",
"data": {
"v": 2,
"id": "llm_9b7ab",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型-讽刺口吻",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是文章撰写专家，请将以下内容用讽刺的口吻改写一遍。"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "{{#input_c0564.text_input#}}",
"varZh": {
"input_c0564.text_input": "输入/text_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2382.1177403285865,
"y": 716.8035163539979
},
"selected": true
},
{
"id": "end_c54b8",
"data": {
"id": "end_c54b8",
"name": "结束",
"type": "end",
"description": "工作流运行到此结束。",
"group_params": []
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2808.1668882050744,
"y": -42.43221090899556
},
"selected": false
},
{
"id": "end_d3a29",
"data": {
"id": "end_d3a29",
"name": "结束2",
"type": "end",
"description": "工作流运行到此结束。",
"group_params": []
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2814.629079906689,
"y": 129.02998079261903
},
"selected": false
},
{
"id": "end_95a69",
"data": {
"id": "end_95a69",
"name": "结束3",
"type": "end",
"description": "工作流运行到此结束。",
"group_params": []
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2807.319452039774,
"y": 319.3509041526474
},
"selected": false
},
{
"id": "end_b740c",
"data": {
"id": "end_b740c",
"name": "结束4",
"type": "end",
"description": "工作流运行到此结束。",
"group_params": []
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2790.9481643515182,
"y": 514.5879452405269
},
"selected": false
},
{
"id": "end_7a9d0",
"data": {
"id": "end_7a9d0",
"name": "结束5",
"type": "end",
"description": "工作流运行到此结束。",
"group_params": []
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 134
},
"position": {
"x": 2805.2370410878793,
"y": 703.8147945606022
},
"selected": false
},
{
"id": "input_c0564",
"data": {
"v": "3",
"id": "input_c0564",
"tab": {
"value": "form_input",
"options": [
{
"key": "dialog_input",
"help": "true",
"label": "true"
},
{
"key": "form_input",
"help": "true",
"label": "true"
}
]
},
"name": "输入",
"type": "input",
"description": "接收用户在会话页面的输入，支持 2 种形式：对话框输入，表单输入。",
"group_params": [
{
"name": "接收文本",
"params": [
{
"key": "user_input",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
}
]
},
{
"name": "",
"params": [
{
"key": "user_input_file",
"tab": "dialog_input",
"value": true,
"groupTitle": true
},
{
"key": "file_parse_mode",
"tab": "dialog_input",
"type": "select_parsemode",
"value": "extract_text"
},
{
"key": "dialog_files_content",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_files_content_size",
"min": 0,
"tab": "dialog_input",
"type": "char_number",
"label": "true",
"value": 15000
},
{
"key": "dialog_file_accept",
"tab": "dialog_input",
"type": "select_fileaccept",
"label": "true",
"value": "all"
},
{
"key": "dialog_image_files",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_file_paths",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
}
],
"groupKey": "inputfile"
},
{
"name": "",
"params": [
{
"key": "recommended_questions_flag",
"tab": "dialog_input",
"help": "true",
"label": "true",
"value": false,
"hidden": "true",
"groupTitle": true
},
{
"key": "recommended_llm",
"tab": "dialog_input",
"type": "bisheng_model",
"label": "true",
"value": 261,
"required": true,
"placeholder": "true"
},
{
"key": "recommended_system_prompt",
"tab": "dialog_input",
"type": "var_textarea",
"label": "true",
"value": "# Role\n你是一个极具洞察力的\"对话延续预测专家\"。你的任务是根据当前的对话历史，预测用户接下来最有可能输入的 3 条短语或问题。\n\n# Context Awareness\n在生成预测前，请执行以下深度思考步骤：\n1. **意图分析**：用户当前是在寻求事实、表达情绪、请求建议，还是在尝试结束对话？\n2. **逻辑推演**：根据 AI 上一条回复的内容，用户最自然产生的联想或疑问是什么？\n3. **风格对齐**：观察用户的用词习惯（专业、随性、简短还是详细），预测结果需保持语气一致。\n\n# Goals\n- **多样性**：提供不同方向的后续（例如：一个深入探究的问题，一个相关的新话题，一个反馈性评价）。\n- **简洁性**：预测内容应短小精悍，通常不超过 15 个字，方便用户点击。\n- **前瞻性**：能够预判用户的\"痛点\"，直接提供解决方案的入口。\n\n# Output Format\n请严格遵守 JSON 格式输出。\n返回一个包含单一键名 \"suggestions\" 的 JSON 对象。\n- 所有字符串必须使用**双引号**。\n- 不要输出 Markdown 代码块标记（如 ```json），直接输出 JSON 字符串。\n**示例：**\n{\"suggestions\": [\"你能举个例子吗？\", \"这个方案的成本是多少？\", \"听起来不错，怎么开始？\"]}",
"required": true
},
{
"key": "recommended_history_num",
"tab": "dialog_input",
"help": "true",
"step": 1,
"type": "slide",
"label": "true",
"scope": [
1,
10
],
"value": 2
}
],
"groupKey": "custom"
},
{
"name": "",
"params": [
{
"key": "form_input",
"tab": "form_input",
"type": "form",
"label": "true",
"value": [
{
"key": "category",
"type": "select",
"value": "帮您做什么",
"options": [
{
"id": "55078e1a",
"text": "检查错误",
"type": ""
},
{
"id": "314f10d4",
"text": "改写优化",
"type": ""
},
{
"id": "beea95fb",
"text": "起一个标题",
"type": ""
}
],
"multiple": false,
"required": true,
"file_path": "file_path",
"file_type": "all",
"image_file": "image_file",
"file_content": "file_content",
"file_parse_mode": "extract_text",
"file_content_size": 15000
},
{
"key": "text_input",
"type": "text",
"value": "需要改写的内容\n\n",
"options": [],
"multiple": false,
"required": true,
"file_path": "file_path",
"file_type": "all",
"image_file": "image_file",
"file_content": "file_content",
"file_parse_mode": "extract_text",
"file_content_size": 15000
}
],
"global": "item:form_input"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 356
},
"position": {
"x": 210.10966500379396,
"y": -57.43844277479269
},
"selected": false
},
{
"id": "input_93a57",
"data": {
"v": "3",
"id": "input_93a57",
"tab": {
"value": "form_input",
"options": [
{
"key": "dialog_input",
"help": "true",
"label": "true"
},
{
"key": "form_input",
"help": "true",
"label": "true"
}
]
},
"name": "选择优化方向",
"type": "input",
"description": "接收用户在会话页面的输入，支持 2 种形式：对话框输入，表单输入。",
"group_params": [
{
"name": "接收文本",
"params": [
{
"key": "user_input",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
}
]
},
{
"name": "",
"params": [
{
"key": "user_input_file",
"tab": "dialog_input",
"value": true,
"groupTitle": true
},
{
"key": "file_parse_mode",
"tab": "dialog_input",
"type": "select_parsemode",
"value": "extract_text"
},
{
"key": "dialog_files_content",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_files_content_size",
"min": 0,
"tab": "dialog_input",
"type": "char_number",
"label": "true",
"value": 15000
},
{
"key": "dialog_file_accept",
"tab": "dialog_input",
"type": "select_fileaccept",
"label": "true",
"value": "all"
},
{
"key": "dialog_image_files",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_file_paths",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
}
],
"groupKey": "inputfile"
},
{
"name": "",
"params": [
{
"key": "recommended_questions_flag",
"tab": "dialog_input",
"help": "true",
"label": "true",
"value": false,
"hidden": "true",
"groupTitle": true
},
{
"key": "recommended_llm",
"tab": "dialog_input",
"type": "bisheng_model",
"label": "true",
"value": 261,
"required": true,
"placeholder": "true"
},
{
"key": "recommended_system_prompt",
"tab": "dialog_input",
"type": "var_textarea",
"label": "true",
"value": "# Role\n你是一个极具洞察力的\"对话延续预测专家\"。你的任务是根据当前的对话历史，预测用户接下来最有可能输入的 3 条短语或问题。\n\n# Context Awareness\n在生成预测前，请执行以下深度思考步骤：\n1. **意图分析**：用户当前是在寻求事实、表达情绪、请求建议，还是在尝试结束对话？\n2. **逻辑推演**：根据 AI 上一条回复的内容，用户最自然产生的联想或疑问是什么？\n3. **风格对齐**：观察用户的用词习惯（专业、随性、简短还是详细），预测结果需保持语气一致。\n\n# Goals\n- **多样性**：提供不同方向的后续（例如：一个深入探究的问题，一个相关的新话题，一个反馈性评价）。\n- **简洁性**：预测内容应短小精悍，通常不超过 15 个字，方便用户点击。\n- **前瞻性**：能够预判用户的\"痛点\"，直接提供解决方案的入口。\n\n# Output Format\n请严格遵守 JSON 格式输出。\n返回一个包含单一键名 \"suggestions\" 的 JSON 对象。\n- 所有字符串必须使用**双引号**。\n- 不要输出 Markdown 代码块标记（如 ```json），直接输出 JSON 字符串。\n**示例：**\n{\"suggestions\": [\"你能举个例子吗？\", \"这个方案的成本是多少？\", \"听起来不错，怎么开始？\"]}",
"required": true
},
{
"key": "recommended_history_num",
"tab": "dialog_input",
"help": "true",
"step": 1,
"type": "slide",
"label": "true",
"scope": [
1,
10
],
"value": 2
}
],
"groupKey": "custom"
},
{
"name": "",
"params": [
{
"key": "form_input",
"tab": "form_input",
"type": "form",
"label": "true",
"value": [
{
"key": "category2",
"type": "select",
"value": "需要优化的方向",
"options": [
{
"id": "1df7e6d8",
"text": "变长",
"type": ""
},
{
"id": "e8ce5e3e",
"text": "变短",
"type": ""
},
{
"id": "723fa2e1",
"text": "改成讽刺口吻",
"type": ""
}
],
"multiple": false,
"required": true,
"file_path": "file_path",
"file_type": "all",
"image_file": "image_file",
"file_content": "file_content",
"file_parse_mode": "extract_text",
"file_content_size": 15000
}
],
"global": "item:form_input"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 311
},
"position": {
"x": 1301.034049020076,
"y": 362.6108337624355
},
"selected": false
}
],
"viewport": {
"x": 168.41000286983763,
"y": 188.33690890750776,
"zoom": 0.39845595187329674
}
},
"order_num": 13221748,
"guide_word": null,
"flow_type": 10
},
{
"update_time": "2026-02-05 21:29:08",
"parameters": null,
"name": "大模型竞技场 ",
"description": "功能简介 这是 “大模型竞技场” 工作流，支持 “对话框 / 表单” 2 种输入形式，调用 gpt40mini、混元、qwen 2.5 三个大模型并行处理任务，最后通过 “输出 2” 节点，以无交互、选择型交互或输入型交互方式，反馈模型处理结果 。\n使用说明\n在 “输入” 节点选交互形式（对话框 / 表单），填入内容、上传文件（可选），运行流程。三个大模型会同时处理输入，完成后在 “输出 2” 节点，根据需求选交互类型，查看 / 使用模型输出结果 。",
"flow_id": "fba9c65151184d629348219f05a70e27",
"api_parameters": null,
"create_time": "2026-02-05 21:29:08",
"id": 374,
"data": {
"edges": [
{
"id": "xy-edge__start_68c76right_handle-input_ac896left_handle",
"type": "customEdge",
"source": "start_68c76",
"target": "input_ac896",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_ac896right_handle-llm_1c15aleft_handle",
"type": "customEdge",
"source": "input_ac896",
"target": "llm_1c15a",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_ac896right_handle-llm_0bbf3left_handle",
"type": "customEdge",
"source": "input_ac896",
"target": "llm_0bbf3",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_ac896right_handle-llm_5b4ebleft_handle",
"type": "customEdge",
"source": "input_ac896",
"target": "llm_5b4eb",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_1c15aright_handle-output_074afleft_handle",
"type": "customEdge",
"source": "llm_1c15a",
"target": "output_074af",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_0bbf3right_handle-output_074afleft_handle",
"type": "customEdge",
"source": "llm_0bbf3",
"target": "output_074af",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__llm_5b4ebright_handle-output_074afleft_handle",
"type": "customEdge",
"source": "llm_5b4eb",
"target": "output_074af",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__output_074afright_handle-input_ac896left_handle",
"type": "customEdge",
"source": "output_074af",
"target": "input_ac896",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
}
],
"nodes": [
{
"id": "start_68c76",
"data": {
"v": 3,
"id": "start_68c76",
"name": "开始",
"type": "start",
"description": "工作流运行的起始节点",
"group_params": [
{
"name": "开场引导",
"params": [
{
"key": "guide_word",
"type": "textarea",
"label": "开场白",
"value": "大模型竞技场"
},
{
"key": "guide_question",
"help": "为用户提供推荐问题，引导用户输入，超过3个时将随机选取3个",
"type": "input_list",
"label": "引导问题",
"value": [
"你是谁~",
""
]
}
]
},
{
"name": "全局变量",
"params": [
{
"key": "user_info",
"type": "var",
"label": "true",
"value": "",
"global": "key"
},
{
"key": "current_time",
"type": "var",
"label": "当前时间",
"value": "",
"global": "key"
},
{
"key": "chat_history",
"type": "chat_history_num",
"value": 10,
"global": "key"
},
{
"key": "preset_question",
"type": "input_list",
"label": "预置问题列表",
"value": [
{
"key": "18b2f7",
"value": ""
}
],
"global": "item:input_list"
},
{
"key": "custom_variables",
"help": "true",
"type": "global_var",
"label": "true",
"value": [],
"global": "item:input_list"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 753
},
"position": {
"x": -182.1450323475628,
"y": -13.21802815370333
},
"selected": false
},
{
"id": "llm_1c15a",
"data": {
"v": 2,
"id": "llm_1c15a",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型-gpt4omini ",
"type": "llm",
"expand": true,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "友好的助手"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "用户：{{#input_ac896.user_input#}}你的回复：",
"varZh": {
"input_ac896.user_input": "输入/user_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": false
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 836
},
"position": {
"x": 660.2576202218669,
"y": -7.589292442416962
},
"selected": false
},
{
"id": "llm_0bbf3",
"data": {
"v": 2,
"id": "llm_0bbf3",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型-混元",
"type": "llm",
"expand": false,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "友好的助手"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "用户：{{#input_ac896.user_input#}}你的回复：",
"varZh": {
"input_ac896.user_input": "输入/user_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": false
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 135
},
"position": {
"x": 663.5982291778208,
"y": 160.9498553634532
},
"selected": false
},
{
"id": "llm_5b4eb",
"data": {
"v": 2,
"id": "llm_5b4eb",
"tab": {
"value": "single",
"options": [
{
"key": "single",
"label": "单次运行"
},
{
"key": "batch",
"label": "批量运行"
}
]
},
"name": "大模型-qwen 2.5",
"type": "llm",
"expand": true,
"description": "调用大模型回答用户问题或者处理任务。",
"group_params": [
{
"params": [
{
"key": "batch_variable",
"tab": "batch",
"help": "选择需要批处理的变量，将会多次运行本节点，每次运行时从选择的变量中取一项赋值给batch_variable进行处理。",
"type": "user_question",
"label": "批处理变量",
"value": [],
"global": "self",
"linkage": "output",
"required": true,
"placeholder": "请选择批处理变量"
}
]
},
{
"name": "模型设置",
"params": [
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "提示词",
"params": [
{
"key": "system_prompt",
"test": "var",
"type": "var_textarea",
"label": "系统提示词",
"value": "友好的助手"
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "用户：{{#input_ac896.user_input#}}你的回复：",
"varZh": {
"input_ac896.user_input": "输入/user_input"
},
"required": true
},
{
"key": "image_prompt",
"help": "当使用多模态大模型时，可通过此功能传入图片，结合图像内容进行问答",
"type": "image_prompt",
"label": "视觉",
"value": []
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景可开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": false
},
{
"key": "output",
"type": "var",
"label": "输出变量",
"value": [],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 836
},
"position": {
"x": 666.4815464404269,
"y": 386.324186907184
},
"selected": false
},
{
"id": "input_ac896",
"data": {
"v": "3",
"id": "input_ac896",
"tab": {
"value": "dialog_input",
"options": [
{
"key": "dialog_input",
"help": "true",
"label": "true"
},
{
"key": "form_input",
"help": "true",
"label": "true"
}
]
},
"name": "输入",
"type": "input",
"description": "接收用户在会话页面的输入，支持 2 种形式：对话框输入，表单输入。",
"group_params": [
{
"name": "接收文本",
"params": [
{
"key": "user_input",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
}
]
},
{
"name": "",
"params": [
{
"key": "user_input_file",
"tab": "dialog_input",
"value": true,
"groupTitle": true
},
{
"key": "file_parse_mode",
"tab": "dialog_input",
"type": "select_parsemode",
"value": "extract_text"
},
{
"key": "dialog_files_content",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_files_content_size",
"min": 0,
"tab": "dialog_input",
"type": "char_number",
"label": "true",
"value": 15000
},
{
"key": "dialog_file_accept",
"tab": "dialog_input",
"type": "select_fileaccept",
"label": "true",
"value": "all"
},
{
"key": "dialog_image_files",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_file_paths",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
}
],
"groupKey": "inputfile"
},
{
"name": "",
"params": [
{
"key": "recommended_questions_flag",
"tab": "dialog_input",
"help": "true",
"label": "true",
"value": false,
"hidden": "true",
"groupTitle": true
},
{
"key": "recommended_llm",
"tab": "dialog_input",
"type": "bisheng_model",
"label": "true",
"value": 261,
"required": true,
"placeholder": "true"
},
{
"key": "recommended_system_prompt",
"tab": "dialog_input",
"type": "var_textarea",
"label": "true",
"value": "# Role\n你是一个极具洞察力的\"对话延续预测专家\"。你的任务是根据当前的对话历史，预测用户接下来最有可能输入的 3 条短语或问题。\n\n# Context Awareness\n在生成预测前，请执行以下深度思考步骤：\n1. **意图分析**：用户当前是在寻求事实、表达情绪、请求建议，还是在尝试结束对话？\n2. **逻辑推演**：根据 AI 上一条回复的内容，用户最自然产生的联想或疑问是什么？\n3. **风格对齐**：观察用户的用词习惯（专业、随性、简短还是详细），预测结果需保持语气一致。\n\n# Goals\n- **多样性**：提供不同方向的后续（例如：一个深入探究的问题，一个相关的新话题，一个反馈性评价）。\n- **简洁性**：预测内容应短小精悍，通常不超过 15 个字，方便用户点击。\n- **前瞻性**：能够预判用户的\"痛点\"，直接提供解决方案的入口。\n\n# Output Format\n请严格遵守 JSON 格式输出。\n返回一个包含单一键名 \"suggestions\" 的 JSON 对象。\n- 所有字符串必须使用**双引号**。\n- 不要输出 Markdown 代码块标记（如 ```json），直接输出 JSON 字符串。\n**示例：**\n{\"suggestions\": [\"你能举个例子吗？\", \"这个方案的成本是多少？\", \"听起来不错，怎么开始？\"]}",
"required": true
},
{
"key": "recommended_history_num",
"tab": "dialog_input",
"help": "true",
"step": 1,
"type": "slide",
"label": "true",
"scope": [
1,
10
],
"value": 2
}
],
"groupKey": "custom"
},
{
"name": "",
"params": [
{
"key": "form_input",
"tab": "form_input",
"type": "form",
"label": "true",
"value": [],
"global": "item:form_input"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 572
},
"position": {
"x": 241.1814647549988,
"y": -10.967656001848963
},
"selected": false
},
{
"id": "output_074af",
"data": {
"v": "2",
"id": "output_074af",
"name": "输出",
"type": "output",
"description": "可向用户发送文本和文件，并且支持进行更丰富的交互，例如请求用户批准进行某项敏感操作、允许用户在模型输出内容的基础上直接修改并提交。",
"group_params": [
{
"params": [
{
"key": "message",
"type": "var_textarea_file",
"label": "true",
"value": {
"msg": "\n# GPT-4o-mini \n{{#llm_1c15a.output#}}\n\n\n# 混元 \n{{#llm_0bbf3.output#}} \n\n\n# 千问 \n{{#llm_5b4eb.output#}} \n\n\n\n",
"files": []
},
"varZh": {
"llm_0bbf3.output": "大模型-混元/output",
"llm_1c15a.output": "大模型-gpt4omini/output",
"llm_5b4eb.output": "大模型-qwen 2.5/output"
},
"global": "key",
"required": true,
"placeholder": "true"
},
{
"key": "output_result",
"type": "output_form",
"label": "用户输入内容",
"value": {
"type": "",
"value": ""
},
"global": "value.type=input",
"options": [],
"required": false
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 465
},
"position": {
"x": 1116.4196010933445,
"y": -4.140951751970215
},
"selected": true
}
],
"viewport": {
"x": -397.1989979664372,
"y": 59.476134021867665,
"zoom": 1.0694829253528593
}
},
"order_num": 13287283,
"guide_word": null,
"flow_type": 10
},
{
"update_time": "2026-02-05 21:44:36",
"parameters": null,
"name": "先检索 QA 后检索文档知识库问答",
"description": "功能简介\n这是 “先检索 QA 后检索文档知识” 工作流，通过 “开始” 引导对话，“输入” 模块采集用户内容，先经 “QA 知识检索” 查问答库，无结果则触发 “文档知识问答” 补充回答，保障信息覆盖。\n使用说明\n在 “开始” 填引导语、示例问题；“输入” 选录入方式，设置文件参数。\n运行流程，优先用 “QA 知识检索” 查问答库；无结果自动走 “文档知识问答”。\n测试无误后点 “上线”，即可接入业务，接收用户提问并智能回答 。",
"flow_id": "6445ac25c5c0433091df2827963a762e",
"api_parameters": null,
"create_time": "2026-02-05 21:44:36",
"id": 376,
"data": {
"edges": [
{
"id": "xy-edge__condition_6c26160b32025-output_bf694left_handle",
"type": "customEdge",
"source": "condition_6c261",
"target": "output_bf694",
"animated": true,
"sourceHandle": "60b32025",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__condition_6c261right_handle-rag_01223left_handle",
"type": "customEdge",
"source": "condition_6c261",
"target": "rag_01223",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__qa_retriever_3bb15right_handle-condition_6c261left_handle",
"type": "customEdge",
"source": "qa_retriever_3bb15",
"target": "condition_6c261",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__start_d8f27right_handle-input_46939left_handle",
"type": "customEdge",
"source": "start_d8f27",
"target": "input_46939",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__input_46939right_handle-qa_retriever_3bb15left_handle",
"type": "customEdge",
"source": "input_46939",
"target": "qa_retriever_3bb15",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__output_bf694right_handle-input_46939left_handle",
"type": "customEdge",
"source": "output_bf694",
"target": "input_46939",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
},
{
"id": "xy-edge__rag_01223right_handle-input_46939left_handle",
"type": "customEdge",
"source": "rag_01223",
"target": "input_46939",
"animated": true,
"sourceHandle": "right_handle",
"targetHandle": "left_handle"
}
],
"nodes": [
{
"id": "start_d8f27",
"data": {
"v": 3,
"id": "start_d8f27",
"name": "开始",
"type": "start",
"description": "工作流运行的起始节点。",
"group_params": [
{
"name": "开场引导",
"params": [
{
"key": "guide_word",
"type": "textarea",
"label": "开场白",
"value": "先检索QA库中是否有类似问答，如果没有找到答案则进行文档知识库问答",
"placeholder": "每次工作流开始执行时向用户发送此消息，支持 Markdown 格式，为空时不发送。"
},
{
"key": "guide_question",
"help": "为用户提供推荐问题，引导用户输入，超过3个时将随机选取3个。",
"type": "input_list",
"label": "引导问题",
"value": [
"毕昇是什么？",
"大模型应用开发平台有哪些？",
""
],
"placeholder": "请输入引导问题"
}
]
},
{
"name": "全局变量",
"params": [
{
"key": "user_info",
"type": "var",
"label": "true",
"value": "",
"global": "key"
},
{
"key": "current_time",
"type": "var",
"label": "当前时间",
"value": "",
"global": "key"
},
{
"key": "chat_history",
"type": "chat_history_num",
"value": 10,
"global": "key"
},
{
"key": "preset_question",
"help": "适合文档审核、报告生成等场景，利用提前预置的问题批量进行 RAG 问答。",
"type": "input_list",
"label": "预置问题列表",
"value": [
{
"key": "0023a4",
"value": ""
}
],
"global": "item:input_list",
"placeholder": "输入批量预置问题"
},
{
"key": "custom_variables",
"help": "true",
"type": "global_var",
"label": "true",
"value": [],
"global": "item:input_list"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 793
},
"position": {
"x": -322.001475694825,
"y": -74.5314588248279
},
"selected": false
},
{
"id": "condition_6c261",
"data": {
"id": "condition_6c261",
"name": "条件分支",
"type": "condition",
"description": "根据条件表达式执行不同的分支。",
"group_params": [
{
"params": [
{
"key": "condition",
"type": "condition",
"label": "",
"value": [
{
"id": "60b32025",
"operator": "and",
"conditions": [
{
"id": "37b199ff",
"left_var": "qa_retriever_3bb15.retrieved_result",
"left_label": "retrieved_result",
"right_label": "",
"right_value": "",
"right_value_type": "input",
"comparison_operation": "is_not_empty"
}
]
}
]
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 322,
"height": 348
},
"position": {
"x": 1041.11058653272,
"y": -61.43751998557724
},
"selected": false
},
{
"id": "output_bf694",
"data": {
"v": 2,
"id": "output_bf694",
"name": "输出",
"type": "output",
"description": "可向用户发送消息，并且支持进行更丰富的交互，例如请求用户批准进行某项敏感操作、允许用户在模型输出内容的基础上直接修改并提交。",
"group_params": [
{
"params": [
{
"key": "message",
"type": "var_textarea_file",
"label": "消息内容",
"value": {
"msg": "{{#qa_retriever_3bb15.retrieved_result#}}",
"files": []
},
"varZh": {
"qa_retriever_12db4.retrieved_result": "QA知识库检索2/retrieved_result",
"qa_retriever_3bb15.retrieved_result": "QA知识库检索/retrieved_result",
"qa_retriever_913d3.retrieval_result": "QA知识库检索/retrieval_result"
},
"global": "key",
"required": true,
"placeholder": "输入需要发送给用户的消息，例如“接下来我将执行 XX 操作，请您确认”，“以下是我的初版草稿，您可以在其基础上进行修改”"
},
{
"key": "output_result",
"type": "output_form",
"label": "用户输入内容",
"value": {
"type": "",
"value": ""
},
"global": "value.type=input",
"options": [],
"required": false
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 465
},
"position": {
"x": 1615.0030925136334,
"y": -284.24496879143226
},
"selected": false
},
{
"id": "rag_01223",
"data": {
"v": 3,
"id": "rag_01223",
"name": "文档知识库问答",
"type": "rag",
"expand": true,
"description": "根据用户问题从知识库中检索相关内容，结合检索结果调用大模型生成最终结果，支持多个问题并行执行。",
"group_params": [
{
"name": "知识库检索设置",
"params": [
{
"key": "user_question",
"help": "当选择多个问题时，将会多次运行本节点，每次运行时从批量问题中取一项进行处理。",
"test": "var",
"type": "user_question",
"label": "用户问题",
"value": [
"input_46939.user_input"
],
"varZh": {
"input_46939.user_input": "输入/user_input",
"input_eb5c4.user_input": "输入/user_input"
},
"global": "self=user_prompt",
"linkage": "output_user_input",
"required": true,
"placeholder": "请选择用户问题"
},
{
"key": "knowledge",
"type": "knowledge_select_multi",
"label": "检索范围",
"value": {
"type": "knowledge",
"value": [
{
"key": 2941,
"label": "230 release知识库回归"
}
]
},
"required": true,
"placeholder": "请选择知识库"
},
{
"key": "metadata_filter",
"type": "metadata_filter",
"label": "true",
"value": {
"enabled": false
}
},
{
"key": "advanced_retrieval_switch",
"type": "search_switch",
"label": "true",
"value": {
"user_auth": false,
"rerank_flag": false,
"rerank_model": "",
"search_switch": true,
"vector_weight": 0.5,
"keyword_weight": 0.5,
"max_chunk_size": 15000
}
},
{
"key": "retrieved_result",
"type": "var",
"label": "检索结果",
"global": "self=user_prompt"
}
]
},
{
"name": "AI回复生成设置",
"params": [
{
"key": "system_prompt",
"type": "var_textarea",
"label": "系统提示词",
"value": "你是一个知识库问答助手： \n1.用中文回答用户问题，并且答案要严谨专业。\n2.你需要依据【参考文本】中的内容来回答，当【参考文本】中有明确与用户问题相关的内容时才进行回答，不可根据自己的知识来回答。\n3.由于【参考文本】可能包含多个来自不同信息源的信息，所以根据这些不同的信息源可能得出有差异甚至冲突的答案，当发现这种情况时，这些答案都列举出来；如果没有冲突或差异，则只需要给出一个最终结果。\n4.若【参考文本】中内容与用户问题不相关则回复“没有找到相关内容”。",
"required": true
},
{
"key": "user_prompt",
"test": "var",
"type": "var_textarea",
"label": "用户提示词",
"value": "用户问题：{{#rag_01223.user_question#}}\n参考文本：{{#rag_01223.retrieved_result#}}\n你的回答：",
"varZh": {
"rag_01223.user_question": "user_question",
"rag_01223.retrieved_result": "retrieved_result"
},
"required": true
},
{
"key": "model_id",
"type": "bisheng_model",
"label": "模型",
"value": 261,
"required": true,
"placeholder": "请选择模型"
},
{
"key": "temperature",
"step": 0.1,
"type": "slide",
"label": "温度",
"scope": [
0,
2
],
"value": 0.7
}
]
},
{
"name": "输出",
"params": [
{
"key": "output_user",
"help": "一般在问答等场景开启，文档审核、报告生成等场景可关闭。",
"type": "switch",
"label": "将输出结果展示在会话中",
"value": true
},
{
"key": "output_user_input",
"type": "var",
"label": "输出变量",
"value": [
{
"key": "output_user_input",
"label": "output_user_input"
}
],
"global": "code:value.map(el => ({ label: el.label, value: el.key }))"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 1278
},
"position": {
"x": 1640.164090703263,
"y": 286.0127309008938
},
"selected": false
},
{
"id": "qa_retriever_3bb15",
"data": {
"id": "qa_retriever_3bb15",
"name": "QA知识库检索",
"type": "qa_retriever",
"expand": true,
"description": "从 QA 知识库中检索问题以及对应的答案。",
"group_params": [
{
"name": "检索设置",
"params": [
{
"key": "user_question",
"test": "var",
"type": "var_select",
"label": "输入变量",
"value": "input_46939.user_input",
"varZh": {
"input_46939.user_input": "输入/user_input",
"input_eb5c4.user_input": "输入/user_input"
},
"required": true,
"placeholder": "请选择检索问题"
},
{
"key": "qa_knowledge_id",
"type": "qa_select_multi",
"label": "QA知识库",
"value": [
{
"key": 1175,
"label": "毕昇 QA 知识库"
}
],
"required": true,
"placeholder": "请选择QA知识库"
},
{
"key": "score",
"help": "低于阈值的结果将会被过滤。",
"step": 0.01,
"type": "slide",
"label": "相似度阈值",
"scope": [
0.01,
0.99
],
"value": 0.9
}
]
},
{
"name": "输出",
"params": [
{
"key": "retrieved_result",
"type": "var",
"label": "检索结果",
"value": "",
"global": "key"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 497
},
"position": {
"x": 573.66669672053,
"y": -60.24872646754291
},
"selected": false
},
{
"id": "input_46939",
"data": {
"v": "3",
"id": "input_46939",
"tab": {
"value": "dialog_input",
"options": [
{
"key": "dialog_input",
"help": "true",
"label": "true"
},
{
"key": "form_input",
"help": "true",
"label": "true"
}
]
},
"name": "输入",
"type": "input",
"description": "接收用户在会话页面的输入，支持 2 种形式：对话框输入，表单输入。",
"group_params": [
{
"name": "接收文本",
"params": [
{
"key": "user_input",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
}
]
},
{
"name": "",
"params": [
{
"key": "user_input_file",
"tab": "dialog_input",
"value": true,
"groupTitle": true
},
{
"key": "file_parse_mode",
"tab": "dialog_input",
"type": "select_parsemode",
"value": "extract_text"
},
{
"key": "dialog_files_content",
"tab": "dialog_input",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_files_content_size",
"min": 0,
"tab": "dialog_input",
"type": "char_number",
"label": "true",
"value": 15000
},
{
"key": "dialog_file_accept",
"tab": "dialog_input",
"type": "select_fileaccept",
"label": "true",
"value": "all"
},
{
"key": "dialog_image_files",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
},
{
"key": "dialog_file_paths",
"tab": "dialog_input",
"help": "true",
"type": "var",
"label": "true",
"global": "key"
}
],
"groupKey": "inputfile"
},
{
"name": "",
"params": [
{
"key": "recommended_questions_flag",
"tab": "dialog_input",
"help": "true",
"label": "true",
"value": false,
"hidden": "true",
"groupTitle": true
},
{
"key": "recommended_llm",
"tab": "dialog_input",
"type": "bisheng_model",
"label": "true",
"value": 261,
"required": true,
"placeholder": "true"
},
{
"key": "recommended_system_prompt",
"tab": "dialog_input",
"type": "var_textarea",
"label": "true",
"value": "# Role\n你是一个极具洞察力的\"对话延续预测专家\"。你的任务是根据当前的对话历史，预测用户接下来最有可能输入的 3 条短语或问题。\n\n# Context Awareness\n在生成预测前，请执行以下深度思考步骤：\n1. **意图分析**：用户当前是在寻求事实、表达情绪、请求建议，还是在尝试结束对话？\n2. **逻辑推演**：根据 AI 上一条回复的内容，用户最自然产生的联想或疑问是什么？\n3. **风格对齐**：观察用户的用词习惯（专业、随性、简短还是详细），预测结果需保持语气一致。\n\n# Goals\n- **多样性**：提供不同方向的后续（例如：一个深入探究的问题，一个相关的新话题，一个反馈性评价）。\n- **简洁性**：预测内容应短小精悍，通常不超过 15 个字，方便用户点击。\n- **前瞻性**：能够预判用户的\"痛点\"，直接提供解决方案的入口。\n\n# Output Format\n请严格遵守 JSON 格式输出。\n返回一个包含单一键名 \"suggestions\" 的 JSON 对象。\n- 所有字符串必须使用**双引号**。\n- 不要输出 Markdown 代码块标记（如 ```json），直接输出 JSON 字符串。\n**示例：**\n{\"suggestions\": [\"你能举个例子吗？\", \"这个方案的成本是多少？\", \"听起来不错，怎么开始？\"]}",
"required": true
},
{
"key": "recommended_history_num",
"tab": "dialog_input",
"help": "true",
"step": 1,
"type": "slide",
"label": "true",
"scope": [
1,
10
],
"value": 2
}
],
"groupKey": "custom"
},
{
"name": "",
"params": [
{
"key": "form_input",
"tab": "form_input",
"type": "form",
"label": "true",
"value": [],
"global": "item:form_input"
}
]
}
]
},
"type": "flowNode",
"dragging": false,
"measured": {
"width": 334,
"height": 572
},
"position": {
"x": 143.54412024155707,
"y": -86.67644166492184
},
"selected": false
}
],
"viewport": {
"x": 596.1815863554245,
"y": 166.1856841363313,
"zoom": 0.4404147755670133
}
},
"order_num": 13418353,
"guide_word": null,
"flow_type": 10
}
]
================================================
FILE: src/backend/bisheng/database/models/__init__.py
================================================
================================================
FILE: src/backend/bisheng/database/models/assistant.py
================================================
from datetime import datetime
from enum import Enum
from typing import List, Optional, Tuple
from sqlalchemy import JSON, Column, DateTime, Text, and_, func, or_, text
from sqlmodel import Field, select, col
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum
from bisheng.common.models.base import SQLModelSerializable
from bisheng.common.services import telemetry_service
from bisheng.core.database import get_sync_db_session, get_async_db_session
from bisheng.core.logger import trace_id_var
from bisheng.database.models.role_access import AccessType, RoleAccess
from bisheng.utils import generate_uuid
class AssistantStatus(Enum):
OFFLINE = 1
ONLINE = 2
class AssistantBase(SQLModelSerializable):
id: Optional[str] = Field(default_factory=generate_uuid, nullable=False, primary_key=True,
description='Uniqueness quantificationID')
name: str = Field(default='', description='The assistant name.')
logo: str = Field(default='', description='logoimage URL')
desc: str = Field(default='', sa_column=Column(Text), description='Assistant description')
system_prompt: str = Field(default='', sa_column=Column(Text), description='System Prompt')
prompt: str = Field(default='', sa_column=Column(Text), description='User Visible Descriptor')
guide_word: Optional[str] = Field(default='', sa_column=Column(Text), description='Ice Breaker ')
guide_question: Optional[List] = Field(default_factory=list, sa_column=Column(JSON),
description='Facilitation Questions')
model_name: str = Field(default='', description='Corresponds to the only model in the model managementID')
temperature: float = Field(default=0.5, description='Model Temperature')
max_token: int = Field(default=32000, description='MaxtokenQuantity')
status: int = Field(default=AssistantStatus.OFFLINE.value, description='Whether the assistant is online')
user_id: int = Field(default=0, description='Create UserID')
is_delete: int = Field(default=0, description='Remove logo')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class AssistantLinkBase(SQLModelSerializable):
id: Optional[int] = Field(default=None, nullable=False, primary_key=True, description='Uniqueness quantificationID')
assistant_id: Optional[str] = Field(default=0, index=True, description='assistantID')
tool_id: Optional[int] = Field(default=0, index=True, description='ToolsID')
flow_id: Optional[str] = Field(default='', index=True, description='SkillID')
knowledge_id: Optional[int] = Field(default=0, index=True, description='The knowledge base uponID')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class Assistant(AssistantBase, table=True):
id: str = Field(default_factory=generate_uuid, primary_key=True, unique=True)
class AssistantLink(AssistantLinkBase, table=True):
pass
class AssistantDao(AssistantBase):
@classmethod
def create_assistant(cls, data: Assistant) -> Assistant:
with get_sync_db_session() as session:
session.add(data)
session.commit()
session.refresh(data)
return data
@classmethod
def update_assistant(cls, data: Assistant) -> Assistant:
with get_sync_db_session() as session:
session.add(data)
session.commit()
session.refresh(data)
return data
@classmethod
def delete_assistant(cls, data: Assistant) -> Assistant:
with get_sync_db_session() as session:
data.is_delete = 1
session.add(data)
session.commit()
session.refresh(data)
return data
@classmethod
def get_one_assistant(cls, assistant_id: str) -> Assistant:
with get_sync_db_session() as session:
statement = select(Assistant).where(Assistant.id == assistant_id)
return session.exec(statement).first()
@classmethod
async def aget_one_assistant(cls, assistant_id: str) -> Assistant:
statement = select(Assistant).where(Assistant.id == assistant_id)
async with get_async_db_session() as session:
return (await session.exec(statement)).first()
@classmethod
def get_assistants_by_ids(cls, assistant_ids: List[str]) -> List[Assistant]:
if not assistant_ids:
return []
with get_sync_db_session() as session:
statement = select(Assistant).where(Assistant.id.in_(assistant_ids))
return session.exec(statement).all()
@classmethod
async def aget_assistants_by_ids(cls, assistant_ids: List[str]) -> List[Assistant]:
if not assistant_ids:
return []
statement = select(Assistant).where(col(Assistant.id).in_(assistant_ids))
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def get_assistant_by_name_user_id(cls, name: str, user_id: int) -> Assistant:
with get_sync_db_session() as session:
statement = select(Assistant).filter(Assistant.name == name,
Assistant.user_id == user_id,
Assistant.is_delete == 0)
return session.exec(statement).first()
@classmethod
def get_assistants(cls, user_id: int, name: str, assistant_ids_extra: List[str],
status: Optional[int], page: int, limit: int, assistant_ids: List[str] = None) -> (
List[Assistant], int):
with get_sync_db_session() as session:
count_statement = session.query(func.count(
Assistant.id)).where(Assistant.is_delete == 0)
statement = select(Assistant).where(Assistant.is_delete == 0)
if assistant_ids_extra:
# Membutuhkanor Requirements to join
statement = statement.where(
or_(Assistant.id.in_(assistant_ids_extra), Assistant.user_id == user_id))
count_statement = count_statement.where(
or_(Assistant.id.in_(assistant_ids_extra), Assistant.user_id == user_id))
else:
statement = statement.where(Assistant.user_id == user_id)
count_statement = count_statement.where(Assistant.user_id == user_id)
if assistant_ids:
statement = statement.where(Assistant.id.in_(assistant_ids))
count_statement = count_statement.where(Assistant.id.in_(assistant_ids))
if name:
statement = statement.where(or_(
Assistant.name.like(f'%{name}%'),
Assistant.desc.like(f'%{name}%')
))
count_statement = count_statement.where(or_(
Assistant.name.like(f'%{name}%'),
Assistant.desc.like(f'%{name}%')
))
if status is not None:
statement = statement.where(Assistant.status == status)
count_statement = count_statement.where(Assistant.status == status)
if limit == 0 and page == 0:
# Get all, no pagination
statement = statement.order_by(Assistant.update_time.desc())
else:
statement = statement.offset(
(page - 1) * limit).limit(limit).order_by(Assistant.update_time.desc())
return session.exec(statement).all(), session.exec(count_statement).scalar()
@classmethod
def get_all_online_assistants(cls, flow_ids: List[str]) -> List[Assistant]:
""" Get all live assistants """
statement = select(Assistant).filter(Assistant.status == AssistantStatus.ONLINE.value,
Assistant.is_delete == 0)
if flow_ids:
statement = statement.where(Assistant.flow_id.in_(flow_ids))
statement = statement.order_by(Assistant.update_time.desc())
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
def get_all_assistants(cls, name: str, page: int, limit: int, assistant_ids: List[str] = None,
status: int = None) -> (List[Assistant], int):
with get_sync_db_session() as session:
statement = select(Assistant).where(Assistant.is_delete == 0)
count_statement = session.query(func.count(
Assistant.id)).where(Assistant.is_delete == 0)
if name:
statement = statement.where(or_(
Assistant.name.like(f'%{name}%'),
Assistant.desc.like(f'%{name}%')
))
count_statement = count_statement.where(or_(
Assistant.name.like(f'%{name}%'),
Assistant.desc.like(f'%{name}%')
))
if assistant_ids:
statement = statement.where(Assistant.id.in_(assistant_ids))
count_statement = count_statement.where(Assistant.id.in_(assistant_ids))
if status is not None:
statement = statement.where(Assistant.status == status)
count_statement = count_statement.where(Assistant.status == status)
if page and limit:
statement = statement.offset(
(page - 1) * limit
).limit(limit)
statement = statement.order_by(Assistant.update_time.desc())
return session.exec(statement).all(), session.exec(count_statement).scalar()
@classmethod
def get_assistants_by_access(cls, role_id: int, name: str, page_size: int,
page_num: int) -> List[Tuple[Assistant, RoleAccess]]:
statment = select(Assistant,
RoleAccess).join(RoleAccess,
and_(RoleAccess.role_id == role_id,
RoleAccess.type == AccessType.ASSISTANT_READ.value,
RoleAccess.third_id == Assistant.id),
isouter=True).where(Assistant.is_delete == 0)
if name:
statment = statment.where(Assistant.name.like('%' + name + '%'))
if page_num and page_size and page_num != 'undefined':
page_num = int(page_num)
statment = statment.order_by(RoleAccess.type.desc()).order_by(
Assistant.update_time.desc()).offset((page_num - 1) * page_size).limit(page_size)
with get_sync_db_session() as session:
return session.exec(statment).all()
@classmethod
def get_count_by_filters(cls, filters: List) -> int:
with get_sync_db_session() as session:
count_statement = session.query(func.count(Assistant.id))
filters.append(Assistant.is_delete == 0)
return session.exec(count_statement.where(*filters)).scalar()
@classmethod
def filter_assistant_by_id(cls, assistant_ids: List[str], keywords: str = None, page: int = 0,
limit: int = 0) -> (List[Assistant], int):
"""
Based on keywords and assistantsidFilter out corresponding assistants
"""
statement = select(Assistant).where(Assistant.is_delete == 0)
count_statement = select(func.count(Assistant.id)).where(Assistant.is_delete == 0)
if assistant_ids:
statement = statement.where(Assistant.id.in_(assistant_ids))
count_statement = count_statement.where(Assistant.id.in_(assistant_ids))
if keywords:
statement = statement.where(or_(
Assistant.name.like(f'%{keywords}%'),
Assistant.desc.like(f'%{keywords}%')
))
count_statement = count_statement.where(or_(
Assistant.name.like(f'%{keywords}%'),
Assistant.desc.like(f'%{keywords}%')
))
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
statement = statement.order_by(Assistant.update_time.desc())
with get_sync_db_session() as session:
result = session.exec(statement).all()
return result, session.scalar(count_statement)
class AssistantLinkDao(AssistantLink):
@classmethod
def insert_batch(cls,
assistant_id: str,
tool_list: List[int] = None,
flow_list: List[str] = None):
if not tool_list and not flow_list:
return []
with get_sync_db_session() as session:
if tool_list:
for one in tool_list:
if one == 0:
continue
session.add(AssistantLink(assistant_id=assistant_id, tool_id=one))
if flow_list:
for one in flow_list:
if not one:
continue
session.add(AssistantLink(assistant_id=assistant_id, flow_id=one))
session.commit()
@classmethod
async def get_assistant_link(cls, assistant_id: str) -> List[AssistantLink]:
async with get_async_db_session() as session:
statement = select(AssistantLink).where(AssistantLink.assistant_id == assistant_id)
result = await session.exec(statement)
return result.all()
@classmethod
def update_assistant_tool(cls, assistant_id: str, tool_list: List[int]):
with get_sync_db_session() as session:
session.query(AssistantLink).filter(AssistantLink.assistant_id == assistant_id,
AssistantLink.tool_id != 0).delete()
for one in tool_list:
if one == 0:
continue
session.add(AssistantLink(assistant_id=assistant_id, tool_id=one))
session.commit()
@classmethod
def update_assistant_flow(cls, assistant_id: str, flow_list: List[str]):
with get_sync_db_session() as session:
session.query(AssistantLink).filter(AssistantLink.assistant_id == assistant_id,
AssistantLink.flow_id != '',
AssistantLink.knowledge_id == 0).delete()
for one in flow_list:
if not one:
continue
session.add(AssistantLink(assistant_id=assistant_id, flow_id=one))
session.commit()
@classmethod
def update_assistant_knowledge(cls, assistant_id: str, knowledge_list: List[int],
flow_id: str):
# Must have skills when saving knowledge base associationsID
with get_sync_db_session() as session:
session.query(AssistantLink).filter(AssistantLink.assistant_id == assistant_id,
AssistantLink.knowledge_id != 0).delete()
for one in knowledge_list:
if one == 0:
continue
session.add(
AssistantLink(assistant_id=assistant_id, knowledge_id=one, flow_id=flow_id))
session.commit()
================================================
FILE: src/backend/bisheng/database/models/audit_log.py
================================================
from datetime import datetime
from enum import Enum
from typing import List, Optional
from sqlmodel import Field, select, Column, DateTime, text, Text, func, or_, JSON, and_
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session, get_async_db_session
from bisheng.utils import generate_uuid
# System Module Enumeration
class SystemId(Enum):
CHAT = "chat" # Sessions
BUILD = "build" # Build.
KNOWLEDGE = "knowledge" # The knowledge base upon
SYSTEM = "system" # System
DASHBOARD = "dashboard" # KANBAN
# Action Behavior Enumeration
class EventType(Enum):
CREATE_CHAT = "create_chat" # New Session
DELETE_CHAT = "delete_chat" # Delete Thread
CREATE_BUILD = "create_build" # New App
UPDATE_BUILD = "update_build" # Edit App Page
DELETE_BUILD = "delete_build" # Delete App?
CREATE_KNOWLEDGE = "create_knowledge" # New Knowledge Base
DELETE_KNOWLEDGE = "delete_knowledge" # Delete Knowledge Base
UPLOAD_FILE = "upload_file" # Knowledge Base Upload Files
DELETE_FILE = "delete_file" # Knowledge Base Delete File
UPDATE_USER = "update_user" # Edit account
FORBID_USER = "forbid_user" # Deactivate user
RECOVER_USER = "recover_user" # Enable User
CREATE_USER_GROUP = "create_user_group" # Add Usergroup
DELETE_USER_GROUP = "delete_user_group" # Can delete existing usergroups
UPDATE_USER_GROUP = "update_user_group" # Can edit existing usergroups
CREATE_ROLE = "create_role" # Create a Role
DELETE_ROLE = "delete_role" # Delete a Role
UPDATE_ROLE = "update_role" # Edit a Role
ADD_TOOL = "add_tool" # Add Widget
UPDATE_TOOL = "update_tool"
DELETE_TOOL = "delete_tool"
USER_LOGIN = "user_login" # Login Pengguna
CREATE_DASHBOARD = "create_dashboard"
UPDATE_DASHBOARD = "update_dashboard"
DELETE_DASHBOARD = "delete_dashboard"
# Action object type enumeration
class ObjectType(Enum):
NONE = "none" # W/O
FLOW = "flow" # Skill
WORK_FLOW = "work_flow" # The Workflow
ASSISTANT = "assistant" # assistant
WORKSTATION = "workstation" # Workstation
KNOWLEDGE = "knowledge" # The knowledge base upon
FILE = "file" # Doc.
USER_CONF = "user_conf" # User Configuration
USER_GROUP_CONF = "user_group_conf" # User Group Configuration
ROLE_CONF = "role_conf" # Configuration of user roles
TOOL = "tool"
DASHBOARD = "dashboard" # KANBAN
class AuditLogBase(SQLModelSerializable):
"""
Audit Log Table
"""
operator_id: int = Field(index=True, description="Operating User'sID")
operator_name: Optional[str] = Field(description="Username")
group_ids: Optional[List[int | str]] = Field(sa_column=Column(JSON),
description="Belongs to a user groupIDVertical")
system_id: Optional[str] = Field(index=True, description="Module Item")
event_type: Optional[str] = Field(index=True, description="Operation behaviors")
object_type: Optional[str] = Field(index=True, description="Action object type")
object_id: Optional[str] = Field(index=True, description="Operation ObjectID")
object_name: Optional[str] = Field(sa_column=Column(Text), description="Action object name")
note: Optional[str] = Field(sa_column=Column(Text), description="Action notes")
ip_address: Optional[str] = Field(index=True,
description="Client's at time of operationIPAddress:")
create_time: Optional[datetime] = Field(sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')), description="operate time")
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class AuditLog(AuditLogBase, table=True):
# id = 2 Represents the default user group
id: str = Field(default_factory=generate_uuid, primary_key=True, index=True, description="primary keyuuidFormat")
class AuditLogDao(AuditLogBase):
@classmethod
async def get_audit_logs(cls, group_ids: List[int], operator_ids: List[int] = 0,
start_time: datetime = None,
end_time: datetime = None, system_id: str = None, event_type: str = None,
page: int = 0, limit: int = 0) -> (List[AuditLog], int):
"""
Filter logs by user group
"""
statement = select(AuditLog)
count_statement = select(func.count(AuditLog.id))
if group_ids:
group_filters = []
for one in group_ids:
group_filters.append(func.json_contains(AuditLog.group_ids, str(one)))
statement = statement.where(or_(*group_filters))
count_statement = count_statement.where(or_(*group_filters))
if operator_ids:
statement = statement.where(AuditLog.operator_id.in_(operator_ids))
count_statement = count_statement.where(AuditLog.operator_id.in_(operator_ids))
if start_time:
statement = statement.where(AuditLog.create_time >= start_time)
count_statement = count_statement.where(AuditLog.create_time >= start_time)
if end_time:
statement = statement.where(AuditLog.create_time <= end_time)
count_statement = count_statement.where(AuditLog.create_time <= end_time)
if system_id:
statement = statement.where(AuditLog.system_id == system_id)
count_statement = count_statement.where(AuditLog.system_id == system_id)
if event_type:
statement = statement.where(AuditLog.event_type == event_type)
count_statement = count_statement.where(AuditLog.event_type == event_type)
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit).order_by(AuditLog.create_time.desc())
with get_sync_db_session() as session:
return session.exec(statement).all(), session.scalar(count_statement)
@classmethod
def insert_audit_logs(cls, audit_logs: List[AuditLog]):
with get_sync_db_session() as session:
session.add_all(audit_logs)
session.commit()
@classmethod
async def ainsert_audit_logs(cls, audit_logs: List[AuditLog]):
async with get_async_db_session() as session:
session.add_all(audit_logs)
await session.commit()
@classmethod
def get_all_operators(cls, group_ids: List[int]):
statement = select(AuditLog.operator_id, AuditLog.operator_name).distinct()
if group_ids:
group_filters = []
for one in group_ids:
group_filters.append(func.json_contains(AuditLog.group_ids, str(one)))
statement = statement.where(or_(*group_filters))
with get_sync_db_session() as session:
return session.exec(statement).all()
================================================
FILE: src/backend/bisheng/database/models/component.py
================================================
from datetime import datetime
from typing import Any, List, Optional
from sqlmodel import JSON, Column, DateTime, Field, select, text
from bisheng.core.database import get_sync_db_session
from bisheng.common.models.base import SQLModelSerializable
from bisheng.utils import generate_uuid
class ComponentBase(SQLModelSerializable):
name: str = Field(max_length=50, index=True, description='Saved Component Name')
description: Optional[str] = Field(default='', description='DESCRIPTION')
data: Optional[Any] = Field(default=None, description='Component Data')
version: str = Field(default='', index=True, description='Component Revision')
user_id: int = Field(default=None, index=True, description='creatorID')
user_name: str = Field(default=None, description='creatorName')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class Component(ComponentBase, table=True):
id: str = Field(default_factory=generate_uuid, primary_key=True, unique=True)
data: Optional[Any] = Field(default=None, sa_column=Column(JSON))
class ComponentDao(ComponentBase):
@classmethod
def get_user_components(cls, user_id: int) -> List[Component]:
with get_sync_db_session() as session:
statement = select(Component).where(
Component.user_id == user_id
).order_by(Component.create_time.desc())
return session.exec(statement).all()
@classmethod
def get_component_by_name(cls, user_id: int, name: str) -> Component | None:
with get_sync_db_session() as session:
statement = select(Component).where(Component.user_id == user_id, Component.name == name)
return session.exec(statement).first()
@classmethod
def insert_component(cls, component: Component) -> Component:
with get_sync_db_session() as session:
session.add(component)
session.commit()
session.refresh(component)
return component
@classmethod
def update_component(cls, component: Component) -> Component:
with get_sync_db_session() as session:
session.add(component)
session.commit()
session.refresh(component)
return component
@classmethod
def delete_component(cls, component: Component) -> Component:
with get_sync_db_session() as session:
session.delete(component)
session.commit()
return component
================================================
FILE: src/backend/bisheng/database/models/dataset.py
================================================
from datetime import datetime
from typing import Any, List, Optional
from sqlalchemy import Column, DateTime, delete, text
from sqlmodel import Field, select
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session
class DatasetBase(SQLModelSerializable):
user_id: Optional[int] = Field(index=True, description='Create Userid')
name: str = Field(index=True, description='Dataset Name')
type: str = Field(index=False, default=0, description='Reserved Fields')
description: Optional[str] = Field(default=None, index=False, description='Dataset description')
object_name: Optional[str] = Field(default=None, index=False, description='data setS3Part Name')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class Dataset(DatasetBase, table=True):
id: Optional[int] = Field(default=None, primary_key=True)
class DatasetCreate(DatasetBase):
pass
class DatasetRead(DatasetBase):
user_name: Optional[str]
url: Optional[str]
class DatasetUpdate(DatasetBase):
pass
class DatasetDao(DatasetBase):
@classmethod
def filter_dataset_by_ids(cls,
dataset_ids: List[int],
keyword: str = None,
page: int = 0,
limit: int = 0) -> List[Dataset]:
with get_sync_db_session() as session:
query = select(Dataset)
if dataset_ids:
query = query.where(Dataset.id.in_(dataset_ids))
if keyword:
query = query.where(Dataset.name.like(f'%{keyword}%'))
if page and limit:
query = query.order_by(Dataset.update_time.desc()).offset(
(page - 1) * limit).limit(limit)
return session.exec(query).all()
@classmethod
def get_count_by_filter(cls, filters: List[Any]) -> int:
with get_sync_db_session() as session:
return session.scalar(select(Dataset.id).where(*filters))
@classmethod
def insert(cls, dataset: DatasetCreate):
with get_sync_db_session() as session:
db_insert = Dataset.validate(dataset)
session.add(db_insert)
session.commit()
session.refresh(db_insert)
return db_insert
@classmethod
def get_dataset_by_name(cls, name: str):
with get_sync_db_session() as session:
return session.exec(select(Dataset).where(Dataset.name == name)).all()
@classmethod
def update(cls, data: Dataset):
with get_sync_db_session() as session:
session.add(data)
session.commit()
session.refresh(data)
return data
@classmethod
def delete(cls, dataset_id: int):
with get_sync_db_session() as session:
session.exec(delete(Dataset).where(Dataset.id == dataset_id))
session.commit()
================================================
FILE: src/backend/bisheng/database/models/evaluation.py
================================================
from datetime import datetime
from enum import Enum
from typing import List, Optional, Dict
from sqlalchemy import Column, DateTime, Text, text, func, and_, JSON
from sqlmodel import Field, select
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session
class ExecType(Enum):
FLOW = 'flow'
ASSISTANT = 'assistant'
WORKFLOW = 'workflow'
class EvaluationTaskStatus(Enum):
running = 1
failed = 2
success = 3
class EvaluationBase(SQLModelSerializable):
user_id: int = Field(default=None, index=True)
file_name: str = Field(default='', description='Uploaded filename')
file_path: str = Field(default='', description='Doc. minio Address:')
exec_type: str = Field(description='Execute subject categories. Assistants, Skills, Workflows, ReferenceExecTypeEnum')
unique_id: str = Field(index=True, description='Unique to the executing entityID')
version: Optional[int] = Field(default=None, description='Version of workflow or skillID')
status: int = Field(index=True, default=1, description='Task Execution Status: 1:Executing "{0}" 2: execute fail 3:execute success')
prompt: str = Field(default='', sa_column=Column(Text), description='Evaluation Instruction Text')
result_file_path: str = Field(default='', description='of the assessment results minio Address:')
result_score: Optional[Dict | str] = Field(default=None, sa_column=Column(JSON), description='Final Assessment Score')
description: str = Field(default='', sa_column=Column(Text), description='Error description information')
is_delete: int = Field(default=0, description='whether delete')
create_time: Optional[datetime] = Field(default=None,
sa_column=Column(DateTime, nullable=False,
server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class Evaluation(EvaluationBase, table=True):
id: int = Field(default=None, primary_key=True, unique=True)
class EvaluationRead(EvaluationBase):
id: int
user_name: Optional[str] = None
class EvaluationCreate(EvaluationBase):
pass
class EvaluationDao(EvaluationBase):
@classmethod
def get_my_evaluations(cls, user_id: int, page: int, limit: int) -> (List[Evaluation], int):
with get_sync_db_session() as session:
statement = select(Evaluation).where(and_(Evaluation.is_delete == 0, Evaluation.user_id == user_id))
count_statement = session.query(func.count(
Evaluation.id)).where(and_(Evaluation.is_delete == 0, Evaluation.user_id == user_id))
statement = statement.offset(
(page - 1) * limit
).limit(limit).order_by(
Evaluation.update_time.desc()
)
return session.exec(statement).all(), session.exec(count_statement).scalar()
@classmethod
def delete_evaluation(cls, data: Evaluation) -> Evaluation:
with get_sync_db_session() as session:
data.is_delete = 1
session.add(data)
session.commit()
return data
@classmethod
def get_user_one_evaluation(cls, user_id: int, evaluation_id: int) -> Evaluation:
with get_sync_db_session() as session:
statement = select(Evaluation).where(and_(Evaluation.id == evaluation_id, Evaluation.user_id == user_id))
return session.exec(statement).first()
@classmethod
def get_one_evaluation(cls, evaluation_id: int) -> Evaluation:
with get_sync_db_session() as session:
statement = select(Evaluation).where(Evaluation.id == evaluation_id)
return session.exec(statement).first()
@classmethod
def update_evaluation(cls, evaluation: Evaluation) -> Evaluation:
with get_sync_db_session() as session:
session.add(evaluation)
session.commit()
session.refresh(evaluation)
return evaluation
================================================
FILE: src/backend/bisheng/database/models/flow.py
================================================
# Path: src/backend/bisheng/database/models/flow.py
from datetime import datetime
from enum import Enum
from typing import Dict, List, Optional, Tuple, Union
from pydantic import field_validator
from sqlalchemy import Column, DateTime, String, and_, func, or_, text
from sqlmodel import JSON, Field, select, update, col
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum, ApplicationTypeEnum
from bisheng.common.models.base import SQLModelSerializable
from bisheng.common.schemas.telemetry.event_data_schema import NewApplicationEventData
from bisheng.common.services import telemetry_service
from bisheng.core.database import get_sync_db_session, get_async_db_session
from bisheng.core.logger import trace_id_var
from bisheng.database.models.assistant import Assistant
from bisheng.database.models.role_access import AccessType, RoleAccess, RoleAccessDao
from bisheng.user.domain.models.user_role import UserRoleDao
from bisheng.utils import generate_uuid
# if TYPE_CHECKING:
class FlowStatus(Enum):
OFFLINE = 1
ONLINE = 2
class FlowType(Enum):
FLOW = 1
ASSISTANT = 5
WORKFLOW = 10
WORKSTATION = 15
LINSIGHT = 20 # Inspiration Mode
class AppEnum(Enum):
Flow = 'flow'
ASSISTANT = 'assistant'
WORKFLOW = 'workflow'
class UserLinkType(Enum):
app = AppEnum
class FlowBase(SQLModelSerializable):
name: str = Field(index=True)
user_id: Optional[int] = Field(default=None, index=True)
description: Optional[str] = Field(default=None, sa_column=Column(String(length=1000)))
data: Optional[Dict] = Field(default=None)
logo: Optional[str] = Field(default=None, index=False)
status: Optional[int] = Field(index=False, default=1)
flow_type: Optional[int] = Field(index=False, default=1)
guide_word: Optional[str] = Field(default=None, sa_column=Column(String(length=1000)))
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
@field_validator('data', mode='before')
@classmethod
def validate_json(cls, v):
if not v:
return v
if not isinstance(v, dict):
raise ValueError('Flow must be a valid JSON')
# data must contain nodes and edges
if 'nodes' not in v.keys():
raise ValueError('Flow must have nodes')
if 'edges' not in v.keys():
raise ValueError('Flow must have edges')
return v
class Flow(FlowBase, table=True):
id: str = Field(default_factory=generate_uuid, primary_key=True, unique=True)
data: Optional[Dict] = Field(default=None, sa_column=Column(JSON))
class FlowCreate(FlowBase):
flow_id: Optional[str] = None
class FlowRead(FlowBase):
id: str
user_name: Optional[str] = None
version_id: Optional[int] = None
class FlowReadWithStyle(FlowRead):
# style: Optional['FlowStyleRead'] = None
total: Optional[int] = None
class FlowUpdate(SQLModelSerializable):
name: Optional[str] = None
logo: Optional[str] = None
description: Optional[str] = None
data: Optional[Dict] = None
status: Optional[int] = None
guide_word: Optional[str] = None
class FlowDao(FlowBase):
@classmethod
def create_flow(cls, flow_info: Flow, flow_type: Optional[int]) -> Flow:
from bisheng.database.models.flow_version import FlowVersion
with get_sync_db_session() as session:
session.add(flow_info)
# Create a default version
flow_version = FlowVersion(name='v0',
is_current=1,
data=flow_info.data,
flow_id=flow_info.id,
create_time=datetime.now(),
user_id=flow_info.user_id,
flow_type=flow_type)
session.add(flow_version)
session.commit()
session.refresh(flow_info)
if flow_type == FlowType.FLOW.value:
app_type = ApplicationTypeEnum.SKILL
elif flow_type == FlowType.WORKFLOW.value:
app_type = ApplicationTypeEnum.WORKFLOW
elif flow_type == FlowType.ASSISTANT.value:
app_type = ApplicationTypeEnum.ASSISTANT
elif flow_type == FlowType.LINSIGHT.value:
app_type = ApplicationTypeEnum.LINSIGHT
else:
app_type = ApplicationTypeEnum.DAILY_CHAT
# RecordTelemetryJournal
telemetry_service.log_event_sync(user_id=flow_info.user_id,
event_type=BaseTelemetryTypeEnum.NEW_APPLICATION,
trace_id=trace_id_var.get(),
event_data=NewApplicationEventData(
app_id=flow_info.id,
app_name=flow_info.name,
app_type=app_type.value
))
return flow_info
@classmethod
def delete_flow(cls, flow_info: Flow) -> Flow:
from bisheng.database.models.flow_version import FlowVersion
with get_sync_db_session() as session:
session.delete(flow_info)
# Delete the corresponding version information
update_statement = update(FlowVersion).where(
FlowVersion.flow_id == flow_info.id).values(is_delete=1)
session.exec(update_statement)
session.commit()
return flow_info
@classmethod
def get_flow_by_id(cls, flow_id: str) -> Optional[Flow]:
with get_sync_db_session() as session:
statement = select(Flow).where(Flow.id == flow_id)
return session.exec(statement).first()
@classmethod
async def aget_flow_by_id(cls, flow_id: str) -> Optional[Flow]:
async with get_async_db_session() as session:
statement = select(Flow).where(Flow.id == flow_id)
result = await session.exec(statement)
return result.first()
@classmethod
def get_flow_by_idstr(cls, flow_id: str) -> Optional[Flow]:
with get_sync_db_session() as session:
statement = select(Flow).where(Flow.id == flow_id)
return session.exec(statement).first()
@classmethod
def get_flow_by_ids(cls, flow_ids: List[str]) -> List[Flow]:
if not flow_ids:
return []
with get_sync_db_session() as session:
statement = select(Flow).where(Flow.id.in_(flow_ids))
return session.exec(statement).all()
@classmethod
async def aget_flow_by_ids(cls, flow_ids: List[str]) -> List[Flow]:
if not flow_ids:
return []
async with get_async_db_session() as session:
statement = select(Flow).where(col(Flow.id).in_(flow_ids))
result = await session.exec(statement)
return result.all()
@classmethod
def get_flow_by_user(cls, user_id: int) -> List[Flow]:
with get_sync_db_session() as session:
statement = select(Flow).where(Flow.user_id == user_id)
return session.exec(statement).all()
@classmethod
def get_flow_by_name(cls, user_id: int, name: str) -> Optional[Flow]:
with get_sync_db_session() as session:
statement = select(Flow).where(Flow.user_id == user_id, Flow.name == name)
return session.exec(statement).first()
@classmethod
def get_flow_list_by_name(cls, name: str) -> List[Flow]:
with get_sync_db_session() as session:
statement = select(Flow).where(Flow.name.like('%{}%'.format(name)))
return session.exec(statement).all()
@classmethod
def get_flow_by_access(cls, role_id: int, name: str, page_size: int,
page_num: int) -> List[Tuple[Flow, RoleAccess]]:
statment = select(Flow, RoleAccess).join(RoleAccess,
and_(RoleAccess.role_id == role_id,
RoleAccess.type == AccessType.FLOW.value,
RoleAccess.third_id == Flow.id),
isouter=True)
if name:
statment = statment.where(Flow.name.like('%' + name + '%'))
if page_num and page_size and page_num != 'undefined':
page_num = int(page_num)
statment = statment.order_by(RoleAccess.type.desc()).order_by(
Flow.update_time.desc()).offset((page_num - 1) * page_size).limit(page_size)
with get_sync_db_session() as session:
return session.exec(statment).all()
@classmethod
def get_count_by_filters(cls, filters) -> int:
with get_sync_db_session() as session:
count_statement = session.query(func.count(Flow.id))
return session.exec(count_statement.where(*filters)).scalar()
@classmethod
def get_flows(cls,
user_id: Optional[int],
extra_ids: Union[List[str], str],
name: str,
status: Optional[int] = None,
flow_ids: List[str] = None,
page: int = 0,
limit: int = 0,
flow_type: Optional[int] = None) -> List[Flow]:
with get_sync_db_session() as session:
# data The amount of data is too large, yesmysql Influential
statement = select(Flow.id, Flow.user_id, Flow.name, Flow.status, Flow.create_time,
Flow.logo, Flow.update_time, Flow.description, Flow.guide_word,
Flow.flow_type)
if extra_ids and isinstance(extra_ids, List):
statement = statement.where(or_(Flow.id.in_(extra_ids), Flow.user_id == user_id))
elif not extra_ids:
statement = statement.where(Flow.user_id == user_id)
if name:
statement = statement.where(
or_(Flow.name.like(f'%{name}%'), Flow.description.like(f'%{name}%')))
if status is not None:
statement = statement.where(Flow.status == status)
if flow_type is not None:
statement = statement.where(Flow.flow_type == flow_type)
if flow_ids:
statement = statement.where(Flow.id.in_(flow_ids))
statement = statement.order_by(Flow.update_time.desc())
if page > 0 and limit > 0:
statement = statement.offset((page - 1) * limit).limit(limit)
flows = session.exec(statement)
flows_partial = flows.mappings().all()
return [Flow.model_validate(f) for f in flows_partial]
@classmethod
def count_flows(cls,
user_id: Optional[int],
extra_ids: Union[List[str], str],
name: str,
status: Optional[int] = None,
flow_ids: List[str] = None,
flow_type: Optional[int] = None) -> int:
with get_sync_db_session() as session:
count_statement = session.query(func.count(Flow.id))
if extra_ids and isinstance(extra_ids, List):
count_statement = count_statement.filter(
or_(Flow.id.in_(extra_ids), Flow.user_id == user_id))
elif not extra_ids:
count_statement = count_statement.filter(Flow.user_id == user_id)
if name:
count_statement = count_statement.filter(
or_(Flow.name.like(f'%{name}%'), Flow.description.like(f'%{name}%')))
if flow_type is not None:
count_statement = count_statement.where(Flow.flow_type == flow_type)
if flow_ids:
count_statement = count_statement.filter(Flow.id.in_(flow_ids))
if status is not None:
count_statement = count_statement.filter(Flow.status == status)
return count_statement.scalar()
@classmethod
def get_all_online_flows(cls,
keyword: str = None,
flow_ids: List[str] = None,
flow_type: int = FlowType.FLOW.value) -> List[Flow]:
with get_sync_db_session() as session:
statement = select(Flow.id, Flow.user_id, Flow.name, Flow.status, Flow.create_time,
Flow.logo, Flow.update_time, Flow.description,
Flow.guide_word).where(Flow.status == FlowStatus.ONLINE.value)
if flow_ids:
statement = statement.where(Flow.id.in_(flow_ids))
if keyword:
statement = statement.where(
or_(Flow.name.like(f'%{keyword}%'), Flow.description.like(f'%{keyword}%')))
result = session.exec(statement).mappings().all()
return [Flow.model_validate(f) for f in result]
@classmethod
def get_user_access_online_flows(cls,
user_id: int,
page: int = 0,
limit: int = 0,
keyword: str = None,
flow_ids: List[str] = None,
flow_type: int = FlowType.FLOW.value) -> List[Flow]:
user_role = UserRoleDao.get_user_roles(user_id)
flow_id_extra = []
if user_role:
role_ids = [role.role_id for role in user_role]
if 1 in role_ids:
# admin
flow_id_extra = 'admin'
else:
role_access = RoleAccessDao.get_role_access(role_ids, AccessType.FLOW)
if role_access:
flow_id_extra = [access.third_id for access in role_access]
return FlowDao.get_flows(user_id,
flow_id_extra,
keyword,
FlowStatus.ONLINE.value,
flow_ids=flow_ids,
page=page,
limit=limit,
flow_type=flow_type)
@classmethod
def filter_flows_by_ids(cls, flow_ids: List[str], keyword: str = None,
page: int = 0, limit: int = 0, flow_type: int = FlowType.FLOW.value) \
-> (List[Flow], int):
"""
Via SkillsIDFilter the list of skills by returning only brief information and not includingdata
"""
statement = select(Flow.id, Flow.user_id, Flow.name, Flow.status, Flow.create_time,
Flow.update_time, Flow.description, Flow.guide_word)
count_statement = select(func.count(Flow.id))
if flow_ids:
statement = statement.where(Flow.id.in_(flow_ids))
count_statement = count_statement.where(Flow.id.in_(flow_ids))
if keyword:
statement = statement.where(
or_(Flow.name.like(f'%{keyword}%'), Flow.description.like(f'%{keyword}%')))
count_statement = count_statement.where(
or_(Flow.name.like(f'%{keyword}%'), Flow.description.like(f'%{keyword}%')))
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
statement = statement.where(Flow.flow_type == flow_type)
statement = statement.order_by(Flow.update_time.desc())
with get_sync_db_session() as session:
result = session.exec(statement).mappings().all()
return [Flow.model_validate(f) for f in result], session.scalar(count_statement)
@classmethod
def update_flow(cls, flow: Flow) -> Flow:
with get_sync_db_session() as session:
session.add(flow)
session.commit()
session.refresh(flow)
return flow
@classmethod
async def aupdate_flow(cls, flow: Flow) -> Flow:
async with get_async_db_session() as session:
session.add(flow)
await session.commit()
await session.refresh(flow)
return flow
@classmethod
def get_all_apps(cls,
name: str = None,
status: int = None,
id_list: list = None,
flow_type: int = None,
user_id: int = None,
id_extra: list = None,
id_list_not_in: list = None,
page: int = 0,
limit: int = 0) -> (List[Dict], int):
""" Get all apps Contains skills, assistants, workflows """
sub_query = select(
Flow.id, Flow.name, Flow.description, Flow.flow_type, Flow.logo, Flow.user_id,
Flow.status, Flow.create_time, Flow.update_time).union_all(
select(Assistant.id, Assistant.name, Assistant.desc, FlowType.ASSISTANT.value,
Assistant.logo, Assistant.user_id, Assistant.status, Assistant.create_time,
Assistant.update_time).where(Assistant.is_delete == 0)).subquery()
statement = select(sub_query.c.id, sub_query.c.name, sub_query.c.description,
sub_query.c.flow_type, sub_query.c.logo, sub_query.c.user_id,
sub_query.c.status, sub_query.c.create_time, sub_query.c.update_time)
count_statement = select(func.count(sub_query.c.id))
if name:
statement = statement.where(sub_query.c.name.like(f'%{name}%'))
count_statement = count_statement.where(sub_query.c.name.like(f'%{name}%'))
if status is not None:
statement = statement.where(sub_query.c.status == status)
count_statement = count_statement.where(sub_query.c.status == status)
if id_list:
statement = statement.where(sub_query.c.id.in_(id_list))
count_statement = count_statement.where(sub_query.c.id.in_(id_list))
if flow_type is not None:
statement = statement.where(sub_query.c.flow_type == flow_type)
count_statement = count_statement.where(sub_query.c.flow_type == flow_type)
if user_id is not None:
if id_extra:
statement = statement.where(
or_(sub_query.c.user_id == user_id, sub_query.c.id.in_(id_extra)))
count_statement = count_statement.where(
or_(sub_query.c.user_id == user_id, sub_query.c.id.in_(id_extra)))
else:
statement = statement.where(sub_query.c.user_id == user_id)
count_statement = count_statement.where(sub_query.c.user_id == user_id)
if id_list_not_in:
statement = statement.where(~sub_query.c.id.in_(id_list_not_in))
count_statement = count_statement.where(~sub_query.c.id.in_(id_list_not_in))
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
statement = statement.order_by(sub_query.c.update_time.desc())
with (get_sync_db_session() as session):
ret = session.exec(statement).all()
total = session.scalar(count_statement)
data = []
for one in ret:
data.append({
'id': one[0],
'name': one[1],
'description': one[2],
'flow_type': one[3],
'logo': one[4],
'user_id': one[5],
'status': one[6],
'create_time': one[7],
'update_time': one[8]
})
return data, total
@classmethod
async def aget_all_apps(cls,
name: str = None,
status: int = None,
id_list: list = None,
flow_type: int = None,
user_id: int = None,
id_extra: list = None,
id_list_not_in: list = None,
page: int = 0,
limit: int = 0) -> (List[Dict], int):
"""
Get all apps Contains skills, assistants, workflows
Args:
name:
status:
id_list:
flow_type:
user_id:
id_extra:
id_list_not_in:
page:
limit:
Returns:
(List[Dict], int)
"""
sub_query = select(
Flow.id, Flow.name, Flow.description, Flow.flow_type, Flow.logo, Flow.user_id,
Flow.status, Flow.create_time, Flow.update_time).union_all(
select(Assistant.id, Assistant.name, Assistant.desc, FlowType.ASSISTANT.value,
Assistant.logo, Assistant.user_id, Assistant.status, Assistant.create_time,
Assistant.update_time).where(Assistant.is_delete == 0)).subquery()
statement = select(sub_query.c.id, sub_query.c.name, sub_query.c.description,
sub_query.c.flow_type, sub_query.c.logo, sub_query.c.user_id,
sub_query.c.status, sub_query.c.create_time, sub_query.c.update_time)
count_statement = select(func.count(sub_query.c.id))
if name:
statement = statement.where(sub_query.c.name.like(f'%{name}%'))
count_statement = count_statement.where(sub_query.c.name.like(f'%{name}%'))
if status is not None:
statement = statement.where(sub_query.c.status == status)
count_statement = count_statement.where(sub_query.c.status == status)
if id_list:
statement = statement.where(sub_query.c.id.in_(id_list))
count_statement = count_statement.where(sub_query.c.id.in_(id_list))
if flow_type is not None:
statement = statement.where(sub_query.c.flow_type == flow_type)
count_statement = count_statement.where(sub_query.c.flow_type == flow_type)
if user_id is not None:
if id_extra:
statement = statement.where(
or_(sub_query.c.user_id == user_id, sub_query.c.id.in_(id_extra)))
count_statement = count_statement.where(
or_(sub_query.c.user_id == user_id, sub_query.c.id.in_(id_extra)))
else:
statement = statement.where(sub_query.c.user_id == user_id)
count_statement = count_statement.where(sub_query.c.user_id == user_id)
if id_list_not_in:
statement = statement.where(~sub_query.c.id.in_(id_list_not_in))
count_statement = count_statement.where(~sub_query.c.id.in_(id_list_not_in))
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
statement = statement.order_by(sub_query.c.update_time.desc())
async with get_async_db_session() as session:
result = await session.exec(statement)
ret = result.all()
total_result = await session.exec(count_statement)
total = total_result.first()
data = []
for one in ret:
data.append({
'id': one[0],
'name': one[1],
'description': one[2],
'flow_type': one[3],
'logo': one[4],
'user_id': one[5],
'status': one[6],
'create_time': one[7],
'update_time': one[8]
})
return data, total
@classmethod
async def get_one_flow_simple(cls, flow_id: str) -> Optional[Flow]:
""" get simple info of one flow by id. not contain data field """
statement = select(Flow.id, Flow.name, Flow.description, Flow.flow_type, Flow.logo, Flow.user_id,
Flow.status, Flow.create_time, Flow.update_time).where(Flow.id == flow_id)
async with get_async_db_session() as session:
result = await session.exec(statement)
one = result.first()
if not one:
return None
return Flow(**{
'id': one[0],
'name': one[1],
'description': one[2],
'flow_type': one[3],
'logo': one[4],
'user_id': one[5],
'status': one[6],
'create_time': one[7],
'update_time': one[8]
})
@classmethod
def get_one_flow_simple_sync(cls, flow_id: str) -> Optional[Flow]:
""" get simple info of one flow by id. not contain data field """
statement = select(Flow.id, Flow.name, Flow.description, Flow.flow_type, Flow.logo, Flow.user_id,
Flow.status, Flow.create_time, Flow.update_time).where(Flow.id == flow_id)
with get_sync_db_session() as session:
result = session.exec(statement)
one = result.first()
if not one:
return None
return Flow(**{
'id': one[0],
'name': one[1],
'description': one[2],
'flow_type': one[3],
'logo': one[4],
'user_id': one[5],
'status': one[6],
'create_time': one[7],
'update_time': one[8]
})
@classmethod
def get_all_app_by_time_range_sync(cls, start_time: datetime, end_time: datetime, page: int = 0,
page_size: int = 0):
sub_query = select(
Flow.id, Flow.name, Flow.description, Flow.flow_type, Flow.logo, Flow.user_id,
Flow.status, Flow.create_time, Flow.update_time).union_all(
select(Assistant.id, Assistant.name, Assistant.desc, FlowType.ASSISTANT.value,
Assistant.logo, Assistant.user_id, Assistant.status, Assistant.create_time,
Assistant.update_time).where(Assistant.is_delete == 0)).subquery()
statement = select(sub_query.c.id, sub_query.c.name, sub_query.c.description,
sub_query.c.flow_type, sub_query.c.logo, sub_query.c.user_id,
sub_query.c.status, sub_query.c.create_time, sub_query.c.update_time)
statement = statement.where(and_(sub_query.c.create_time >= start_time,
sub_query.c.create_time < end_time))
if page and page_size:
statement = statement.offset((page - 1) * page_size).limit(page_size)
with get_sync_db_session() as session:
result = session.exec(statement).all()
data = []
for one in result:
data.append({
'id': one[0],
'name': one[1],
'description': one[2],
'flow_type': one[3],
'logo': one[4],
'user_id': one[5],
'status': one[6],
'create_time': one[7],
'update_time': one[8]
})
return data
@classmethod
def get_first_app(cls):
sub_query = select(
Flow.id, Flow.name, Flow.description, Flow.flow_type, Flow.logo, Flow.user_id,
Flow.status, Flow.create_time, Flow.update_time).union_all(
select(Assistant.id, Assistant.name, Assistant.desc, FlowType.ASSISTANT.value,
Assistant.logo, Assistant.user_id, Assistant.status, Assistant.create_time,
Assistant.update_time).where(Assistant.is_delete == 0)).subquery()
statement = select(sub_query.c.id, sub_query.c.name, sub_query.c.description,
sub_query.c.flow_type, sub_query.c.logo, sub_query.c.user_id,
sub_query.c.status, sub_query.c.create_time, sub_query.c.update_time)
statement = statement.order_by(sub_query.c.create_time.asc()).limit(1)
with get_sync_db_session() as session:
result = session.exec(statement).all()
data = []
for one in result:
data.append({
'id': one[0],
'name': one[1],
'description': one[2],
'flow_type': one[3],
'logo': one[4],
'user_id': one[5],
'status': one[6],
'create_time': one[7],
'update_time': one[8]
})
return data[0] if data else None
================================================
FILE: src/backend/bisheng/database/models/flow_version.py
================================================
# Path: src/backend/bisheng/database/models/flow.py
from datetime import datetime
from typing import Dict, List, Optional
# if TYPE_CHECKING:
from pydantic import field_validator
from sqlalchemy import func, String
from sqlmodel import JSON, Field, select, update, text, Column, DateTime
from bisheng.core.database import get_sync_db_session, get_async_db_session
from bisheng.common.models.base import SQLModelSerializable
from bisheng.database.models.flow import Flow
class FlowVersionBase(SQLModelSerializable):
id: Optional[int] = Field(default=None, primary_key=True, unique=True)
flow_id: str = Field(index=True, max_length=32, description="Belonging SkillsID")
name: str = Field(index=True, description="Version Name")
data: Optional[Dict] = Field(default=None, description="Version Data")
description: Optional[str] = Field(default=None, sa_column=Column(String(length=1000)))
user_id: Optional[int] = Field(default=None, index=True, description="creator")
flow_type: Optional[int] = Field(default=1, description="Type of version")
is_current: Optional[int] = Field(default=0, description="Is version in use")
is_delete: Optional[int] = Field(default=0, description="whether delete")
original_version_id: Optional[int] = Field(default=None, description="Source Version ofID")
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
@field_validator('data')
@classmethod
def validate_json(cls, v):
# dict_keys(['description', 'name', 'id', 'data'])
if not v:
return v
if not isinstance(v, dict):
raise ValueError('Flow must be a valid JSON')
# data must contain nodes and edges
if 'nodes' not in v.keys():
raise ValueError('Flow must have nodes')
if 'edges' not in v.keys():
raise ValueError('Flow must have edges')
return v
class FlowVersion(FlowVersionBase, table=True):
data: Optional[Dict] = Field(default=None, sa_column=Column(JSON), description="Version Data")
class FlowVersionRead(FlowVersionBase):
pass
class FlowVersionDao(FlowVersionBase):
@classmethod
def create_version(cls, version: FlowVersion) -> FlowVersion:
"""
Create New Version
"""
with get_sync_db_session() as session:
session.add(version)
session.commit()
session.refresh(version)
return version
@classmethod
def update_version(cls, version: FlowVersion) -> FlowVersion:
"""
Update the version information while updating the Skill SheetdataDATA
"""
with get_sync_db_session() as session:
session.add(version)
session.commit()
# Update the data in the skill sheet if it is the current version
if version.is_current == 1:
# Update Skill SheetdataDATA
update_flow = update(Flow).where(Flow.id == version.flow_id).values(data=version.data)
session.exec(update_flow)
session.commit()
session.refresh(version)
return version
@classmethod
async def aupdate_version(cls, version: FlowVersion) -> FlowVersion:
"""
Update version information asynchronously while updating the Skill SheetdataDATA
"""
async with get_async_db_session() as session:
session.add(version)
await session.commit()
# Update the data in the skill sheet if it is the current version
if version.is_current == 1:
# Update Skill SheetdataDATA
update_flow = update(Flow).where(Flow.id == version.flow_id).values(data=version.data)
await session.exec(update_flow)
await session.commit()
await session.refresh(version)
return version
@classmethod
def get_version_by_name(cls, flow_id: str, name: str) -> Optional[FlowVersion]:
"""
By SkillIDand version name for version information
"""
with get_sync_db_session() as session:
statement = select(FlowVersion).where(FlowVersion.flow_id == flow_id,
FlowVersion.name == name,
FlowVersion.is_delete == 0)
return session.exec(statement).first()
@classmethod
def get_version_by_id(cls, version_id: int, include_delete: bool = False) -> Optional[FlowVersion]:
"""
According to versionIDGet information on the skill version
"""
with get_sync_db_session() as session:
statement = select(FlowVersion).where(FlowVersion.id == version_id)
if not include_delete:
statement = statement.where(FlowVersion.is_delete == 0)
return session.exec(statement).first()
@classmethod
async def aget_version_by_id(cls, version_id: int, include_delete: bool = False) -> Optional[FlowVersion]:
"""
According to versionIDGet skill version information (asynchronous)
"""
async with get_async_db_session() as session:
statement = select(FlowVersion).where(FlowVersion.id == version_id)
if not include_delete:
statement = statement.where(FlowVersion.is_delete == 0)
result = await session.exec(statement)
return result.first()
@classmethod
def get_version_by_flow(cls, flow_id: str) -> Optional[FlowVersion]:
"""
By SkillIDGet information on the current skill version
"""
with get_sync_db_session() as session:
statement = select(FlowVersion).where(FlowVersion.flow_id == flow_id,
FlowVersion.is_current == 1,
FlowVersion.is_delete == 0)
return session.exec(statement).first()
@classmethod
def get_list_by_ids(cls, ids: List[int]) -> List[FlowVersion]:
"""
accordingIDList for all version details
"""
with get_sync_db_session() as session:
statement = select(FlowVersion).where(FlowVersion.id.in_(ids))
return session.exec(statement).all()
@classmethod
def get_list_by_flow(cls, flow_id: str) -> List[FlowVersionRead]:
"""
By SkillID Get all skill versions
"""
with get_sync_db_session() as session:
statement = select(FlowVersion.id, FlowVersion.flow_id, FlowVersion.name, FlowVersion.description,
FlowVersion.is_current, FlowVersion.create_time, FlowVersion.update_time).where(
FlowVersion.flow_id == flow_id, FlowVersion.is_delete == 0).order_by(FlowVersion.id.desc())
ret = session.exec(statement).mappings().all()
return [FlowVersionRead.model_validate(f) for f in ret]
@classmethod
def count_list_by_flow(cls, flow_id: str, include_delete: bool = False) -> int:
"""
By SkillID Number of Skill Versions
"""
with get_sync_db_session() as session:
count_statement = session.query(func.count()).where(FlowVersion.flow_id == flow_id)
if not include_delete:
count_statement = count_statement.where(FlowVersion.is_delete == 0)
return count_statement.scalar()
@classmethod
def get_list_by_flow_ids(cls, flow_ids: List[str]) -> List[FlowVersionRead]:
"""
By SkillIDVertical Get all versions of all skills
"""
with get_sync_db_session() as session:
statement = select(FlowVersion.id, FlowVersion.flow_id, FlowVersion.name, FlowVersion.description,
FlowVersion.is_current, FlowVersion.create_time, FlowVersion.update_time).where(
FlowVersion.flow_id.in_(flow_ids), FlowVersion.is_delete == 0).order_by(FlowVersion.id.desc())
ret = session.exec(statement).mappings().all()
return [FlowVersionRead.model_validate(f) for f in ret]
@classmethod
def delete_flow_version(cls, version_id: int) -> None:
"""
Deleting a version, the version in use cannot be deleted
"""
with get_sync_db_session() as session:
update_statement = update(FlowVersion).where(
FlowVersion.id == version_id, FlowVersion.is_current == 0).values(is_delete=1)
session.exec(update_statement)
session.commit()
@classmethod
async def change_current_version(cls, flow_id: str, new_version_info: FlowVersion) -> bool:
"""
Modify the current version of the skill, Determine if the current version exists
Also modify the Skill SheetdataDATA
"""
async with get_async_db_session() as session:
# Set current version
set_statement = update(FlowVersion).where(
FlowVersion.flow_id == flow_id,
FlowVersion.id == new_version_info.id,
FlowVersion.is_delete == 0,
).values(is_current=1)
update_ret = await session.exec(set_statement)
if update_ret.rowcount == 0:
# If the update is not successful, the current version of the previous setting is not canceled
return False
# Update Skill SheetdataDATA
update_flow = update(Flow).where(Flow.id == flow_id).values(data=new_version_info.data)
await session.exec(update_flow)
await session.commit()
# Modify another version to not the current version
statement = update(FlowVersion).where(
FlowVersion.flow_id == flow_id,
FlowVersion.id != new_version_info.id,
FlowVersion.is_current == 1).values(
is_current=0)
await session.exec(statement)
await session.commit()
return True
================================================
FILE: src/backend/bisheng/database/models/group.py
================================================
from datetime import datetime
from typing import Dict, List, Optional
from sqlalchemy import Column, DateTime, delete, text, update
from sqlmodel import Field, select
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session, get_async_db_session
# Default User GroupID
DefaultGroup = 2
class GroupBase(SQLModelSerializable):
group_name: str = Field(index=False, description='Frontend Display Name', unique=True)
remark: Optional[str] = Field(default=None, index=False)
create_user: Optional[int] = Field(default=None, index=True, description="Creating a user'sID")
update_user: Optional[int] = Field(default=None, description="Update user'sID")
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class Group(GroupBase, table=True):
# id = 2 Represents the default user group
id: Optional[int] = Field(default=None, primary_key=True)
class GroupRead(GroupBase):
id: Optional[int] = None
group_admins: Optional[List[Dict]] = None
class GroupUpdate(GroupBase):
role_name: Optional[str] = None
remark: Optional[str] = None
class GroupCreate(GroupBase):
group_admins: Optional[List[int]] = None
class GroupDao(GroupBase):
@classmethod
def get_user_group(cls, group_id: int) -> Group | None:
with get_sync_db_session() as session:
statement = select(Group).where(Group.id == group_id)
return session.exec(statement).first()
@classmethod
def insert_group(cls, group: GroupCreate) -> Group:
with get_sync_db_session() as session:
group_add = Group.validate(group)
session.add(group_add)
session.commit()
session.refresh(group_add)
return group_add
@classmethod
def get_all_group(cls) -> list[Group]:
with get_sync_db_session() as session:
statement = select(Group).order_by(Group.update_time.desc())
return session.exec(statement).all()
@classmethod
def get_group_by_ids(cls, ids: List[int]) -> list[Group]:
if not ids:
raise ValueError('ids is empty')
with get_sync_db_session() as session:
statement = select(Group).where(Group.id.in_(ids)).order_by(Group.update_time.desc())
return session.exec(statement).all()
@classmethod
async def aget_group_by_ids(cls, ids: List[int]) -> list[Group]:
if not ids:
raise ValueError('ids is empty')
async with get_async_db_session() as session:
statement = select(Group).where(Group.id.in_(ids)).order_by(Group.update_time.desc())
result = await session.exec(statement)
return result.all()
@classmethod
def delete_group(cls, group_id: int):
with get_sync_db_session() as session:
session.exec(delete(Group).where(Group.id == group_id))
session.commit()
@classmethod
def update_group(cls, group: Group) -> Group:
with get_sync_db_session() as session:
session.add(group)
session.commit()
session.refresh(group)
return group
@classmethod
def update_group_update_user(cls, group_id: int, user_id: int):
with get_sync_db_session() as session:
statement = update(Group).where(Group.id == group_id).values(update_user=user_id,
update_time=datetime.now())
session.exec(statement)
session.commit()
================================================
FILE: src/backend/bisheng/database/models/group_resource.py
================================================
from datetime import datetime
from enum import Enum
from typing import Dict, List, Optional
from sqlalchemy import Column, DateTime, text, delete
from sqlmodel import Field, select
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session, get_async_db_session
class ResourceTypeEnum(Enum):
KNOWLEDGE = 1
FLOW = 2
ASSISTANT = 3
GPTS_TOOL = 4
WORK_FLOW = 5
DASHBOARD = 6 # KANBAN
WORKSTATION = 7 # Workstation
class GroupResourceBase(SQLModelSerializable):
group_id: str = Field(index=True)
third_id: str = Field(index=False)
type: int = Field(index=False, description='Resource Categories 1:The knowledge base upon 2:Skill 3:assistant 4:Tools 5:The Workflow')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class GroupResource(GroupResourceBase, table=True):
id: Optional[int] = Field(default=None, primary_key=True)
class GroupResourceRead(GroupResourceBase):
id: Optional[int] = None
group_admins: Optional[List[Dict]] = None
class GroupResourceUpdate(GroupResourceBase):
role_name: Optional[str] = None
remark: Optional[str] = None
class GroupResourceCreate(GroupResourceBase):
pass
class GroupResourceDao(GroupResourceBase):
@classmethod
def insert_group(cls, group_resource: GroupResource) -> GroupResource:
with get_sync_db_session() as session:
session.add(group_resource)
session.commit()
session.refresh(group_resource)
return group_resource
@classmethod
def insert_group_batch(cls, group_resources: List[GroupResource]) -> List[GroupResource]:
with get_sync_db_session() as session:
session.add_all(group_resources)
session.commit()
return group_resources
@classmethod
async def ainsert_group_batch(cls, group_resources: List[GroupResource]) -> List[GroupResource]:
async with get_async_db_session() as session:
session.add_all(group_resources)
await session.commit()
return group_resources
@classmethod
def get_group_resource(cls,
group_id: int,
resource_type: ResourceTypeEnum,
name: str = None,
page_size: int = None,
page_num: int = None) -> list[GroupResource]:
with get_sync_db_session() as session:
statement = select(GroupResource).where(GroupResource.group_id == group_id,
GroupResource.type == resource_type.value)
if name:
statement = statement.where(GroupResource.third_id.like(f'%{name}%'))
if page_num and page_size:
statement = statement.offset(page_size * (page_num - 1)).limit(page_size)
return session.exec(statement).all()
@classmethod
async def aget_group_resources(cls, group_id: int, resource_type: ResourceTypeEnum) -> List[GroupResource]:
statement = select(GroupResource).where(
GroupResource.group_id == group_id,
GroupResource.type == resource_type.value
)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
async def get_groups_resource(cls,
group_ids: List[int],
resource_types: List[ResourceTypeEnum] = None,
name: str = None,
page_size: int = None,
page_num: int = None) -> list[GroupResource]:
async with get_async_db_session() as session:
statement = select(GroupResource).where(GroupResource.group_id.in_(group_ids))
if resource_types:
statement = statement.where(GroupResource.type.in_([r.value for r in resource_types]))
if name:
statement = statement.where(GroupResource.third_id.like(f'%{name}%'))
if page_num and page_size:
statement = statement.offset(page_size * (page_num - 1)).limit(page_size)
result = await session.exec(statement)
return result.all()
@classmethod
def get_resource_group(cls, resource_type: ResourceTypeEnum, third_id: str) -> list[GroupResource]:
"""
Get the group to which the resource belongs
"""
with get_sync_db_session() as session:
statement = select(GroupResource).where(GroupResource.third_id == third_id,
GroupResource.type == resource_type.value)
return session.exec(statement).all()
@classmethod
async def aget_resource_group(cls, resource_type: ResourceTypeEnum, third_id: str) -> list[GroupResource]:
"""
Get the group to which the resource belongs
"""
async with get_async_db_session() as session:
statement = select(GroupResource).where(GroupResource.third_id == third_id,
GroupResource.type == resource_type.value)
result = await session.exec(statement)
return result.all()
@classmethod
def get_resources_group(cls, resource_type: ResourceTypeEnum | None, third_ids: List[str]) -> list[GroupResource]:
"""
Get the grouping to which the bulk resource belongs
"""
statement = select(GroupResource).where(GroupResource.third_id.in_(third_ids))
if resource_type:
statement = statement.where(GroupResource.type == resource_type.value)
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
def delete_group_resource_by_third_id(cls, third_id: str, resource_type: ResourceTypeEnum) -> None:
with (get_sync_db_session() as session):
statement = delete(GroupResource).where(
GroupResource.third_id == third_id).where(
GroupResource.type == resource_type.value)
session.exec(statement)
session.commit()
@classmethod
def delete_group_resource_by_group_id(cls, group_id: int):
with (get_sync_db_session() as session):
statement = delete(GroupResource).where(GroupResource.group_id == group_id)
session.exec(statement)
session.commit()
@classmethod
def get_group_all_resource(cls, group_id: int) -> List[GroupResource]:
"""
Get all resources under a group
"""
with get_sync_db_session() as session:
return session.exec(
select(GroupResource).where(GroupResource.group_id == group_id)).all()
@classmethod
def update_group_resource(cls, group_resources: List[GroupResource]) -> List[GroupResource]:
with (get_sync_db_session() as session):
session.add_all(group_resources)
session.commit()
return group_resources
================================================
FILE: src/backend/bisheng/database/models/invite_code.py
================================================
from datetime import datetime
from typing import Optional, List
from sqlmodel import Field, Column, text, DateTime, select, update
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_async_db_session
class InviteCodeBase(SQLModelSerializable):
"""
Invitation code model for storing invitation code information.
"""
code: str = Field(..., index=True, unique=True, description='Invitation Code')
batch_id: str = Field(..., index=True, description='BatchesID')
batch_name: str = Field(..., description='Batch')
limit: int = Field(..., description='Usage Limits')
used: Optional[int] = Field(default=0, description='Used times')
bind_user: Optional[int] = Field(default=0, index=True, description='Linked UsersID')
created_id: Optional[int] = Field(default=None, index=True, description='creatorID')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class InviteCode(InviteCodeBase, table=True):
id: Optional[int] = Field(default=None, index=True, primary_key=True, description='Uniqueness quantificationID')
class InviteCodeDao(InviteCodeBase):
"""
The invitation code data access object, which is used to manipulate the invitation code data.
"""
@classmethod
async def insert_invite_code(cls, invite_code: List[InviteCode]) -> List[InviteCode]:
async with get_async_db_session() as session:
session.add_all(invite_code)
await session.commit()
return invite_code
@classmethod
async def get_user_bind_code(cls, bind_user: int) -> list[InviteCode]:
"""
Get a valid invitation code bound by the user
"""
statement = select(InviteCode).where(InviteCode.bind_user == bind_user).where(
InviteCode.used < InviteCode.limit).order_by(InviteCode.id.asc())
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
async def get_user_all_code(cls, bind_user: int) -> list[InviteCode]:
"""
Get all invitation codes linked by a user
"""
statement = select(InviteCode).where(InviteCode.bind_user == bind_user).order_by(InviteCode.id.desc())
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
async def bind_invite_code(cls, user_id: int, code: str) -> bool:
"""
Binding Invitation Code
"""
statement = update(InviteCode).where(InviteCode.code == code).where(InviteCode.bind_user == 0).values(
bind_user=user_id
)
async with get_async_db_session() as session:
result = await session.exec(statement)
await session.commit()
if result.rowcount > 0:
return True
return False
@classmethod
async def use_invite_code(cls, user_id: int, code: str) -> bool:
statement = update(InviteCode).where(InviteCode.code == code).where(InviteCode.bind_user == user_id).values(
used=InviteCode.used + 1
).where(InviteCode.used < InviteCode.limit)
async with get_async_db_session() as session:
result = await session.exec(statement)
await session.commit()
if result.rowcount > 0:
return True
return False
@classmethod
async def revoke_invite_code_used(cls, user_id: int, code: str) -> bool:
"""
Revoke Invitation Code
"""
statement = update(InviteCode).where(InviteCode.code == code).where(InviteCode.bind_user == user_id).values(
used=InviteCode.used - 1
).where(InviteCode.used > 0)
async with get_async_db_session() as session:
result = await session.exec(statement)
await session.commit()
if result.rowcount > 0:
return True
return False
================================================
FILE: src/backend/bisheng/database/models/mark_app_user.py
================================================
from datetime import datetime
from typing import List, Optional
# if TYPE_CHECKING:
from sqlalchemy import Column, DateTime, text
from sqlmodel import Field
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session
class MarkAppUserBase(SQLModelSerializable):
app_id: str = Field(index=True)
user_id: int = Field(index=True)
task_id: int = Field(index=True)
create_id: int = Field(index=True)
status: Optional[int] = Field(index=False, default=1)
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class MarkAppUser(MarkAppUserBase, table=True):
id: Optional[int] = Field(default=None, primary_key=True)
class MarkAppUserDao(MarkAppUserBase):
@classmethod
def create_task(cls, task_info: List[MarkAppUser]) -> List[MarkAppUser]:
with get_sync_db_session() as session:
session.add_all(task_info)
session.commit()
return task_info
================================================
FILE: src/backend/bisheng/database/models/mark_record.py
================================================
from datetime import datetime
from enum import Enum
from typing import List, Optional
# if TYPE_CHECKING:
from sqlalchemy import Column, DateTime, delete, text
from sqlmodel import Field, select
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session
class MarkRecordStatus(Enum):
DEFAULT = 1
DONE = 2
NO = 3
class MarkRecordBase(SQLModelSerializable):
create_user: str = Field(index=True)
flow_type: int = Field(index=True)
create_id: int = Field(index=True)
app_id: int = Field(index=True, nullable=True)
task_id: int = Field(index=True)
session_id: str = Field(index=True)
status: int = Field(index=False, default=1)
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class MarkRecord(MarkRecordBase, table=True):
id: Optional[int] = Field(default=None, primary_key=True)
class MarkRecordDao(MarkRecordBase):
@classmethod
def update_record(cls, record_info: MarkRecord) -> MarkRecord:
with get_sync_db_session() as session:
session.add(record_info)
session.commit()
session.refresh(record_info)
return record_info
@classmethod
def get_prev_task(cls, user_id: int, task_id: int):
with get_sync_db_session() as session:
statement = select(MarkRecord).where(MarkRecord.create_id == user_id).where(
MarkRecord.task_id == task_id).order_by(MarkRecord.id)
return session.exec(statement).all()
@classmethod
def create_record(cls, record_info: MarkRecord) -> MarkRecord:
with get_sync_db_session() as session:
session.add(record_info)
session.commit()
session.refresh(record_info)
return record_info
@classmethod
def del_record(cls, task_id: int):
with get_sync_db_session() as session:
st = delete(MarkRecord).where(MarkRecord.task_id == task_id)
session.exec(st)
session.commit()
return
@classmethod
def del_task_chat(cls, task_id: int, session_id: str):
with get_sync_db_session() as session:
st = delete(MarkRecord).where(MarkRecord.task_id == task_id).where(MarkRecord.session_id == session_id)
session.exec(st)
session.commit()
return
@classmethod
def get_list_by_taskid(cls, task_id: int):
with get_sync_db_session() as session:
statement = select(MarkRecord).where(MarkRecord.task_id == task_id)
return session.exec(statement).all()
@classmethod
def get_count(cls, task_id: int):
with get_sync_db_session() as session:
sql = text(
"select create_user,count(*) as user_count,create_id from markrecord where task_id=:task_id group by create_id,create_user")
query = session.execute(sql, {"task_id": task_id}).fetchall()
return query
@classmethod
def get_record(cls, task_id: int, session_id: str) -> MarkRecord:
with get_sync_db_session() as session:
statement = select(MarkRecord).where(MarkRecord.task_id == task_id).where(
MarkRecord.session_id == session_id)
return session.exec(statement).first()
@classmethod
def filter_records(cls, task_id: int, chat_ids: list[str] = None, status: int = None, mark_user: int = None) -> \
List[MarkRecord]:
statement = select(MarkRecord).where(MarkRecord.task_id == task_id)
if chat_ids:
statement = statement.where(MarkRecord.session_id.in_(chat_ids))
if status is not None:
statement = statement.where(MarkRecord.status == status)
if mark_user is not None:
statement = statement.where(MarkRecord.create_user == str(mark_user))
with get_sync_db_session() as session:
return session.exec(statement).all()
================================================
FILE: src/backend/bisheng/database/models/mark_task.py
================================================
from datetime import datetime
from enum import Enum
from typing import List, Optional
# if TYPE_CHECKING:
from sqlalchemy import Column, DateTime, and_, delete, func, or_, text
from sqlmodel import Field, select, update
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session
class MarkTaskStatus(Enum):
DEFAULT = 1
DONE = 2
ING = 3
class MarkTaskBase(SQLModelSerializable):
create_user: str = Field(index=True)
create_id: int = Field(index=True)
app_id: str = Field(index=False, max_length=2048)
process_users: str = Field(index=False) # 23,2323
mark_user: Optional[str] = Field(default=None, index=True, nullable=True)
status: Optional[int] = Field(index=False, default=1)
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class MarkTask(MarkTaskBase, table=True):
id: Optional[int] = Field(default=None, primary_key=True)
class MarkTaskRead(MarkTaskBase):
id: Optional[int] = None
mark_process: Optional[List[str]] = None
class MarkTaskDao(MarkTaskBase):
@classmethod
def create_task(cls, task_info: MarkTask) -> MarkTask:
with get_sync_db_session() as session:
session.add(task_info)
session.commit()
session.refresh(task_info)
return task_info
@classmethod
def delete_task(cls, task_id: int):
with get_sync_db_session() as session:
st = delete(MarkTask).where(MarkTask.id == task_id)
session.exec(st)
session.commit()
@classmethod
def get_task_byid(cls, task_id: int) -> MarkTask:
with get_sync_db_session() as session:
statement = select(MarkTask).where(MarkTask.id == task_id)
return session.exec(statement).first()
@classmethod
def get_task(cls, user_id: int) -> MarkTask:
with get_sync_db_session() as session:
statement = select(MarkTask).where(MarkTask.process_users.like('%{}%'.format(user_id)))
return session.exec(statement).first()
@classmethod
def get_task_list_byuid(cls, user_id: int, task_id: int) -> MarkTask:
with get_sync_db_session() as session:
statement = select(MarkTask).where(MarkTask.id == task_id)
return session.exec(statement).first()
@classmethod
def update_task(cls, task_id: int, status: int):
with get_sync_db_session() as session:
st = update(MarkTask).where(MarkTask.id == task_id).values(status=status)
session.exec(st)
session.commit()
@classmethod
def get_all_task(
cls,
page_size: int = 10,
page_num: int = 1,
):
with get_sync_db_session() as session:
statement = select(MarkTask)
total_count_query = select(func.count()).select_from(statement.alias('subquery'))
statement = statement.order_by(MarkTask.create_time.desc())
total_count = session.execute(total_count_query).scalar()
statement = statement.limit(page_size).offset((page_num - 1) * page_size)
return session.exec(statement).all(), total_count
@classmethod
def get_task_list(
cls,
status: int,
create_id: Optional[int],
user_id: Optional[int],
page_size: int = 10,
page_num: int = 1,
):
with get_sync_db_session() as session:
statement = select(MarkTask)
filter = []
filter_or = []
if status:
filter.append(MarkTask.status == status)
if user_id:
filter.append(func.find_in_set(user_id, MarkTask.process_users) > 0)
if create_id:
filter.append(MarkTask.create_id == create_id)
if user_id:
filter_or.append(func.find_in_set(user_id, MarkTask.process_users) > 0)
# filter_or.append(MarkTask.process_users.like('%{}%'.format(user_id)))
if status:
filter_or.append(MarkTask.status == status)
if filter and not filter_or:
statement = statement.filter(*filter)
if filter_or:
statement = statement.filter(or_(and_(*filter), and_(*filter_or)))
# Calculate total tasks
total_count_query = select(func.count()).select_from(statement.alias('subquery'))
statement = statement.order_by(MarkTask.create_time.desc())
total_count = session.execute(total_count_query).scalar()
statement = statement.limit(page_size).offset((page_num - 1) * page_size)
return session.exec(statement).all(), total_count
================================================
FILE: src/backend/bisheng/database/models/message.py
================================================
from datetime import datetime
from enum import Enum
from typing import Dict, List, Optional, Tuple
from loguru import logger
from sqlalchemy.dialects.mysql import LONGTEXT
from sqlmodel import (JSON, Column, DateTime, Field, String, Text, case, delete, func, not_, or_,
select, text, update, col)
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session, get_async_db_session
class LikedType(Enum):
UNRATED = 0 # Not assessed
LIKED = 1 # Love
DISLIKED = 2 # don't like}
class MessageBase(SQLModelSerializable):
is_bot: bool = Field(index=False, description='Chat Role')
source: Optional[int] = Field(default=None, index=False, description='Whether traceability is supported')
mark_status: Optional[int] = Field(index=False, default=1, description='Tag status')
mark_user: Optional[int] = Field(default=None, index=False, description='Flagging User')
mark_user_name: Optional[str] = Field(default=None, index=False, description='Flagging User')
message: Optional[str] = Field(default=None, sa_column=Column(LONGTEXT), description='Chat Message')
extra: Optional[str] = Field(default=None, sa_column=Column(Text), description='Connection information, etc.')
type: str = Field(index=False, description='Type of Message')
category: str = Field(index=False, max_length=32, description='Message category, questionetc.')
flow_id: str = Field(index=True, description='Corresponding Skillsid')
chat_id: Optional[str] = Field(default=None, index=True, description='chat_id, Frontend Generation')
user_id: Optional[int] = Field(default=None, index=True, description='Usersid')
liked: Optional[int] = Field(index=False, default=0, description="Whether the user likes it or 0Not assessed/1 Love/2 don't like}")
solved: Optional[int] = Field(index=False, default=0, description='Whether the user likes it or 0Not assessed/1 Solution/2 Unresolve')
copied: Optional[int] = Field(index=False, default=0, description='Is the user copying 0: Not copied 1Copied: ')
sensitive_status: Optional[int] = Field(index=False, default=1, description='Sensitive Word Status 1Pass 2violates regulation')
sender: Optional[str] = Field(index=False, default='', description='autogen Sender')
receiver: Optional[Dict] = Field(index=False, default=None, description='autogen Sender')
intermediate_steps: Optional[str] = Field(default=None, sa_column=Column(Text), description='Process Log')
files: Optional[str] = Field(default=None, sa_column=Column(String(length=4096)), description='Uploaded documents, etc.')
remark: Optional[str] = Field(default=None, sa_column=Column(String(length=4096)),
description='Note. break_answer: Interrupted response inactionhistoryPass to Model')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class ChatMessage(MessageBase, table=True):
id: Optional[int] = Field(default=None, primary_key=True)
receiver: Optional[Dict] = Field(default=None, sa_column=Column(JSON))
# Key: Set table level character set to utf8mb4
__table_args__ = {
"mysql_charset": "utf8mb4",
"mysql_collate": "utf8mb4_unicode_ci"
}
class ChatMessageRead(MessageBase):
id: Optional[int] = None
class ChatMessageQuery(MessageBase):
id: Optional[int] = None
receiver: Optional[Dict] = None
class ChatMessageCreate(MessageBase):
pass
class MessageDao(MessageBase):
@classmethod
def static_msg_liked(cls, liked: int, flow_id: str, create_time_begin: datetime,
create_time_end: datetime):
base_condition = select(func.count(ChatMessage.id)).where(ChatMessage.liked == liked)
if flow_id:
base_condition = base_condition.where(ChatMessage.flow_id == flow_id)
if create_time_begin and create_time_end:
base_condition = base_condition.where(ChatMessage.create_time > create_time_begin,
ChatMessage.create_time < create_time_end)
with get_sync_db_session() as session:
return session.scalar(base_condition)
@classmethod
def app_list_group_by_chat_id(
cls,
page_size: int,
page_num: int,
flow_ids: Optional[list[str]],
user_ids: Optional[list[int]],
) -> Tuple[List[Dict], int]:
with get_sync_db_session() as session:
count_stat = select(func.count(func.distinct(ChatMessage.chat_id)))
sql = select(
ChatMessage.chat_id,
ChatMessage.user_id,
ChatMessage.flow_id,
func.max(ChatMessage.create_time).label('create_time'),
func.sum(case((ChatMessage.liked == 1, 1), else_=0)),
func.sum(case((ChatMessage.liked == 2, 1), else_=0)),
func.sum(case((ChatMessage.copied == 1, 1), else_=0)),
)
if flow_ids:
count_stat = count_stat.where(ChatMessage.flow_id.in_(flow_ids))
sql = sql.where(ChatMessage.flow_id.in_(flow_ids))
if user_ids:
count_stat = count_stat.where(
or_(
ChatMessage.mark_user.in_(user_ids),
ChatMessage.mark_status == 1,
))
sql = sql.where(
or_(ChatMessage.mark_user.in_(user_ids), ChatMessage.mark_status == 1))
sql = sql.group_by(ChatMessage.chat_id, ChatMessage.user_id,
ChatMessage.flow_id).order_by(
func.max(ChatMessage.create_time).desc()).offset(
page_size * (page_num - 1)).limit(page_size)
res_list = session.exec(sql).all()
total_count = session.scalar(count_stat)
dict_res = [{
'chat_id': chat_id,
'user_id': user_id,
'flow_id': flow_id,
'like_count': like_num,
'dislike_count': dislike_num,
'copied_count': copied_num,
'create_time': create_time
} for chat_id, user_id, flow_id, create_time, like_num, dislike_num, copied_num in
res_list]
logger.info(res_list)
return dict_res, total_count
class ChatMessageDao(MessageBase):
@classmethod
def get_latest_message_by_chatid(cls, chat_id: str):
statement = select(ChatMessage).where(ChatMessage.chat_id == chat_id).order_by(
ChatMessage.id.desc()).limit(1)
with get_sync_db_session() as session:
res = session.exec(statement).all()
if res:
return res[0]
else:
return None
@classmethod
async def aget_latest_message_by_chatid(cls, chat_id: str):
statement = select(ChatMessage).where(ChatMessage.chat_id == chat_id).order_by(
ChatMessage.id.desc()).limit(1)
async with get_async_db_session() as session:
res = await session.exec(statement)
res = res.all()
if res:
return res[0]
else:
return None
@classmethod
def get_latest_message_by_chat_ids(cls,
chat_ids: list[str],
category: str = None,
exclude_category: str = None):
"""
Get the most recent message content for each session
"""
statement = select(ChatMessage.chat_id,
func.max(ChatMessage.id)).where(ChatMessage.chat_id.in_(chat_ids))
if category:
statement = statement.where(ChatMessage.category == category)
if exclude_category:
statement = statement.where(ChatMessage.category != exclude_category)
statement = statement.group_by(ChatMessage.chat_id)
with get_sync_db_session() as session:
# Get the latestidVertical
res = session.exec(statement).all()
ids = [one[1] for one in res]
# Get the details of your message
statement = select(ChatMessage).where(ChatMessage.id.in_(ids))
return session.exec(statement).all()
@classmethod
def get_messages_by_chat_id(cls,
chat_id: str,
category_list: list = None,
limit: int = 10) -> List[ChatMessage]:
with get_sync_db_session() as session:
statement = select(ChatMessage).where(ChatMessage.chat_id == chat_id)
if category_list:
statement = statement.where(ChatMessage.category.in_(category_list))
statement = statement.limit(limit).order_by(ChatMessage.create_time.asc())
return session.exec(statement).all()
@classmethod
async def aget_messages_by_chat_id(cls,
chat_id: str,
category_list: list = None,
limit: int = 10) -> List[ChatMessage]:
async with get_async_db_session() as session:
statement = select(ChatMessage).where(ChatMessage.chat_id == chat_id)
if category_list:
statement = statement.where(ChatMessage.category.in_(category_list))
statement = statement.limit(limit).order_by(ChatMessage.create_time.asc())
result = await session.exec(statement)
return result.all()
@classmethod
def get_last_msg_by_flow_id(cls, flow_id: List[str], chat_id: List[str]):
with get_sync_db_session() as session:
statement = select(ChatMessage.chat_id,
ChatMessage.flow_id).where(ChatMessage.flow_id.in_(flow_id)).where(
not_(ChatMessage.chat_id.in_(chat_id))).group_by(
ChatMessage.chat_id, ChatMessage.flow_id)
return session.exec(statement).all()
@classmethod
def get_msg_by_chat_id(cls, chat_id: str):
with get_sync_db_session() as session:
statement = select(ChatMessage).where(ChatMessage.chat_id == chat_id)
return session.exec(statement).all()
@classmethod
def get_msg_by_flow(cls, flow_id: str):
with get_sync_db_session() as session:
# sql = text("select chat_id,count(*) as chat_count from chatmessage where flow_id=:flow_id group by chat_id") # noqa
st = select(ChatMessage.chat_id).where(ChatMessage.flow_id == flow_id).group_by(
ChatMessage.chat_id)
return session.exec(st).all()
@classmethod
def get_msg_by_flows(cls, flow_id: List[str]):
with get_sync_db_session() as session:
st = select(ChatMessage.chat_id).where(ChatMessage.flow_id.in_(flow_id)).group_by(
ChatMessage.chat_id)
return session.exec(st).all()
@classmethod
def delete_by_user_chat_id(cls, user_id: int, chat_id: str):
if user_id is None or chat_id is None:
logger.info('delete_param_error user_id={} chat_id={}', user_id, chat_id)
return False
statement = delete(ChatMessage).where(ChatMessage.chat_id == chat_id,
ChatMessage.user_id == user_id)
with get_sync_db_session() as session:
session.exec(statement)
session.commit()
return True
@classmethod
def delete_by_message_id(cls, user_id: int, message_id: str):
if user_id is None or message_id is None:
logger.info('delete_param_error user_id={} chat_id={}', user_id, message_id)
return False
statement = delete(ChatMessage).where(ChatMessage.id == message_id,
ChatMessage.user_id == user_id)
with get_sync_db_session() as session:
session.exec(statement)
session.commit()
return True
@classmethod
def insert_one(cls, message: ChatMessage) -> ChatMessage:
with get_sync_db_session() as session:
session.add(message)
session.commit()
session.refresh(message)
return message
@classmethod
async def ainsert_one(cls, message: ChatMessage) -> ChatMessage:
async with get_async_db_session() as session:
session.add(message)
await session.commit()
await session.refresh(message)
return message
@classmethod
def insert_batch(cls, messages: List[ChatMessage]):
with get_sync_db_session() as session:
session.add_all(messages)
session.commit()
ret = []
for one in messages:
session.refresh(one)
ret.append(one)
return ret
@classmethod
def get_message_by_id(cls, message_id: int) -> Optional[ChatMessage]:
with get_sync_db_session() as session:
return session.exec(select(ChatMessage).where(ChatMessage.id == message_id)).first()
@classmethod
async def aget_message_by_id(cls, message_id: int) -> Optional[ChatMessage]:
async with get_async_db_session() as session:
result = await session.exec(select(ChatMessage).where(ChatMessage.id == message_id))
return result.first()
@classmethod
def update_message(cls, message_id: int, user_id: int, message: str):
with get_sync_db_session() as session:
statement = update(ChatMessage).where(ChatMessage.id == message_id).where(
ChatMessage.user_id == user_id).values(message=message)
session.exec(statement)
session.commit()
@classmethod
def update_message_model(cls, message: ChatMessage):
with get_sync_db_session() as session:
session.add(message)
session.commit()
session.refresh(message)
return message
@classmethod
async def aupdate_message_model(cls, message: ChatMessage):
async with get_async_db_session() as session:
session.add(message)
await session.commit()
await session.refresh(message)
return message
@classmethod
def update_message_copied(cls, message_id: int, copied: int):
with get_sync_db_session() as session:
statement = update(ChatMessage).where(ChatMessage.id == message_id).values(
copied=copied)
session.exec(statement)
session.commit()
@classmethod
def update_message_mark(cls, chat_id: str, status: int):
with get_sync_db_session() as session:
statement = update(ChatMessage).where(ChatMessage.chat_id == chat_id).values(
mark_status=status)
session.exec(statement)
session.commit()
@classmethod
async def get_all_message_by_chat_ids(cls, chat_ids: List[str]) -> List[ChatMessage]:
statement = select(ChatMessage).where(ChatMessage.chat_id.in_(chat_ids)).order_by(
ChatMessage.create_time.asc())
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
async def afilter_message_by_chat_id(cls, chat_id: str, flow_id: str, message_id: int = None, page_size: int = 20) \
-> List[ChatMessage]:
statement = select(ChatMessage).where(ChatMessage.chat_id == chat_id).where(ChatMessage.flow_id == flow_id)
if message_id:
statement = statement.where(ChatMessage.id < message_id)
statement = statement.order_by(col(ChatMessage.id).desc()).limit(page_size)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
================================================
FILE: src/backend/bisheng/database/models/recall_chunk.py
================================================
from datetime import datetime
from typing import Optional
from sqlalchemy import Column, DateTime, Text, text
from sqlmodel import Field
from bisheng.common.models.base import SQLModelSerializable
class RecallBase(SQLModelSerializable):
message_id: Optional[int] = Field(default=None, index=True, unique=False)
chat_id: str = Field(index=False)
keywords: str = Field(sa_column=Column(Text))
chunk: Optional[str] = Field(default=None, sa_column=Column(Text))
meta_data: Optional[str] = Field(default=None, sa_column=Column(Text))
file_id: Optional[int] = Field(default=None, index=False)
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class RecallChunk(RecallBase, table=True):
id: Optional[int] = Field(default=None, primary_key=True)
class RecallChunkRead(RecallBase):
id: Optional[int] = None
score: Optional[int] = None
class RecallChunkQuery(SQLModelSerializable):
id: Optional[int] = None
server: Optional[str] = None
class RecallChunkCreate(RecallBase):
pass
================================================
FILE: src/backend/bisheng/database/models/report.py
================================================
from datetime import datetime
from typing import Optional
from sqlalchemy import Column, DateTime, text
from sqlmodel import Field
from bisheng.common.models.base import SQLModelSerializable
class ReportBase(SQLModelSerializable):
# ```
# Useflow_id Path to get the latest template in reverse order by update time
# Records of the final report generated by the template are stored`
flow_id: str = Field(index=False, description='Skill name')
file_name: Optional[str] = Field(default=None, index=False, description='Generate Report Name')
template_name: Optional[str] = Field(default=None, index=False, description='Report Template Data Storage Path')
version_key: Optional[str] = Field(default=None, index=True, unique=True, description='Frontend Template Uniquekey')
newversion_key: Optional[str] = Field(index=False, default=None, description='Frontend Template Nextkey')
object_name: Optional[str] = Field(default=None, index=False, description='Report Template Data Storage Path')
del_yn: Optional[int] = Field(index=False, default=0, description='Delete Status 1Show Delete')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class Report(ReportBase, table=True):
__tablename__ = 't_report'
id: Optional[int] = Field(default=None, primary_key=True)
class ReportRead(ReportBase):
id: Optional[int] = None
class RoleUpdate(ReportBase):
role_name: Optional[str] = None
remark: Optional[str] = None
class RoleCreate(ReportBase):
pass
================================================
FILE: src/backend/bisheng/database/models/role.py
================================================
from datetime import datetime
from typing import List, Optional
from sqlalchemy import Column, DateTime, text, func, delete, and_, UniqueConstraint
from sqlmodel import Field, select
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session, get_async_db_session
from bisheng.database.constants import AdminRole
from bisheng.database.models.role_access import RoleAccess
from bisheng.user.domain.models.user_role import UserRole
class RoleBase(SQLModelSerializable):
role_name: str = Field(index=False, description='Frontend Display Name')
group_id: Optional[int] = Field(default=None, index=True)
remark: Optional[str] = Field(default=None, index=False)
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class Role(RoleBase, table=True):
__table_args__ = (UniqueConstraint('group_id', 'role_name', name='group_role_name_uniq'),)
id: Optional[int] = Field(default=None, primary_key=True)
class RoleRead(RoleBase):
id: Optional[int] = None
class RoleUpdate(RoleBase):
role_name: Optional[str] = None
remark: Optional[str] = None
class RoleCreate(RoleBase):
pass
class RoleDao(RoleBase):
@classmethod
def get_role_by_groups(cls, group: List[int], keyword: str = None, page: int = 0, limit: int = 0) -> List[Role]:
"""
Get a list of roles within a user group, Does not contain the System Administrator role
params:
group: User GroupsIDVertical
page: Page
limit: Listings Per Page
return: Role List
"""
statement = select(Role).where(Role.id > AdminRole)
if group:
statement = statement.where(Role.group_id.in_(group))
if keyword:
statement = statement.filter(Role.role_name.like(f'%{keyword}%'))
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
statement = statement.order_by(Role.create_time.desc())
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
def count_role_by_groups(cls, group: List[int], keyword: str = None) -> int:
"""
Count the number of roles in the user group, the parameters are as above
"""
statement = select(func.count(Role.id)).where(Role.id > AdminRole)
if group:
statement = statement.where(Role.group_id.in_(group))
if keyword:
statement = statement.filter(Role.role_name.like(f'%{keyword}%'))
with get_sync_db_session() as session:
return session.scalar(statement)
@classmethod
def insert_role(cls, role: RoleCreate):
with get_sync_db_session() as session:
session.add(role)
session.commit()
session.refresh(role)
return role
@classmethod
def delete_role(cls, role_id: int):
with get_sync_db_session() as session:
session.exec(delete(Role).where(Role.id == role_id))
session.exec(delete(UserRole).where(UserRole.role_id == role_id))
session.exec(delete(RoleAccess).where(RoleAccess.role_id == role_id))
session.commit()
@classmethod
def get_role_by_ids(cls, role_ids: List[int]) -> List[Role]:
with get_sync_db_session() as session:
return session.query(Role).filter(Role.id.in_(role_ids)).all()
@classmethod
def get_role_by_id(cls, role_id: int) -> Role:
with get_sync_db_session() as session:
return session.query(Role).filter(Role.id == role_id).first()
@classmethod
async def aget_role_by_id(cls, role_id: int) -> Role:
async with get_async_db_session() as session:
result = await session.execute(select(Role).where(Role.id == role_id))
return result.scalars().first()
@classmethod
def delete_role_by_group_id(cls, group_id: int):
"""
Delete all roles under the grouping and clean up the corresponding roles for the user
"""
from bisheng.user.domain.models.user_role import UserRole
with get_sync_db_session() as session:
# Clean corresponding users
all_user = select(UserRole, Role).join(
Role, and_(UserRole.role_id == Role.id,
Role.group_id == group_id)).group_by(UserRole.id)
all_user = session.exec(all_user).all()
session.exec(delete(UserRole).where(UserRole.id.in_([one.UserRole.id for one in all_user])))
session.exec(delete(Role).where(Role.group_id == group_id))
session.commit()
================================================
FILE: src/backend/bisheng/database/models/role_access.py
================================================
from datetime import datetime
from enum import Enum
from typing import List, Optional, Union
from pydantic import BaseModel
from sqlalchemy import Column, DateTime, text
from sqlmodel import Field, select, delete, col
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session, get_async_db_session
class RoleAccessBase(SQLModelSerializable):
role_id: int = Field(index=True)
third_id: str = Field(index=False)
type: int = Field(index=False)
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class RoleAccess(RoleAccessBase, table=True):
id: Optional[int] = Field(default=None, primary_key=True)
class RoleAccessRead(RoleAccessBase):
id: Optional[int] = None
class RoleAccessCreate(RoleAccessBase):
pass
class WebMenuResource(Enum):
"""Front-end menu bar resources"""
BUILD = 'build' # Build Menu
KNOWLEDGE = 'knowledge' # Knowledge Menu
MODEL = 'model' # Model Menu
EVALUATION = 'evaluation' # Evaluation Menu
FRONTEND = 'frontend' # Front-end permissions
BACKEND = 'backend' # Backend Access
CREATE_DASHBOARD = 'create_dashboard' # Create board permissions
class AccessType(Enum):
"""Type of the role_access"""
KNOWLEDGE = 1 # Knowledge Base Reading Permissions
FLOW = 2 # Skill Read Permissions
KNOWLEDGE_WRITE = 3 # Knowledge Base Write Permissions
FLOW_WRITE = 4 # Skill Write Permissions
ASSISTANT_READ = 5 # Assistant Read Permissions
ASSISTANT_WRITE = 6 # Assistant Write Permissions
GPTS_TOOL_READ = 7 # Tool Read Permissions
GPTS_TOOL_WRITE = 8 # Tool Write Permissions
WORKFLOW = 9 # Workflow Read Permissions
WORKFLOW_WRITE = 10 # Workflow write permissions
DASHBOARD = 11 # Kanban Reading Permissions
DASHBOARD_WRITE = 12 # Kanban writing permissions
WEB_MENU = 99 # Frontend Menu Bar Permission Restrictions
class RoleRefresh(BaseModel):
role_id: int
access_id: list[Union[str, int]]
type: int
class RoleAccessDao(RoleAccessBase):
@classmethod
def get_role_access(cls, role_ids: List[int], access_type: AccessType) -> List[RoleAccess]:
with get_sync_db_session() as session:
if access_type:
return session.exec(
select(RoleAccess).where(RoleAccess.role_id.in_(role_ids),
RoleAccess.type == access_type.value)).all()
return session.exec(select(RoleAccess).where(RoleAccess.role_id.in_(role_ids))).all()
@classmethod
async def aget_role_access(cls, role_ids: List[int], access_type: AccessType = None) -> List[RoleAccess]:
statement = select(RoleAccess).where(RoleAccess.role_id.in_(role_ids))
if access_type:
statement = statement.where(RoleAccess.type == access_type.value)
async with get_async_db_session() as session:
return (await session.exec(statement)).all()
@classmethod
def get_role_access_batch(cls, role_ids: List[int], access_type: List[AccessType]) -> List[RoleAccess]:
with get_sync_db_session() as session:
if access_type:
return session.exec(
select(RoleAccess).where(RoleAccess.role_id.in_(role_ids),
RoleAccess.type.in_([x.value for x in access_type]))).all()
@classmethod
async def aget_role_access_batch(cls, role_ids: List[int], access_type: List[AccessType]) -> List[RoleAccess]:
statement = select(RoleAccess).where(col(RoleAccess.role_id).in_(role_ids))
if access_type:
statement = statement.where(col(RoleAccess.type).in_([x.value for x in access_type]))
async with get_async_db_session() as session:
return (await session.exec(statement)).all()
@classmethod
def judge_role_access(cls, role_ids: List[int], third_id: str, access_type: AccessType) -> Optional[RoleAccess]:
with get_sync_db_session() as session:
return session.exec(select(RoleAccess).filter(
RoleAccess.role_id.in_(role_ids),
RoleAccess.type == access_type.value,
RoleAccess.third_id == third_id
)).first()
@classmethod
async def ajudge_role_access(cls, role_ids: List[int], third_id: str, access_type: AccessType) -> Optional[
RoleAccess]:
statement = select(RoleAccess).filter(
col(RoleAccess.role_id).in_(role_ids),
col(RoleAccess.type) == access_type.value,
col(RoleAccess.third_id) == third_id
)
async with get_async_db_session() as session:
return (await session.exec(statement)).first()
@classmethod
def find_role_access(cls, role_ids: List[int], third_ids: List[str], access_type: AccessType) -> List[RoleAccess]:
with get_sync_db_session() as session:
if access_type:
return session.exec(
select(RoleAccess).where(RoleAccess.role_id.in_(role_ids),
RoleAccess.third_id.in_(third_ids),
RoleAccess.type == access_type.value)).all()
return session.exec(select(RoleAccess).where(RoleAccess.role_id.in_(role_ids))).all()
@classmethod
async def afind_role_access(cls, role_ids: List[int], third_ids: List[str], access_type: AccessType) -> List[
RoleAccess]:
statement = select(RoleAccess).where(
col(RoleAccess.role_id).in_(role_ids),
col(RoleAccess.third_id).in_(third_ids)
)
if access_type:
statement = statement.where(col(RoleAccess.type) == access_type.value)
async with get_async_db_session() as session:
return (await session.exec(statement)).all()
@classmethod
async def update_role_access_all(cls, role_id: int, access_type: AccessType,
access_ids: List[Union[str, int]]) -> None:
"""
Update the role's permissions, delete it first and add it later
"""
async with get_async_db_session() as session:
# Clear all old permissions first
statement = delete(RoleAccess).where(col(RoleAccess.role_id) == str(role_id),
col(RoleAccess.type) == access_type.value)
await session.exec(statement)
# Add New Permission
for access_id in access_ids:
role_access = RoleAccess(role_id=role_id, third_id=str(access_id), type=access_type.value)
session.add(role_access)
await session.commit()
================================================
FILE: src/backend/bisheng/database/models/session.py
================================================
from datetime import datetime
from enum import Enum
from typing import Optional, List
from sqlalchemy import JSON
from sqlmodel import Field, Column, DateTime, text, select, func, update
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session, get_async_db_session
from bisheng.database.models.user_group import UserGroupDao
class SensitiveStatus(Enum):
PASS = 1 # Setuju
VIOLATIONS = 2 # violates regulation
class MessageSessionBase(SQLModelSerializable):
""" Conversation table """
chat_id: str = Field(default=None, primary_key=True, description='Session UniqueID')
flow_id: str = Field(index=True, description='Apply UniqueID')
flow_type: int = Field(description='App type. Skills, assistants, workflows')
flow_name: str = Field(index=True, description='Application name')
flow_description: Optional[str] = Field(default=None, description='App Description')
flow_logo: Optional[str] = Field(default=None, description='Applicationslogo')
user_id: int = Field(index=True, description='User who created the sessionID')
group_ids: Optional[List[int]] = Field(default=None, sa_column=Column(JSON),
description="Belongs to a user groupIDVertical")
is_delete: Optional[bool] = Field(default=False,
description='Whether the corresponding skill or the session itself was deleted')
like: Optional[int] = Field(default=0, description='Number of liked messages')
dislike: Optional[int] = Field(default=0, description='Number of messages clicked')
copied: Optional[int] = Field(default=0, description='Number of messages copied')
sensitive_status: int = Field(default=SensitiveStatus.PASS.value, description='Review Status')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class MessageSession(MessageSessionBase, table=True):
__tablename__ = 'message_session'
class MessageSessionDao(MessageSessionBase):
@classmethod
def insert_one(cls, data: MessageSession) -> MessageSession:
if not data.group_ids:
user_groups = UserGroupDao.get_user_group(data.user_id)
data.group_ids = [ug.group_id for ug in user_groups]
with get_sync_db_session() as session:
session.add(data)
session.commit()
session.refresh(data)
return data
@classmethod
async def async_insert_one(cls, data: MessageSession) -> MessageSession:
if not data.group_ids:
user_groups = await UserGroupDao.aget_user_group(data.user_id)
data.group_ids = [ug.group_id for ug in user_groups]
async with get_async_db_session() as session:
session.add(data)
await session.commit()
await session.refresh(data)
return data
@classmethod
async def delete_session(cls, chat_id: str):
statement = update(MessageSession).where(MessageSession.chat_id == chat_id).values(is_delete=True)
with get_async_db_session() as session:
await session.exec(statement)
await session.commit()
@classmethod
def get_one(cls, chat_id: str) -> MessageSession | None:
statement = select(MessageSession).where(MessageSession.chat_id == chat_id)
with get_sync_db_session() as session:
return session.exec(statement).first()
@classmethod
async def async_get_one(cls, chat_id: str) -> MessageSession | None:
statement = select(MessageSession).where(MessageSession.chat_id == chat_id)
async with get_async_db_session() as session:
return (await session.exec(statement)).first()
@classmethod
def generate_filter_session_statement(cls,
statement,
chat_ids: List[str] = None,
sensitive_status: List[SensitiveStatus] = None,
flow_ids: List[str] = None,
user_ids: List[int] = None,
feedback: str = None,
start_date: datetime = None,
end_date: datetime = None,
include_delete: bool = True,
exclude_chats: List[str] = None,
flow_type: int = None):
if chat_ids:
statement = statement.where(MessageSession.chat_id.in_(chat_ids))
if flow_ids:
statement = statement.where(MessageSession.flow_id.in_(flow_ids))
if user_ids:
statement = statement.where(MessageSession.user_id.in_(user_ids))
if feedback == 'like':
statement = statement.where(MessageSession.like > 0)
elif feedback == 'dislike':
statement = statement.where(MessageSession.dislike > 0)
elif feedback == 'copied':
statement = statement.where(MessageSession.copied > 0)
if not include_delete:
statement = statement.where(MessageSession.is_delete == False) # noqa
if exclude_chats:
statement = statement.where(MessageSession.chat_id.not_in(exclude_chats))
if start_date:
statement = statement.where(MessageSession.create_time >= start_date)
if end_date:
statement = statement.where(MessageSession.create_time <= end_date)
if sensitive_status:
statement = statement.where(
MessageSession.sensitive_status.in_([one.value for one in sensitive_status]))
if flow_type:
statement = statement.where(MessageSession.flow_type.in_(flow_type))
return statement
@classmethod
def filter_session(cls,
chat_ids: List[str] = None,
sensitive_status: List[SensitiveStatus] = None,
flow_ids: List[str] = None,
user_ids: List[int] = None,
feedback: str = None,
start_date: datetime = None,
end_date: datetime = None,
include_delete: bool = True,
exclude_chats: List[str] = None,
page: int = 0,
limit: int = 0,
flow_type: List[int] = None) -> List[MessageSession]:
statement = select(MessageSession)
statement = cls.generate_filter_session_statement(statement,
chat_ids,
sensitive_status,
flow_ids,
user_ids,
feedback,
start_date,
end_date,
include_delete,
exclude_chats,
flow_type=flow_type)
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
statement = statement.order_by(MessageSession.create_time.desc())
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
async def afilter_session(cls,
chat_ids: List[str] = None,
sensitive_status: List[SensitiveStatus] = None,
flow_ids: List[str] = None,
user_ids: List[int] = None,
feedback: str = None,
start_date: datetime = None,
end_date: datetime = None,
include_delete: bool = True,
exclude_chats: List[str] = None,
page: int = 0,
limit: int = 0,
flow_type: List[int] = None) -> List[MessageSession]:
statement = select(MessageSession)
statement = cls.generate_filter_session_statement(statement,
chat_ids,
sensitive_status,
flow_ids,
user_ids,
feedback,
start_date,
end_date,
include_delete,
exclude_chats,
flow_type=flow_type)
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
statement = statement.order_by(MessageSession.create_time.desc())
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
async def get_statement_results(cls, statement, page: int = 0, limit: int = 0) -> List[MessageSession]:
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
statement = statement.order_by(MessageSession.create_time.desc())
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
async def get_statement_count(cls, statement) -> int:
count_statement = select(func.count()).select_from(statement.subquery())
async with get_async_db_session() as session:
result = await session.exec(count_statement)
return result.first()
@classmethod
async def filter_session_count(cls,
chat_ids: List[str] = None,
sensitive_status: List[SensitiveStatus] = None,
flow_ids: List[str] = None,
user_ids: List[int] = None,
feedback: str = None,
start_date: datetime = None,
end_date: datetime = None,
include_delete: bool = True,
exclude_chats: List[str] = None,
flow_type: List[int] = None) -> int:
statement = select(func.count(MessageSession.chat_id))
statement = cls.generate_filter_session_statement(statement, chat_ids, sensitive_status,
flow_ids, user_ids, feedback, start_date,
end_date, include_delete, exclude_chats, flow_type=flow_type)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.first()
@classmethod
def update_sensitive_status(cls, chat_id: str, sensitive_status: SensitiveStatus):
statement = update(MessageSession).where(MessageSession.chat_id == chat_id).values(
sensitive_status=sensitive_status.value)
with get_sync_db_session() as session:
session.exec(statement)
session.commit()
@classmethod
def add_like_count(cls, chat_id: str, like_count: int):
if like_count == 0:
return
statement = update(MessageSession).where(MessageSession.chat_id == chat_id).values(
like=MessageSession.like + like_count)
with get_sync_db_session() as session:
session.exec(statement)
session.commit()
@classmethod
def add_dislike_count(cls, chat_id: str, dislike_count: int):
if dislike_count == 0:
return
statement = update(MessageSession).where(MessageSession.chat_id == chat_id).values(
dislike=MessageSession.dislike + dislike_count)
with get_sync_db_session() as session:
session.exec(statement)
session.commit()
@classmethod
def add_copied_count(cls, chat_id: str, copied_count: int):
if copied_count == 0:
return
statement = update(MessageSession).where(MessageSession.chat_id == chat_id).values(
copied=MessageSession.copied + copied_count)
with get_sync_db_session() as session:
session.exec(statement)
session.commit()
@classmethod
def update_session_info_by_flow(cls, name: str, description: str, logo: str, flow_id: str, flow_type: int):
statement = update(MessageSession).where(
MessageSession.flow_id == flow_id,
MessageSession.flow_type == flow_type
).values(
flow_name=name,
flow_description=description,
flow_logo=logo
)
with get_sync_db_session() as session:
session.exec(statement)
session.commit()
================================================
FILE: src/backend/bisheng/database/models/tag.py
================================================
from datetime import datetime
from typing import Optional, List, Dict
from sqlalchemy import Column, DateTime, UniqueConstraint
from sqlmodel import Field, select, delete, and_, func, text
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session
from bisheng.database.models.group_resource import ResourceTypeEnum
class TagBase(SQLModelSerializable):
"""
Tag Form
"""
name: Optional[str] = Field(default=None, index=True, unique=True, description="Label Name")
user_id: int = Field(default=0, description='Create UserID')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')), description="Creation Time")
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class Tag(TagBase, table=True):
id: Optional[int] = Field(default=None, index=True, primary_key=True, description="Tag UniqueID")
class TagLinkBase(SQLModelSerializable):
"""
Label Association Table
"""
tag_id: int = Field(index=True, description="labelID")
resource_id: str = Field(description="Resource UniqueID")
resource_type: int = Field(description="Resource Type") # Usegroup_resource.ResourceTypeEnumEnumeration
user_id: int = Field(default=0, description='Create UserID')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')), description="Creation Time")
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class TagLink(TagLinkBase, table=True):
__table_args__ = (UniqueConstraint('resource_id', 'resource_type', 'tag_id', name='resource_tag_uniq'),)
id: Optional[int] = Field(default=None, index=True, primary_key=True, description="Tag Association UniqueID")
class TagDao(Tag):
@classmethod
def search_tags(cls, keyword: str = None, page: int = 0, limit: int = 0) -> List[Tag]:
""" Get all tags by default Paginable """
statement = select(Tag)
if keyword:
statement = statement.where(Tag.name.like(f'%{keyword}%'))
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
def count_tags(cls, keyword: str = None) -> int:
""" Count the number of tags """
statement = select(func.count(Tag.id))
if keyword:
statement = statement.where(Tag.name.like(f'%{keyword}%'))
with get_sync_db_session() as session:
return session.scalar(statement)
@classmethod
def insert_tag(cls, data: Tag) -> Tag:
""" Insert a new label data """
with get_sync_db_session() as session:
session.add(data)
session.commit()
session.refresh(data)
return data
@classmethod
def delete_tag(cls, tag_id: int) -> bool:
""" Delete a label data """
with get_sync_db_session() as session:
# Delete tag data
session.exec(delete(Tag).where(Tag.id == tag_id))
# Delete data associated with tags
session.exec(delete(TagLink).where(TagLink.tag_id == tag_id))
session.commit()
return True
@classmethod
def get_tag_by_name(cls, name: str) -> Tag:
""" Find tags by tag name """
with get_sync_db_session() as session:
statement = select(Tag).where(Tag.name == name)
return session.exec(statement).first()
@classmethod
def get_tag_by_id(cls, tag_id: int) -> Tag:
""" by TagIDFind Tags """
with get_sync_db_session() as session:
statement = select(Tag).where(Tag.id == tag_id)
return session.exec(statement).first()
@classmethod
def get_tags_by_ids(cls, tag_ids: List[int]) -> List[Tag]:
""" by TagIDFind Tags """
with get_sync_db_session() as session:
statement = select(Tag).where(Tag.id.in_(tag_ids))
return session.exec(statement).all()
@classmethod
def get_tags_by_resource(cls, resource_type: ResourceTypeEnum | None, resource_ids: list[str]) -> Dict[
str, List[Tag]]:
""" Query all tags under resources """
if resource_type is None:
statement = select(Tag.id, Tag.name, TagLink.resource_id).join(TagLink,
and_(
Tag.id == TagLink.tag_id,
TagLink.resource_id.in_(resource_ids)))
else:
statement = select(Tag.id, Tag.name, TagLink.resource_id).join(TagLink,
and_(
Tag.id == TagLink.tag_id,
TagLink.resource_id.in_(resource_ids),
TagLink.resource_type == resource_type.value))
with get_sync_db_session() as session:
result = session.exec(statement).all()
ret = {}
for one in result:
if one[2] not in ret:
ret[one[2]] = []
ret[one[2]].append(Tag(id=one[0], name=one[1]))
return ret
@classmethod
def get_tags_by_resource_batch(cls, resource_type: List[ResourceTypeEnum], resource_ids: list[str]) -> Dict[
str, List[Tag]]:
""" Query all tags under resources """
with get_sync_db_session() as session:
statement = select(Tag.id, Tag.name, TagLink.resource_id).join(TagLink,
and_(
Tag.id == TagLink.tag_id,
TagLink.resource_id.in_(resource_ids),
TagLink.resource_type.in_(
[x.value for x in resource_type])))
result = session.exec(statement).all()
ret = {}
for one in result:
if one[2] not in ret:
ret[one[2]] = []
ret[one[2]].append(Tag(id=one[0], name=one[1]))
return ret
@classmethod
def get_resources_by_tags(cls, tag_ids: List[int], resource_type: ResourceTypeEnum) -> List[TagLink]:
""" Query all resources under tags """
statement = select(TagLink).where(TagLink.tag_id.in_(tag_ids), TagLink.resource_type == resource_type.value)
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
def get_resources_by_tags_batch(cls, tag_ids: List[int], resource_type: List[ResourceTypeEnum]) -> List[TagLink]:
""" Query all resources under tags """
statement = select(TagLink).where(TagLink.tag_id.in_(tag_ids),
TagLink.resource_type.in_([x.value for x in resource_type]))
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
def insert_tag_link(cls, tag_link: TagLink) -> TagLink:
""" Insert Label Associated Data """
with get_sync_db_session() as session:
session.add(tag_link)
session.commit()
session.refresh(tag_link)
return tag_link
@classmethod
def get_tag_link(cls, tag_link_id: int) -> TagLink:
""" Associate via tagsIDFind Label Associated Data """
with get_sync_db_session() as session:
statement = select(TagLink).where(TagLink.id == tag_link_id)
return session.exec(statement).first()
@classmethod
def delete_tag_link(cls, tag_link_id: int) -> bool:
""" Delete tag association data """
with get_sync_db_session() as session:
# Delete tag association data
session.exec(delete(TagLink).where(TagLink.id == tag_link_id))
session.commit()
return True
@classmethod
def delete_resource_tag(cls, tag_id: int, resource_id: str, resource_type: ResourceTypeEnum) -> bool:
""" Delete tag association data """
statement = delete(TagLink).where(
TagLink.tag_id == tag_id,
TagLink.resource_id == resource_id,
TagLink.resource_type == resource_type.value
)
with get_sync_db_session() as session:
# Delete tag association data
session.exec(statement)
session.commit()
return True
================================================
FILE: src/backend/bisheng/database/models/template.py
================================================
from datetime import datetime
from typing import Dict, Optional
from pydantic import model_validator
from sqlalchemy import JSON, Column, DateTime, text, String
from sqlmodel import Field
from bisheng.common.models.base import SQLModelSerializable
class TemplateSkillBase(SQLModelSerializable):
name: str = Field(index=True)
description: Optional[str] = Field(default=None, sa_column=Column(String(length=1000)))
data: Optional[Dict] = Field(default=None, sa_column=Column(JSON))
order_num: Optional[int] = Field(default=True, index=True)
# 1 flow 5 assistant 10 workflow
flow_type: Optional[int] = Field(default=1)
flow_id: Optional[str] = Field(default=None, index=False)
guide_word: Optional[str] = Field(default=None, sa_column=Column(String(length=1000)))
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class Template(TemplateSkillBase, table=True):
id: Optional[int] = Field(default=None, primary_key=True)
class TemplateRead(TemplateSkillBase):
id: int
name: str
class TemplateCreate(TemplateSkillBase):
# 1 flow 5 assistant 10 workflow
# flow_type: int = 1
pass
class TemplateUpdate(SQLModelSerializable):
name: Optional[str] = None
description: Optional[str] = None
data: Optional[Dict] = None
order_num: Optional[int] = None
guide_word: Optional[str] = None
@model_validator(mode="before")
@classmethod
def convert_field(cls, values: dict) -> dict:
if values.get("order_num", None):
values['order_num'] = int(float(values['order_num']))
return values
================================================
FILE: src/backend/bisheng/database/models/user_group.py
================================================
from datetime import datetime
from typing import List, Optional
from sqlalchemy import Column, DateTime, delete, text, INT
from sqlmodel import Field, select
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session, get_async_db_session
from bisheng.database.models.group import DefaultGroup
class UserGroupBase(SQLModelSerializable):
user_id: Optional[int] = Field(
default=None,
foreign_key="user.user_id",
primary_key=True,
ondelete="CASCADE"
)
group_id: Optional[int] = Field(
default=None,
foreign_key="group.id",
primary_key=True,
ondelete="CASCADE"
)
is_group_admin: bool = Field(default=False, index=False, description='Is Group Admin') # Admin does not belong to this user group
remark: Optional[str] = Field(default=None, index=False)
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class UserGroup(UserGroupBase, table=True):
id: Optional[int] = Field(default=None, sa_column=Column(INT, primary_key=True, autoincrement=True))
class UserGroupRead(UserGroupBase):
id: Optional[int] = None
class UserGroupUpdate(UserGroupBase):
user_id: Optional[int] = None
group_id: Optional[int] = None
is_group_admin: Optional[bool] = None
remark: Optional[str] = None
class UserGroupCreate(UserGroupBase):
pass
class UserGroupDao(UserGroupBase):
@classmethod
def get_user_group(cls, user_id: int) -> List[UserGroup]:
"""
Get the user group the user is in
"""
with get_sync_db_session() as session:
statement = select(UserGroup).where(UserGroup.user_id == user_id).where(UserGroup.is_group_admin == 0)
return session.exec(statement).all()
@classmethod
async def aget_user_group(cls, user_id: int) -> List[UserGroup]:
"""
Get the user group the user is in asynchronously
"""
statement = select(UserGroup).where(UserGroup.user_id == user_id).where(UserGroup.is_group_admin == 0)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def get_user_admin_group(cls, user_id: int) -> List[UserGroup]:
"""
Get user groups where the user is an administrator
"""
with get_sync_db_session() as session:
statement = select(UserGroup).where(UserGroup.user_id == user_id).where(UserGroup.is_group_admin == 1)
return session.exec(statement).all()
@classmethod
async def aget_user_admin_group(cls, user_id: int, group_id: int = None) -> List[UserGroup]:
"""
Get user groups where the user is an administrator, If it passes,group_idOnly the group's admin information will be retrieved
"""
statement = select(UserGroup).where(UserGroup.user_id == user_id).where(UserGroup.is_group_admin == 1)
if group_id:
statement = statement.where(UserGroup.group_id == group_id)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def insert_user_group(cls, user_group: UserGroupCreate) -> UserGroup:
with get_sync_db_session() as session:
user_group = UserGroup.validate(user_group)
session.add(user_group)
session.commit()
session.refresh(user_group)
return user_group
@classmethod
def insert_user_group_admin(cls, user_id: int, group_id: int) -> UserGroup:
"""
Set user as group administrator
"""
with get_sync_db_session() as session:
user_group = UserGroup(user_id=user_id, group_id=group_id, is_group_admin=True)
session.add(user_group)
session.commit()
session.refresh(user_group)
return user_group
@classmethod
def delete_user_group(cls, user_id: int, group_id: int) -> None:
with get_sync_db_session() as session:
statement = select(UserGroup).where(UserGroup.user_id == user_id,
UserGroup.group_id == group_id)
user_group = session.exec(statement).first()
session.delete(user_group)
session.commit()
@classmethod
def delete_user_groups(cls, user_id: int, group_ids: List[int]):
"""
Remove users from certain groups
"""
with get_sync_db_session() as session:
# Empty all old user groups first
statement = delete(UserGroup).where(
UserGroup.user_id == user_id).where(
UserGroup.is_group_admin == 0).where(
UserGroup.group_id.in_(group_ids)
)
session.exec(statement)
session.commit()
@classmethod
def add_user_groups(cls, user_id: int, group_ids: List[int]):
"""
Add users to some groups
"""
with get_sync_db_session() as session:
for group_id in group_ids:
user_group = UserGroup(user_id=user_id, group_id=group_id, is_group_admin=0)
session.add(user_group)
session.commit()
@classmethod
def get_group_user(cls,
group_id: int,
page_size: str = None,
page_num: str = None) -> List[UserGroup]:
"""
Get all users in a group, excluding admins
"""
with get_sync_db_session() as session:
statement = select(UserGroup).where(UserGroup.group_id == group_id).where(UserGroup.is_group_admin == 0)
if page_num and page_size:
statement = statement.offset(
(int(page_num) - 1) * int(page_size)).limit(int(page_size))
return session.exec(statement).all()
@classmethod
def get_groups_user(cls,
group_ids: List[int],
page: int = 0,
limit: int = 0) -> List[UserGroup]:
"""
Batch Get Users Under Grouping
"""
with get_sync_db_session() as session:
statement = select(UserGroup).where(UserGroup.group_id.in_(group_ids)).where(
UserGroup.is_group_admin == 0)
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
return session.exec(statement).all()
@classmethod
async def aget_group_users(cls,
group_ids: List[int],
page: int = 0,
limit: int = 0) -> List[UserGroup]:
"""
Batch Get Users Under Grouping
Args:
group_ids:
page:
limit:
Returns:
"""
statement = select(UserGroup).where(UserGroup.group_id.in_(group_ids))
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def is_users_in_group(cls, group_id: int, user_ids: List[int]) -> List[UserGroup]:
with get_sync_db_session() as session:
statement = select(UserGroup).where(UserGroup.group_id == group_id,
UserGroup.user_id.in_(user_ids))
return session.exec(statement).all()
@classmethod
def get_groups_admins(cls, group_ids: List[int]) -> List[UserGroup]:
with get_sync_db_session() as session:
statement = select(UserGroup).where(UserGroup.group_id.in_(group_ids),
UserGroup.is_group_admin == 1)
return session.exec(statement).all()
@classmethod
def update_user_groups(cls, user_groups: List[UserGroup]) -> List[UserGroup]:
with get_sync_db_session() as session:
session.add_all(user_groups)
session.commit()
return user_groups
@classmethod
async def add_default_user_group(cls, user_id: int) -> None:
"""
Add users to the default user group
"""
async with get_async_db_session() as session:
user_group = UserGroup(user_id=user_id, group_id=DefaultGroup, is_group_admin=False)
session.add(user_group)
await session.commit()
@classmethod
def delete_group_admins(cls, group_id: int, admin_ids: List[int]) -> None:
"""
Bulk delete user groupsadmin
"""
with get_sync_db_session() as session:
statement = delete(UserGroup).where(
UserGroup.group_id == group_id).where(
UserGroup.user_id.in_(admin_ids)).where(
UserGroup.is_group_admin == 1)
session.exec(statement)
session.commit()
@classmethod
def delete_group_all_admin(cls, group_id: int) -> None:
"""
Remove all administrators under the user group
"""
with get_sync_db_session() as session:
statement = delete(UserGroup).where(
UserGroup.group_id == group_id).where(
UserGroup.is_group_admin == 1)
session.exec(statement)
session.commit()
================================================
FILE: src/backend/bisheng/database/models/user_link.py
================================================
from datetime import datetime
from typing import List
from typing import Optional
from sqlalchemy import text, and_, delete
from sqlalchemy.exc import IntegrityError
from sqlmodel import Field, Column, DateTime, select
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session
class UserLinkBase(SQLModelSerializable):
user_id: int = Field(index=True)
type: str = Field(index=True, max_length=32)
type_detail: str = Field(index=True, max_length=255, description='typeRelated Information')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class UserLink(UserLinkBase, table=True):
__tablename__ = 'user_link'
id: Optional[int] = Field(default=None, primary_key=True)
class UserLinkDao(UserLinkBase):
@classmethod
def get_user_link(cls, user_id: int, types: list) -> List[UserLink]:
with get_sync_db_session() as session:
statement = select(UserLink).where(and_(UserLink.user_id == user_id, UserLink.type.in_(types)))
statement = statement.order_by(UserLink.create_time.desc()) # Sort by creation time descending, newest added first
return session.exec(statement).all()
@classmethod
def add_user_link(cls, user_id: int, type: str, type_detail: str) -> tuple[UserLink, bool]:
"""
Add user link
Returns:
tuple[UserLink, bool]: (User Linked Objects, Whether to add for new)
"""
with get_sync_db_session() as session:
try:
# Try inserting directly, leveraging the database's unique constraints
user_link = UserLink(user_id=user_id, type=type, type_detail=type_detail, create_time=datetime.now(),
update_time=datetime.now())
session.add(user_link)
session.commit()
return user_link, True # Add New
except IntegrityError:
# Rollback and query existing records if unique constraints are violated
session.rollback()
existing = session.exec(
select(UserLink).where(
and_(
UserLink.user_id == user_id,
UserLink.type == type,
UserLink.type_detail == type_detail
)
)
).first()
return existing, False # already exists
@classmethod
def delete_user_link(cls, user_id: int, type: str, type_detail: str) -> None:
with get_sync_db_session() as session:
statement = delete(UserLink).where(
and_(UserLink.user_id == user_id, UserLink.type == type, UserLink.type_detail == type_detail))
session.exec(statement)
session.commit()
return True
================================================
FILE: src/backend/bisheng/database/models/variable_value.py
================================================
from datetime import datetime
from typing import Optional, List
# if TYPE_CHECKING:
from pydantic import field_validator
from sqlalchemy import Column, DateTime, text
from sqlmodel import Field, select
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session
class VariableBase(SQLModelSerializable):
flow_id: str = Field(index=True, description='Belonging Skills')
version_id: int = Field(description='Version of the skill to which it belongs')
node_id: str = Field(index=True, description='belong tonode')
variable_name: Optional[str] = Field(index=True, default=None, description='Variables')
value_type: int = Field(index=False, description='Variable type1=Text 2=list 3=file')
is_option: int = Field(index=False, default=1, description='Required? 1=Required 0=Price is not required')
value: str = Field(index=False, default=0, description='variable value, the length of the incoming text when the text is')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
@field_validator('variable_name')
@classmethod
def validate_length(cls, v):
# dict_keys(['description', 'name', 'id', 'data'])
if not v:
return v
if len(v) > 50:
v = v[:50]
return v
@field_validator('value')
@classmethod
def validate_value(cls, v):
# dict_keys(['description', 'name', 'id', 'data'])
if not v:
return v
# Deduplication keeps the original order
v_list = v.split(',')
res = []
for one in v_list:
if one not in res:
res.append(one)
return ','.join(res)
class Variable(VariableBase, table=True):
__tablename__ = 't_variable_value'
id: Optional[int] = Field(default=None, primary_key=True)
class VariableCreate(VariableBase):
pass
class VariableRead(VariableBase):
id: int
class VariableDao(Variable):
@classmethod
def create_variable(cls, variable: Variable) -> Variable:
"""
Create New variant
"""
with get_sync_db_session() as session:
db_variable = Variable.from_orm(variable)
session.add(db_variable)
session.commit()
session.refresh(db_variable)
return db_variable
@classmethod
def get_variables(cls, flow_id: str, node_id: str, variable_name: str, version_id: int) -> List[Variable]:
with get_sync_db_session() as session:
query = select(Variable).where(Variable.flow_id == flow_id)
if node_id:
query = query.where(Variable.node_id == node_id)
if variable_name:
query = query.where(Variable.variable_name == variable_name)
if version_id:
query = query.where(Variable.version_id == version_id)
return session.exec(query.order_by(Variable.id.asc())).all()
@classmethod
def copy_variables(cls, flow_id: str, old_version_id: int, version_id: int) -> List[Variable]:
"""
Copy the version of the form data to In the new version
"""
with get_sync_db_session() as session:
query = select(Variable).where(Variable.flow_id == flow_id, Variable.version_id == old_version_id)
old_version = session.exec(query).all()
for one in old_version:
new_version = Variable.from_orm(one)
new_version.id = None
new_version.version_id = version_id
session.add(new_version)
session.commit()
return old_version
================================================
FILE: src/backend/bisheng/default_node.yaml
================================================
autogen_roles:
AutoGenAssistant:
documentation: ""
AutoGenGroupChatManager:
documentation: ""
AutoGenUser:
documentation: ""
AutoGenCoder:
documentation: ""
AutoGenCustomRole:
documentation: ""
agents:
LLMFunctionsAgent:
documentation: ""
ZeroShotAgent:
documentation: "https://python.langchain.com/docs/modules/agents/how_to/custom_mrkl_agent"
JsonAgent:
documentation: "https://python.langchain.com/docs/modules/agents/toolkits/openapi"
CSVAgent:
documentation: "https://python.langchain.com/docs/modules/agents/toolkits/csv"
AgentInitializer:
documentation: "https://python.langchain.com/docs/modules/agents/agent_types/"
VectorStoreAgent:
documentation: ""
VectorStoreRouterAgent:
documentation: ""
SQLAgent:
documentation: ""
ChatglmFunctionsAgent:
documentation: ""
OpenAIToolsAgent:
documentation: ""
chains:
RetrievalChain:
documentation: ""
RuleBasedRouter:
documentation: ""
MultiRuleChain:
documentation: ""
MultiPromptChain:
documentation: ""
LLMRouterChain:
documentation: ""
AutoGenChain:
documentation: ""
APIChain:
documentation: ""
TransformChain:
documentation: ""
QAGenerationChainV2:
documentation: ""
SimpleSequentialChain:
documentation: ""
SequentialChain:
documentation: ""
LLMChain:
documentation: "https://python.langchain.com/docs/modules/chains/foundational/llm_chain"
LLMMathChain:
documentation: "https://python.langchain.com/docs/modules/chains/additional/llm_math"
LLMCheckerChain:
documentation: "https://python.langchain.com/docs/modules/chains/additional/llm_checker"
ConversationChain:
documentation: ""
SeriesCharacterChain:
documentation: ""
MidJourneyPromptChain:
documentation: ""
TimeTravelGuideChain:
documentation: ""
SQLDatabaseChain:
documentation: ""
RetrievalQA:
documentation: "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
BishengRetrievalQA:
documentation: ""
RetrievalQAWithSourcesChain:
documentation: ""
ConversationalRetrievalChain:
documentation: "https://python.langchain.com/docs/modules/chains/popular/chat_vector_db"
CombineDocsChain:
documentation: ""
# SummarizeDocsChain:
# documentation: ""
LoaderOutputChain:
documentation: ""
DalleGeneratorChain:
documentation: ""
documentloaders:
FireCrawlLoader:
documentation: ""
CustomKVLoader:
documentation: ""
UniversalKVLoader:
documentation: ""
ElemUnstructuredLoaderV0:
documentation: ""
AirbyteJSONLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/airbyte_json"
CoNLLULoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/conll-u"
CSVLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/csv"
EverNoteLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/evernote"
FacebookChatLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/facebook_chat"
GutenbergLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/gutenberg"
BSHTMLLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/how_to/html"
PyPDFDirectoryLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/how_to/pdf"
PyPDFLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/how_to/pdf"
SRTLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/subtitle"
TelegramChatLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/telegram"
TextLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/"
WebBaseLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/web_base"
AZLyricsLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/azlyrics"
CollegeConfidentialLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/college_confidential"
HNLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/hacker_news"
IFixitLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/ifixit"
IMSDbLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/imsdb"
GitbookLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/gitbook"
ReadTheDocsLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/readthedocs_documentation"
SlackDirectoryLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/slack"
NotionDirectoryLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/notion"
DirectoryLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/how_to/file_directory"
GitLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/git"
PDFWithSemanticLoader:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/git"
embeddings:
BishengEmbedding:
documentation: ""
OpenAIProxyEmbedding:
documentation: ""
OpenAIEmbeddings:
documentation: "https://python.langchain.com/docs/modules/data_connection/text_embedding/integrations/openai"
HuggingFaceEmbeddings:
documentation: "https://python.langchain.com/docs/modules/data_connection/text_embedding/integrations/sentence_transformers"
CohereEmbeddings:
documentation: "https://python.langchain.com/docs/modules/data_connection/text_embedding/integrations/cohere"
WenxinEmbeddings:
documentation: ""
HostEmbeddings:
documentation: ""
CustomHostEmbedding:
documentation: ""
llms:
BishengLLM:
documentation: ""
AzureChatOpenAI:
documentation: ""
OpenAI:
documentation: "https://python.langchain.com/docs/modules/model_io/models/llms/integrations/openai"
ChatOpenAI:
documentation: "https://python.langchain.com/docs/modules/model_io/models/chat/integrations/openai"
LlamaCpp:
documentation: "https://python.langchain.com/docs/modules/model_io/models/llms/integrations/llamacpp"
CTransformers:
documentation: "https://python.langchain.com/docs/modules/model_io/models/llms/integrations/ctransformers"
Cohere:
documentation: "https://python.langchain.com/docs/modules/model_io/models/llms/integrations/cohere"
Anthropic:
documentation: ""
ChatAnthropic:
documentation: "https://python.langchain.com/docs/modules/model_io/models/chat/integrations/anthropic"
HuggingFaceHub:
documentation: "https://python.langchain.com/docs/modules/model_io/models/llms/integrations/huggingface_hub"
VertexAI:
documentation: "https://python.langchain.com/docs/modules/model_io/models/llms/integrations/google_vertex_ai_palm"
ProxyChatLLM:
documentation: ""
ChatMinimaxAI:
documentation: ""
ChatWenxin:
documentation: ""
ChatQWen:
documentation: ""
ChatZhipuAI:
documentation: ""
ChatXunfeiAI:
documentation: ""
HostChatGLM:
documentation: ""
HostBaichuanChat:
documentation: ""
HostLlama2Chat:
documentation: ""
HostQwenChat:
documentation: ""
HostQwen1_5Chat:
documentation: ""
CustomLLMChat:
documentation: ""
SenseChat:
documentation: ""
HostYuanChat:
documentation: ""
HostYiChat:
documentation: ""
###
# There's a bug in this component deactivating until we get it sorted: _language_models.py", line 804, in send_message
# is_blocked=safety_attributes.get("blocked", False),
# AttributeError: 'list' object has no attribute 'get'
# ChatVertexAI:
# documentation: "https://python.langchain.com/docs/modules/model_io/models/chat/integrations/google_vertex_ai_palm"
###
memories:
# https://github.com/supabase-community/supabase-py/issues/482
# ZepChatMessageHistory:
# documentation: "https://python.langchain.com/docs/modules/memory/integrations/zep_memory"
ConversationEntityMemory:
documentation: "https://python.langchain.com/docs/modules/memory/integrations/entity_memory_with_sqlite"
# https://github.com/hwchase17/langchain/issues/6091
# SQLiteEntityStore:
# documentation: "https://python.langchain.com/docs/modules/memory/integrations/entity_memory_with_sqlite"
PostgresChatMessageHistory:
documentation: "https://python.langchain.com/docs/modules/memory/integrations/postgres_chat_message_history"
ConversationBufferMemory:
documentation: "https://python.langchain.com/docs/modules/memory/how_to/buffer"
ConversationSummaryMemory:
documentation: "https://python.langchain.com/docs/modules/memory/how_to/summary"
ConversationKGMemory:
documentation: "https://python.langchain.com/docs/modules/memory/how_to/kg"
ConversationBufferWindowMemory:
documentation: "https://python.langchain.com/docs/modules/memory/how_to/buffer_window"
VectorStoreRetrieverMemory:
documentation: "https://python.langchain.com/docs/modules/memory/how_to/vectorstore_retriever_memory"
MongoDBChatMessageHistory:
documentation: "https://python.langchain.com/docs/modules/memory/integrations/mongodb_chat_message_history"
ConversationRedisMemory:
documentation: ""
prompts:
ChatMessagePromptTemplate:
documentation: "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/msg_prompt_templates"
HumanMessagePromptTemplate:
documentation: "https://python.langchain.com/docs/modules/model_io/models/chat/how_to/prompts"
SystemMessagePromptTemplate:
documentation: "https://python.langchain.com/docs/modules/model_io/models/chat/how_to/prompts"
ChatPromptTemplate:
documentation: "https://python.langchain.com/docs/modules/model_io/models/chat/how_to/prompts"
PromptTemplate:
documentation: "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/"
MessagesPlaceholder:
documentation: ""
textsplitters:
CharacterTextSplitter:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_transformers/text_splitters/character_text_splitter"
RecursiveCharacterTextSplitter:
documentation: "https://python.langchain.com/docs/modules/data_connection/document_transformers/text_splitters/recursive_text_splitter"
toolkits:
OpenAPIToolkit:
documentation: ""
JsonToolkit:
documentation: ""
VectorStoreInfo:
documentation: ""
VectorStoreRouterToolkit:
documentation: ""
VectorStoreToolkit:
documentation: ""
tools:
Search:
documentation: ""
PAL-MATH:
documentation: ""
Calculator:
documentation: ""
Serper Search:
documentation: ""
Tool:
documentation: ""
PythonFunctionTool:
documentation: ""
PythonFunction:
documentation: ""
JsonSpec:
documentation: ""
News API:
documentation: ""
TMDB API:
documentation: ""
Podcast API:
documentation: ""
QuerySQLDataBaseTool:
documentation: ""
InfoSQLDatabaseTool:
documentation: ""
ListSQLDatabaseTool:
documentation: ""
BingSearchRun:
documentation: ""
GoogleSearchRun:
documentation: ""
GoogleSearchResults:
documentation: ""
GoogleSerperRun:
documentation: ""
JsonListKeysTool:
documentation: ""
JsonGetValueTool:
documentation: ""
PythonREPLTool:
documentation: ""
PythonAstREPLTool:
documentation: ""
RequestsGetTool:
documentation: ""
RequestsPostTool:
documentation: ""
RequestsPatchTool:
documentation: ""
RequestsPutTool:
documentation: ""
RequestsDeleteTool:
documentation: ""
WikipediaQueryRun:
documentation: ""
WolframAlphaQueryRun:
documentation: ""
utilities:
BingSearchAPIWrapper:
documentation: ""
GoogleSearchAPIWrapper:
documentation: ""
GoogleSerperAPIWrapper:
documentation: ""
SearxResults:
documentation: ""
SearxSearchWrapper:
documentation: ""
SerpAPIWrapper:
documentation: ""
WikipediaAPIWrapper:
documentation: ""
WolframAlphaAPIWrapper:
documentation: ""
retrievers:
MixEsVectorRetriever:
documentation: ""
MultiQueryRetriever:
documentation: "https://python.langchain.com/docs/modules/data_connection/retrievers/how_to/MultiQueryRetriever"
# https://github.com/supabase-community/supabase-py/issues/482
# ZepRetriever:
# documentation: "https://python.langchain.com/docs/modules/data_connection/retrievers/integrations/zep_memorystore"
vectorstores:
ElasticKeywordsSearch:
documentation: "http://192.168.106.116:8030"
ElasticsearchStore:
documentation: https://python.langchain.com/docs/integrations/vectorstores/elasticsearch
Milvus:
documentation: "http://192.168.106.116:8030"
MilvusWithPermissionCheck:
documentation: "http://192.168.106.116:8030"
ElasticsearchWithPermissionCheck:
documentation: "http://192.168.106.116:8030"
Chroma:
documentation: "https://python.langchain.com/docs/modules/data_connection/vectorstores/integrations/chroma"
Qdrant:
documentation: "https://python.langchain.com/docs/modules/data_connection/vectorstores/integrations/qdrant"
Weaviate:
documentation: "https://python.langchain.com/docs/modules/data_connection/vectorstores/integrations/weaviate"
FAISS:
documentation: "https://python.langchain.com/docs/modules/data_connection/vectorstores/integrations/faiss"
Pinecone:
documentation: "https://python.langchain.com/docs/modules/data_connection/vectorstores/integrations/pinecone"
SupabaseVectorStore:
documentation: "https://python.langchain.com/docs/modules/data_connection/vectorstores/integrations/supabase"
MongoDBAtlasVectorSearch:
documentation: "https://python.langchain.com/docs/modules/data_connection/vectorstores/integrations/mongodb_atlas"
# Requires docarray >=0.32.0 but langchain-serve requires jina 3.15.2 which doesn't support docarray >=0.32.0
# DocArrayInMemorySearch:
# documentation: "https://python.langchain.com/docs/modules/data_connection/vectorstores/integrations/docarray_in_memory"
wrappers:
RequestsWrapper:
documentation: ""
SQLDatabase:
documentation: ""
output_parsers:
StructuredOutputParser:
documentation: "https://python.langchain.com/docs/modules/model_io/output_parsers/structured"
ResponseSchema:
documentation: "https://python.langchain.com/docs/modules/model_io/output_parsers/structured"
RouterOutputParser:
documentation: ""
input_output:
VariableNode:
documentation: ""
InputNode:
documentation: ""
Output:
documentation: ""
InputFileNode:
documentation: ""
================================================
FILE: src/backend/bisheng/field_typing/__init__.py
================================================
from typing import Any
from .constants import (AgentExecutor, BaseChatMemory, BaseLanguageModel, BaseLLM, BaseLoader,
BaseMemory, BaseOutputParser, BasePromptTemplate, BaseRetriever, Callable,
Chain, ChatPromptTemplate, Data, Document, Embeddings, NestedDict, Object,
Prompt, PromptTemplate, TextSplitter, Tool, VectorStore)
from .range_spec import RangeSpec
def _import_template_field():
from bisheng.template.field.base import TemplateField
return TemplateField
def __getattr__(name: str) -> Any:
# This is to avoid circular imports
if name == 'TemplateField':
return _import_template_field()
elif name == 'RangeSpec':
return RangeSpec
# The other names should work as if they were imported from constants
# Import the constants module langflow.field_typing.constants
from . import constants
return getattr(constants, name)
__all__ = [
'NestedDict',
'Data',
'Tool',
'PromptTemplate',
'Chain',
'BaseChatMemory',
'BaseLLM',
'BaseLanguageModel',
'BaseLoader',
'BaseMemory',
'BaseOutputParser',
'BaseRetriever',
'VectorStore',
'Embeddings',
'TextSplitter',
'Document',
'AgentExecutor',
'Object',
'Callable',
'BasePromptTemplate',
'ChatPromptTemplate',
'Prompt',
'RangeSpec',
'TemplateField',
]
================================================
FILE: src/backend/bisheng/field_typing/constants.py
================================================
from typing import Callable, Dict, Union
from langchain.agents.agent import AgentExecutor
from langchain.chains.base import Chain
from langchain.document_loaders.base import BaseLoader
from langchain.llms.base import BaseLLM
from langchain.memory.chat_memory import BaseChatMemory
from langchain.prompts import BasePromptTemplate, ChatPromptTemplate, PromptTemplate
from langchain.schema import BaseOutputParser, BaseRetriever, Document
from langchain.schema.embeddings import Embeddings
from langchain.schema.language_model import BaseLanguageModel
from langchain.schema.memory import BaseMemory
from langchain.text_splitter import TextSplitter
from langchain.tools import Tool
from langchain_community.vectorstores import VectorStore
# Type alias for more complex dicts
NestedDict = Dict[str, Union[str, Dict]]
class Object:
pass
class Data:
pass
class Prompt:
pass
LANGCHAIN_BASE_TYPES = {
'Chain': Chain,
'AgentExecutor': AgentExecutor,
'Tool': Tool,
'BaseLLM': BaseLLM,
'BaseLanguageModel': BaseLanguageModel,
'PromptTemplate': PromptTemplate,
'ChatPromptTemplate': ChatPromptTemplate,
'BasePromptTemplate': BasePromptTemplate,
'BaseLoader': BaseLoader,
'Document': Document,
'TextSplitter': TextSplitter,
'VectorStore': VectorStore,
'Embeddings': Embeddings,
'BaseRetriever': BaseRetriever,
'BaseOutputParser': BaseOutputParser,
'BaseMemory': BaseMemory,
'BaseChatMemory': BaseChatMemory,
}
# Langchain base types plus Python base types
CUSTOM_COMPONENT_SUPPORTED_TYPES = {
**LANGCHAIN_BASE_TYPES,
'NestedDict': NestedDict,
'Data': Data,
'Object': Object,
'Callable': Callable,
'Prompt': Prompt,
}
================================================
FILE: src/backend/bisheng/field_typing/range_spec.py
================================================
from pydantic import field_validator, BaseModel, model_validator
class RangeSpec(BaseModel):
min: float = -1.0
max: float = 1.0
step: float = 0.1
@model_validator(mode='before')
@classmethod
def max_must_be_greater_than_min(cls, values):
if 'min' in values and values['max'] <= values['min']:
raise ValueError('max must be greater than min')
return values
@field_validator('step')
@classmethod
def step_must_be_positive(cls, v):
if v <= 0:
raise ValueError('step must be positive')
return v
================================================
FILE: src/backend/bisheng/finetune/README.md
================================================
# 微调和数据集模块
================================================
FILE: src/backend/bisheng/finetune/__init__.py
================================================
================================================
FILE: src/backend/bisheng/finetune/api/__init__.py
================================================
================================================
FILE: src/backend/bisheng/finetune/api/finetune.py
================================================
import json
import tempfile
from typing import Optional
from fastapi import APIRouter, Body, Depends, File, Query, UploadFile
from loguru import logger
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.schemas.api import resp_200, PageData
from bisheng.core.cache.utils import async_file_download
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync
from bisheng.finetune.domain.models.finetune import Finetune, FinetuneChangeModelName, FinetuneList
from bisheng.finetune.domain.services.finetune import FinetuneService
from bisheng.finetune.domain.services.finetune_file import FinetuneFileService
from bisheng.knowledge.domain.models.knowledge import KnowledgeDao
from bisheng.knowledge.domain.models.knowledge_file import QAKnoweldgeDao
from ..schemas import FinetuneCreateReq
router = APIRouter(prefix='/finetune', tags=['Finetune'], dependencies=[Depends(UserPayload.get_login_user)])
# create finetune job
@router.post('/job')
async def create_job(*, finetune: FinetuneCreateReq, login_user: UserPayload = Depends(UserPayload.get_login_user)):
# get login user
finetune = Finetune(**finetune.model_dump(exclude={'method'}),
method=finetune.method.value,
user_id=login_user.user_id,
user_name=login_user.user_name)
ret = await FinetuneService.create_job(finetune)
return resp_200(ret)
# Delete training task
@router.delete('/job')
async def delete_job(*, job_id: str, login_user: UserPayload = Depends(UserPayload.get_login_user)):
# get login user
await FinetuneService.delete_job(job_id, login_user)
return resp_200(None)
# Abort training mission
@router.post('/job/cancel')
async def cancel_job(*, job_id: str, login_user: UserPayload = Depends(UserPayload.get_login_user)):
# get login user
ret = await FinetuneService.cancel_job(job_id, login_user)
return resp_200(ret)
# Publish Training Tasks
@router.post('/job/publish')
async def publish_job(*, job_id: str, login_user: UserPayload = Depends(UserPayload.get_login_user)):
# get login user
ret = await FinetuneService.publish_job(job_id, login_user)
return resp_200(ret)
@router.post('/job/publish/cancel')
async def cancel_publish_job(*, job_id: str, login_user: UserPayload = Depends(UserPayload.get_login_user)):
# get login user
ret = await FinetuneService.cancel_publish_job(job_id, login_user)
return resp_200(ret)
# Get a list of training tasks, support pagination
@router.get('/job')
async def get_job(*,
server: str = Query(default=None, description='RelatedRTService name'),
status: str = Query(
default='',
title='Multiple commas,separator',
description='the status of the training task,1: Training 2: Training failed 3: Task Aborted 4: Training Successful 5: Publication complete'),
model_name: Optional[str] = Query(default='', description='Model Name,Fuzzy search'),
page: Optional[int] = Query(default=1, description='Page'),
limit: Optional[int] = Query(default=10, description='Listings Per Page'),
login_user: UserPayload = Depends(UserPayload.get_login_user)):
status_list = []
if status.strip():
status_list = [int(one) for one in status.strip().split(',')]
req_data = FinetuneList(server_name=server,
status=status_list,
model_name=model_name,
page=page,
limit=limit)
if not login_user.is_admin():
req_data.user_id = login_user.user_id
data, total = await FinetuneService.get_all_job(req_data)
return resp_200(data=PageData(data=data, total=total))
# Get the latest details of the task, this interface will synchronize the querySFT-backendThe side updates the task status to the latest
@router.get('/job/info')
async def get_job_info(*, job_id: str, login_user: UserPayload = Depends(UserPayload.get_login_user)):
# get login user
ret = await FinetuneService.get_job_info(job_id)
return resp_200(ret)
@router.patch('/job/model')
async def update_job(*, req_data: FinetuneChangeModelName,
login_user: UserPayload = Depends(UserPayload.get_login_user)):
# get login user
ret = await FinetuneService.change_job_model_name(req_data)
return resp_200(ret)
@router.post('/job/file')
async def upload_file(*,
files: list[UploadFile] = File(description='Training File List'),
login_user: UserPayload = Depends(UserPayload.get_login_user)):
try:
ret = await FinetuneFileService.upload_file(files, False, login_user)
return resp_200(ret)
except Exception as e:
logger.error(f'upload finetune file error: {e}')
raise e
finally:
if files:
for file in files:
await file.close()
@router.post('/job/file/preset')
async def upload_preset_file(*,
files: Optional[str] = Body(default=None, description='Preset Training File List'),
name: Optional[str] = Body(description='Dataset Name'),
qa_list: Optional[list[int]] = Body(default=None,
description='QAThe knowledge base upon'),
login_user: UserPayload = Depends(UserPayload.get_login_user)):
ret = None
if files:
filepath, file_name = await async_file_download(files)
logger.info(f'dataset upload_file_name: {file_name}')
ret = await FinetuneFileService.upload_preset_file(name, 0, filepath, login_user)
elif qa_list:
# will beqaData byfinetuneFormat for file storage
qa_knowledge_db = await KnowledgeDao.aget_list_by_ids(qa_list)
qa_knowledge_db_ids = [qa_knowledge.id for qa_knowledge in qa_knowledge_db]
qa_db_list = await QAKnoweldgeDao.aget_qa_knowledge_by_knowledge_ids(qa_knowledge_db_ids)
qa_list = []
for qa in qa_db_list:
qa_list.extend([{
'instruction': question,
'input': '',
'output': json.loads(qa.answers)[0]
} for question in qa.questions])
with tempfile.NamedTemporaryFile(mode='w+', suffix='.json') as filepath:
json.dump(qa_list, filepath, ensure_ascii=False, indent=2)
filepath.seek(0)
ret = await FinetuneFileService.upload_preset_file(name, 1, filepath.name, login_user)
return resp_200(ret)
# Get a list of preset training files
@router.get('/job/file/preset')
async def get_preset_file(*,
page_size: Optional[int] = None,
page_num: Optional[int] = None,
keyword: Optional[str] = None,
login_user: UserPayload = Depends(UserPayload.get_login_user)):
ret = await FinetuneFileService.get_preset_file(keyword, page_size, page_num)
return resp_200(ret)
@router.delete('/job/file/preset')
async def delete_preset_file(*, file_id: str, login_user: UserPayload = Depends(UserPayload.get_login_user)):
# get login user
await FinetuneFileService.delete_preset_file(file_id, login_user)
return resp_200()
@router.get('/job/file/download')
async def get_download_url(*, file_url: str, login_user: UserPayload = Depends(UserPayload.get_login_user)):
minio_client = get_minio_storage_sync()
download_url = await minio_client.get_share_link(file_url, clear_host=False)
return resp_200(data={'url': download_url})
@router.get('/server/filters')
async def get_server_filters(*, login_user: UserPayload = Depends(UserPayload.get_login_user)):
ret = await FinetuneService.get_server_filters()
return resp_200(data=ret)
@router.get('/model/list')
async def get_model_list(login_user: UserPayload = Depends(UserPayload.get_login_user),
server_id: int = Query(..., description='ftService UniqueID')):
""" DapatkanftList of all models under the service """
ret = await FinetuneService.get_model_list(server_id)
return resp_200(data=ret)
@router.get('/gpu')
async def get_gpu_info(*, login_user: UserPayload = Depends(UserPayload.get_login_user)):
# get login user
ret = await FinetuneService.get_gpu_info()
return resp_200(data=ret)
================================================
FILE: src/backend/bisheng/finetune/api/server.py
================================================
from fastapi import APIRouter, Depends
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.schemas.api import resp_200
from bisheng.finetune.domain.models.model_deploy import ModelDeployDao
from bisheng.finetune.domain.models.server import Server, ServerCreate, ServerDao
from ...common.errcode.http_error import NotFoundError
# build router
router = APIRouter(prefix='/server', tags=['server'], dependencies=[Depends(UserPayload.get_login_user)])
@router.post('/add')
async def add_server(*, server: ServerCreate):
db_server = Server.model_validate(server)
db_server = await ServerDao.insert(db_server)
return resp_200(db_server)
@router.get('/list_server')
async def list_server():
rt_server = await ServerDao.find_all_server()
rt_server = rt_server or []
return resp_200(rt_server)
@router.delete('/{server_id}', status_code=200)
async def delete_server(*, server_id: int):
await ServerDao.delete(server_id)
return resp_200(None)
@router.get('/model/{deploy_id}')
async def get_model_deploy(*, deploy_id: int):
model_deploy = await ModelDeployDao.find_model(deploy_id)
if not ModelDeployDao:
raise NotFoundError()
return resp_200(data=model_deploy)
================================================
FILE: src/backend/bisheng/finetune/domain/__init__.py
================================================
================================================
FILE: src/backend/bisheng/finetune/domain/models/__init__.py
================================================
================================================
FILE: src/backend/bisheng/finetune/domain/models/finetune.py
================================================
from datetime import datetime
from enum import Enum
from typing import Any, Dict, List, Optional
from pydantic import field_validator, BaseModel, model_validator
from sqlalchemy import Select
from sqlalchemy.dialects.mysql import LONGTEXT
from sqlmodel import JSON, Column, DateTime, Field, func, select, text, update, col
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_async_db_session
from bisheng.utils import generate_uuid
class TrainMethod(Enum):
FULL = 'full'
FREEZE = 'freeze'
LORA = 'lora'
class FinetuneStatus(Enum):
# Training
TRAINING = 1
# Training failed
FAILED = 2
# Task Aborted
CANCEL = 3
# Training Successful
SUCCESS = 4
# Publication complete
PUBLISHED = 5
class FinetuneBase(SQLModelSerializable):
id: str = Field(default=None, nullable=False, primary_key=True, description='Uniqueness quantificationID')
server: int = Field(default=0, index=True, description='RelatedRTSERVICESID')
server_name: str = Field(default='', index=True, description='RTService name')
rt_endpoint: str = Field(default='', description='RTService Address')
sft_endpoint: str = Field(default='', description='FTService Address')
base_model: int = Field(default=0, index=True, description='Foundation ModelID')
base_model_name: str = Field(max_length=50, description='Base Model Name')
root_model_name: str = Field(default='', description='Root underlying model name, which is the initial model name')
model_id: int = Field(default=0, index=True, description='Published Training ModelsID')
model_name: str = Field(index=True, max_length=50, description='Name of the training model')
method: str = Field(default=TrainMethod.FULL.value, nullable=False, max_length=20, description='Training Methods')
extra_params: Dict = Field(sa_column=Column(JSON), description='Additional parameters required for training tasks')
train_data: Optional[List[Dict]] = Field(default=None, sa_column=Column(JSON), description='Personal Training Dataset Information')
preset_data: Optional[List[Dict]] = Field(default=None, sa_column=Column(JSON), description='Preset training dataset information')
status: int = Field(default=FinetuneStatus.TRAINING.value, index=True, description='Status of the training task')
reason: Optional[str] = Field(default='', sa_column=Column(LONGTEXT), description='Task Failure Reason')
log_path: Optional[str] = Field(default='', max_length=512, description='Training log inminioPath on')
report: Optional[Dict] = Field(default=None, sa_column=Column(JSON), description='Assessment report data for training tasks')
user_id: int = Field(default=None, index=True, description='creatorID')
user_name: str = Field(default=None, description='creatorName')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
# Check training set data format
@classmethod
def validate_train(cls, v: Any):
if v is None:
return v
if v is None or not isinstance(v, list):
raise ValueError('Finetune.train_data must be a list')
for one in v:
if not (one.get('name', None) and one.get('url', None)):
raise ValueError('Finetune.train_data each item must be {name:"",url:"",num:0}')
return v
@model_validator(mode='before')
@classmethod
def validate_all(cls, values: Dict[str, Any]) -> Dict[str, Any]:
extra_params = values.get('extra_params', None)
if extra_params is None or not isinstance(extra_params, dict):
raise ValueError('Finetune.extra_params must be a valid json')
train_data = values.get('train_data', None)
cls.validate_train(train_data)
preset_data = values.get('preset_data', None)
cls.validate_train(preset_data)
return values
class Finetune(FinetuneBase, table=True):
id: str = Field(default_factory=generate_uuid, primary_key=True, unique=True)
class FinetuneList(BaseModel):
server: Optional[int] = Field(None, description='RelatedRTSERVICESID')
server_name: Optional[str] = Field(None, description='RelatedRTService name')
status: Optional[List[int]] = Field(None, description='Status list of training tasks')
model_name: Optional[str] = Field(None, description='Model Name, Fuzzy search')
page: Optional[int] = Field(default=1, description='Page')
limit: Optional[int] = Field(default=10, description='Listings Per Page')
user_id: Optional[int] = Field(None, description='UsersID')
def get_select_statement(self) -> (Select, Select):
"""Generate the select and count statements based on the filters."""
statement = select(Finetune)
count_statement = select(func.count(Finetune.id))
if self.server:
statement = statement.where(Finetune.server == self.server)
count_statement = count_statement.where(Finetune.server == self.server)
if self.server_name:
statement = statement.where(Finetune.server_name == self.server_name)
count_statement = count_statement.where(Finetune.server_name == self.server_name)
if self.status:
statement = statement.where(col(Finetune.status).in_(self.status))
count_statement = count_statement.where(col(Finetune.status).in_(self.status))
if self.model_name:
statement = statement.where(col(Finetune.model_name).like(f'%{self.model_name}%'))
count_statement = count_statement.where(col(Finetune.model_name).like(f'%{self.model_name}%'))
if self.user_id:
statement = statement.where(Finetune.user_id == self.user_id)
count_statement = count_statement.where(Finetune.user_id == self.user_id)
if self.page and self.limit:
offset = (self.page - 1) * self.limit
statement = statement.offset(offset).limit(self.limit).order_by(col(Finetune.create_time).desc())
return statement, count_statement
class FinetuneChangeModelName(BaseModel):
id: str = Field(description='Training Mission UniqueID')
model_name: str
class FinetuneExtraParams(BaseModel):
gpus: str = Field(..., description='Needs to be usedGPUCard #')
val_ratio: float = Field(0.1, ge=0, le=1, description='Proportion of validation sets')
per_device_train_batch_size: int = Field(1, description='Batch size')
learning_rate: float = Field(0.00005, ge=0, le=1, description='Learning rate ')
num_train_epochs: int = Field(3, gt=0, description='Iteration rounds')
max_seq_len: int = Field(8192, gt=0, description='Maximum List Length')
cpu_load: str = Field('false', description='Y/NcpuLoad')
@field_validator('per_device_train_batch_size', mode='before')
@classmethod
def validate_batch_size(cls, v: str):
try:
batch_size = int(v)
if batch_size != 1:
if batch_size % 2 != 0:
raise ValueError('per_device_train_batch_size must be 1 or even number')
return batch_size
except Exception as e:
raise ValueError(f'per_device_train_batch_size must be an integer {e}')
@field_validator('gpus', mode='before')
@classmethod
def validate_gpus(cls, v: str):
try:
gpu_list = v.split(',')
gpus = ''
for one in gpu_list:
if not one:
continue
if not one.isdigit():
raise ValueError('gpus number must be integer')
gpus += one + ','
gpus = gpus[:-1]
if not gpus:
raise ValueError('gpus must not be empty')
return gpus
except Exception as e:
raise ValueError(f'gpus must be an str {e}')
class FinetuneDao(FinetuneBase):
@classmethod
async def insert_one(cls, data: Finetune) -> Finetune:
async with get_async_db_session() as session:
session.add(data)
await session.commit()
await session.refresh(data)
return data
@classmethod
async def update_job(cls, finetune: Finetune) -> Finetune:
if not finetune.id:
raise ValueError('Finetune.id must not be empty when update')
async with get_async_db_session() as session:
session.add(finetune)
await session.commit()
await session.refresh(finetune)
return finetune
@classmethod
async def find_job(cls, job_id: str) -> Finetune | None:
async with get_async_db_session() as session:
statement = select(Finetune).where(Finetune.id == job_id)
return (await session.exec(statement)).first()
@classmethod
async def find_job_by_model_name(cls, model_name: str) -> Finetune | None:
async with get_async_db_session() as session:
statement = select(Finetune).where(Finetune.model_name == model_name)
return (await session.exec(statement)).first()
@classmethod
async def find_job_by_model_id(cls, model_id: int) -> Finetune | None:
async with get_async_db_session() as session:
statement = select(Finetune).where(Finetune.model_id == model_id)
return (await session.exec(statement)).first()
@classmethod
async def change_status(cls, job_id: str, old_status: int, status: int) -> bool:
update_statement = update(Finetune).where(col(Finetune.id) == job_id,
col(Finetune.status) == old_status).values(status=status)
async with get_async_db_session() as session:
update_ret = await session.exec(update_statement)
await session.commit()
return update_ret.rowcount != 0
@classmethod
async def delete_job(cls, job: Finetune) -> bool:
if not job or not job.id:
raise ValueError('Finetune job to delete must not be None, and id must not be empty')
async with get_async_db_session() as session:
await session.delete(job)
await session.commit()
return True
@classmethod
async def find_jobs(cls, finetune_list: FinetuneList) -> (List[Finetune], int):
select_statement, count_statement = finetune_list.get_select_statement()
async with get_async_db_session() as session:
return (await session.exec(select_statement)).all(), await session.scalar(count_statement)
@classmethod
async def get_server_filters(cls) -> List[str]:
async with get_async_db_session() as session:
statement = select(Finetune.server_name).distinct()
return (await session.exec(statement)).all()
================================================
FILE: src/backend/bisheng/finetune/domain/models/model_deploy.py
================================================
from datetime import datetime
from typing import Optional, List
from sqlalchemy import Column, DateTime, String, UniqueConstraint, delete, text
from sqlmodel import Field, select, col
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_async_db_session
class ModelDeployBase(SQLModelSerializable):
endpoint: str = Field(index=False, unique=False)
server: str = Field(index=True)
model: str = Field(index=False)
config: Optional[str] = Field(default=None, sa_column=Column(String(length=512)))
status: Optional[str] = Field(default=None, index=False)
remark: Optional[str] = Field(default=None, sa_column=Column(String(length=4096)))
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class ModelDeploy(ModelDeployBase, table=True):
__table_args__ = (UniqueConstraint('model', 'server', name='model_server_uniq'),)
id: Optional[int] = Field(default=None, primary_key=True)
class ModelDeployDao(ModelDeployBase):
@classmethod
async def find_model(cls, model_id: int) -> ModelDeploy | None:
async with get_async_db_session() as session:
statement = select(ModelDeploy).where(ModelDeploy.id == model_id)
return (await session.exec(statement)).first()
@classmethod
async def find_model_by_server(cls, server_id: str) -> List[ModelDeploy]:
async with get_async_db_session() as session:
statement = select(ModelDeploy).where(ModelDeploy.server == server_id)
return (await session.exec(statement)).all()
@classmethod
async def find_model_by_server_and_name(cls, server: str, model: str) -> ModelDeploy | None:
async with get_async_db_session() as session:
statement = select(ModelDeploy).where(ModelDeploy.server == server, ModelDeploy.model == model)
return (await session.exec(statement)).first()
@classmethod
async def find_model_by_name(cls, model: str) -> ModelDeploy | None:
async with get_async_db_session() as session:
statement = select(ModelDeploy).where(ModelDeploy.model == model)
return (await session.exec(statement)).first()
@classmethod
async def delete_model(cls, model: ModelDeploy) -> bool:
async with get_async_db_session() as session:
await session.delete(model)
await session.commit()
return True
@classmethod
async def delete_model_by_id(cls, model_id: int):
async with get_async_db_session() as session:
statement = delete(ModelDeploy).where(col(ModelDeploy.id) == model_id)
await session.exec(statement)
await session.commit()
@classmethod
async def insert_one(cls, model: ModelDeploy) -> ModelDeploy:
async with get_async_db_session() as session:
session.add(model)
await session.commit()
await session.refresh(model)
return model
@classmethod
async def update_model(cls, model: ModelDeploy) -> ModelDeploy:
async with get_async_db_session() as session:
session.add(model)
await session.commit()
await session.refresh(model)
return model
class ModelDeployRead(ModelDeployBase):
id: Optional[int]
class ModelDeployQuery(SQLModelSerializable):
id: Optional[int]
server: Optional[str]
class ModelDeployCreate(ModelDeployBase):
pass
class ModelDeployUpdate(SQLModelSerializable):
id: int
config: Optional[str] = None
class ModelDeployInfo(ModelDeploy):
sft_support: bool = Field(default=False, description='Whether to support fine-tuning training')
================================================
FILE: src/backend/bisheng/finetune/domain/models/preset_train.py
================================================
from datetime import datetime
from typing import List, Optional, Tuple
from sqlalchemy import func
from sqlmodel import Column, DateTime, Field, select, text, col
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_async_db_session
from bisheng.utils import generate_uuid
# FinetunePreset training set for tasks
class PresetTrainBase(SQLModelSerializable):
id: str = Field(default=None, primary_key=True, description='Preset Training Files UniqueID')
url: str = Field(default='', description='minIoFile link on')
name: str = Field(default='', index=True, description='Uploaded file name')
user_id: str = Field(default='', index=True, description='creatorID')
user_name: str = Field(default='', index=True, description='creatorName')
type: int = Field(default=0, index=True, description='0 Doc. 1 QA')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class PresetTrain(PresetTrainBase, table=True):
id: str = Field(default_factory=generate_uuid, primary_key=True, unique=True)
class PresetTrainDao(PresetTrainBase):
@classmethod
async def insert_batch(cls, models: List[PresetTrain]) -> List[PresetTrain]:
async with get_async_db_session() as session:
for one in models:
session.add(one)
await session.commit()
for one in models:
await session.refresh(one)
return models
@classmethod
async def delete_one(cls, model: PresetTrain) -> bool:
async with get_async_db_session() as session:
await session.delete(model)
await session.commit()
return True
@classmethod
async def find_one(cls, file_id: str) -> PresetTrain | None:
async with get_async_db_session() as session:
statement = select(PresetTrain).where(PresetTrain.id == file_id)
return (await session.exec(statement)).first()
@classmethod
async def find_all(cls) -> List[PresetTrain]:
async with get_async_db_session() as session:
statement = select(PresetTrain)
return (await session.exec(statement)).all()
@classmethod
async def search_name(cls,
keyword: str = None,
page_size: int = None,
page_num: int = None) -> Tuple[List[PresetTrain], int]:
async with get_async_db_session() as session:
statement = select(PresetTrain)
count = select(func.count(PresetTrain.id))
if keyword:
statement = statement.where(col(PresetTrain.name).like('%{}%'.format(keyword)))
count = count.where(col(PresetTrain.name).like('%{}%'.format(keyword)))
if page_num and page_size:
statement = statement.offset((page_num - 1) * page_size).limit(page_size)
statement = statement.order_by(col(PresetTrain.create_time).desc())
return (await session.exec(statement)).all(), await session.scalar(count)
================================================
FILE: src/backend/bisheng/finetune/domain/models/server.py
================================================
from datetime import datetime
from typing import Optional
from sqlalchemy import Column, DateTime, text, delete
from sqlmodel import Field, select, col
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_async_db_session
class ServerBase(SQLModelSerializable):
endpoint: str = Field(index=False)
sft_endpoint: str = Field(default='', index=False, description='FinetuneService Address')
server: str = Field(index=True)
remark: Optional[str] = Field(default=None, index=False)
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class Server(ServerBase, table=True):
id: Optional[int] = Field(default=None, primary_key=True)
# Package business operations
class ServerDao(ServerBase):
@classmethod
async def find_server(cls, server_id: int) -> Server | None:
async with get_async_db_session() as session:
statement = select(Server).where(Server.id == server_id)
return (await session.exec(statement)).first()
@classmethod
async def find_all_server(cls):
async with get_async_db_session() as session:
statement = select(Server)
return (await session.exec(statement)).all()
@classmethod
async def insert(cls, server: Server) -> Server:
async with get_async_db_session() as session:
session.add(server)
await session.commit()
await session.refresh(server)
return server
@classmethod
async def delete(cls, server_id: int) -> None:
statement = delete(Server).where(col(Server.id) == server_id)
async with get_async_db_session() as session:
await session.exec(statement)
await session.commit()
class ServerRead(ServerBase):
id: Optional[int] = None
class ServerQuery(ServerBase):
id: Optional[int] = None
server: Optional[str] = None
class ServerCreate(ServerBase):
pass
================================================
FILE: src/backend/bisheng/finetune/domain/models/sft_model.py
================================================
from datetime import datetime
from typing import Optional
from sqlalchemy import Column, DateTime, delete, text, update
from sqlmodel import Field, select, col
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_async_db_session
# Available for trainingmodelVertical
class SftModelBase(SQLModelSerializable):
id: int = Field(default=None, nullable=False, primary_key=True, description='Uniqueness quantificationID')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class SftModel(SftModelBase, table=True):
model_name: str = Field(index=True, description='Model name that can be used for fine-tuning training')
class SftModelDao(SftModel):
@classmethod
async def get_sft_model(cls, model_name: str) -> SftModel | None:
async with get_async_db_session() as session:
statement = select(SftModel).where(SftModel.model_name == model_name)
return (await session.exec(statement)).first()
@classmethod
async def get_all_sft_model(cls):
async with get_async_db_session() as session:
statement = select(SftModel)
return (await session.exec(statement)).all()
@classmethod
async def insert_sft_model(cls, model_name: str) -> SftModel:
async with get_async_db_session() as session:
model = SftModel(model_name=model_name)
session.add(model)
await session.commit()
await session.refresh(model)
return model
@classmethod
async def delete_sft_model(cls, model_name: str) -> bool:
async with get_async_db_session() as session:
statement = delete(SftModel).where(col(SftModel.model_name) == model_name)
await session.exec(statement)
await session.commit()
return True
@classmethod
async def change_sft_model(cls, old_model_name, model_name) -> bool:
async with get_async_db_session() as session:
update_statement = update(SftModel).where(SftModel.model_name == old_model_name).values(
model_name=model_name)
update_ret = await session.exec(update_statement)
await session.commit()
return update_ret.rowcount != 0
================================================
FILE: src/backend/bisheng/finetune/domain/services/__init__.py
================================================
================================================
FILE: src/backend/bisheng/finetune/domain/services/finetune.py
================================================
import asyncio
import io
import json
from typing import Dict, List, Any
from loguru import logger
from pydantic import ValidationError
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.finetune import (CancelJobError, ChangeModelNameError, CreateFinetuneError,
DeleteJobError, ExportJobError, InvalidExtraParamsError, JobStatusError,
ModelNameExistsError, NotFoundJobError,
TrainDataNoneError, UnExportJobError, GetModelError)
from bisheng.common.errcode.model_deploy import NotFoundModelError
from bisheng.common.errcode.server import NoSftServerError
from bisheng.common.schemas.api import UnifiedResponseModel
from bisheng.core.cache import InMemoryCache
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync, get_minio_storage
from ..models.finetune import (Finetune, FinetuneChangeModelName, FinetuneDao,
FinetuneExtraParams, FinetuneList, FinetuneStatus)
from ..models.model_deploy import ModelDeploy, ModelDeployDao, ModelDeployInfo
from ..models.server import Server, ServerDao
from ..models.sft_model import SftModelDao
from ..sft_backend import SFTBackend
from ...utils import parse_gpus, parse_server_host
class FinetuneService:
ServerCache: InMemoryCache = InMemoryCache()
@classmethod
def validate_params(cls, finetune: Finetune) -> UnifiedResponseModel | None:
""" Check request parameters, returningNoneIndicates check passed """
# Individual Training Sets and Preset Training Sets Use at least one
if not finetune.train_data and not finetune.preset_data:
raise TrainDataNoneError()
try:
# Check Extra Parameter Values
FinetuneExtraParams(**finetune.extra_params.copy())
except ValidationError as e:
logger.error(f'Finetune extra_params is invalid {e}')
raise InvalidExtraParamsError()
return None
@classmethod
def parse_command_params(cls, finetune: Finetune, base_model: ModelDeploy) -> Dict:
""" Parse the request parameters and combine them into a training instructioncommandParameters """
params = finetune.extra_params.copy()
# Needs to be inSFT-backendThe service willmodel_nameAbsolute path to the model
params['model_name_or_path'] = base_model.model
params['model_template'] = finetune.root_model_name
params['finetuning_type'] = finetune.method
# Special treatedcpu_loadformat, because the transmission method is different --cpu_load means isTrue, no additional parameter values are required
if params.get('cpu_load') == 'false':
del params['cpu_load']
elif params.get('cpu_load') == 'true':
params['cpu_load'] = ''
# Stitching training set parameters
params['dataset'] = []
params['each_max_samples'] = []
cls.parse_command_train_file(finetune.train_data, params)
cls.parse_command_train_file(finetune.preset_data, params)
params['dataset'] = ','.join(params['dataset'])
params['each_max_samples'] = ','.join(params['each_max_samples'])
return params
@classmethod
def parse_command_train_file(cls, train_data: List[Dict], params: Dict):
""" DapatkanminioA public link to the file onSFT-BackendDownload training files """
if train_data is None:
return
minio_client = get_minio_storage_sync()
for i in train_data:
params['dataset'].append(minio_client.get_share_link_sync(i['url'], clear_host=False))
params['each_max_samples'].append(str(i.get('num', 0)))
@classmethod
def validate_status(cls, finetune: Finetune, new_status: int) -> None:
""" Verify that the state change is logical ReturnNoneIndicates success"""
# Training can only To Training Success, Training Failure, Task Abort
if finetune.status == FinetuneStatus.TRAINING.value:
if new_status not in [FinetuneStatus.SUCCESS.value, FinetuneStatus.FAILED.value,
FinetuneStatus.CANCEL.value]:
raise JobStatusError.http_exception(msg='In training, it can only be changed to training success, training failure, task abort')
# Training failed can only Becoming in training
elif finetune.status == FinetuneStatus.FAILED.value:
if new_status != FinetuneStatus.TRAINING.value:
raise JobStatusError.http_exception(msg='A training failure can only turn into a training')
# Task Aborted can only Becoming in training
elif finetune.status == FinetuneStatus.CANCEL.value:
if new_status != FinetuneStatus.TRAINING.value:
raise JobStatusError.http_exception(msg='Task abort can only become in training')
# Training Successful can only Change to Publish Complete
elif finetune.status == FinetuneStatus.SUCCESS.value:
if new_status != FinetuneStatus.PUBLISHED.value:
raise JobStatusError.http_exception(msg='Training success can only change to release completion')
# Publication complete can only Become Training Successful
elif finetune.status == FinetuneStatus.PUBLISHED.value:
if new_status != FinetuneStatus.SUCCESS.value:
raise JobStatusError.http_exception(msg='Release completion can only be converted into training success')
return None
@classmethod
async def get_sft_server(cls, server_id: int) -> Server | None:
server = await cls.get_server_by_cache(server_id)
if not server:
logger.warning('not found rt server data by id: %s', server_id)
return None
if not server.sft_endpoint:
logger.warning('not found sft endpoint by id: %s', server_id)
return None
return server
@classmethod
async def verify_job_model_name(cls, model_name: str) -> bool:
""" Correctionmodel_name Already exists? """
if await ModelDeployDao.find_model_by_name(model_name):
raise ModelNameExistsError()
if await FinetuneDao.find_job_by_model_name(model_name):
raise ModelNameExistsError()
return True
@classmethod
async def create_job(cls, finetune: Finetune) -> Finetune:
# Verify Extra Parameters
_ = cls.validate_params(finetune)
# CariSFTWhether the service exists
server = await cls.get_sft_server(finetune.server)
if not server:
raise NoSftServerError()
# Verify that the model name already exists
await cls.verify_job_model_name(finetune.model_name)
# Find out if the underlying model exists
base_model = await ModelDeployDao.find_model(finetune.base_model)
if not base_model:
raise NotFoundModelError()
root_model_name = base_model.model
# Can find instructions for retraining based on trained and completed models
if base_job := await FinetuneDao.find_job_by_model_name(base_model.model):
root_model_name = base_job.root_model_name
finetune.server_name = server.server
finetune.rt_endpoint = server.endpoint
finetune.sft_endpoint = server.sft_endpoint
finetune.base_model_name = base_model.model
finetune.root_model_name = root_model_name
# RecallSFT-backendright of privacyAPIAdd New
logger.info(f'start create sft job: {finetune.id}')
# Required for stitching instructionscommandParameters
command_params = cls.parse_command_params(finetune, base_model)
sft_ret = await SFTBackend.create_job(host=parse_server_host(finetune.sft_endpoint),
job_id=finetune.id, params=command_params)
if not sft_ret[0]:
logger.error(f'create sft job error: job_id: {finetune.id}, err: {sft_ret[1]}')
raise CreateFinetuneError()
# Insert into Database
finetune = await FinetuneDao.insert_one(finetune)
logger.info('create sft job success')
return finetune
@classmethod
async def cancel_job(cls, job_id: str, user: UserPayload) -> Finetune:
# ViewjobTask Information
finetune = await FinetuneDao.find_job(job_id)
if not finetune:
raise NotFoundJobError()
# Verify task status change
new_status = FinetuneStatus.CANCEL.value
cls.validate_status(finetune, new_status)
# RecallSFT-backendright of privacyAPICancel Task
logger.info(f'start cancel job_id: {job_id}, user: {user.user_name}')
sft_ret = await SFTBackend.cancel_job(host=parse_server_host(finetune.sft_endpoint), job_id=job_id)
if not sft_ret[0]:
logger.error(f'cancel sft job error: job_id: {job_id}, err: {sft_ret[1]}')
raise CancelJobError()
logger.info('change sft job status')
await FinetuneDao.change_status(job_id, finetune.status, new_status)
finetune.status = new_status
logger.info('cancel sft job success')
return finetune
@classmethod
async def delete_job(cls, job_id: str, user: UserPayload) -> Finetune:
# ViewjobTask Information
finetune = await FinetuneDao.find_job(job_id)
if not finetune:
raise NotFoundJobError()
model_name = await cls.delete_published_model(finetune)
# Call the interface to delete the training task
logger.info(f'start delete sft job: {job_id}, user: {user.user_name}')
sft_ret = await SFTBackend.delete_job(host=parse_server_host(finetune.sft_endpoint), job_id=job_id,
model_name=model_name)
if not sft_ret[0]:
logger.error(f'delete sft job error: job_id: {job_id}, err: {sft_ret[1]}')
raise DeleteJobError()
# Delete training task data
logger.info('delete sft job data')
# CleanedminioLog file on
await FinetuneDao.delete_job(finetune)
logger.info(f'delete sft job success, data: {finetune.model_dump()}')
asyncio.create_task(cls.delete_job_log(finetune))
return finetune
@classmethod
async def delete_job_log(cls, finetune: Finetune):
minio_client = await get_minio_storage()
await minio_client.remove_object(object_name=f'/finetune/log/{finetune.id}')
@classmethod
async def upload_job_log(cls, finetune: Finetune, log_data: io.BytesIO, length: int) -> str:
minio_client = await get_minio_storage()
log_path = f'finetune/log/{finetune.id}'
await minio_client.put_object(bucket_name=minio_client.bucket, object_name=log_path, file=log_data,
length=length)
return log_path
@classmethod
async def get_job_log(cls, finetune: Finetune) -> str | None:
minio_client = await get_minio_storage()
resp = await minio_client.get_object(object_name=finetune.log_path)
if resp is None:
return None
return resp.decode('utf-8')
@classmethod
async def delete_published_model(cls, finetune: Finetune) -> str | None:
"""
Offline the published model, delete the published model data, and return the published model name
param finetune: Training Tasks
"""
# Determine training task status
if finetune.status != FinetuneStatus.PUBLISHED.value:
return finetune.model_name
# View published modelsID
published_model = await ModelDeployDao.find_model(finetune.model_id)
if not published_model:
return finetune.model_name
# Delete Published Model Data
await ModelDeployDao.delete_model(published_model)
logger.info(f'delete published model: {published_model.model}, id: {published_model.id}')
return published_model.model
@classmethod
async def publish_job(cls, job_id: str, user: UserPayload) -> Finetune:
# ViewjobTask Information
finetune = await FinetuneDao.find_job(job_id)
if not finetune:
raise NotFoundJobError()
new_status = FinetuneStatus.PUBLISHED.value
cls.validate_status(finetune, new_status)
# RecallSFT-backendright of privacyAPIInterfaces
logger.info(f'start export sft job: {job_id}, user: {user.user_name}')
sft_ret = await SFTBackend.publish_job(host=parse_server_host(finetune.sft_endpoint), job_id=job_id,
model_name=finetune.model_name)
if not sft_ret[0]:
logger.error(f'export sft job error: job_id: {job_id}, err: {sft_ret[1]}')
raise ExportJobError()
# Create Published Model Data
logger.info('create published model')
published_model = ModelDeploy(model=finetune.model_name,
server=str(finetune.server),
endpoint=f'http://{finetune.rt_endpoint}/v2.1/models')
try:
published_model = await ModelDeployDao.insert_one(published_model)
except Exception as e:
logger.error(f'create published model error: {str(e)}')
published_model = await ModelDeployDao.find_model_by_server_and_name(published_model.server,
published_model.model)
# Record the name of the model that can be used for training
await SftModelDao.insert_sft_model(published_model.model)
# Update training task status
logger.info('update sft job data')
finetune.status = new_status
finetune.model_id = published_model.id
await FinetuneDao.update_job(finetune)
logger.info('export sft job success')
return finetune
@classmethod
async def cancel_publish_job(cls, job_id: str, user: UserPayload) -> Finetune:
# ViewjobTask Information
finetune = await FinetuneDao.find_job(job_id)
if not finetune:
raise NotFoundJobError()
new_status = FinetuneStatus.SUCCESS.value
cls.validate_status(finetune, new_status)
await cls.delete_published_model(finetune)
# RecallSFT-backendright of privacyAPIInterfaces
logger.info(f'start cancel export sft job: {job_id}, user: {user.user_name}')
sft_ret = await SFTBackend.cancel_publish_job(host=parse_server_host(finetune.sft_endpoint), job_id=job_id,
model_name=finetune.model_name)
if not sft_ret[0]:
logger.error(f'cancel export sft job error: job_id: {job_id}, err: {sft_ret[1]}')
raise UnExportJobError()
await SftModelDao.delete_sft_model(finetune.model_name)
# Delete published model information
logger.info(f'delete published model: {finetune.model_id}')
await ModelDeployDao.delete_model_by_id(finetune.model_id)
logger.info('update finetune status')
finetune.status = new_status
finetune.model_id = 0
await FinetuneDao.update_job(finetune)
logger.info('cancel export sft job success')
return finetune
@classmethod
async def get_server_by_cache(cls, server_id: int):
# Fetch from memory first
cache_server = cls.ServerCache.get(server_id)
if cache_server:
return cls.ServerCache.get(server_id)
# Retrieve from database
server = await ServerDao.find_server(server_id)
if server:
cls.ServerCache.set(server_id, server)
return server
@classmethod
async def get_all_job(cls, req_data: FinetuneList) -> (List[Finetune], int):
job_list, total = await FinetuneDao.find_jobs(req_data)
# Asynchronous thread update task status
asyncio.create_task(cls.sync_all_job_status(job_list))
return job_list, total
@classmethod
async def sync_all_job_status(cls, job_list: List[Finetune]) -> None:
# Asynchronous threads update the status of bulk tasks
for finetune in job_list:
await cls.sync_job_status(finetune, finetune.sft_endpoint)
@classmethod
async def get_job_info(cls, job_id: str) -> Dict[str, Any]:
""" Get real-time information on tasks in training """
# ViewjobTask Information
finetune = await FinetuneDao.find_job(job_id)
if not finetune:
raise NotFoundJobError()
# Sync Task Execution
await cls.sync_job_status(finetune, finetune.sft_endpoint)
# Get log file
log_data = None
res_data = list()
if finetune.log_path:
log_data = await cls.get_job_log(finetune)
res_data = cls.parse_log_data(log_data)
return {
'finetune': finetune,
'log': log_data if finetune.status != FinetuneStatus.FAILED.value else finetune.reason,
'loss_data': res_data, # like [{"step": 10, "loss": 0.5}, {"step": 20, "loss": 0.3}]
'report': finetune.report if finetune.report else None,
}
@classmethod
def parse_log_data(cls, log_data: str) -> List[Dict[str, str]]:
if log_data is None:
return []
res_data = []
contents = log_data.split('\n')
for elem in contents:
if elem.strip() == '':
continue
sub_data = {'step': None, 'loss': None}
elem = elem.strip()
elem_data = json.loads(elem)
if elem_data.get('loss', None) is None:
continue
sub_data['step'] = elem_data['current_steps']
sub_data['loss'] = elem_data['loss']
res_data.append(sub_data)
return res_data
@classmethod
async def sync_job_status(cls, finetune: Finetune, sft_endpoint: str) -> bool:
""" FROMSFT-backendService Synchronization Task Status """
if finetune.status != FinetuneStatus.TRAINING.value:
return True
logger.info(f'start sync job status: {finetune.id}')
sft_ret = await SFTBackend.get_job_status(host=parse_server_host(sft_endpoint), job_id=finetune.id)
if not sft_ret[0]:
logger.error(f'get sft job status error: job_id: {finetune.id}, err: {sft_ret[1]}')
return False
if sft_ret[1]['status'] == SFTBackend.JOB_FINISHED:
finetune.status = FinetuneStatus.SUCCESS.value
await FinetuneDao.change_status(finetune.id, finetune.status, FinetuneStatus.SUCCESS.value)
elif sft_ret[1]['status'] == SFTBackend.JOB_FAILED:
finetune.status = FinetuneStatus.FAILED.value
finetune.reason = sft_ret[1]['reason']
await FinetuneDao.update_job(finetune)
# Execution failed without querying logs and reports
if finetune.status == FinetuneStatus.FAILED.value:
logger.info('sft job status failed, no need exec log and report')
return False
# Query task execution logs and reports
logger.info('start query sft job log and report')
sft_ret = await SFTBackend.get_job_log(host=parse_server_host(sft_endpoint), job_id=finetune.id)
if not sft_ret[0]:
logger.error(f'get sft job log error: job_id: {finetune.id}, err: {sft_ret[1]}')
log_data = sft_ret[1]['log_data'].encode('utf-8')
# Upload log file tominioUpward
log_path = await cls.upload_job_log(finetune, io.BytesIO(log_data), len(log_data))
finetune.log_path = log_path
# Query Task Evaluation Report
logger.info('start query sft job report')
sft_ret = await SFTBackend.get_job_metrics(host=parse_server_host(sft_endpoint), job_id=finetune.id)
if not sft_ret[0]:
logger.error(f'get sft job report error: job_id: {finetune.id}, err: {sft_ret[1]}')
else:
finetune.report = sft_ret[1]['report']
# Changelog and Report Data
await FinetuneDao.update_job(finetune)
return True
@classmethod
async def change_job_model_name(cls, req: FinetuneChangeModelName) -> Finetune:
""" Modify the model name of the training task """
finetune = await FinetuneDao.find_job(req.id)
if not finetune:
raise NotFoundJobError()
# Verify that the model name already exists
await cls.verify_job_model_name(req.model_name)
# Modify Published Model Name
if not await cls.change_published_model_name(finetune, req.model_name):
raise ChangeModelNameError()
# Update training task'smodel_name
finetune.model_name = req.model_name
await FinetuneDao.update_job(finetune)
return finetune
@classmethod
async def change_published_model_name(cls, finetune: Finetune, model_name: str) -> bool:
""" Modify the model name of the training task """
# Unpublished training tasks do not require actionablemodel
if finetune.status != FinetuneStatus.PUBLISHED.value:
return True
published_model = await ModelDeployDao.find_model(finetune.model_id)
if not published_model:
logger.error(f'published model not found, job_id: {finetune.id}, model_id: {finetune.model_id}')
return False
# Call the interface to modify the name of the published model
sft_ret = await SFTBackend.change_model_name(parse_server_host(finetune.sft_endpoint), finetune.id,
published_model.model, model_name)
if not sft_ret[0]:
logger.error(f'change model name error: job_id: {finetune.id}, err: {sft_ret[1]}')
return False
# Modify pre-trained model name
await SftModelDao.change_sft_model(published_model.model, model_name)
# Updating the published model'smodel_name
published_model.model = model_name
await ModelDeployDao.update_model(published_model)
return True
@classmethod
async def get_server_filters(cls) -> List[Dict[str, Any]]:
""" DapatkanftServer Filter Criteria """
server_filters = await FinetuneDao.get_server_filters()
res = []
for one in server_filters:
res.append({
'id': one,
'server_name': one,
})
return res
@classmethod
async def get_model_list(cls, server_id: int) -> List[ModelDeploy]:
""" DapatkanftList of all models under the service """
server_info = await ServerDao.find_server(server_id)
if not server_info:
raise NoSftServerError()
flag, model_name_list = await SFTBackend.get_all_model(parse_server_host(server_info.sft_endpoint))
if not flag:
logger.error(f'get model list error: server_id: {server_id}, err: {model_name_list}')
raise GetModelError()
ret = []
db_model = await ModelDeployDao.find_model_by_server(str(server_id))
for one in db_model:
if one.model in model_name_list:
ret.append(one)
model_name_list.remove(one.model)
for one in model_name_list:
ret.append(await ModelDeployDao.insert_one(ModelDeploy(server=str(server_id),
model=one,
endpoint=f'http://{server_info.endpoint}/v2.1/models')))
res = []
for one in ret:
res.append(ModelDeployInfo(**one.dict(), sft_support=True))
return res
@classmethod
async def get_gpu_info(cls) -> List[Dict]:
""" DapatkanGPUMessage """
all_server = await ServerDao.find_all_server()
res = []
for server in all_server:
if not server.sft_endpoint:
continue
sft_ret = await SFTBackend.get_gpu_info(parse_server_host(server.sft_endpoint))
if not sft_ret[0]:
logger.error(f'get gpu info error: server_id: {server.id}, err: {sft_ret[1]}')
continue
gpu_info = parse_gpus(sft_ret[1])
for one in gpu_info:
one['server'] = server.server
res.append(one)
return res
================================================
FILE: src/backend/bisheng/finetune/domain/services/finetune_file.py
================================================
import os.path
from typing import List
from fastapi import UploadFile
from loguru import logger
from pydantic import BaseModel
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.finetune import TrainFileNotExistError
from bisheng.common.schemas.api import PageList
from bisheng.core.storage.minio.minio_manager import get_minio_storage
from bisheng.utils import generate_uuid
from ..models.preset_train import PresetTrain, PresetTrainDao
class FinetuneFileService(BaseModel):
""" Training Tasks File management """
@classmethod
async def upload_file(cls, files: List[UploadFile], is_preset: bool,
user: UserPayload) -> List[PresetTrain]:
if len(files) == 0:
raise TrainFileNotExistError()
# Upload training files tominio
file_root = cls.get_upload_file_root(is_preset)
file_list = await cls.upload_file_to_minio(files, file_root, user)
# Store preset data in database
if is_preset:
logger.info(f'save preset file : {file_list}')
file_list = await PresetTrainDao.insert_batch(file_list)
return file_list
@classmethod
async def upload_preset_file(cls, name: str, preset_type: int, file_path: str,
user: UserPayload) -> PresetTrain:
# Upload training files tominio
file_root = cls.get_upload_file_root(False)
file_id = generate_uuid()
file_ext = os.path.basename(file_path).split('.')[-1]
object_name = f'{file_root}/{file_id}.{file_ext}'
minio_client = await get_minio_storage()
await minio_client.put_object(bucket_name=minio_client.bucket, object_name=object_name, file=file_path)
# Store preset data in database
file_info = PresetTrain(id=file_id,
name=name,
url=object_name,
type=preset_type,
user_id=user.user_id,
user_name=user.user_name)
await PresetTrainDao.insert_batch([file_info])
return file_info
@classmethod
def get_upload_file_root(cls, is_preset: bool) -> str:
if is_preset:
return 'finetune/train_file/preset'
else:
return 'finetune/train_file/personal'
@classmethod
async def upload_file_to_minio(cls, files: List[UploadFile], file_root: str,
user: UserPayload) -> List[PresetTrain]:
minio_client = await get_minio_storage()
ret = []
for file in files:
file_id = generate_uuid()
file_ext = os.path.basename(file.filename).split('.')[-1]
file_info = PresetTrain(id=file_id,
name=file.filename,
url=f'{file_root}/{file_id}.{file_ext}',
user_id=user.user_id,
user_name=user.user_name)
await minio_client.put_object(bucket_name=minio_client.bucket, object_name=file_info.url,
file=file.file, content_type=file.content_type, length=file.size)
ret.append(file_info)
return ret
@classmethod
async def get_preset_file(cls,
keyword: str = None,
page_size: int = None,
page_num: int = None) -> PageList:
list_res, total_count = await PresetTrainDao.search_name(keyword, page_size, page_num)
return PageList(list=list_res, total=total_count)
@classmethod
async def delete_preset_file(cls, file_id: str, user: UserPayload) -> None:
file_data = await PresetTrainDao.find_one(file_id)
if not file_data:
raise TrainFileNotExistError()
logger.info(f'delete preset train file, user: {user}; file: {file_data.model_dump()}')
await PresetTrainDao.delete_one(file_data)
logger.info('delete preset train file success')
return None
================================================
FILE: src/backend/bisheng/finetune/domain/sft_backend.py
================================================
import json
from typing import Dict, List, Literal
from bisheng.core.external.http_client.http_client_manager import get_http_client
class SFTBackend:
""" Packaging andSFT-BackendInteraction """
# of fine-tuning training instructionsoptionsParameters Listing
CMD_OPTIONS = ['train']
# jobStatus Misi
JOB_FINISHED = 'FINISHED'
JOB_FAILED = 'FAILED'
@classmethod
async def handle_response(cls, res) -> (bool, str | None | Dict):
content = await res.content.read()
if res.status != 200:
return False, content.decode('utf-8')
result = json.loads(content)
if result.get('status_code') != 200:
return False, content.decode('utf-8')
return True, result.get('data', None)
@classmethod
async def _base_request(cls, method: Literal['get', 'post'], *args, **kwargs) -> (bool, str | Dict):
client = await get_http_client()
client = await client.get_aiohttp_client()
if method == 'get':
res = await client.get(*args, **kwargs)
else:
res = await client.post(*args, **kwargs)
return await cls.handle_response(res)
@classmethod
async def create_job(cls, host: str, job_id: str, params: Dict) -> (bool, str | Dict):
"""
host RTof servicehostAddress:
job_id is unique to the instructionid，UUIDFormat
options Is InstructionoptionsParameters
params Is InstructionalcommandParameter Parameter
"""
uri = '/v2.1/sft/job'
url = '/v2.1/models/sft_elem/infer'
return await cls._base_request('post',
f'{host}{url}',
json={'uri': uri,
'job_id': job_id,
'options': cls.CMD_OPTIONS,
'params': params})
@classmethod
async def cancel_job(cls, host: str, job_id: str) -> (bool, str | Dict):
""" Cancel training task """
uri = '/v2.1/sft/job/cancel'
url = '/v2.1/models/sft_elem/infer'
return await cls._base_request('post', f'{host}{url}', json={'uri': uri, 'job_id': job_id})
@classmethod
async def delete_job(cls, host: str, job_id: str, model_name: str) -> (bool, str | Dict):
""" Delete training task """
uri = '/v2.1/sft/job/delete'
url = '/v2.1/models/sft_elem/infer'
return await cls._base_request('post', f'{host}{url}',
json={'uri': uri, 'job_id': job_id, 'model_name': model_name})
@classmethod
async def publish_job(cls, host: str, job_id: str, model_name: str) -> (bool, str | Dict):
""" Publish Training Tasks From training path to formal path"""
uri = '/v2.1/sft/job/publish'
url = '/v2.1/models/sft_elem/infer'
return await cls._base_request('post', f'{host}{url}',
json={'uri': uri, 'job_id': job_id, 'model_name': model_name})
@classmethod
async def cancel_publish_job(cls, host: str, job_id: str, model_name: str) -> (bool, str | Dict):
""" Deactivate Training Task Published Model """
uri = '/v2.1/sft/job/publish/cancel'
url = '/v2.1/models/sft_elem/infer'
return await cls._base_request('post', f'{host}{url}',
json={'uri': uri, 'job_id': job_id, 'model_name': model_name})
@classmethod
async def get_job_status(cls, host: str, job_id: str) -> (bool, str | Dict):
"""
Get training task status
Interface return format:
{
"status": "FINISHED",
"reason": "Failure Reason"
}
"""
uri = '/v2.1/sft/job/status'
url = '/v2.1/models/sft_elem/infer'
return await cls._base_request('post', f'{host}{url}', json={'uri': uri, 'job_id': job_id})
@classmethod
async def get_job_log(cls, host: str, job_id: str) -> (bool, str | Dict):
"""
Get the training task log and use it temporarilydictFormat - Returns the contents of the
TODO zgq Subsequent AdoptionhttpStandard File Transfer Format
Data format returned by the interface
{
"log_data": Ref.bisheng-ftProduction Training Log File Contents
}
"""
uri = '/v2.1/sft/job/log'
url = '/v2.1/models/sft_elem/infer'
return await cls._base_request('post', f'{host}{url}', json={'uri': uri, 'job_id': job_id})
@classmethod
async def get_job_metrics(cls, host: str, job_id: str) -> (bool, str | Dict):
"""
Get training mission final report
Interface return data format
{
"report": {}
}
"""
uri = '/v2.1/sft/job/metrics'
url = '/v2.1/models/sft_elem/infer'
return await cls._base_request('post', f'{host}{url}', json={'uri': uri, 'job_id': job_id})
@classmethod
async def change_model_name(cls, host, job_id: str, old_model_name: str, model_name: str) -> (bool, str):
""" Modify Model Name """
uri = '/v2.1/sft/job/model_name'
url = '/v2.1/models/sft_elem/infer'
return await cls._base_request('post', f'{host}{url}',
json={'uri': uri, 'job_id': job_id, 'old_model_name': old_model_name,
'model_name': model_name})
@classmethod
async def get_all_model(cls, host) -> (bool, List[str]):
""" Get a list of all models """
url = '/v2.1/sft/model'
return await cls._base_request('get', f'{host}{url}')
@classmethod
async def get_gpu_info(cls, host) -> (bool, str):
""" DapatkanGPUMessage """
url = '/v2.1/sft/gpu'
return await cls._base_request('get', f'{host}{url}')
================================================
FILE: src/backend/bisheng/finetune/schemas.py
================================================
from typing import Dict, Optional, List
from pydantic import BaseModel, Field
from bisheng.finetune.domain.models.finetune import TrainMethod
class FinetuneCreateReq(BaseModel):
server: int = Field(description='RelatedRTSERVICESID')
base_model: int = Field(description='Foundation ModelID')
model_name: str = Field(max_length=50, description='Model Name')
method: TrainMethod = Field(description='Training Methods')
extra_params: Dict = Field(default_factory=dict, description='Additional parameters required for training tasks')
train_data: Optional[List[Dict]] = Field(default=None, description='Personal training data')
preset_data: Optional[List[Dict]] = Field(default=None, description='Preset training data')
================================================
FILE: src/backend/bisheng/finetune/utils.py
================================================
import xml.dom.minidom
from typing import List, Dict
# will be nvidia-smi -q -x 's output is parsed as visual data
def parse_gpus(gpu_str: str) -> List[Dict]:
dom_tree = xml.dom.minidom.parseString(gpu_str)
collections = dom_tree.documentElement
gpus = collections.getElementsByTagName('gpu')
res = []
for one in gpus:
fb_mem_elem = one.getElementsByTagName('fb_memory_usage')[0]
gpu_uuid_elem = one.getElementsByTagName('uuid')[0]
gpu_id_elem = one.getElementsByTagName('minor_number')[0]
gpu_total_mem = fb_mem_elem.getElementsByTagName('total')[0]
free_mem = fb_mem_elem.getElementsByTagName('free')[0]
gpu_utility_elem = one.getElementsByTagName('utilization')[0].getElementsByTagName(
'gpu_util')[0]
res.append({
'gpu_uuid':
gpu_uuid_elem.firstChild.data,
'gpu_id':
gpu_id_elem.firstChild.data,
'gpu_total_mem':
'%.2f G' % (float(gpu_total_mem.firstChild.data.split(' ')[0]) / 1024),
'gpu_used_mem':
'%.2f G' % (float(free_mem.firstChild.data.split(' ')[0]) / 1024),
'gpu_utility':
round(float(gpu_utility_elem.firstChild.data.split(' ')[0]) / 100, 2)
})
return res
def parse_server_host(endpoint: str):
""" Put the data in the databaseendpointsResolve tohttpRequestedhost """
endpoint = endpoint.replace('http://', '').split('/')[0]
return f'http://{endpoint}'
================================================
FILE: src/backend/bisheng/graph/__init__.py
================================================
from bisheng.graph.edge.base import Edge
from bisheng.graph.graph.base import Graph
from bisheng.graph.vertex.base import Vertex
from bisheng.graph.vertex.types import (AgentVertex, ChainVertex,
DocumentLoaderVertex, EmbeddingVertex,
LLMVertex, MemoryVertex, PromptVertex,
RetrieverVertex, TextSplitterVertex,
ToolkitVertex, ToolVertex,
VectorStoreVertex, WrapperVertex)
__all__ = [
'Graph',
'Vertex',
'Edge',
'AgentVertex',
'ChainVertex',
'DocumentLoaderVertex',
'EmbeddingVertex',
'LLMVertex',
'MemoryVertex',
'PromptVertex',
'TextSplitterVertex',
'ToolVertex',
'ToolkitVertex',
'VectorStoreVertex',
'WrapperVertex',
'RetrieverVertex',
]
================================================
FILE: src/backend/bisheng/graph/edge/__init__.py
================================================
================================================
FILE: src/backend/bisheng/graph/edge/base.py
================================================
from typing import TYPE_CHECKING, List, Optional
from loguru import logger
from pydantic import BaseModel, Field
if TYPE_CHECKING:
from bisheng.graph.vertex.base import Vertex
class SourceHandle(BaseModel):
baseClasses: List[str] = Field(..., description='List of base classes for the source handle.')
dataType: str = Field(..., description='Data type for the source handle.')
id: str = Field(..., description='Unique identifier for the source handle.')
class TargetHandle(BaseModel):
fieldName: str = Field(..., description='Field name for the target handle.')
id: str = Field(..., description='Unique identifier for the target handle.')
inputTypes: Optional[List[str]] = Field(
None, description='List of input types for the target handle.')
type: str = Field(..., description='Type of the target handle.')
class Edge:
def __init__(self, source: 'Vertex', target: 'Vertex', edge: dict):
self.source_id: str = source.id if source else ''
self.target_id: str = target.id if target else ''
if data := edge.get('data', {}):
self._source_handle = data.get('sourceHandle', {})
self._target_handle = data.get('targetHandle', {})
self.source_handle: SourceHandle = SourceHandle(**self._source_handle)
self.target_handle: TargetHandle = TargetHandle(**self._target_handle)
self.target_param = self.target_handle.fieldName
# validate handles
self.validate_handles(source, target)
else:
# Logging here because this is a breaking change
# logger.error('Edge data is empty')
self._source_handle = edge.get('sourceHandle', '')
self._target_handle = edge.get('targetHandle', '')
# 'BaseLoader;BaseOutputParser|documents|PromptTemplate-zmTlD'
# target_param is documents Compatible processing nestinggroup
self.target_param = self._target_handle.split('|')[1].split(f'_{self.target_id}')[0]
# 'BaseLoader;BaseOutputParser|documents|PromptTemplate-zmTlD'
# target_param is documents
self.validate_edge(source, target)
def validate_handles(self, source, target) -> None:
if self.target_handle.inputTypes is None:
self.valid_handles = self.target_handle.type in self.source_handle.baseClasses
else:
self.valid_handles = (any(baseClass in self.target_handle.inputTypes
for baseClass in self.source_handle.baseClasses)
or self.target_handle.type in self.source_handle.baseClasses)
if not self.valid_handles:
logger.debug(self.source_handle)
logger.debug(self.target_handle)
raise ValueError(f'Edge between {source.vertex_type} and {target.vertex_type} '
f'has invalid handles')
def __setstate__(self, state):
self.source_id = state['source_id']
self.target_id = state['target_id']
self.target_param = state['target_param']
self.source_handle = state.get('source_handle')
self.target_handle = state.get('target_handle')
def validate_edge(self, source, target) -> None:
# Validate that the outputs of the source node are valid inputs
# for the target node
self.source_types = source.output
self.target_reqs = target.required_inputs + target.optional_inputs
# Both lists contain strings and sometimes a string contains the value we are
# looking for e.g. comgin_out=["Chain"] and target_reqs=["LLMChain"]
# so we need to check if any of the strings in source_types is in target_reqs
self.valid = any(output in target_req for output in self.source_types
for target_req in self.target_reqs)
# Get what type of input the target node is expecting
self.matched_type = next(
(output for output in self.source_types if output in self.target_reqs),
None,
)
no_matched_type = self.matched_type is None
if no_matched_type:
logger.debug(self.source_types)
logger.debug(self.target_reqs)
if no_matched_type:
raise ValueError(f'Edge between {source.vertex_type} and {target.vertex_type} '
f'has no matched type')
def __repr__(self) -> str:
return (
f'Edge(source={self.source_id}, target={self.target_id}, target_param={self.target_param}'
f', matched_type={self.matched_type})')
def __hash__(self) -> int:
return hash(self.__repr__())
def __eq__(self, __value: object) -> bool:
return (self.__repr__() == __value.__repr__() if isinstance(__value, Edge) else False)
================================================
FILE: src/backend/bisheng/graph/graph/__init__.py
================================================
================================================
FILE: src/backend/bisheng/graph/graph/base.py
================================================
from typing import Dict, Generator, List, Type, Union
from bisheng.graph.edge.base import Edge
from bisheng.graph.graph.constants import lazy_load_vertex_dict
from bisheng.graph.utils import process_flow
from bisheng.graph.vertex.base import Vertex
from bisheng.graph.vertex.types import FileToolVertex, LLMVertex, ToolkitVertex
from bisheng.interface.tools.constants import FILE_TOOLS
from bisheng.utils import payload
from langchain.chains.base import Chain
from loguru import logger
class Graph:
"""A class representing a graph of nodes and edges."""
def __init__(
self,
nodes: List[Dict[str, Union[str, Dict[str, Union[str, List[str]]]]]],
edges: List[Dict[str, str]],
) -> None:
self._vertices = nodes
self._nodes = nodes
self._edges = edges
self.raw_graph_data = {'nodes': nodes, 'edges': edges}
self.top_level_vertices = []
for vertex in self._vertices:
if vertex_id := vertex.get('id'):
self.top_level_vertices.append(vertex_id)
self._graph_data = process_flow(self.raw_graph_data)
self._vertices = self._graph_data['nodes']
self._edges = self._graph_data['edges']
self._build_graph()
def __getstate__(self):
return self.raw_graph_data
def __setstate__(self, state):
self.__init__(**state)
def __eq__(self, other: object) -> bool:
if not isinstance(other, Graph):
return False
return self.__repr__() == other.__repr__()
@classmethod
def from_payload(cls, payload: Dict) -> 'Graph':
"""
Creates a graph from a payload.
Args:
payload (Dict): The payload to create the graph from.˜`
Returns:
Graph: The created graph.
"""
if 'data' in payload:
payload = payload['data']
try:
nodes = payload['nodes']
if len(nodes) == 0:
raise Exception("At least one node is required")
edges = payload['edges']
return cls(nodes, edges)
except KeyError as exc:
import traceback
traceback.print_exc()
raise ValueError(
f"Invalid payload. Expected keys 'nodes' and 'edges'. Found {list(payload.keys())}"
) from exc
def _build_graph(self) -> None:
"""Builds the graph from the nodes and edges."""
self.vertices = self._build_vertices()
self.vertex_map = {vertex.id: vertex for vertex in self.vertices}
self.edges = self._build_edges()
# This is a hack to make sure that the LLM node is sent to
# the toolkit node
self._build_vertex_params()
# remove invalid vertices
self._validate_vertices()
def _build_vertex_params(self) -> None:
"""Identifies and handles the LLM vertex within the graph."""
llm_vertex = None
for vertex in self.vertices:
vertex._build_params()
if isinstance(vertex, LLMVertex):
llm_vertex = vertex
if llm_vertex:
for vertex in self.vertices:
if isinstance(vertex, ToolkitVertex):
vertex.params['llm'] = llm_vertex
def _validate_vertices(self) -> None:
"""Check that all vertices have edges"""
if len(self.vertices) == 1:
return
for vertex in self.vertices:
if not self._validate_vertex(vertex):
raise ValueError(f'{vertex.vertex_type} is not connected to any other components')
def _validate_vertex(self, vertex: Vertex) -> bool:
"""Validates a vertex."""
# All vertices that do not have edges are invalid
return len(self.get_vertex_edges(vertex.id)) > 0
def get_vertex(self, vertex_id: str) -> Union[None, Vertex]:
"""Returns a vertex by id."""
return self.vertex_map.get(vertex_id)
def get_vertex_edges(self, vertex_id: str) -> List[Edge]:
"""Returns a list of edges for a given vertex."""
return [
edge for edge in self.edges
if edge.source_id == vertex_id or edge.target_id == vertex_id
]
def get_vertices_with_target(self, vertex_id: str) -> List[Vertex]:
"""Returns the vertices connected to a vertex."""
vertices: List[Vertex] = []
for edge in self.edges:
if edge.target_id == vertex_id:
vertex = self.get_vertex(edge.source_id)
if vertex is None:
continue
vertices.append(vertex)
return vertices
def get_input_nodes(self) -> List[Vertex]:
"""Builds the graph."""
# Get root node
input_node = payload.get_root_node(self)
if input_node is None:
raise ValueError('No input root node found')
return input_node
async def abuild(self) -> Chain:
"""Builds the graph."""
# Get root node
root_vertex = payload.get_root_vertex(self)
if root_vertex is None:
raise ValueError('No root node vertex found')
return await root_vertex.build()
def topological_sort(self) -> List[Vertex]:
"""
Performs a topological sort of the vertices in the graph.
Returns:
List[Vertex]: A list of vertices in topological order.
Raises:
ValueError: If the graph contains a cycle.
"""
# States: 0 = unvisited, 1 = visiting, 2 = visited
state = {node: 0 for node in self.vertices}
sorted_vertices = []
def dfs(node):
if state[node] == 1:
# We have a cycle
raise ValueError('Graph contains a cycle, cannot perform topological sort')
if state[node] == 0:
state[node] = 1
for edge in node.edges:
if edge.source_id == node.id:
dfs(self.get_vertex(edge.target_id))
state[node] = 2
sorted_vertices.append(node)
# Visit each node
for node in self.vertices:
if state[node] == 0:
dfs(node)
return list(reversed(sorted_vertices))
def generator_build(self) -> Generator:
"""Builds each vertex in the graph and yields it."""
sorted_vertices = self.topological_sort()
logger.debug('There are %s vertices in the graph', len(sorted_vertices))
yield from sorted_vertices
def get_vertex_neighbors(self, vertex: Vertex) -> Dict[Vertex, int]:
"""Returns the neighbors of a vertex."""
neighbors: Dict[Vertex, int] = {}
for edge in self.edges:
if edge.source_id == vertex.id:
neighbor = self.get_vertex(edge.target_id)
if neighbor is None:
continue
if neighbor not in neighbors:
neighbors[neighbor] = 0
neighbors[neighbor] += 1
elif edge.target_id == vertex.id:
neighbor = self.get_vertex(edge.source_id)
if neighbor is None:
continue
if neighbor not in neighbors:
neighbors[neighbor] = 0
neighbors[neighbor] += 1
return neighbors
def _build_edges(self) -> List[Edge]:
"""Builds the edges of the graph."""
# Edge takes two nodes as arguments, so we need to build the nodes first
# and then build the edges
# if we can't find a node, we raise an error
edges: List[Edge] = []
for edge in self._edges:
source = self.get_vertex(edge['source'])
target = self.get_vertex(edge['target'])
if source is None:
raise ValueError(f"Source node {edge['source']} not found")
if target is None:
raise ValueError(f"Target node {edge['target']} not found")
edges.append(Edge(source, target, edge))
return edges
def _get_vertex_class(self, node_type: str, vertex_base_type: str) -> Type[Vertex]:
"""Returns the node class based on the node type."""
if node_type in FILE_TOOLS:
return FileToolVertex
if vertex_base_type == 'CustomComponent':
return lazy_load_vertex_dict.get_custom_component_vertex_type()
if vertex_base_type in lazy_load_vertex_dict.VERTEX_TYPE_MAP:
return lazy_load_vertex_dict.VERTEX_TYPE_MAP[vertex_base_type]
return (lazy_load_vertex_dict.VERTEX_TYPE_MAP[node_type]
if node_type in lazy_load_vertex_dict.VERTEX_TYPE_MAP else Vertex)
def _build_vertices(self) -> List[Vertex]:
"""Builds the vertices of the graph."""
nodes: List[Vertex] = []
for node in self._vertices:
node_data = node['data']
node_type: str = node_data['type'] # type: ignore
vertex_base_type: str = node_data['node']['template']['_type'] # type: ignore
VertexClass = self._get_vertex_class(node_type, vertex_base_type)
vertex_instance = VertexClass(node, graph=self)
vertex_instance.set_top_level(self.top_level_vertices)
nodes.append(vertex_instance)
return nodes
def get_children_by_node_type(self, node: Vertex, node_type: str) -> List[Vertex]:
"""Returns the children of a node based on the node type."""
children = []
node_types = [node.data['type']]
if 'node' in node.data:
node_types += node.data['node']['base_classes']
if node_type in node_types:
children.append(node)
return children
def __repr__(self):
node_ids = [node.id for node in self.vertices]
edges_repr = '\n'.join([f'{edge.source_id} --> {edge.target_id}' for edge in self.edges])
return f'Graph:\nNodes: {node_ids}\nConnections:\n{edges_repr}'
================================================
FILE: src/backend/bisheng/graph/graph/constants.py
================================================
from bisheng.graph.vertex import types
from bisheng.interface.agents.base import agent_creator
from bisheng.interface.chains.base import chain_creator
from bisheng.interface.document_loaders.base import documentloader_creator
from bisheng.interface.embeddings.base import embedding_creator
from bisheng.interface.llms.base import llm_creator
from bisheng.interface.memories.base import memory_creator
from bisheng.interface.output_parsers.base import output_parser_creator
from bisheng.interface.prompts.base import prompt_creator
from bisheng.interface.retrievers.base import retriever_creator
from bisheng.interface.text_splitters.base import textsplitter_creator
from bisheng.interface.toolkits.base import toolkits_creator
from bisheng.interface.tools.base import tool_creator
from bisheng.interface.vector_store.base import vectorstore_creator
from bisheng.interface.wrappers.base import wrapper_creator
from bisheng.utils.lazy_load import LazyLoadDictBase
class VertexTypesDict(LazyLoadDictBase):
def __init__(self):
self._all_types_dict = None
@property
def VERTEX_TYPE_MAP(self):
return self.all_types_dict
def _build_dict(self):
langchain_types_dict = self.get_type_dict()
return {
**langchain_types_dict,
'Custom': ['Custom Tool', 'Python Function'],
}
def get_custom_component_vertex_type(self):
return types.CustomComponentVertex
def get_type_dict(self):
return {
**{
t: types.PromptVertex
for t in prompt_creator.to_list()
},
**{
t: types.AgentVertex
for t in agent_creator.to_list()
},
**{
t: types.ChainVertex
for t in chain_creator.to_list()
},
**{
t: types.ToolVertex
for t in tool_creator.to_list()
},
**{
t: types.ToolkitVertex
for t in toolkits_creator.to_list()
},
**{
t: types.WrapperVertex
for t in wrapper_creator.to_list()
},
**{
t: types.LLMVertex
for t in llm_creator.to_list()
},
**{
t: types.MemoryVertex
for t in memory_creator.to_list()
},
**{
t: types.EmbeddingVertex
for t in embedding_creator.to_list()
},
**{
t: types.VectorStoreVertex
for t in vectorstore_creator.to_list()
},
**{
t: types.DocumentLoaderVertex
for t in documentloader_creator.to_list()
},
**{
t: types.TextSplitterVertex
for t in textsplitter_creator.to_list()
},
**{
t: types.OutputParserVertex
for t in output_parser_creator.to_list()
},
# **{t: types.CustomComponentVertex for t in custom_component_creator.to_list()},
**{
t: types.RetrieverVertex
for t in retriever_creator.to_list()
},
}
lazy_load_vertex_dict = VertexTypesDict()
================================================
FILE: src/backend/bisheng/graph/graph/utils.py
================================================
import copy
from collections import deque
from typing import Dict, List
def find_last_node(nodes, edges):
"""
This function receives a flow and returns the last node.
"""
return next((n for n in nodes if all(e['source'] != n['id'] for e in edges)), None)
def add_parent_node_id(nodes, parent_node_id):
"""
This function receives a list of nodes and adds a parent_node_id to each node.
"""
for node in nodes:
node['parent_node_id'] = parent_node_id
def ungroup_node(group_node_data, base_flow):
template, flow = (
group_node_data['node']['template'],
group_node_data['node']['flow'],
)
parent_node_id = group_node_data['id']
g_nodes = flow['data']['nodes']
add_parent_node_id(g_nodes, parent_node_id)
g_edges = flow['data']['edges']
# Redirect edges to the correct proxy node
updated_edges = get_updated_edges(base_flow, g_nodes, g_edges, group_node_data['id'])
# Update template values
update_template(template, g_nodes)
nodes = [n for n in base_flow['nodes'] if n['id'] != group_node_data['id']] + g_nodes
edges = ([
e for e in base_flow['edges']
if e['target'] != group_node_data['id'] and e['source'] != group_node_data['id']
] + g_edges + updated_edges)
base_flow['nodes'] = nodes
base_flow['edges'] = edges
return nodes
def raw_topological_sort(nodes, edges) -> List[Dict]:
# Redefine the above function but using the nodes and self._edges
# which are dicts instead of Vertex and Edge objects
# nodes have an id, edges have a source and target keys
# return a list of node ids in topological order
# States: 0 = unvisited, 1 = visiting, 2 = visited
state = {node['id']: 0 for node in nodes}
nodes_dict = {node['id']: node for node in nodes}
sorted_vertices = []
def dfs(node):
if state[node] == 1:
# We have a cycle
raise ValueError('Graph contains a cycle, cannot perform topological sort')
if state[node] == 0:
state[node] = 1
for edge in edges:
if edge['source'] == node:
dfs(edge['target'])
state[node] = 2
sorted_vertices.append(node)
# Visit each node
for node in nodes:
if state[node['id']] == 0:
dfs(node['id'])
reverse_sorted = list(reversed(sorted_vertices))
return [nodes_dict[node_id] for node_id in reverse_sorted]
def process_flow(flow_object):
cloned_flow = copy.deepcopy(flow_object)
processed_nodes = set() # To keep track of processed nodes
def process_node(node):
node_id = node.get('id')
# If node already processed, skip
if node_id in processed_nodes:
return
if node.get('data') and node['data'].get('node') and node['data']['node'].get('flow'):
process_flow(node['data']['node']['flow']['data'])
new_nodes = ungroup_node(node['data'], cloned_flow)
# Add new nodes to the queue for future processing
nodes_to_process.extend(new_nodes)
# Mark node as processed
processed_nodes.add(node_id)
sorted_nodes_list = raw_topological_sort(cloned_flow['nodes'], cloned_flow['edges'])
nodes_to_process = deque(sorted_nodes_list)
while nodes_to_process:
node = nodes_to_process.popleft()
process_node(node)
return cloned_flow
def update_template(template, g_nodes):
"""
Updates the template of a node in a graph with the given template.
Args:
template (dict): The new template to update the node with.
g_nodes (list): The list of nodes in the graph.
Returns:
None
"""
for _, value in template.items():
if not value.get('proxy'):
continue
proxy_dict = value['proxy']
field, id_ = proxy_dict['field'], proxy_dict['id']
node_index = next((i for i, n in enumerate(g_nodes) if n['id'] == id_), -1)
if node_index != -1:
display_name = None
show = g_nodes[node_index]['data']['node']['template'][field]['show']
advanced = g_nodes[node_index]['data']['node']['template'][field]['advanced']
if 'display_name' in g_nodes[node_index]['data']['node']['template'][field]:
display_name = g_nodes[node_index]['data']['node']['template'][field][
'display_name']
else:
display_name = g_nodes[node_index]['data']['node']['template'][field]['name']
g_nodes[node_index]['data']['node']['template'][field] = value
g_nodes[node_index]['data']['node']['template'][field]['show'] = show
g_nodes[node_index]['data']['node']['template'][field]['advanced'] = advanced
g_nodes[node_index]['data']['node']['template'][field]['display_name'] = display_name
def update_target_handle(
new_edge,
g_nodes,
group_node_id,
):
"""
Updates the target handle of a given edge if it is a proxy node.
Args:
new_edge (dict): The edge to update.
g_nodes (list): The list of nodes in the graph.
group_node_id (str): The ID of the group node.
Returns:
dict: The updated edge.
"""
target_handle = new_edge['data']['targetHandle']
if target_handle.get('proxy'):
proxy_id = target_handle['proxy']['id']
if node := next((n for n in g_nodes if n['id'] == proxy_id), None):
set_new_target_handle(proxy_id, new_edge, target_handle, node)
else:
raise ValueError(
f'Group node {group_node_id} has an invalid target proxy node {proxy_id}')
return new_edge
def set_new_target_handle(proxy_id, new_edge, target_handle, node):
"""
Sets a new target handle for a given edge.
Args:
proxy_id (str): The ID of the proxy.
new_edge (dict): The new edge to be created.
target_handle (dict): The target handle of the edge.
node (dict): The node containing the edge.
Returns:
None
"""
new_edge['target'] = proxy_id
_type = target_handle.get('type')
if _type is None:
raise KeyError("The 'type' key must be present in target_handle.")
field = target_handle['proxy']['field']
new_target_handle = {
'fieldName': field,
'type': _type,
'id': proxy_id,
}
if node['data']['node'].get('flow'):
new_target_handle['proxy'] = {
'field': node['data']['node']['template'][field]['proxy']['field'],
'id': node['data']['node']['template'][field]['proxy']['id'],
}
if input_types := target_handle.get('inputTypes'):
new_target_handle['inputTypes'] = input_types
new_edge['data']['targetHandle'] = new_target_handle
def update_source_handle(new_edge, g_nodes, g_edges):
"""
Updates the source handle of a given edge to the last node in the flow data.
Args:
new_edge (dict): The edge to update.
flow_data (dict): The flow data containing the nodes and edges.
Returns:
dict: The updated edge with the new source handle.
"""
last_node = copy.deepcopy(find_last_node(g_nodes, g_edges))
new_edge['source'] = last_node['id']
new_source_handle = new_edge['data']['sourceHandle']
new_source_handle['id'] = last_node['id']
new_edge['data']['sourceHandle'] = new_source_handle
return new_edge
def get_updated_edges(base_flow, g_nodes, g_edges, group_node_id):
"""
Given a base flow, a list of graph nodes and a group node id, returns a list of updated edges.
An updated edge is an edge that has its target or source handle updated based on the group node id.
Args:
base_flow (dict): The base flow containing a list of edges.
g_nodes (list): A list of graph nodes.
group_node_id (str): The id of the group node.
Returns:
list: A list of updated edges.
"""
updated_edges = []
for edge in base_flow['edges']:
new_edge = copy.deepcopy(edge)
if new_edge['target'] == group_node_id:
new_edge = update_target_handle(new_edge, g_nodes, group_node_id)
if new_edge['source'] == group_node_id:
new_edge = update_source_handle(new_edge, g_nodes, g_edges)
if edge['target'] == group_node_id or edge['source'] == group_node_id:
updated_edges.append(new_edge)
return updated_edges
================================================
FILE: src/backend/bisheng/graph/utils.py
================================================
import copy
from collections import deque
from typing import Any, Dict, List, Union
from bisheng.interface.utils import extract_input_variables_from_prompt
class UnbuiltObject:
pass
def validate_prompt(prompt: str):
"""Validate prompt."""
if extract_input_variables_from_prompt(prompt):
return prompt
return fix_prompt(prompt)
def fix_prompt(prompt: str):
"""Fix prompt."""
return prompt + ' {input}'
def flatten_list(list_of_lists: list[Union[list, Any]]) -> list:
"""Flatten list of lists."""
new_list = []
for item in list_of_lists:
if isinstance(item, list):
new_list.extend(item)
else:
new_list.append(item)
return new_list
def find_last_node(nodes, edges):
"""
This function receives a flow and returns the last node.
"""
return next((n for n in nodes if all(e['source'] != n['id'] for e in edges)), None)
def add_parent_node_id(nodes, parent_node_id):
"""
This function receives a list of nodes and adds a parent_node_id to each node.
"""
for node in nodes:
node['parent_node_id'] = parent_node_id
def ungroup_node(group_node_data, base_flow):
template, flow = (
group_node_data['node']['template'],
group_node_data['node']['flow'],
)
parent_node_id = group_node_data['id']
g_nodes = flow['data']['nodes']
add_parent_node_id(g_nodes, parent_node_id)
g_edges = flow['data']['edges']
# Redirect edges to the correct proxy node
updated_edges = get_updated_edges(base_flow, g_nodes, g_edges, group_node_data['id'])
# Update template values
update_template(template, g_nodes)
nodes = [n for n in base_flow['nodes'] if n['id'] != group_node_data['id']] + g_nodes
edges = ([
e for e in base_flow['edges']
if e['target'] != group_node_data['id'] and e['source'] != group_node_data['id']
] + g_edges + updated_edges)
base_flow['nodes'] = nodes
base_flow['edges'] = edges
return nodes
def raw_topological_sort(nodes, edges) -> List[Dict]:
# Redefine the above function but using the nodes and self._edges
# which are dicts instead of Vertex and Edge objects
# nodes have an id, edges have a source and target keys
# return a list of node ids in topological order
# States: 0 = unvisited, 1 = visiting, 2 = visited
state = {node['id']: 0 for node in nodes}
nodes_dict = {node['id']: node for node in nodes}
sorted_vertices = []
def dfs(node):
if state[node] == 1:
# We have a cycle
raise ValueError('Graph contains a cycle, cannot perform topological sort')
if state[node] == 0:
state[node] = 1
for edge in edges:
if edge['source'] == node:
dfs(edge['target'])
state[node] = 2
sorted_vertices.append(node)
# Visit each node
for node in nodes:
if state[node['id']] == 0:
dfs(node['id'])
reverse_sorted = list(reversed(sorted_vertices))
return [nodes_dict[node_id] for node_id in reverse_sorted]
def process_flow(flow_object):
cloned_flow = copy.deepcopy(flow_object)
processed_nodes = set() # To keep track of processed nodes
def process_node(node):
node_id = node.get('id')
# If node already processed, skip
if node_id in processed_nodes:
return
if node.get('data') and node['data'].get('node') and node['data']['node'].get('flow'):
process_flow(node['data']['node']['flow']['data'])
new_nodes = ungroup_node(node['data'], cloned_flow)
# Add new nodes to the queue for future processing
nodes_to_process.extend(new_nodes)
# Mark node as processed
processed_nodes.add(node_id)
sorted_nodes_list = raw_topological_sort(cloned_flow['nodes'], cloned_flow['edges'])
nodes_to_process = deque(sorted_nodes_list)
while nodes_to_process:
node = nodes_to_process.popleft()
process_node(node)
return cloned_flow
def update_template(template, g_nodes):
"""
Updates the template of a node in a graph with the given template.
Args:
template (dict): The new template to update the node with.
g_nodes (list): The list of nodes in the graph.
Returns:
None
"""
for _, value in template.items():
if not value.get('proxy'):
continue
proxy_dict = value['proxy']
field, id_ = proxy_dict['field'], proxy_dict['id']
node_index = next((i for i, n in enumerate(g_nodes) if n['id'] == id_), -1)
if node_index != -1:
display_name = None
show = g_nodes[node_index]['data']['node']['template'][field].get('show', False)
advanced = g_nodes[node_index]['data']['node']['template'][field]['advanced']
if 'display_name' in g_nodes[node_index]['data']['node']['template'][field]:
display_name = g_nodes[node_index]['data']['node']['template'][field][
'display_name']
else:
display_name = g_nodes[node_index]['data']['node']['template'][field]['name']
g_nodes[node_index]['data']['node']['template'][field] = value
g_nodes[node_index]['data']['node']['template'][field]['show'] = show
g_nodes[node_index]['data']['node']['template'][field]['advanced'] = advanced
g_nodes[node_index]['data']['node']['template'][field]['display_name'] = display_name
def update_target_handle(
new_edge,
g_nodes,
group_node_id,
):
"""
Updates the target handle of a given edge if it is a proxy node.
Args:
new_edge (dict): The edge to update.
g_nodes (list): The list of nodes in the graph.
group_node_id (str): The ID of the group node.
Returns:
dict: The updated edge.
"""
# Compatibility Logic
for node in g_nodes:
if node['id'] in new_edge['targetHandle']:
new_edge['target'] = node['id']
break
return new_edge
def set_new_target_handle(proxy_id, new_edge, target_handle, node):
"""
Sets a new target handle for a given edge.
Args:
proxy_id (str): The ID of the proxy.
new_edge (dict): The new edge to be created.
target_handle (dict): The target handle of the edge.
node (dict): The node containing the edge.
Returns:
None
"""
new_edge['target'] = proxy_id
_type = target_handle.get('type')
if _type is None:
raise KeyError("The 'type' key must be present in target_handle.")
field = target_handle['proxy']['field']
new_target_handle = {
'fieldName': field,
'type': _type,
'id': proxy_id,
}
if node['data']['node'].get('flow'):
new_target_handle['proxy'] = {
'field': node['data']['node']['template'][field]['proxy']['field'],
'id': node['data']['node']['template'][field]['proxy']['id'],
}
if input_types := target_handle.get('inputTypes'):
new_target_handle['inputTypes'] = input_types
if not new_edge.get('data'):
new_edge['data'] = {}
new_edge['data']['targetHandle'] = new_target_handle
def update_source_handle(new_edge, g_nodes, g_edges):
"""
Updates the source handle of a given edge to the last node in the flow data.
Args:
new_edge (dict): The edge to update.
flow_data (dict): The flow data containing the nodes and edges.
Returns:
dict: The updated edge with the new source handle.
"""
last_node = copy.deepcopy(find_last_node(g_nodes, g_edges))
new_edge['sourceHandle'] = new_edge['sourceHandle'].replace(new_edge['source'],
last_node['id'])
new_edge['source'] = last_node['id']
return new_edge
def get_updated_edges(base_flow, g_nodes, g_edges, group_node_id):
"""
Given a base flow, a list of graph nodes and a group node id, returns a list of updated edges.
An updated edge is an edge that has its target or source handle updated based on the group node id.
Args:
base_flow (dict): The base flow containing a list of edges.
g_nodes (list): A list of graph nodes.
group_node_id (str): The id of the group node.
Returns:
list: A list of updated edges.
"""
updated_edges = []
for edge in base_flow['edges']:
new_edge = copy.deepcopy(edge)
if new_edge['target'] == group_node_id:
new_edge = update_target_handle(new_edge, g_nodes, group_node_id)
if new_edge['source'] == group_node_id:
new_edge = update_source_handle(new_edge, g_nodes, g_edges)
if edge['target'] == group_node_id or edge['source'] == group_node_id:
updated_edges.append(new_edge)
return updated_edges
def find_next_node(graph_data: Dict, node_id: str) -> List[Dict]:
"""
Finds the next node in the graph data based on the given node id.
"""
nodes = graph_data.get('nodes', [])
edges = graph_data.get('edges', [])
edges_ = [e['target'] for e in edges if e['source'] == node_id]
return [n for n in nodes if n['id'] in edges_]
def cut_graph_bynode(graph_data: Dict, node_id: str) -> List[Dict]:
"""
Setujunode_id Find andnodeAll dependent nodes involved.
"""
nodes = graph_data.get('nodes', [])
edges = graph_data.get('edges', [])
nodes_new_list = []
edges_new_list = []
iflast = True
for e in edges:
if e['target'] == node_id:
iflast = False
node_list, edge_list = cut_graph_bynode(graph_data, e['source'])
nodes_new_list.extend(node_list)
edges_new_list.extend(edge_list)
nodes_new_list.append([n for n in nodes if n['id'] == node_id][0])
edges_new_list.append(e)
if iflast:
for node in nodes:
if node['id'] == node_id:
return [node], []
return nodes_new_list, edges_new_list
================================================
FILE: src/backend/bisheng/graph/vertex/__init__.py
================================================
================================================
FILE: src/backend/bisheng/graph/vertex/base.py
================================================
import ast
import inspect
import json
import types
from typing import TYPE_CHECKING, Any, Coroutine, Dict, List, Optional
from bisheng.graph.utils import UnbuiltObject
from bisheng.interface.initialize import loading
from bisheng.interface.listing import lazy_load_dict
from bisheng.utils.constants import DIRECT_TYPES, NODE_ID_DICT, PRESET_QUESTION
from bisheng.utils.util import sync_to_async
from loguru import logger
if TYPE_CHECKING:
from bisheng.graph.edge.base import Edge
from bisheng.graph.graph.base import Graph
class Vertex:
def __init__(
self,
data: Dict,
graph: 'Graph',
base_type: Optional[str] = None,
is_task: bool = False,
params: Optional[Dict] = None,
) -> None:
self.graph = graph
self.id: str = data['id']
self._data = data
self.base_type: Optional[str] = base_type
self._parse_data()
self._built_object = UnbuiltObject()
self._built = False
self.artifacts: Dict[str, Any] = {}
self.task_id: Optional[str] = None
self.is_task = is_task
self.params = params or {}
self.parent_node_id: Optional[str] = self._data.get('parent_node_id')
self.parent_is_top_level = False
@property
def edges(self) -> List['Edge']:
return self.graph.get_vertex_edges(self.id)
def __getstate__(self):
return {
'_data': self._data,
'params': {},
'base_type': self.base_type,
'is_task': self.is_task,
'id': self.id,
'_built_object': UnbuiltObject(),
'_built': False,
'parent_node_id': self.parent_node_id,
'parent_is_top_level': self.parent_is_top_level,
}
def __setstate__(self, state):
self._data = state['_data']
self.params = state['params']
self.base_type = state['base_type']
self.is_task = state['is_task']
self.id = state['id']
self._parse_data()
if '_built_object' in state:
self._built_object = state['_built_object']
self._built = state['_built']
else:
self._built_object = UnbuiltObject()
self._built = False
self.artifacts: Dict[str, Any] = {}
self.task_id: Optional[str] = None
self.parent_node_id = state['parent_node_id']
self.parent_is_top_level = state['parent_is_top_level']
def set_top_level(self, top_level_vertices: List[str]) -> None:
self.parent_is_top_level = self.parent_node_id in top_level_vertices
def _parse_data(self) -> None:
self.data = self._data['data']
self.output = self.data['node']['base_classes']
template_dicts = {
key: value
for key, value in self.data['node']['template'].items() if isinstance(value, dict)
}
self.required_inputs = [
template_dicts[key]['type'] for key, value in template_dicts.items()
if value['required']
]
self.optional_inputs = [
template_dicts[key]['type'] for key, value in template_dicts.items()
if not value['required']
]
# Add the template_dicts[key]["input_types"] to the optional_inputs
self.optional_inputs.extend([
input_type for value in template_dicts.values()
for input_type in value.get('input_types', [])
])
template_dict = self.data['node']['template']
self.vertex_type = (self.data['type'] if 'Tool' not in self.output
or template_dict['_type'].islower() else template_dict['_type'])
if self.base_type is None:
for base_type, value in lazy_load_dict.ALL_TYPES_DICT.items():
if self.vertex_type in value:
self.base_type = base_type
break
def get_task(self):
# using the task_id, get the task from celery
# and return it
from celery.result import AsyncResult # type: ignore
return AsyncResult(self.task_id)
def _build_params(self):
# sourcery skip: merge-list-append, remove-redundant-if
# Some params are required, some are optional
# but most importantly, some params are python base classes
# like str and others are LangChain objects like LLMChain, BasePromptTemplate
# so we need to be able to distinguish between the two
# The dicts with "type" == "str" are the ones that are python base classes
# and most likely have a "value" key
# So for each key besides "_type" in the template dict, we have a dict
# with a "type" key. If the type is not "str", then we need to get the
# edge that connects to that node and get the Node with the required data
# and use that as the value for the param
# If the type is "str", then we need to get the value of the "value" key
# and use that as the value for the param
if self.graph is None:
raise ValueError('Graph not found')
template_dict = {
key: value
for key, value in self.data['node']['template'].items() if isinstance(value, dict)
}
params = {}
for edge in self.edges:
if not hasattr(edge, 'target_param'):
continue
param_key = edge.target_param
# If the param_key is in the template_dict and the edge.target_id is the current node
# We check this to make sure params with the same name but different target_id
# don't get overwritten
if param_key in template_dict and edge.target_id == self.id:
if template_dict[param_key]['list']:
if param_key not in params:
params[param_key] = []
params[param_key].append(self.graph.get_vertex(edge.source_id))
elif edge.target_id == self.id:
params[param_key] = self.graph.get_vertex(edge.source_id)
# for report, should get the source of source
for inner_edge in self.graph.get_vertex(edge.source_id).edges:
source_type = inner_edge.target_param
if (source_type == 'input_node' and inner_edge.source_id != self.id
and inner_edge.target_id != self.id):
if self.graph.get_vertex(inner_edge.source_id).vertex_type == 'InputNode':
# for extra params,
if PRESET_QUESTION not in params:
params[PRESET_QUESTION] = {}
params[PRESET_QUESTION].update({
inner_edge.target_id:
(inner_edge.source_id, self.graph.get_vertex(inner_edge.source_id))
})
elif (source_type == 'documents' and inner_edge.source_id != self.id
and inner_edge.target_id != self.id and self.graph.get_vertex(
inner_edge.target_id).vertex_type == 'LoaderOutputChain'):
if self.graph.get_vertex(inner_edge.source_id).vertex_type in {
'UniversalKVLoader', 'CustomKVLoader'
}:
for key, value in self.graph.get_vertex(
inner_edge.source_id).data['node']['template'].items():
if key in {'schemas', 'schema'}:
schema = value['value'].split('|')
if PRESET_QUESTION not in params:
params[PRESET_QUESTION] = {}
if inner_edge.target_id in params[PRESET_QUESTION]:
params[PRESET_QUESTION][inner_edge.target_id].append(
(inner_edge.source_id, schema))
else:
params[PRESET_QUESTION].update(
{inner_edge.target_id: [(inner_edge.source_id, schema)]})
for key, value in template_dict.items():
if key in params:
continue
if key == '_type' or (not value.get('show')
and not value.get('value')) and key != 'code':
continue
if value.get('collection_id') and 'collection_id' not in template_dict:
params['collection_id'] = value.get('collection_id')
# If the type is not transformable to a python base class
# then we need to get the edge that connects to this node
if value.get('type') == 'file':
# Load the type in value.get('suffixes') using
# what is inside value.get('content')
# value.get('value') is the file name
file_path = value.get('file_path')
params['file_name'] = value.get('value')
params[key] = file_path
elif value.get('type') in DIRECT_TYPES and params.get(key) is None:
val = value.get('value')
if value.get('type') == 'code':
try:
params[key] = ast.literal_eval(val) if val else None
except Exception as exc:
logger.debug(f'Error parsing code: {exc}')
params[key] = val
elif value.get('type') in ['dict', 'NestedDict']:
# When dict comes from the frontend it comes as a
# list of dicts, so we need to convert it to a dict
# before passing it to the build method
if isinstance(val, list):
params[key] = {
k: v
for item in value.get('value', [])
for k, v in item.items()
}
elif isinstance(val, dict):
params[key] = val
elif isinstance(val, str):
params[key] = json.loads(val) if val else {}
elif value.get('type') == 'int' and val is not None:
try:
params[key] = int(val)
except ValueError:
params[key] = val
elif value.get('type') == 'float' and val is not None:
try:
params[key] = float(val)
except ValueError:
params[key] = val
else:
params[key] = val
if not value.get('required') and params.get(key) is None:
if value.get('default'):
params[key] = value.get('default')
else:
params.pop(key, None)
# Add _type to params
self._raw_params = params
self.params = params
async def _build(self, user_id=None):
"""
Initiate the build process.
"""
logger.debug(f'Building {self.vertex_type}')
# keep node_id in params
self.params[NODE_ID_DICT] = {}
await self._build_each_node_in_params_dict(user_id)
await self._get_and_instantiate_class(user_id)
self._validate_built_object()
self._built = True
async def _build_each_node_in_params_dict(self, user_id=None):
"""
Iterates over each node in the params dictionary and builds it.
"""
for key, value in self.params.copy().items():
if self._is_node(value):
if value == self:
del self.params[key]
continue
await self._build_node_and_update_params(key, value, user_id)
elif isinstance(value, list) and self._is_list_of_nodes(value):
await self._build_list_of_nodes_and_update_params(key, value, user_id)
elif isinstance(value, dict) and self._is_dict_of_nodes(value):
await self._build_dict_of_nodes_and_update_params(key, value, user_id)
def _is_node(self, value):
"""
Checks if the provided value is an instance of Vertex.
"""
return isinstance(value, Vertex)
def _is_list_of_nodes(self, value):
"""
Checks if the provided value is a list of Vertex instances.
"""
return all(self._is_node(node) for node in value)
def _is_dict_of_nodes(self, value):
nodes = [node for node in value.values() if isinstance(node, tuple)]
if nodes:
return any(self._is_node(node[1]) for node in nodes)
else:
return False
async def get_result(self, user_id=None, timeout=None) -> Any:
# Check if the Vertex was built already
if self._built:
return self._built_object
if self.is_task and self.task_id is not None:
task = self.get_task()
result = task.get(timeout=timeout)
if isinstance(result, Coroutine):
result = await result
if result is not None: # If result is ready
self._update_built_object_and_artifacts(result)
return self._built_object
else:
# Handle the case when the result is not ready (retry, throw exception, etc.)
pass
# If there's no task_id, build the vertex locally
await self.build(user_id=user_id)
return self._built_object
async def _build_node_and_update_params(self, key, node, user_id=None):
"""
Builds a given node and updates the params dictionary accordingly.
"""
result = await node.get_result(user_id)
self._handle_func(key, result)
if isinstance(result, list):
self._extend_params_list_with_result(key, result)
self.params[key] = result
self.params[NODE_ID_DICT].update({key: node.id})
async def _build_list_of_nodes_and_update_params(self,
key,
nodes: List['Vertex'],
user_id=None):
"""
Iterates over a list of nodes, builds each and updates the params dictionary.
"""
self.params[key] = []
key_list = []
for node in nodes:
key_list.append(node.id)
built = await node.get_result(user_id)
if isinstance(built, list):
if key not in self.params:
self.params[key] = []
self.params[key].extend(built)
else:
self.params[key].append(built)
self.params[NODE_ID_DICT].update({key: key_list})
async def _build_dict_of_nodes_and_update_params(self, key, dicts, user_id=None):
self.params[key] = {}
for k, v in dicts.items():
if isinstance(v, list):
# loaderOutput
for k1, v1 in v:
if self._is_node(v1):
result = await v1.get_result(user_id)
self.params[key][k] = (k1, result)
else:
self.params[key][k] = (k1, v1)
elif self._is_node(v[1]):
result = await v[1].get_result(user_id)
self.params[key][k] = (v[0], result)
else:
self.params[key][k] = (v[0], v[1])
def _handle_func(self, key, result):
"""
Handles 'func' key by checking if the result is a function and setting it as coroutine.
"""
if key == 'func':
if not isinstance(result, types.FunctionType):
func_ = ''
if hasattr(result, 'arun'):
self.params['coroutine'] = result.arun
func_ = result.arun
if hasattr(result, 'run'):
result = result.run
func_ = func_ or result.run # type: ignore
elif hasattr(result, 'get_function'):
func_ = result.get_function() # type: ignore
elif inspect.iscoroutinefunction(result):
self.params['coroutine'] = result
else:
self.params['coroutine'] = sync_to_async(result)
return result
def _extend_params_list_with_result(self, key, result):
"""
Extends a list in the params dictionary with the given result if it exists.
"""
if isinstance(self.params[key], list):
self.params[key].extend(result)
async def _get_and_instantiate_class(self, user_id=None):
"""
Gets the class from a dictionary and instantiates it with the params.
"""
if self.base_type is None:
raise ValueError(f'Base type for node {self.vertex_type} not found')
try:
result = await loading.instantiate_class(node_type=self.vertex_type,
base_type=self.base_type,
params=self.params,
user_id=user_id)
self._update_built_object_and_artifacts(result)
except Exception as exc:
logger.exception("Error building node")
raise ValueError(f'Error building node {self.vertex_type}: {str(exc)}') from exc
def _update_built_object_and_artifacts(self, result):
"""
Updates the built object and its artifacts.
"""
if isinstance(result, tuple):
self._built_object, self.artifacts = result
else:
self._built_object = result
def _validate_built_object(self):
"""
Checks if the built object is None and raises a ValueError if so.
"""
if isinstance(self._built_object, UnbuiltObject):
raise ValueError(f'{self.vertex_type}: {self._built_object_repr()}')
elif self._built_object is None:
message = f'{self.vertex_type} returned None.'
if self.base_type == 'custom_components':
message += ' Make sure your build method returns a component.'
logger.warning(message)
raise ValueError(f'Node type {self.vertex_type} not found')
async def build(self, force: bool = False, user_id=None, *args, **kwargs) -> Any:
if not self._built or force:
await self._build(user_id, *args, **kwargs)
return self._built_object
def add_edge(self, edge: 'Edge') -> None:
if edge not in self.edges:
self.edges.append(edge)
def __repr__(self) -> str:
return f'Vertex(id={self.id}, data={self.data})'
def __eq__(self, __o: object) -> bool:
try:
return self.id == __o.id if isinstance(__o, Vertex) else False
except AttributeError:
return False
def __hash__(self) -> int:
return id(self)
def _built_object_repr(self):
# Add a message with an emoji, stars for sucess,
return f'Built successfully {self.id}' if self._built_object else f'Failed to build {self.id}'
================================================
FILE: src/backend/bisheng/graph/vertex/constants.py
================================================
================================================
FILE: src/backend/bisheng/graph/vertex/types.py
================================================
import ast
from typing import Any, Dict, List, Optional, Union
from bisheng.graph.utils import UnbuiltObject, flatten_list
from bisheng.graph.vertex.base import Vertex
from bisheng.interface.utils import extract_input_variables_from_prompt
class AgentVertex(Vertex):
def __init__(self, data: Dict, graph, params: Optional[Dict] = None):
super().__init__(data, graph=graph, base_type='agents', params=params)
self.tools: List[Union[ToolkitVertex, ToolVertex]] = []
self.chains: List[ChainVertex] = []
def __getstate__(self):
state = super().__getstate__()
state['tools'] = self.tools
state['chains'] = self.chains
return state
def __setstate__(self, state):
self.tools = state['tools']
self.chains = state['chains']
super().__setstate__(state)
def _set_tools_and_chains(self) -> None:
for edge in self.edges:
if not hasattr(edge, 'source_id'):
continue
source_node = self.graph.get_vertex(edge.source_id)
if isinstance(source_node, (ToolVertex, ToolkitVertex)):
self.tools.append(source_node)
elif isinstance(source_node, ChainVertex):
self.chains.append(source_node)
async def build(self, force: bool = False, user_id=None, *args, **kwargs) -> Any:
if not self._built or force:
self._set_tools_and_chains()
# First, build the tools
for tool_node in self.tools:
await tool_node.build(user_id=user_id)
# Next, build the chains and the rest
for chain_node in self.chains:
await chain_node.build(tools=self.tools, user_id=user_id)
await self._build(user_id=user_id)
return self._built_object
class ToolVertex(Vertex):
def __init__(
self,
data: Dict,
graph,
params: Optional[Dict] = None,
):
super().__init__(data, graph=graph, base_type='tools', params=params)
class LLMVertex(Vertex):
built_node_type = None
class_built_object = None
def __init__(self, data: Dict, graph, params: Optional[Dict] = None):
super().__init__(data, graph=graph, base_type='llms', params=params)
async def build(self, force: bool = False, user_id=None, *args, **kwargs) -> Any:
# LLM is different because some models might take up too much memory
# or time to load. So we only load them when we need them.ß
if self.vertex_type == self.built_node_type:
return self.class_built_object
if not self._built or force:
await self._build(user_id=user_id)
self.built_node_type = self.vertex_type
self.class_built_object = self._built_object
# Avoid deepcopying the LLM
# that are loaded from a file
return self._built_object
class ToolkitVertex(Vertex):
def __init__(self, data: Dict, graph, params=None):
super().__init__(data, graph=graph, base_type='toolkits', params=params)
class FileToolVertex(ToolVertex):
def __init__(self, data: Dict, graph, params=None):
super().__init__(data, graph=graph, params=params)
class WrapperVertex(Vertex):
def __init__(self, data: Dict, graph):
super().__init__(data, graph=graph, base_type='wrappers')
async def build(self, force: bool = False, user_id=None, *args, **kwargs) -> Any:
if not self._built or force:
if 'headers' in self.params:
self.params['headers'] = ast.literal_eval(self.params['headers'])
await self._build(user_id=user_id)
return self._built_object
class DocumentLoaderVertex(Vertex):
def __init__(self, data: Dict, graph, params: Optional[Dict] = None):
super().__init__(data, graph=graph, base_type='documentloaders', params=params)
def _built_object_repr(self):
# This built_object is a list of documents. Maybe we should
# show how many documents are in the list?
if self._built_object and not isinstance(self._built_object, UnbuiltObject):
avg_length = sum(
len(doc.page_content) for doc in self._built_object
if hasattr(doc, 'page_content')) / len(self._built_object)
return f"""{self.vertex_type}({len(self._built_object)} documents)
\nAvg. Document Length (characters): {int(avg_length)}
Documents: {self._built_object[:3]}..."""
return f'{self.vertex_type}()'
class EmbeddingVertex(Vertex):
def __init__(self, data: Dict, graph, params: Optional[Dict] = None):
super().__init__(data, graph=graph, base_type='embeddings', params=params)
class VectorStoreVertex(Vertex):
def __init__(self, data: Dict, graph, params=None):
super().__init__(data, graph=graph, base_type='vectorstores')
self.params = params or {}
# VectorStores may contain databse connections
# so we need to define the __reduce__ method and the __setstate__ method
# to avoid pickling errors
def remove_docs_and_texts_from_params(self):
# remove documents and texts from params
# so that we don't try to pickle a database connection
self.params.pop('documents', None)
self.params.pop('texts', None)
# def __getstate__(self):
# # We want to save the params attribute
# # and if "documents" or "texts" are in the params
# # we want to remove them because they have already
# # been processed.
# params = self.params.copy()
# params.pop("documents", None)
# params.pop("texts", None)
# return super().__getstate__()
def __setstate__(self, state):
super().__setstate__(state)
self.remove_docs_and_texts_from_params()
class MemoryVertex(Vertex):
def __init__(self, data: Dict, graph):
super().__init__(data, graph=graph, base_type='memory')
class RetrieverVertex(Vertex):
def __init__(self, data: Dict, graph):
super().__init__(data, graph=graph, base_type='retrievers')
class TextSplitterVertex(Vertex):
def __init__(self, data: Dict, graph, params: Optional[Dict] = None):
super().__init__(data, graph=graph, base_type='textsplitters', params=params)
def _built_object_repr(self):
# This built_object is a list of documents. Maybe we should
# show how many documents are in the list?
if self._built_object and not isinstance(self._built_object, UnbuiltObject):
avg_length = sum(len(doc.page_content)
for doc in self._built_object) / len(self._built_object)
return f"""{self.vertex_type}({len(self._built_object)} documents)
\nAvg. Document Length (characters): {int(avg_length)}
\nDocuments: {self._built_object[:3]}..."""
return f'{self.vertex_type}()'
class ChainVertex(Vertex):
def __init__(self, data: Dict, graph):
super().__init__(data, graph=graph, base_type='chains')
async def build(
self,
force: bool = False,
user_id=None,
*args,
**kwargs,
) -> Any:
if not self._built or force:
# Temporarily remove the code from the params
self.params.pop('code', None)
# Check if the chain requires a PromptVertex
# Temporarily remove "code" from the params
self.params.pop('code', None)
for key, value in self.params.items():
if isinstance(value, PromptVertex):
# Build the PromptVertex, passing the tools if available
tools = kwargs.get('tools', None)
self.params[key] = await value.build(tools=tools, force=force)
await self._build(user_id=user_id)
return self._built_object
class PromptVertex(Vertex):
def __init__(self, data: Dict, graph):
super().__init__(data, graph=graph, base_type='prompts')
async def build(
self,
force: bool = False,
user_id=None,
tools: Optional[List[Union[ToolkitVertex, ToolVertex]]] = None,
*args,
**kwargs,
) -> Any:
if not self._built or force:
if 'input_variables' not in self.params or self.params['input_variables'] is None:
self.params['input_variables'] = []
# Check if it is a ZeroShotPrompt and needs a tool
if 'ShotPrompt' in self.vertex_type:
tools = [await tool_node.build(user_id=user_id)
for tool_node in tools] if tools is not None else []
# flatten the list of tools if it is a list of lists
# first check if it is a list
if tools and isinstance(tools, list) and isinstance(tools[0], list):
tools = flatten_list(tools)
self.params['tools'] = tools
prompt_params = [
key for key, value in self.params.items()
if isinstance(value, str) and key != 'format_instructions'
]
else:
prompt_params = ['template']
if 'prompt' not in self.params and 'messages' not in self.params:
for param in prompt_params:
prompt_text = self.params[param]
variables = extract_input_variables_from_prompt(prompt_text)
self.params['input_variables'].extend(variables)
self.params['input_variables'] = list(set(self.params['input_variables']))
elif isinstance(self.params, dict):
self.params.pop('input_variables', None)
await self._build(user_id=user_id)
return self._built_object
def _built_object_repr(self):
if not self.artifacts or self._built_object is None or not hasattr(
self._built_object, 'format'):
return super()._built_object_repr()
# We'll build the prompt with the artifacts
# to show the user what the prompt looks like
# with the variables filled in
artifacts = self.artifacts.copy()
# Remove the handle_keys from the artifacts
# so the prompt format doesn't break
artifacts.pop('handle_keys', None)
try:
if (not hasattr(self._built_object, 'template')
and hasattr(self._built_object, 'prompt')
and not isinstance(self._built_object, UnbuiltObject)):
template = self._built_object.prompt.template
elif not isinstance(self._built_object, UnbuiltObject) and hasattr(
self._built_object, 'template'):
template = self._built_object.template
for key, value in artifacts.items():
if value:
replace_key = '{' + key + '}'
template = template.replace(replace_key, value)
return template if isinstance(template, str) else f'{self.vertex_type}({template})'
except KeyError:
return str(self._built_object)
class OutputParserVertex(Vertex):
def __init__(self, data: Dict, graph):
super().__init__(data, graph=graph, base_type='output_parsers')
class CustomComponentVertex(Vertex):
def __init__(self, data: Dict, graph):
super().__init__(data, graph=graph, base_type='custom_components', is_task=False)
def _built_object_repr(self):
if self.task_id and self.is_task:
if task := self.get_task():
return str(task.info)
else:
return f'Task {self.task_id} is not running'
if self.artifacts and 'repr' in self.artifacts:
return self.artifacts['repr'] or super()._built_object_repr()
================================================
FILE: src/backend/bisheng/initdb_config.yaml
================================================
knowledges: # 知识库相关配置
etl4lm:
# 文档解析模型服务配置，包括OCR、版式分析、表格识别、公式识别等
url: "" # http://192.168.106.12:8180/v1/etl4llm/predict
timeout: 600
# OCR SDK服务地址，默认为空则使用ETL4LM自带的轻量OCR模型（速度快，对于困难场景效果一般），若填写OCR SDK服务地址则使用高精度的OCR模型。
ocr_sdk_url: ""
llm_request:
# 控制技能 LLM 组件模型访问的超时配置, 以下是默认值
request_timeout: 600
max_retries: 1
default_operator:
# 使用免登录链接的方式需要配置，因为免登录链接相当于不知道用户信息，我们系统会自动把这些行为记录到某个用户头上，这里用来配置该用户的id
user: 1
enable_guest_access: true # 免登录链接是否可访问
# 密码安全相关配置
password_conf:
# 密码超过X天必须进行修改, 登录提示重新修改密码。大于0策略才生效
password_valid_period: 200
# 登录错误时间窗口,单位分钟。在错误时间窗口内超过最大错误次数会封禁用户，password_valid_period和login_error_time_window都大于0才生效
login_error_time_window: 5
# 最大错误次数，超过后会封禁用户，大于0时生效
max_error_times: 0
system_login_method:
# 是否允许多点登录
allow_multi_login: true
# sso系统登录配置（毕昇商业扩展套件功能，开源版无需配置）
gateway_login: false # 是否开启sso登录
admin_username: admin # 从 SSO/LDAP 注册的管理员用户名
# 登陆页面是否需要输入验证码，可设置为True或False
use_captcha: True
# 会话窗口底部提示文案
dialog_tips:
"内容由AI生成，仅供参考！"
env:
# 聊天窗口快捷搜索功能使用的搜索引擎，默认为百度，可以配置为内部文档搜索
# dialog_quick_search: http://www.baidu.com/s?wd=
# 当用户的环境前面的网关，不能在同一个端口上既有http又有socket时，需要这个配置，将两个请求区分开，默认可以不用
# websocket_url: 192.168.106.120:3003
office_url: http://IP:8701 # onlyoffice 组件地址，需要浏览器能直接访问
# 是否展示前端界面上的github和帮助链接
show_github_and_help: true
# 是否开启注册
enable_registration: true
# 前端上传文件的最大限制，单位MB
uploaded_files_maximum_size: 50
workflow:
# 节点运行最大步数
max_steps: 50
# 等待用户输入的超时时间，单位分钟
timeout: 5
# 灵思模块相关配置
linsight:
# 历史记录中工具消息的最大token，超过后需要总结下历史记录
tool_buffer: 100000
# 单个任务最大执行步骤数，防止死循环
max_steps: 200
# 灵思任务执行过程中模型调用重试次数
retry_num: 3
# 灵思任务执行过程中模型调用重试间隔时间（秒）
retry_sleep: 5
# 生成SOP时，prompt里放的用户上传文件信息的数量
max_file_num: 5
# 生成SOP时，prompt里放的组织知识库的最大数量
max_knowledge_num: 20
# 拆分二级任务时读取文件内容的字符数，超过后会截断
file_content_length: 5000
# 灵思模型默认温度
default_temperature: 0
# 模型json格式失败重试时的温度
retry_temperature: 1
================================================
FILE: src/backend/bisheng/interface/__init__.py
================================================
================================================
FILE: src/backend/bisheng/interface/agents/__init__.py
================================================
from bisheng.interface.agents.base import AgentCreator
__all__ = ['AgentCreator']
================================================
FILE: src/backend/bisheng/interface/agents/base.py
================================================
from typing import ClassVar, Dict, List, Optional
from bisheng.custom.customs import get_custom_nodes
from bisheng.interface.agents.custom import CUSTOM_AGENTS
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.importing.utils import import_class
from bisheng.common.services.config_service import settings
from bisheng.template.frontend_node.agents import AgentFrontendNode
from loguru import logger
from bisheng.utils.util import build_template_from_class, build_template_from_method
from bisheng_langchain import agents as bisheng_agents
from langchain.agents import types
class AgentCreator(LangChainTypeCreator):
type_name: str = 'agents'
from_method_nodes: ClassVar[Dict] = {
'ZeroShotAgent': 'from_llm_and_tools',
'CSVAgent': 'from_toolkit_and_llm',
'SQLAgent': 'from_toolkit_and_llm',
'ChatglmFunctionsAgent': 'from_llm_and_tools',
'LLMFunctionsAgent': 'from_llm_and_tools',
'OpenAIToolsAgent': 'from_llm_and_tools',
}
@property
def frontend_node_class(self) -> type[AgentFrontendNode]:
return AgentFrontendNode
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict = types.AGENT_TO_CLASS
# Add JsonAgent to the list of agents
for name, agent in CUSTOM_AGENTS.items():
# TODO: validate AgentType
self.type_dict[name] = agent # type: ignore
bisheng = {
chain_name: import_class(f'bisheng_langchain.agents.{chain_name}')
for chain_name in bisheng_agents.__all__
}
self.type_dict.update(bisheng)
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
try:
if name in get_custom_nodes(self.type_name).keys():
return get_custom_nodes(self.type_name)[name]
elif name in self.from_method_nodes:
return build_template_from_method(
name,
type_to_cls_dict=self.type_to_loader_dict,
add_function=True,
method_name=self.from_method_nodes[name],
)
return build_template_from_class(name, self.type_to_loader_dict, add_function=True)
except ValueError as exc:
raise ValueError('Agent not found') from exc
except AttributeError as exc:
logger.error(f'Agent {name} not loaded: {exc}')
return None
# Now this is a generator
def to_list(self) -> List[str]:
names = []
for _, agent in self.type_to_loader_dict.items():
agent_name = (agent.function_name()
if hasattr(agent, 'function_name') else agent.__name__)
if agent_name in settings.agents or settings.dev:
names.append(agent_name)
return names
agent_creator = AgentCreator()
================================================
FILE: src/backend/bisheng/interface/agents/custom.py
================================================
from typing import Any, List, Optional, Sequence, Tuple, Union
from bisheng.interface.base import CustomAgentExecutor
from langchain.agents import (AgentExecutor, AgentType, BaseSingleActionAgent, Tool, ZeroShotAgent,
initialize_agent)
from langchain.agents.agent_toolkits.vectorstore.prompt import PREFIX as VECTORSTORE_PREFIX
from langchain.agents.agent_toolkits.vectorstore.prompt import \
ROUTER_PREFIX as VECTORSTORE_ROUTER_PREFIX
from langchain.agents.agent_toolkits.vectorstore.toolkit import (VectorStoreInfo,
VectorStoreRouterToolkit,
VectorStoreToolkit)
from langchain.agents.mrkl.prompt import FORMAT_INSTRUCTIONS
from langchain.agents.openai_functions_agent.base import OpenAIFunctionsAgent
from langchain.agents.openai_tools.base import create_openai_tools_agent
from langchain.base_language import BaseLanguageModel
from langchain.chains import LLMChain
from langchain.memory.buffer import ConversationBufferMemory
from langchain.memory.chat_memory import BaseChatMemory
from langchain_community.agent_toolkits import SQLDatabaseToolkit
from langchain_community.agent_toolkits.json.prompt import JSON_PREFIX, JSON_SUFFIX
from langchain_community.agent_toolkits.json.toolkit import JsonToolkit
from langchain_community.agent_toolkits.sql.prompt import SQL_PREFIX, SQL_SUFFIX
from langchain_community.tools.sql_database.prompt import QUERY_CHECKER
from langchain_community.utilities.sql_database import SQLDatabase
from langchain_core.agents import AgentAction, AgentFinish
from langchain_core.callbacks import BaseCallbackManager, Callbacks
from langchain_core.memory import BaseMemory
from langchain_core.messages import BaseMessage, SystemMessage
from langchain_core.prompts.chat import (BaseMessagePromptTemplate, ChatPromptTemplate,
HumanMessagePromptTemplate, MessagesPlaceholder)
from langchain_core.runnables import RunnableConfig
from langchain_core.tools import BaseTool
from langchain_experimental.agents.agent_toolkits.pandas.prompt import PREFIX as PANDAS_PREFIX
from langchain_experimental.agents.agent_toolkits.pandas.prompt import \
SUFFIX_WITH_DF as PANDAS_SUFFIX
from langchain_experimental.tools.python.tool import PythonAstREPLTool
from pydantic import Field
history_prompt = """Below is a transcript of your chats:
{history}
"""
class JsonAgent(CustomAgentExecutor):
"""Json agent"""
@staticmethod
def function_name():
return 'JsonAgent'
@classmethod
def initialize(cls, *args, **kwargs):
return cls.from_toolkit_and_llm(*args, **kwargs)
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
@classmethod
def from_toolkit_and_llm(cls, toolkit: JsonToolkit, llm: BaseLanguageModel):
tools = toolkit if isinstance(toolkit, list) else toolkit.get_tools()
tool_names = {tool.name for tool in tools}
prompt = ZeroShotAgent.create_prompt(
tools,
prefix=JSON_PREFIX,
suffix=JSON_SUFFIX,
format_instructions=FORMAT_INSTRUCTIONS,
input_variables=None,
)
llm_chain = LLMChain(
llm=llm,
prompt=prompt,
)
agent = ZeroShotAgent(
llm_chain=llm_chain,
allowed_tools=tool_names # type: ignore
)
return cls.from_agent_and_tools(agent=agent, tools=tools, verbose=True)
def run(self, *args, **kwargs):
return super().run(*args, **kwargs)
class CSVAgent(CustomAgentExecutor):
"""CSV agent"""
@staticmethod
def function_name():
return 'CSVAgent'
@classmethod
def initialize(cls, *args, **kwargs):
return cls.from_toolkit_and_llm(*args, **kwargs)
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
@classmethod
def from_toolkit_and_llm(
cls,
path: str,
llm: BaseLanguageModel,
pandas_kwargs: Optional[dict] = None,
prefix: str = PANDAS_PREFIX,
suffix: str = PANDAS_SUFFIX,
format_instructions: str = FORMAT_INSTRUCTIONS,
input_variables: Optional[List[str]] = ['df_head', 'input', 'agent_scratchpad'],
**kwargs: Any):
import pandas as pd # type: ignore
_kwargs = pandas_kwargs or {}
df = pd.read_csv(path, **_kwargs)
tools = [PythonAstREPLTool(locals={'df': df})] # type: ignore
prompt = ZeroShotAgent.create_prompt(
tools,
prefix=prefix,
suffix=suffix,
format_instructions=format_instructions,
input_variables=input_variables,
)
partial_prompt = prompt.partial(df_head=str(df.head()))
llm_chain = LLMChain(
llm=llm,
prompt=partial_prompt,
)
tool_names = {tool.name for tool in tools}
agent = ZeroShotAgent(
llm_chain=llm_chain,
allowed_tools=tool_names,
**kwargs # type: ignore
)
return cls.from_agent_and_tools(agent=agent, tools=tools, verbose=True)
def run(self, *args, **kwargs):
return super().run(*args, **kwargs)
class VectorStoreAgent(CustomAgentExecutor):
"""Vector store agent"""
@staticmethod
def function_name():
return 'VectorStoreAgent'
@classmethod
def initialize(cls, *args, **kwargs):
return cls.from_toolkit_and_llm(*args, **kwargs)
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
@classmethod
def from_toolkit_and_llm(cls, llm: BaseLanguageModel, vectorstoreinfo: VectorStoreInfo,
**kwargs: Any):
"""Construct a vectorstore agent from an LLM and tools."""
toolkit = VectorStoreToolkit(vectorstore_info=vectorstoreinfo, llm=llm)
tools = toolkit.get_tools()
prompt = ZeroShotAgent.create_prompt(tools, prefix=VECTORSTORE_PREFIX)
llm_chain = LLMChain(
llm=llm,
prompt=prompt,
)
tool_names = {tool.name for tool in tools}
agent = ZeroShotAgent(
llm_chain=llm_chain,
allowed_tools=tool_names,
**kwargs # type: ignore
)
return AgentExecutor.from_agent_and_tools(agent=agent,
tools=tools,
verbose=True,
handle_parsing_errors=True)
def run(self, *args, **kwargs):
return super().run(*args, **kwargs)
class SQLAgent(CustomAgentExecutor):
"""SQL agent"""
@staticmethod
def function_name():
return 'SQLAgent'
@classmethod
def initialize(cls, *args, **kwargs):
return cls.from_toolkit_and_llm(*args, **kwargs)
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
@classmethod
def from_toolkit_and_llm(cls,
llm: BaseLanguageModel,
database_uri: str,
top_k: int = 10,
prefix: str = SQL_PREFIX,
suffix: str = SQL_SUFFIX,
format_instructions: str = FORMAT_INSTRUCTIONS,
input_variables: Optional[List[str]] = ['input', 'agent_scratchpad'],
**kwargs: Any):
"""Construct an SQL agent from an LLM and tools."""
db = SQLDatabase.from_uri(database_uri)
toolkit = SQLDatabaseToolkit(db=db, llm=llm)
# The right code should be this, but there is a problem with tools = toolkit.get_tools()
# related to `OPENAI_API_KEY`
# return create_sql_agent(llm=llm, toolkit=toolkit, verbose=True)
from langchain.prompts import PromptTemplate
from langchain.tools.sql_database.tool import (
InfoSQLDatabaseTool,
ListSQLDatabaseTool,
QuerySQLCheckerTool,
QuerySQLDataBaseTool,
)
llmchain = LLMChain(
llm=llm,
prompt=PromptTemplate(template=QUERY_CHECKER, input_variables=['query', 'dialect']),
)
tools = [
QuerySQLDataBaseTool(db=db), # type: ignore
InfoSQLDatabaseTool(db=db), # type: ignore
ListSQLDatabaseTool(db=db), # type: ignore
QuerySQLCheckerTool(db=db, llm_chain=llmchain, llm=llm), # type: ignore
]
prefix = prefix.format(dialect=toolkit.dialect, top_k=top_k)
prompt = ZeroShotAgent.create_prompt(
tools=tools, # type: ignore
prefix=prefix,
suffix=suffix,
format_instructions=format_instructions,
input_variables=input_variables,
)
llm_chain = LLMChain(
llm=llm,
prompt=prompt,
)
tool_names = {tool.name for tool in tools} # type: ignore
agent = ZeroShotAgent(
llm_chain=llm_chain,
allowed_tools=tool_names,
**kwargs # type: ignore
)
return AgentExecutor.from_agent_and_tools(
agent=agent,
tools=tools, # type: ignore
verbose=True,
max_iterations=15,
early_stopping_method='force',
handle_parsing_errors=True,
)
def run(self, *args, **kwargs):
return super().run(*args, **kwargs)
class VectorStoreRouterAgent(CustomAgentExecutor):
"""Vector Store Router Agent"""
@staticmethod
def function_name():
return 'VectorStoreRouterAgent'
@classmethod
def initialize(cls, *args, **kwargs):
return cls.from_toolkit_and_llm(*args, **kwargs)
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
@classmethod
def from_toolkit_and_llm(cls, llm: BaseLanguageModel,
vectorstoreroutertoolkit: VectorStoreRouterToolkit, **kwargs: Any):
"""Construct a vector store router agent from an LLM and tools."""
tools = (vectorstoreroutertoolkit if isinstance(vectorstoreroutertoolkit, list) else
vectorstoreroutertoolkit.get_tools())
prompt = ZeroShotAgent.create_prompt(tools, prefix=VECTORSTORE_ROUTER_PREFIX)
llm_chain = LLMChain(
llm=llm,
prompt=prompt,
)
tool_names = {tool.name for tool in tools}
agent = ZeroShotAgent(
llm_chain=llm_chain,
allowed_tools=tool_names,
**kwargs # type: ignore
)
return AgentExecutor.from_agent_and_tools(agent=agent,
tools=tools,
verbose=True,
handle_parsing_errors=True)
def run(self, *args, **kwargs):
return super().run(*args, **kwargs)
class OpenAIToolsAgent(OpenAIFunctionsAgent):
"""OpenAI Tools Agent"""
memory: BaseMemory = Field(default_factory=ConversationBufferMemory)
agent: Any
@property
def input_keys(self) -> List[str]:
"""Get input keys. Input refers to user input here."""
return ['input', 'history']
def plan(
self,
intermediate_steps: List[Tuple[AgentAction, str]],
callbacks: Callbacks = None,
with_functions: bool = True,
**kwargs: Any,
) -> Union[AgentAction, AgentFinish]:
input_dict = {'intermediate_steps': intermediate_steps}
selected_inputs = {
k: kwargs[k]
for k in self.prompt.input_variables if k != 'agent_scratchpad'
}
input_dict.update(selected_inputs)
return self.agent.invoke(input_dict, config=RunnableConfig(callbacks=callbacks))
async def aplan(
self,
intermediate_steps: List[Tuple[AgentAction, str]],
callbacks: Callbacks = None,
**kwargs: Any,
) -> Union[AgentAction, AgentFinish]:
input_dict = {'intermediate_steps': intermediate_steps}
selected_inputs = {
k: kwargs[k]
for k in self.prompt.input_variables if k != 'agent_scratchpad'
}
input_dict.update(selected_inputs)
return await self.agent.ainvoke(input_dict, config=RunnableConfig(callbacks=callbacks))
@classmethod
def create_prompt(
cls,
system_message: Optional[SystemMessage] = SystemMessage(
content='You are a helpful AI assistant.'),
extra_prompt_messages: Optional[List[BaseMessagePromptTemplate]] = None,
) -> ChatPromptTemplate:
"""Create prompt for this agent.
Args:
system_message: Message to use as the system message that will be the
first in the prompt.
extra_prompt_messages: Prompt messages that will be placed between the
system message and the new human input.
Returns:
A prompt template to pass into this agent.
"""
_prompts = extra_prompt_messages or []
messages: List[Union[BaseMessagePromptTemplate, BaseMessage]]
if system_message:
messages = [system_message]
else:
messages = []
messages.extend([
*_prompts,
MessagesPlaceholder(variable_name='agent_scratchpad'),
])
return ChatPromptTemplate(messages=messages) # type: ignore[arg-type, call-arg]
@classmethod
def from_llm_and_tools(
cls,
llm: BaseLanguageModel,
tools: Sequence[BaseTool],
callback_manager: Optional[BaseCallbackManager] = None,
extra_prompt: str = history_prompt,
human_prompt: str = '{input}',
system_message: str = 'You are a helpful AI assistant.',
**kwargs: Any,
) -> BaseSingleActionAgent:
"""Construct an agent from an LLM and tools.
Args:
llm: The LLM to use as the agent.
tools: The tools to use.
callback_manager: The callback manager to use. Defaults to None.
extra_prompt_messages: Extra prompt messages to use. Defaults to None.
system_message: The system message to use.
Defaults to a default system message.
kwargs: Additional parameters to pass to the agent.
"""
system_message_prompt = SystemMessage(content=system_message)
extra_prompt_messages = [HumanMessagePromptTemplate.from_template(extra_prompt)]
extra_prompt_messages.extend([HumanMessagePromptTemplate.from_template(human_prompt)])
prompt = cls.create_prompt(system_message=system_message_prompt,
extra_prompt_messages=extra_prompt_messages)
agent = create_openai_tools_agent(llm, tools, prompt)
return cls( # type: ignore[call-arg]
llm=llm,
agent=agent,
prompt=prompt,
tools=tools,
callback_manager=callback_manager,
**kwargs,
)
@classmethod
def initialize(
cls,
llm: BaseLanguageModel,
tools: Sequence[BaseTool],
callback_manager: Optional[BaseCallbackManager] = None,
extra_prompt: str = history_prompt,
human_prompt: str = '{input}',
system_message: str = 'You are a helpful AI assistant.',
memory: Optional[BaseChatMemory] = None,
**kwargs: Any,
):
agent = cls.from_llm_and_tools(
llm,
tools,
callback_manager,
extra_prompt,
human_prompt,
system_message,
memory=memory,
**kwargs,
)
return AgentExecutor.from_agent_and_tools(
agent=agent,
tools=tools,
verbose=True,
memory=memory,
return_intermediate_steps=True,
handle_parsing_errors=True,
)
class InitializeAgent(CustomAgentExecutor):
"""Implementation of AgentInitializer function"""
@staticmethod
def function_name():
return 'AgentInitializer'
@classmethod
def initialize(
cls,
llm: BaseLanguageModel,
tools: List[Tool],
agent: str,
memory: Optional[BaseChatMemory] = None,
):
# Find which value in the AgentType enum corresponds to the string
# passed in as agent
if agent == 'openai-tools':
agent = OpenAIToolsAgent.from_llm_and_tools(
llm,
tools,
memory=memory,
return_intermediate_steps=True,
handle_parsing_errors=True,
)
return AgentExecutor.from_agent_and_tools(
agent=agent,
tools=tools,
memory=memory,
return_intermediate_steps=True,
handle_parsing_errors=True,
)
agent = AgentType(agent)
return initialize_agent(
tools=tools,
llm=llm,
# LangChain now uses Enum for agent, but we still support string
agent=agent, # type: ignore
memory=memory,
return_intermediate_steps=True,
handle_parsing_errors=True,
)
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
def run(self, *args, **kwargs):
return super().run(*args, **kwargs)
# custom agents must initialize with initialize method
CUSTOM_AGENTS = {
'JsonAgent': JsonAgent,
'CSVAgent': CSVAgent,
'AgentInitializer': InitializeAgent,
'VectorStoreAgent': VectorStoreAgent,
'VectorStoreRouterAgent': VectorStoreRouterAgent,
'SQLAgent': SQLAgent,
'OpenAIToolsAgent': OpenAIToolsAgent,
}
================================================
FILE: src/backend/bisheng/interface/agents/prebuilt.py
================================================
from langchain.agents import AgentExecutor, ZeroShotAgent
from langchain.agents.agent_toolkits.json.prompt import JSON_PREFIX, JSON_SUFFIX
from langchain.agents.agent_toolkits.json.toolkit import JsonToolkit
from langchain.agents.mrkl.prompt import FORMAT_INSTRUCTIONS
from langchain.base_language import BaseLanguageModel
from langchain.chains import LLMChain
class MalfoyAgent(AgentExecutor):
"""Json agent"""
prefix = 'Malfoy: '
@classmethod
def initialize(cls, *args, **kwargs):
return cls.from_toolkit_and_llm(*args, **kwargs)
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
@classmethod
def from_toolkit_and_llm(cls, toolkit: JsonToolkit, llm: BaseLanguageModel):
tools = toolkit.get_tools()
tool_names = {tool.name for tool in tools}
prompt = ZeroShotAgent.create_prompt(
tools,
prefix=JSON_PREFIX,
suffix=JSON_SUFFIX,
format_instructions=FORMAT_INSTRUCTIONS,
input_variables=None,
)
llm_chain = LLMChain(
llm=llm,
prompt=prompt,
)
agent = ZeroShotAgent(llm_chain=llm_chain, allowed_tools=tool_names) # type: ignore
return cls.from_agent_and_tools(agent=agent, tools=tools, verbose=True)
def run(self, *args, **kwargs):
return super().run(*args, **kwargs)
PREBUILT_AGENTS = {
'MalfoyAgent': MalfoyAgent,
}
================================================
FILE: src/backend/bisheng/interface/autogenRole/__init__.py
================================================
================================================
FILE: src/backend/bisheng/interface/autogenRole/base.py
================================================
from typing import Any, Dict, List, Optional
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.importing.utils import import_class
from bisheng.template.frontend_node.autogenrole import AutogenRoleFrontNode
from loguru import logger
from bisheng.utils.util import build_template_from_class
from bisheng_langchain import autogen_role
class AutogenRole(LangChainTypeCreator):
type_name: str = 'autogen_roles'
@property
def frontend_node_class(self) -> type[AutogenRoleFrontNode]:
return AutogenRoleFrontNode
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict: dict[str, Any] = {
role_name: import_class(f'bisheng_langchain.autogen_role.{role_name}')
for role_name in autogen_role.__all__
}
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
try:
return build_template_from_class(
name, self.type_to_loader_dict, add_function=True
)
except ValueError as exc:
raise ValueError('Agent not found') from exc
except AttributeError as exc:
logger.error(f'Agent {name} not loaded: {exc}')
return None
# Now this is a generator
def to_list(self) -> List[str]:
names = []
for _, role in self.type_to_loader_dict.items():
role_name = (role.function_name()
if hasattr(role, 'function_name') else role.__name__)
names.append(role_name)
return names
autogenrole_creator = AutogenRole()
================================================
FILE: src/backend/bisheng/interface/base.py
================================================
from abc import ABC, abstractmethod
from typing import Any, Dict, List, Optional, Type, Union
from bisheng.common.services.config_service import settings
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
from bisheng.template.template.base import Template
from loguru import logger
from langchain.agents import AgentExecutor
from langchain.chains.base import Chain
from pydantic import BaseModel
# Assuming necessary imports for Field, Template, and FrontendNode classes
skip_llm = {'CombineDocsChain'}
class LangChainTypeCreator(BaseModel, ABC):
type_name: str
type_dict: Optional[Dict] = None
name_docs_dict: Optional[Dict[str, str]] = None
@property
def frontend_node_class(self) -> Type[FrontendNode]:
"""The class type of the FrontendNode created in frontend_node."""
return FrontendNode
@property
def docs_map(self) -> Dict[str, str]:
"""A dict with the name of the component as key and the documentation link as value."""
if self.name_docs_dict is None:
try:
type_settings = getattr(settings, self.type_name)
self.name_docs_dict = {
name: value_dict['documentation']
for name, value_dict in type_settings.items()
}
except AttributeError as exc:
logger.error(exc)
self.name_docs_dict = {}
return self.name_docs_dict
@property
@abstractmethod
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
raise NotImplementedError
return self.type_dict
@abstractmethod
def get_signature(
self, name: str
) -> Union[Optional[Dict[Any, Any]], FrontendNode]:
pass
@abstractmethod
def to_list(self) -> List[str]:
pass
def to_dict(self) -> Dict:
result: Dict = {self.type_name: {}}
for name in self.to_list():
# frontend_node.to_dict() returns a dict with the following structure:
# {name: {template: {fields}, description: str}}
# so we should update the result dict
node = self.frontend_node(name)
if node is not None:
node = node.to_dict() # type: ignore
result[self.type_name].update(node)
return result
def frontend_node(self, name) -> Union[FrontendNode, None]:
signature = self.get_signature(name)
if signature is None:
logger.error(f'Node {name} not loaded')
return signature
if not isinstance(signature, FrontendNode):
fields = [
TemplateField(
name=key,
field_type=value['type'],
required=value.get('required', False),
placeholder=value.get('placeholder', ''),
is_list=value.get('list', False),
show=value.get('show', True),
multiline=value.get('multiline', False),
value=value.get('value', None),
suffixes=value.get('suffixes', []),
file_types=value.get('fileTypes', []),
file_path=value.get('file_path', None),
) for key, value in signature['template'].items()
if key != '_type'
]
template = Template(type_name=name, fields=fields)
signature = self.frontend_node_class(
template=template,
description=signature.get('description', ''),
base_classes=signature['base_classes'],
name=name,
)
# #Determine if it containsinputKeys
if signature.name not in skip_llm:
if name in self.type_to_loader_dict:
class_tmp = self.type_to_loader_dict[name]
else:
for _, cls_ in self.type_to_loader_dict.items():
if hasattr(cls_, 'function_name') and cls_.function_name() == name:
class_tmp = cls_
elif cls_.__name__ == name:
class_tmp = cls_
if class_tmp and hasattr(class_tmp, 'input_keys'):
signature.template.add_field(
TemplateField(
field_type='input',
required=False,
show=True,
name='input_node',
display_name='Preset Question',
)
)
signature.add_extra_fields()
signature.add_extra_base_classes()
signature.set_documentation(self.docs_map.get(name, ''))
return signature
class CustomChain(Chain, ABC):
"""Custom chain"""
@staticmethod
def function_name():
return 'CustomChain'
@classmethod
def initialize(cls, *args, **kwargs):
pass
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
def run(self, *args, **kwargs):
return super().run(*args, **kwargs)
class CustomAgentExecutor(AgentExecutor, ABC):
"""Custom chain"""
@staticmethod
def function_name():
return 'CustomChain'
@classmethod
def initialize(cls, *args, **kwargs):
pass
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
def run(self, *args, **kwargs):
return super().run(*args, **kwargs)
================================================
FILE: src/backend/bisheng/interface/chains/__init__.py
================================================
from bisheng.interface.chains.base import ChainCreator
__all__ = ['ChainCreator']
================================================
FILE: src/backend/bisheng/interface/chains/base.py
================================================
from typing import Any, ClassVar, Dict, List, Optional, Type
from bisheng.custom.customs import get_custom_nodes
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.importing.utils import import_class
from bisheng.common.services.config_service import settings
from bisheng.template.frontend_node.chains import ChainFrontendNode
from loguru import logger
from bisheng.utils.util import build_template_from_class, build_template_from_method
from bisheng_langchain import chains as bisheng_chains
from bisheng_langchain import sql as bisheng_sql
from bisheng_langchain.rag.bisheng_rag_chain import BishengRetrievalQA
from langchain import chains
from langchain_experimental import sql
# Assuming necessary imports for Field, Template, and FrontendNode classes
class ChainCreator(LangChainTypeCreator):
type_name: str = 'chains'
@property
def frontend_node_class(self) -> Type[ChainFrontendNode]:
return ChainFrontendNode
# We need to find a better solution for this
from_method_nodes: ClassVar[Dict] = {
'APIChain': 'from_llm_and_api_docs',
'ConversationalRetrievalChain': 'from_llm',
'LLMCheckerChain': 'from_llm',
'SQLDatabaseChain': 'from_llm',
'LLMRouterChain': 'from_llm',
'BishengRetrievalQA': 'from_llm',
'QAGenerationChain': 'from_llm',
'QAGenerationChainV2': 'from_llm',
}
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
# langchain
self.type_dict: dict[str, Any] = {
chain_name: import_class(f'langchain.chains.{chain_name}')
for chain_name in chains.__all__
}
# bisheng-langchain
bisheng = {
chain_name: import_class(f'bisheng_langchain.chains.{chain_name}')
for chain_name in bisheng_chains.__all__
}
# If configured incustom_chains, it will not start frommethodHow to initialize, resulting in an error
self.type_dict['BishengRetrievalQA'] = BishengRetrievalQA
self.type_dict.update(bisheng)
# sql community
community = {
chain_name: import_class(f'langchain_experimental.sql.{chain_name}')
for chain_name in sql.__all__
}
self.type_dict.update(community)
# sql community
bisheng_sql_add = {
chain_name: import_class(f'bisheng_langchain.sql.{chain_name}')
for chain_name in bisheng_sql.__all__
}
self.type_dict.update(bisheng_sql_add)
from bisheng.interface.chains.custom import CUSTOM_CHAINS
self.type_dict.update(CUSTOM_CHAINS)
# Filter according to settings.chains
self.type_dict = {
name: chain
for name, chain in self.type_dict.items()
if name in settings.chains or settings.dev
}
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
try:
if name in get_custom_nodes(self.type_name).keys():
return get_custom_nodes(self.type_name)[name]
elif name in self.from_method_nodes.keys():
return build_template_from_method(
name,
type_to_cls_dict=self.type_to_loader_dict,
method_name=self.from_method_nodes[name],
add_function=True,
)
return build_template_from_class(name, self.type_to_loader_dict, add_function=True)
except ValueError as exc:
raise ValueError(f'Chain {name} not found: {exc}') from exc
except AttributeError as exc:
logger.error(f'Chain {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
names = []
for _, chain in self.type_to_loader_dict.items():
chain_name = (chain.function_name()
if hasattr(chain, 'function_name') else chain.__name__)
names.append(chain_name)
return names
chain_creator = ChainCreator()
================================================
FILE: src/backend/bisheng/interface/chains/custom.py
================================================
from typing import Any, Dict, List, Optional, Type, Union
from bisheng.interface.base import CustomChain
from bisheng.interface.utils import extract_input_variables_from_prompt
from bisheng_langchain.chains.question_answering import load_qa_chain
from langchain.base_language import BaseLanguageModel
from langchain.callbacks.manager import CallbackManagerForChainRun
from langchain.chains import ConversationChain, LLMChain
from langchain.chains.summarize import load_summarize_chain
from langchain.memory import ConversationBufferMemory
from langchain.prompts import PromptTemplate
from langchain.schema import BaseMemory
from langchain.schema.prompt_template import BasePromptTemplate
from langchain_community.utilities.dalle_image_generator import DallEAPIWrapper
from pydantic import Field, model_validator
DEFAULT_SUFFIX = """"
Current conversation:
{history}
Human: {input}
{ai_prefix}"""
class BaseCustomConversationChain(ConversationChain):
"""BaseCustomChain is a chain you can use to have a conversation with a custom character."""
template: Optional[str]
ai_prefix_value: Optional[str]
"""Field to use as the ai_prefix. It needs to be set and has to be in the template"""
@model_validator(mode='before')
@classmethod
def build_template(cls, values):
format_dict = {}
input_variables = extract_input_variables_from_prompt(values['template'])
if values.get('ai_prefix_value', None) is None:
values['ai_prefix_value'] = values['memory'].ai_prefix
for key in input_variables:
new_value = values.get(key, f'{{{key}}}')
format_dict[key] = new_value
if key == values.get('ai_prefix_value', None):
values['memory'].ai_prefix = new_value
values['template'] = values['template'].format(**format_dict)
values['template'] = values['template']
values['input_variables'] = extract_input_variables_from_prompt(values['template'])
values['prompt'].template = values['template']
values['prompt'].input_variables = values['input_variables']
return values
class SeriesCharacterChain(BaseCustomConversationChain):
"""SeriesCharacterChain is a chain you can use to have a conversation with a character from a series."""
character: str
series: str
template: Optional[str] = """I want you to act like {character} from {series}.
I want you to respond and answer like {character}. do not write any explanations. only answer like {character}.
You must know all of the knowledge of {character}.
Current conversation:
{history}
Human: {input}
{character}:"""
memory: BaseMemory = Field(default_factory=ConversationBufferMemory)
ai_prefix_value: Optional[str] = 'character'
"""Default memory store."""
class MidJourneyPromptChain(BaseCustomConversationChain):
"""MidJourneyPromptChain is a chain you can use to generate new MidJourney prompts."""
template: Optional[
str
] = """I want you to act as a prompt generator for Midjourney's artificial intelligence program.
Your job is to provide detailed and creative descriptions that will inspire unique and interesting images from the AI.
Keep in mind that the AI is capable of understanding a wide range of language and can interpret abstract concepts, so feel free to be as imaginative and descriptive as possible.
For example, you could describe a scene from a futuristic city, or a surreal landscape filled with strange creatures.
The more detailed and imaginative your description, the more interesting the resulting image will be. Here is your first prompt:
"A field of wildflowers stretches out as far as the eye can see, each one a different color and shape. In the distance, a massive tree towers over the landscape, its branches reaching up to the sky like tentacles.\"
Current conversation:
{history}
Human: {input}
AI:""" # noqa: E501
class TimeTravelGuideChain(BaseCustomConversationChain):
template: Optional[
str
] = """I want you to act as my time travel guide. You are helpful and creative. I will provide you with the historical period or future time I want to visit and you will suggest the best events, sights, or people to experience. Provide the suggestions and any necessary information.
Current conversation:
{history}
Human: {input}
AI:""" # noqa: E501
class CombineDocsChain(CustomChain):
"""Implementation of load_qa_chain function"""
@staticmethod
def function_name():
return 'load_qa_chain'
@classmethod
def initialize(cls,
llm: BaseLanguageModel,
chain_type: str,
prompt: BasePromptTemplate = None,
document_prompt: BasePromptTemplate = None,
token_max: str = -1):
if chain_type == 'stuff':
if document_prompt:
return load_qa_chain(llm=llm,
chain_type=chain_type,
prompt=prompt,
token_max=token_max,
document_prompt=document_prompt)
else:
return load_qa_chain(llm=llm,
chain_type=chain_type,
prompt=prompt,
token_max=token_max)
else:
return load_qa_chain(llm=llm, chain_type=chain_type)
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
def run(self, *args, **kwargs):
return super().run(*args, **kwargs)
class SummarizeDocsChain(CustomChain):
"""Implementation of load_sumarize_chain function"""
@staticmethod
def function_name():
return 'load_summarize_chain'
@classmethod
def initialize(cls,
llm: BaseLanguageModel,
chain_type: str,
prompt: str = None,
token_max: str = -1):
if chain_type == 'stuff':
return load_summarize_chain(llm=llm,
chain_type=chain_type,
prompt=prompt,
token_max=token_max)
else:
return load_summarize_chain(llm=llm, chain_type=chain_type)
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
def run(self, *args, **kwargs):
return super().run(*args, **kwargs)
prompt_default = PromptTemplate(
input_variables=['image_desc'],
template="""Generate a detailed prompt to generate an image based on the following description:
{image_desc}""")
class DalleGeneratorChain(CustomChain):
"""Implementation of dall-e generate images"""
dalle: DallEAPIWrapper
llm: Optional[BaseLanguageModel]
prompt: Optional[PromptTemplate]
input_key: str = 'image_desc'
output_key: str = 'response' #: :meta private:
@property
def input_keys(self) -> List[str]:
"""Use this since so some prompt vars come from history."""
return [self.input_key]
@property
def output_keys(self) -> List[str]:
return [self.output_key]
@staticmethod
def function_name():
return 'DalleGeneratorChain'
@classmethod
def initialize(cls, dalle: DallEAPIWrapper, **kwargs):
return DalleGeneratorChain(dalle=dalle, **kwargs)
def run(self, *args, **kwargs):
return super().run(*args, **kwargs)
def _call(self,
inputs: Dict[str, Any],
run_manager: Optional[CallbackManagerForChainRun] = None) -> Dict[str, Any]:
if self.llm:
prompt = self.prompt or prompt_default
llm_chain = LLMChain(llm=self.llm, prompt=prompt)
pic_url = self.dalle.run(llm_chain.run(inputs)).split('\n')
else:
pic_url = self.dalle.run(inputs.get(self.input_key)).split('\n')
if len(pic_url) > 0:
pic_style = []
for url in pic_url:
pic_style.append(f'')
pic_url = ' \n'.join(pic_style)
return {self.output_key: pic_url, 'type': 'image'}
else:
return {self.output_key: f'', 'type': 'image'}
CUSTOM_CHAINS: Dict[str, Type[Union[ConversationChain, CustomChain]]] = {
'CombineDocsChain': CombineDocsChain,
'SummarizeDocsChain': SummarizeDocsChain,
'SeriesCharacterChain': SeriesCharacterChain,
'MidJourneyPromptChain': MidJourneyPromptChain,
'TimeTravelGuideChain': TimeTravelGuideChain,
'DalleGeneratorChain': DalleGeneratorChain,
}
================================================
FILE: src/backend/bisheng/interface/custom/__init__.py
================================================
from bisheng.interface.custom.base import CustomComponentCreator
from bisheng.interface.custom.custom_component import CustomComponent
__all__ = ['CustomComponentCreator', 'CustomComponent']
================================================
FILE: src/backend/bisheng/interface/custom/base.py
================================================
from typing import Any, Dict, List, Optional, Type
from bisheng.interface.base import LangChainTypeCreator
# from langflow.interface.custom.custom import CustomComponent
from bisheng.interface.custom.custom_component import CustomComponent
from bisheng.template.frontend_node.custom_components import CustomComponentFrontendNode
from loguru import logger
# Assuming necessary imports for Field, Template, and FrontendNode classes
class CustomComponentCreator(LangChainTypeCreator):
type_name: str = 'custom_components'
@property
def frontend_node_class(self) -> Type[CustomComponentFrontendNode]:
return CustomComponentFrontendNode
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict: dict[str, Any] = {
'CustomComponent': CustomComponent,
}
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
from bisheng.custom.customs import get_custom_nodes
try:
if name in get_custom_nodes(self.type_name).keys():
return get_custom_nodes(self.type_name)[name]
except ValueError as exc:
raise ValueError(f'CustomComponent {name} not found: {exc}') from exc
except AttributeError as exc:
logger.error(f'CustomComponent {name} not loaded: {exc}')
return None
return None
def to_list(self) -> List[str]:
return list(self.type_to_loader_dict.keys())
custom_component_creator = CustomComponentCreator()
================================================
FILE: src/backend/bisheng/interface/custom/code_parser/__init__.py
================================================
from .code_parser import CodeParser
__all__ = ['CodeParser']
================================================
FILE: src/backend/bisheng/interface/custom/code_parser/code_parser.py
================================================
import ast
import inspect
import operator
import traceback
from typing import Any, Dict, List, Type, Union
from bisheng.interface.custom.schema import CallableCodeDetails, ClassCodeDetails
from cachetools import TTLCache, cachedmethod, keys
from fastapi import HTTPException
class CodeSyntaxError(HTTPException):
pass
def get_data_type():
from bisheng.field_typing import Data
return Data
def imports_key(*args, **kwargs):
imports = kwargs.pop('imports')
key = keys.methodkey(*args, **kwargs)
key += tuple(imports)
return key
class CodeParser:
"""
A parser for Python source code, extracting code details.
"""
def __init__(self, code: Union[str, Type]) -> None:
"""
Initializes the parser with the provided code.
"""
self.cache: TTLCache = TTLCache(maxsize=1024, ttl=60)
if isinstance(code, type):
if not inspect.isclass(code):
raise ValueError('The provided code must be a class.')
# If the code is a class, get its source code
code = inspect.getsource(code)
self.code = code
self.data: Dict[str, Any] = {
'imports': [],
'functions': [],
'classes': [],
'global_vars': [],
}
self.handlers = {
ast.Import: self.parse_imports,
ast.ImportFrom: self.parse_imports,
ast.FunctionDef: self.parse_functions,
ast.ClassDef: self.parse_classes,
ast.Assign: self.parse_global_vars,
}
def get_tree(self):
"""
Parses the provided code to validate its syntax.
It tries to parse the code into an abstract syntax tree (AST).
"""
try:
tree = ast.parse(self.code)
except SyntaxError as err:
raise CodeSyntaxError(
status_code=400,
detail={'error': err.msg, 'traceback': traceback.format_exc()},
) from err
return tree
def parse_node(self, node: Union[ast.stmt, ast.AST]) -> None:
"""
Parses an AST node and updates the data
dictionary with the relevant information.
"""
if handler := self.handlers.get(type(node)): # type: ignore
handler(node) # type: ignore
def parse_imports(self, node: Union[ast.Import, ast.ImportFrom]) -> None:
"""
Extracts "imports" from the code, including aliases.
"""
if isinstance(node, ast.Import):
for alias in node.names:
if alias.asname:
self.data['imports'].append(f'{alias.name} as {alias.asname}')
else:
self.data['imports'].append(alias.name)
elif isinstance(node, ast.ImportFrom):
for alias in node.names:
if alias.asname:
self.data['imports'].append((node.module, f'{alias.name} as {alias.asname}'))
else:
self.data['imports'].append((node.module, alias.name))
def parse_functions(self, node: ast.FunctionDef) -> None:
"""
Extracts "functions" from the code.
"""
self.data['functions'].append(self.parse_callable_details(node))
def parse_arg(self, arg, default):
"""
Parses an argument and its default value.
"""
arg_dict = {'name': arg.arg, 'default': default}
if arg.annotation:
arg_dict['type'] = ast.unparse(arg.annotation)
return arg_dict
@cachedmethod(operator.attrgetter('cache'))
def construct_eval_env(self, return_type_str: str, imports) -> dict:
"""
Constructs an evaluation environment with the necessary imports for the return type,
taking into account module aliases.
"""
eval_env: dict = {}
for import_entry in imports:
if isinstance(import_entry, tuple): # from module import name
module, name = import_entry
if name in return_type_str:
exec(f'import {module}', eval_env)
exec(f'from {module} import {name}', eval_env)
else: # import module
module = import_entry
alias = None
if ' as ' in module:
module, alias = module.split(' as ')
if module in return_type_str or (alias and alias in return_type_str):
exec(f'import {module} as {alias if alias else module}', eval_env)
return eval_env
@cachedmethod(cache=operator.attrgetter('cache'))
def parse_callable_details(self, node: ast.FunctionDef) -> Dict[str, Any]:
"""
Extracts details from a single function or method node.
"""
return_type = None
if node.returns:
return_type_str = ast.unparse(node.returns)
eval_env = self.construct_eval_env(return_type_str, tuple(self.data['imports']))
try:
return_type = eval(return_type_str, eval_env)
except NameError:
# Handle cases where the type is not found in the constructed environment
pass
func = CallableCodeDetails(
name=node.name,
doc=ast.get_docstring(node),
args=self.parse_function_args(node),
body=self.parse_function_body(node),
return_type=return_type or get_data_type(),
has_return=self.parse_return_statement(node),
)
return func.model_dump()
def parse_function_args(self, node: ast.FunctionDef) -> List[Dict[str, Any]]:
"""
Parses the arguments of a function or method node.
"""
args = []
args += self.parse_positional_args(node)
args += self.parse_varargs(node)
args += self.parse_keyword_args(node)
# Commented out because we don't want kwargs
# showing up as fields in the frontend
# args += self.parse_kwargs(node)
return args
def parse_positional_args(self, node: ast.FunctionDef) -> List[Dict[str, Any]]:
"""
Parses the positional arguments of a function or method node.
"""
num_args = len(node.args.args)
num_defaults = len(node.args.defaults)
num_missing_defaults = num_args - num_defaults
missing_defaults = [None] * num_missing_defaults
default_values = [ast.unparse(default).strip("'") if default else None for default in node.args.defaults]
# Now check all default values to see if there
# are any "None" values in the middle
default_values = [None if value == 'None' else value for value in default_values]
defaults = missing_defaults + default_values
args = [self.parse_arg(arg, default) for arg, default in zip(node.args.args, defaults)]
return args
def parse_varargs(self, node: ast.FunctionDef) -> List[Dict[str, Any]]:
"""
Parses the *args argument of a function or method node.
"""
args = []
if node.args.vararg:
args.append(self.parse_arg(node.args.vararg, None))
return args
def parse_keyword_args(self, node: ast.FunctionDef) -> List[Dict[str, Any]]:
"""
Parses the keyword-only arguments of a function or method node.
"""
kw_defaults = [None] * (len(node.args.kwonlyargs) - len(node.args.kw_defaults)) + [
ast.unparse(default) if default else None for default in node.args.kw_defaults
]
args = [self.parse_arg(arg, default) for arg, default in zip(node.args.kwonlyargs, kw_defaults)]
return args
def parse_kwargs(self, node: ast.FunctionDef) -> List[Dict[str, Any]]:
"""
Parses the **kwargs argument of a function or method node.
"""
args = []
if node.args.kwarg:
args.append(self.parse_arg(node.args.kwarg, None))
return args
def parse_function_body(self, node: ast.FunctionDef) -> List[str]:
"""
Parses the body of a function or method node.
"""
return [ast.unparse(line) for line in node.body]
def parse_return_statement(self, node: ast.FunctionDef) -> bool:
"""
Parses the return statement of a function or method node.
"""
return any(isinstance(n, ast.Return) for n in node.body)
def parse_assign(self, stmt):
"""
Parses an Assign statement and returns a dictionary
with the target's name and value.
"""
for target in stmt.targets:
if isinstance(target, ast.Name):
return {'name': target.id, 'value': ast.unparse(stmt.value)}
def parse_ann_assign(self, stmt):
"""
Parses an AnnAssign statement and returns a dictionary
with the target's name, value, and annotation.
"""
if isinstance(stmt.target, ast.Name):
return {
'name': stmt.target.id,
'value': ast.unparse(stmt.value) if stmt.value else None,
'annotation': ast.unparse(stmt.annotation),
}
def parse_function_def(self, stmt):
"""
Parses a FunctionDef statement and returns the parsed
method and a boolean indicating if it's an __init__ method.
"""
method = self.parse_callable_details(stmt)
return (method, True) if stmt.name == '__init__' else (method, False)
def parse_classes(self, node: ast.ClassDef) -> None:
"""
Extracts "classes" from the code, including inheritance and init methods.
"""
class_details = ClassCodeDetails(
name=node.name,
doc=ast.get_docstring(node),
bases=[ast.unparse(base) for base in node.bases],
attributes=[],
methods=[],
init=None,
)
for stmt in node.body:
if isinstance(stmt, ast.Assign):
if attr := self.parse_assign(stmt):
class_details.attributes.append(attr)
elif isinstance(stmt, ast.AnnAssign):
if attr := self.parse_ann_assign(stmt):
class_details.attributes.append(attr)
elif isinstance(stmt, (ast.FunctionDef, ast.AsyncFunctionDef)):
method, is_init = self.parse_function_def(stmt)
if is_init:
class_details.init = method
else:
class_details.methods.append(method)
self.data['classes'].append(class_details.model_dump())
def parse_global_vars(self, node: ast.Assign) -> None:
"""
Extracts global variables from the code.
"""
global_var = {
'targets': [t.id if hasattr(t, 'id') else ast.dump(t) for t in node.targets],
'value': ast.unparse(node.value),
}
self.data['global_vars'].append(global_var)
def parse_code(self) -> Dict[str, Any]:
"""
Runs all parsing operations and returns the resulting data.
"""
tree = self.get_tree()
for node in ast.walk(tree):
self.parse_node(node)
return self.data
================================================
FILE: src/backend/bisheng/interface/custom/code_parser/utils.py
================================================
import re
from types import GenericAlias
from typing import Any
def extract_inner_type(return_type: str) -> str:
"""
Extracts the inner type from a type hint that is a list.
"""
if match := re.match(r'list\[(.*)\]', return_type, re.IGNORECASE):
return match[1]
return return_type
def extract_inner_type_from_generic_alias(return_type: GenericAlias) -> Any:
"""
Extracts the inner type from a type hint that is a list.
"""
if return_type.__origin__ == list:
return list(return_type.__args__)
return return_type
def extract_union_types(return_type: str) -> list[str]:
"""
Extracts the inner type from a type hint that is a list.
"""
# If the return type is a Union, then we need to parse it
return_type = return_type.replace('Union', '').replace('[', '').replace(']', '')
return_types = return_type.split(',')
return [item.strip() for item in return_types]
def extract_union_types_from_generic_alias(return_type: GenericAlias) -> list:
"""
Extracts the inner type from a type hint that is a Union.
"""
return list(return_type.__args__)
================================================
FILE: src/backend/bisheng/interface/custom/custom_component/__init__.py
================================================
from .custom_component import CustomComponent
__all__ = ['CustomComponent']
================================================
FILE: src/backend/bisheng/interface/custom/custom_component/component.py
================================================
import ast
import operator
import warnings
from typing import Any, ClassVar, Optional
import emoji
from bisheng.interface.custom.code_parser import CodeParser
from bisheng.utils import validate
from cachetools import TTLCache, cachedmethod
from fastapi import HTTPException
class ComponentCodeNullError(HTTPException):
pass
class ComponentFunctionEntrypointNameNullError(HTTPException):
pass
class Component:
ERROR_CODE_NULL: ClassVar[str] = 'Python code must be provided.'
ERROR_FUNCTION_ENTRYPOINT_NAME_NULL: ClassVar[str] = 'The name of the entrypoint function must be provided.'
code: Optional[str] = None
_function_entrypoint_name: str = 'build'
field_config: dict = {}
_user_id: Optional[str]
def __init__(self, **data):
self.cache = TTLCache(maxsize=1024, ttl=60)
for key, value in data.items():
if key == 'user_id':
setattr(self, '_user_id', value)
else:
setattr(self, key, value)
# Validate the emoji at the icon field
if hasattr(self, 'icon') and self.icon:
self.icon = self.validate_icon(self.icon)
def __setattr__(self, key, value):
if key == '_user_id' and hasattr(self, '_user_id'):
warnings.warn('user_id is immutable and cannot be changed.')
super().__setattr__(key, value)
@cachedmethod(cache=operator.attrgetter('cache'))
def get_code_tree(self, code: str):
parser = CodeParser(code)
return parser.parse_code()
def get_function(self):
if not self.code:
raise ComponentCodeNullError(
status_code=400,
detail={'error': self.ERROR_CODE_NULL, 'traceback': ''},
)
if not self._function_entrypoint_name:
raise ComponentFunctionEntrypointNameNullError(
status_code=400,
detail={
'error': self.ERROR_FUNCTION_ENTRYPOINT_NAME_NULL,
'traceback': '',
},
)
return validate.create_function(self.code, self._function_entrypoint_name)
def build_template_config(self, attributes) -> dict:
template_config = {}
for item in attributes:
item_name = item.get('name')
if item_value := item.get('value'):
if 'display_name' in item_name:
template_config['display_name'] = ast.literal_eval(item_value)
elif 'description' in item_name:
template_config['description'] = ast.literal_eval(item_value)
elif 'beta' in item_name:
template_config['beta'] = ast.literal_eval(item_value)
elif 'documentation' in item_name:
template_config['documentation'] = ast.literal_eval(item_value)
elif 'icon' in item_name:
icon_str = ast.literal_eval(item_value)
template_config['icon'] = self.validate_icon(icon_str)
return template_config
def validate_icon(self, value: str):
# we are going to use the emoji library to validate the emoji
# emojis can be defined using the :emoji_name: syntax
if not value.startswith(':') or not value.endswith(':'):
warnings.warn('Invalid emoji. Please use the :emoji_name: syntax.')
return value
emoji_value = emoji.emojize(value, variant='emoji_type')
if value == emoji_value:
warnings.warn(f'Invalid emoji. {value} is not a valid emoji.')
return value
return emoji_value
def build(self, *args: Any, **kwargs: Any) -> Any:
raise NotImplementedError
================================================
FILE: src/backend/bisheng/interface/custom/custom_component/custom_component.py
================================================
import operator
from typing import Any, Callable, ClassVar, List, Optional, Union
from uuid import UUID
import yaml
from bisheng.database.models.flow import Flow, FlowDao
from bisheng.interface.custom.code_parser.utils import (extract_inner_type_from_generic_alias,
extract_union_types_from_generic_alias)
from bisheng.interface.custom.custom_component.component import Component
from bisheng.utils import validate
from cachetools import TTLCache, cachedmethod
from fastapi import HTTPException
class CustomComponent(Component):
display_name: Optional[str] = None
"""The display name of the component. Defaults to None."""
description: Optional[str] = None
"""The description of the component. Defaults to None."""
icon: Optional[str] = None
"""The icon of the component. It should be an emoji. Defaults to None."""
code: Optional[str] = None
"""The code of the component. Defaults to None."""
field_config: dict = {}
"""The field configuration of the component. Defaults to an empty dictionary."""
field_order: Optional[List[str]] = None
"""The field order of the component. Defaults to an empty list."""
code_class_base_inheritance: ClassVar[str] = 'CustomComponent'
function_entrypoint_name: ClassVar[str] = 'build'
function: Optional[Callable] = None
repr_value: Optional[Any] = ''
user_id: Optional[Union[UUID, str]] = None
status: Optional[Any] = None
"""The status of the component. This is displayed on the frontend. Defaults to None."""
_tree: Optional[dict] = None
def __init__(self, **data):
self.cache = TTLCache(maxsize=1024, ttl=60)
super().__init__(**data)
def _get_field_order(self):
return self.field_order or list(self.field_config.keys())
def custom_repr(self):
if self.repr_value == '':
self.repr_value = self.status
if isinstance(self.repr_value, dict):
return yaml.dump(self.repr_value)
if isinstance(self.repr_value, str):
return self.repr_value
return str(self.repr_value)
def build_config(self):
return self.field_config
@property
def tree(self):
return self.get_code_tree(self.code or '')
@property
def get_function_entrypoint_args(self) -> list:
build_method = self.get_build_method()
if not build_method:
return []
args = build_method['args']
for arg in args:
if arg.get('type') == 'prompt':
raise HTTPException(
status_code=400,
detail={
'error': 'Type hint Error',
'traceback': (
'Prompt type is not supported in the build method.' ' Try using PromptTemplate instead.'
),
},
)
elif not arg.get('type') and arg.get('name') != 'self':
# Set the type to Data
arg['type'] = 'Data'
return args
@cachedmethod(operator.attrgetter('cache'))
def get_build_method(self):
if not self.code:
return {}
component_classes = [cls for cls in self.tree['classes'] if self.code_class_base_inheritance in cls['bases']]
if not component_classes:
return {}
# Assume the first Component class is the one we're interested in
component_class = component_classes[0]
build_methods = [
method for method in component_class['methods'] if method['name'] == self.function_entrypoint_name
]
return build_methods[0] if build_methods else {}
@property
def get_function_entrypoint_return_type(self) -> List[Any]:
build_method = self.get_build_method()
if not build_method or not build_method.get('has_return'):
return []
return_type = build_method['return_type']
# If list or List is in the return type, then we remove it and return the inner type
if hasattr(return_type, '__origin__') and return_type.__origin__ in [list, List]:
return_type = extract_inner_type_from_generic_alias(return_type)
# If the return type is not a Union, then we just return it as a list
if not hasattr(return_type, '__origin__') or return_type.__origin__ != Union:
return return_type if isinstance(return_type, list) else [return_type]
# If the return type is a Union, then we need to parse itx
return_type = extract_union_types_from_generic_alias(return_type)
return return_type
@property
def get_main_class_name(self):
if not self.code:
return ''
base_name = self.code_class_base_inheritance
method_name = self.function_entrypoint_name
classes = []
for item in self.tree.get('classes', []):
if base_name in item['bases']:
method_names = [method['name'] for method in item['methods']]
if method_name in method_names:
classes.append(item['name'])
# Get just the first item
return next(iter(classes), '')
@property
def template_config(self):
return self.build_template_config()
def build_template_config(self):
if not self.code:
return {}
attributes = [
main_class['attributes']
for main_class in self.tree.get('classes', [])
if main_class['name'] == self.get_main_class_name
]
# Get just the first item
attributes = next(iter(attributes), [])
return super().build_template_config(attributes)
def index(self, value: int = 0):
"""Returns a function that returns the value at the given index in the iterable."""
def get_index(iterable: List[Any]):
return iterable[value] if iterable else iterable
return get_index
@property
def get_function(self):
return validate.create_function(self.code, self.function_entrypoint_name)
async def load_flow(self, flow_id: str, tweaks: Optional[dict] = None) -> Any:
from bisheng.processing.process import build_sorted_vertices, process_tweaks
flow = FlowDao.get_flow_by_id(flow_id)
graph_data = flow.data if flow else None
if not graph_data:
raise ValueError(f'Flow {flow_id} not found')
if tweaks:
graph_data = process_tweaks(graph_data=graph_data, tweaks=tweaks)
return await build_sorted_vertices(graph_data, self.user_id)
def list_flows(self) -> List[Flow]:
if not self._user_id:
raise ValueError('Session is invalid')
try:
return FlowDao.get_flow_by_user(int(self._user_id))
except Exception as e:
raise ValueError('Session is invalid') from e
async def get_flow(
self,
*,
flow_name: Optional[str] = None,
flow_id: Optional[str] = None,
tweaks: Optional[dict] = None,
) -> Flow:
if flow_id:
flow = FlowDao.get_flow_by_id(flow_id)
elif flow_name:
flow = FlowDao.get_flow_by_name(int(self._user_id), flow_name)
else:
raise ValueError('Either flow_name or flow_id must be provided')
if not flow:
raise ValueError(f'Flow {flow_name or flow_id} not found')
return await self.load_flow(flow.id, tweaks)
def build(self, *args: Any, **kwargs: Any) -> Any:
raise NotImplementedError
================================================
FILE: src/backend/bisheng/interface/custom/directory_reader/__init__.py
================================================
from .directory_reader import DirectoryReader
__all__ = ['DirectoryReader']
================================================
FILE: src/backend/bisheng/interface/custom/directory_reader/directory_reader.py
================================================
import ast
import os
import zlib
from bisheng.interface.custom.custom_component import CustomComponent
from loguru import logger
class CustomComponentPathValueError(ValueError):
pass
class StringCompressor:
def __init__(self, input_string):
"""Initialize StringCompressor with a string to compress."""
self.input_string = input_string
def compress_string(self):
"""
Compress the initial string and return the compressed data.
"""
# Convert string to bytes
byte_data = self.input_string.encode('utf-8')
# Compress the bytes
self.compressed_data = zlib.compress(byte_data)
return self.compressed_data
def decompress_string(self):
"""
Decompress the compressed data and return the original string.
"""
# Decompress the bytes
decompressed_data = zlib.decompress(self.compressed_data)
# Convert bytes back to string
return decompressed_data.decode('utf-8')
class DirectoryReader:
# Ensure the base path to read the files that contain
# the custom components from this directory.
base_path = ''
def __init__(self, directory_path, compress_code_field=False):
"""
Initialize DirectoryReader with a directory path
and a flag indicating whether to compress the code.
"""
self.directory_path = directory_path
self.compress_code_field = compress_code_field
def get_safe_path(self):
"""Check if the path is valid and return it, or None if it's not."""
return self.directory_path if self.is_valid_path() else None
def is_valid_path(self) -> bool:
"""Check if the directory path is valid by comparing it to the base path."""
fullpath = os.path.normpath(os.path.join(self.directory_path))
return fullpath.startswith(self.base_path)
def is_empty_file(self, file_content):
"""
Check if the file content is empty.
"""
return len(file_content.strip()) == 0
def filter_loaded_components(self, data: dict, with_errors: bool) -> dict:
from bisheng.interface.custom.utils import build_component
items = []
for menu in data['menu']:
components = []
for component in menu['components']:
try:
if component['error'] if with_errors else not component['error']:
component_tuple = (*build_component(component), component)
components.append(component_tuple)
except Exception as e:
logger.error(f'Error while loading component: {e}')
continue
items.append({'name': menu['name'], 'path': menu['path'], 'components': components})
filtered = [menu for menu in items if menu['components']]
logger.debug(f'Filtered components {"with errors" if with_errors else ""}: {len(filtered)}')
return {'menu': filtered}
def validate_code(self, file_content):
"""
Validate the Python code by trying to parse it with ast.parse.
"""
try:
ast.parse(file_content)
return True
except SyntaxError:
return False
def validate_build(self, file_content):
"""
Check if the file content contains a function named 'build'.
"""
return 'def build' in file_content
def read_file_content(self, file_path):
"""
Read and return the content of a file.
"""
if not os.path.isfile(file_path):
return None
with open(file_path, 'r') as file:
return file.read()
def get_files(self):
"""
Walk through the directory path and return a list of all .py files.
"""
if not (safe_path := self.get_safe_path()):
raise CustomComponentPathValueError(f"The path needs to start with '{self.base_path}'.")
file_list = []
for root, _, files in os.walk(safe_path):
file_list.extend(
os.path.join(root, filename)
for filename in files
if filename.endswith('.py') and not filename.startswith('__')
)
return file_list
def find_menu(self, response, menu_name):
"""
Find and return a menu by its name in the response.
"""
return next(
(menu for menu in response['menu'] if menu['name'] == menu_name),
None,
)
def _is_type_hint_imported(self, type_hint_name: str, code: str) -> bool:
"""
Check if a specific type hint is imported
from the typing module in the given code.
"""
module = ast.parse(code)
return any(
isinstance(node, ast.ImportFrom)
and node.module == 'typing'
and any(alias.name == type_hint_name for alias in node.names)
for node in ast.walk(module)
)
def _is_type_hint_used_in_args(self, type_hint_name: str, code: str) -> bool:
"""
Check if a specific type hint is used in the
function definitions within the given code.
"""
try:
module = ast.parse(code)
for node in ast.walk(module):
if isinstance(node, ast.FunctionDef):
for arg in node.args.args:
if self._is_type_hint_in_arg_annotation(arg.annotation, type_hint_name):
return True
except SyntaxError:
# Returns False if the code is not valid Python
return False
return False
def _is_type_hint_in_arg_annotation(self, annotation, type_hint_name: str) -> bool:
"""
Helper function to check if a type hint exists in an annotation.
"""
return (
annotation is not None
and isinstance(annotation, ast.Subscript)
and isinstance(annotation.value, ast.Name)
and annotation.value.id == type_hint_name
)
def is_type_hint_used_but_not_imported(self, type_hint_name: str, code: str) -> bool:
"""
Check if a type hint is used but not imported in the given code.
"""
try:
return self._is_type_hint_used_in_args(type_hint_name, code) and not self._is_type_hint_imported(
type_hint_name, code
)
except SyntaxError:
# Returns True if there's something wrong with the code
# TODO : Find a better way to handle this
return True
def process_file(self, file_path):
"""
Process a file by validating its content and
returning the result and content/error message.
"""
file_content = self.read_file_content(file_path)
if file_content is None:
return False, f'Could not read {file_path}'
elif self.is_empty_file(file_content):
return False, 'Empty file'
elif not self.validate_code(file_content):
return False, 'Syntax error'
elif not self.validate_build(file_content):
return False, 'Missing build function'
elif self._is_type_hint_used_in_args('Optional', file_content) and not self._is_type_hint_imported(
'Optional', file_content
):
return (
False,
"Type hint 'Optional' is used but not imported in the code.",
)
else:
if self.compress_code_field:
file_content = str(StringCompressor(file_content).compress_string())
return True, file_content
def build_component_menu_list(self, file_paths):
"""
Build a list of menus with their components
from the .py files in the directory.
"""
response = {'menu': []}
logger.debug('-------------------- Building component menu list --------------------')
for file_path in file_paths:
menu_name = os.path.basename(os.path.dirname(file_path))
logger.debug(f'Menu name: {menu_name}')
filename = os.path.basename(file_path)
validation_result, result_content = self.process_file(file_path)
logger.debug(f'Validation result: {validation_result}')
menu_result = self.find_menu(response, menu_name) or {
'name': menu_name,
'path': os.path.dirname(file_path),
'components': [],
}
component_name = filename.split('.')[0]
# This is the name of the file which will be displayed in the UI
# We need to change it from snake_case to CamelCase
# first check if it's already CamelCase
if '_' in component_name:
component_name_camelcase = ' '.join(word.title() for word in component_name.split('_'))
else:
component_name_camelcase = component_name
if validation_result:
try:
output_types = self.get_output_types_from_code(result_content)
except Exception as exc:
logger.exception(f'Error while getting output types from code: {str(exc)}')
output_types = [component_name_camelcase]
else:
output_types = [component_name_camelcase]
component_info = {
'name': 'CustomComponent',
'output_types': output_types,
'file': filename,
'code': result_content if validation_result else '',
'error': '' if validation_result else result_content,
}
menu_result['components'].append(component_info)
logger.debug(f'Component info: {component_info}')
if menu_result not in response['menu']:
response['menu'].append(menu_result)
logger.debug('-------------------- Component menu list built --------------------')
return response
@staticmethod
def get_output_types_from_code(code: str) -> list:
"""
Get the output types from the code.
"""
custom_component = CustomComponent(code=code)
types_list = custom_component.get_function_entrypoint_return_type
# Get the name of types classes
return [type_.__name__ for type_ in types_list if hasattr(type_, '__name__')]
================================================
FILE: src/backend/bisheng/interface/custom/directory_reader/utils.py
================================================
from bisheng.interface.custom.directory_reader import DirectoryReader
from bisheng.template.frontend_node.custom_components import CustomComponentFrontendNode
from loguru import logger
def merge_nested_dicts_with_renaming(dict1, dict2):
for key, value in dict2.items():
if key in dict1 and isinstance(value, dict) and isinstance(dict1.get(key), dict):
for sub_key, sub_value in value.items():
# if sub_key in dict1[key]:
# new_key = get_new_key(dict1[key], sub_key)
# dict1[key][new_key] = sub_value
# else:
dict1[key][sub_key] = sub_value
else:
dict1[key] = value
return dict1
def build_invalid_menu(invalid_components):
"""Build the invalid menu."""
if not invalid_components.get('menu'):
return {}
logger.debug('------------------- INVALID COMPONENTS -------------------')
invalid_menu = {}
for menu_item in invalid_components['menu']:
menu_name = menu_item['name']
invalid_menu[menu_name] = build_invalid_menu_items(menu_item)
return invalid_menu
def build_valid_menu(valid_components):
"""Build the valid menu."""
valid_menu = {}
logger.debug('------------------- VALID COMPONENTS -------------------')
for menu_item in valid_components['menu']:
menu_name = menu_item['name']
valid_menu[menu_name] = build_menu_items(menu_item)
return valid_menu
def build_and_validate_all_files(reader: DirectoryReader, file_list):
"""Build and validate all files"""
data = reader.build_component_menu_list(file_list)
valid_components = reader.filter_loaded_components(data=data, with_errors=False)
invalid_components = reader.filter_loaded_components(data=data, with_errors=True)
return valid_components, invalid_components
def load_files_from_path(path: str):
"""Load all files from a given path"""
reader = DirectoryReader(path, False)
return reader.get_files()
def build_custom_component_list_from_path(path: str):
"""Build a list of custom components for the langchain from a given path"""
file_list = load_files_from_path(path)
reader = DirectoryReader(path, False)
valid_components, invalid_components = build_and_validate_all_files(reader, file_list)
valid_menu = build_valid_menu(valid_components)
invalid_menu = build_invalid_menu(invalid_components)
return merge_nested_dicts_with_renaming(valid_menu, invalid_menu)
def create_invalid_component_template(component, component_name):
"""Create a template for an invalid component."""
component_code = component['code']
component_frontend_node = CustomComponentFrontendNode(
description='ERROR - Check your Python Code',
display_name=f'ERROR - {component_name}',
)
component_frontend_node.error = component.get('error', None)
field = component_frontend_node.template.get_field('code')
field.value = component_code
component_frontend_node.template.update_field('code', field)
return component_frontend_node.model_dump(by_alias=True, exclude_none=True)
def log_invalid_component_details(component):
"""Log details of an invalid component."""
logger.debug(component)
logger.debug(f"Component Path: {component.get('path', None)}")
logger.debug(f"Component Error: {component.get('error', None)}")
def build_invalid_component(component):
"""Build a single invalid component."""
component_name = component['name']
component_template = create_invalid_component_template(component, component_name)
log_invalid_component_details(component)
return component_name, component_template
def build_invalid_menu_items(menu_item):
"""Build invalid menu items for a given menu."""
menu_items = {}
for component in menu_item['components']:
try:
component_name, component_template = build_invalid_component(component)
menu_items[component_name] = component_template
logger.debug(f'Added {component_name} to invalid menu.')
except Exception as exc:
logger.exception(f'Error while creating custom component [{component_name}]: {str(exc)}')
return menu_items
def get_new_key(dictionary, original_key):
counter = 1
new_key = original_key + ' (' + str(counter) + ')'
while new_key in dictionary:
counter += 1
new_key = original_key + ' (' + str(counter) + ')'
return new_key
def determine_component_name(component):
"""Determine the name of the component."""
component_output_types = component['output_types']
if len(component_output_types) == 1:
return component_output_types[0]
else:
file_name = component.get('file').split('.')[0]
return ''.join(word.capitalize() for word in file_name.split('_')) if '_' in file_name else file_name
def build_menu_items(menu_item):
"""Build menu items for a given menu."""
menu_items = {}
for component_name, component_template, component in menu_item['components']:
try:
menu_items[component_name] = component_template
logger.debug(f'Added {component_name} to valid menu.')
except Exception as exc:
logger.error(f"Error loading Component: {component['output_types']}")
logger.exception(f"Error while building custom component {component['output_types']}: {exc}")
return menu_items
================================================
FILE: src/backend/bisheng/interface/custom/schema.py
================================================
from typing import Any, Optional
from pydantic import BaseModel, Field
class ClassCodeDetails(BaseModel):
"""
A dataclass for storing details about a class.
"""
name: str
doc: Optional[str] = None
bases: list
attributes: list
methods: list
init: Optional[dict] = Field(default_factory=dict)
class CallableCodeDetails(BaseModel):
"""
A dataclass for storing details about a callable.
"""
name: str
doc: Optional[str] = None
args: list
body: list
return_type: Optional[Any] = None
has_return: bool = False
================================================
FILE: src/backend/bisheng/interface/custom/utils.py
================================================
import ast
import contextlib
import re
import traceback
import warnings
from typing import Any, Dict, List, Optional, Union
from uuid import UUID
from bisheng.field_typing.range_spec import RangeSpec
from bisheng.interface.custom.code_parser.utils import extract_inner_type
from bisheng.interface.custom.custom_component import CustomComponent
from bisheng.interface.custom.directory_reader.utils import (build_custom_component_list_from_path,
determine_component_name,
merge_nested_dicts_with_renaming)
from bisheng.interface.importing.utils import eval_custom_component_code
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.custom_components import CustomComponentFrontendNode
from bisheng.utils.util import get_base_classes
from fastapi import HTTPException
from loguru import logger
def add_output_types(frontend_node: CustomComponentFrontendNode, return_types: List[str]):
"""Add output types to the frontend node"""
for return_type in return_types:
if return_type is None:
raise HTTPException(
status_code=400,
detail={
'error': ('Invalid return type. Please check your code and try again.'),
'traceback': traceback.format_exc(),
},
)
if hasattr(return_type, '__name__'):
return_type = return_type.__name__
elif hasattr(return_type, '__class__'):
return_type = return_type.__class__.__name__
else:
return_type = str(return_type)
frontend_node.add_output_type(return_type)
def reorder_fields(frontend_node: CustomComponentFrontendNode, field_order: List[str]):
"""Reorder fields in the frontend node based on the specified field_order."""
if not field_order:
return
# Create a dictionary for O(1) lookup time.
field_dict = {field.name: field for field in frontend_node.template.fields}
reordered_fields = [field_dict[name] for name in field_order if name in field_dict]
# Add any fields that are not in the field_order list
for field in frontend_node.template.fields:
if field.name not in field_order:
reordered_fields.append(field)
frontend_node.template.fields = reordered_fields
def add_base_classes(frontend_node: CustomComponentFrontendNode, return_types: List[str]):
"""Add base classes to the frontend node"""
for return_type_instance in return_types:
if return_type_instance is None:
raise HTTPException(
status_code=400,
detail={
'error': ('Invalid return type. Please check your code and try again.'),
'traceback': traceback.format_exc(),
},
)
base_classes = get_base_classes(return_type_instance)
for base_class in base_classes:
frontend_node.add_base_class(base_class)
def extract_type_from_optional(field_type):
"""
Extract the type from a string formatted as "Optional[]".
Parameters:
field_type (str): The string from which to extract the type.
Returns:
str: The extracted type, or an empty string if no type was found.
"""
match = re.search(r'\[(.*?)\]$', field_type)
return match[1] if match else None
def get_field_properties(extra_field):
"""Get the properties of an extra field"""
field_name = extra_field['name']
field_type = extra_field.get('type', 'str')
field_value = extra_field.get('default', '')
field_required = 'optional' not in field_type.lower()
if not field_required:
field_type = extract_type_from_optional(field_type)
if field_value is not None:
with contextlib.suppress(Exception):
field_value = ast.literal_eval(field_value)
return field_name, field_type, field_value, field_required
def process_type(field_type: str):
if field_type.startswith('list') or field_type.startswith('List'):
return extract_inner_type(field_type)
return 'prompt' if field_type == 'Prompt' else field_type
def add_new_custom_field(
frontend_node: CustomComponentFrontendNode,
field_name: str,
field_type: str,
field_value: Any,
field_required: bool,
field_config: dict,
):
# Check field_config if any of the keys are in it
# if it is, update the value
display_name = field_config.pop('display_name', None)
field_type = field_config.pop('field_type', field_type)
field_contains_list = 'list' in field_type.lower()
field_type = process_type(field_type)
field_value = field_config.pop('value', field_value)
field_advanced = field_config.pop('advanced', False)
if field_type == 'bool' and field_value is None:
field_value = False
# If options is a list, then it's a dropdown
# If options is None, then it's a list of strings
is_list = isinstance(field_config.get('options'), list)
field_config['is_list'] = is_list or field_config.get('is_list', False) or field_contains_list
if 'name' in field_config:
warnings.warn("The 'name' key in field_config is used to build the object and can't be changed.")
required = field_config.pop('required', field_required)
placeholder = field_config.pop('placeholder', '')
new_field = TemplateField(
name=field_name,
field_type=field_type,
value=field_value,
show=True,
required=required,
advanced=field_advanced,
placeholder=placeholder,
display_name=display_name,
**sanitize_field_config(field_config),
)
frontend_node.template.upsert_field(field_name, new_field)
if isinstance(frontend_node.custom_fields, dict):
frontend_node.custom_fields[field_name] = None
return frontend_node
def add_extra_fields(frontend_node, field_config, function_args):
"""Add extra fields to the frontend node"""
if not function_args:
return
for extra_field in function_args:
if 'name' not in extra_field or extra_field['name'] == 'self':
continue
field_name, field_type, field_value, field_required = get_field_properties(extra_field)
config = field_config.get(field_name, {})
frontend_node = add_new_custom_field(
frontend_node,
field_name,
field_type,
field_value,
field_required,
config,
)
def get_field_dict(field: Union[TemplateField, dict]):
"""Get the field dictionary from a TemplateField or a dict"""
if isinstance(field, TemplateField):
return field.model_dump(by_alias=True, exclude_none=True)
return field
def run_build_config(
custom_component: CustomComponent,
user_id: Optional[Union[str, UUID]] = None,
update_field=None,
):
"""Build the field configuration for a custom component"""
try:
if custom_component.code is None:
return {}
elif isinstance(custom_component.code, str):
custom_class = eval_custom_component_code(custom_component.code)
else:
raise ValueError('Invalid code type')
except Exception as exc:
logger.error(f'Error while evaluating custom component code: {str(exc)}')
raise HTTPException(
status_code=400,
detail={
'error': ('Invalid type convertion. Please check your code and try again.'),
'traceback': traceback.format_exc(),
},
) from exc
try:
custom_instance = custom_class(user_id=user_id)
build_config: Dict = custom_instance.build_config()
for field_name, field in build_config.items():
# Allow user to build TemplateField as well
# as a dict with the same keys as TemplateField
field_dict = get_field_dict(field)
if update_field is not None and field_name != update_field:
continue
try:
update_field_dict(field_dict)
build_config[field_name] = field_dict
except Exception as exc:
logger.error(f'Error while getting build_config: {str(exc)}')
return build_config, custom_instance
except Exception as exc:
logger.error(f'Error while building field config: {str(exc)}')
raise HTTPException(
status_code=400,
detail={
'error': ('Invalid type convertion. Please check your code and try again.'),
'traceback': traceback.format_exc(),
},
) from exc
def sanitize_template_config(template_config):
"""Sanitize the template config"""
attributes = {
'display_name',
'description',
'beta',
'documentation',
'output_types',
'icon',
}
for key in template_config.copy():
if key not in attributes:
template_config.pop(key, None)
return template_config
def build_frontend_node(template_config):
"""Build a frontend node for a custom component"""
try:
sanitized_template_config = sanitize_template_config(template_config)
return CustomComponentFrontendNode(**sanitized_template_config)
except Exception as exc:
logger.error(f'Error while building base frontend node: {exc}')
raise exc
def add_code_field(frontend_node: CustomComponentFrontendNode, raw_code, field_config):
code_field = TemplateField(
dynamic=True,
required=True,
placeholder='',
multiline=True,
value=raw_code,
password=False,
name='code',
advanced=field_config.pop('advanced', False),
field_type='code',
is_list=False,
)
frontend_node.template.add_field(code_field)
return frontend_node
def build_custom_component_template(
custom_component: CustomComponent,
user_id: Optional[Union[str, UUID]] = None,
update_field: Optional[str] = None,
) -> Optional[Dict[str, Any]]:
"""Build a custom component template for the langchain"""
try:
logger.debug('Building custom component template')
frontend_node = build_frontend_node(custom_component.template_config)
logger.debug('Built base frontend node')
logger.debug('Updated attributes')
field_config, custom_instance = run_build_config(custom_component, user_id=user_id, update_field=update_field)
logger.debug('Built field config')
entrypoint_args = custom_component.get_function_entrypoint_args
add_extra_fields(frontend_node, field_config, entrypoint_args)
logger.debug('Added extra fields')
frontend_node = add_code_field(frontend_node, custom_component.code, field_config.get('code', {}))
logger.debug('Added code field')
add_base_classes(frontend_node, custom_component.get_function_entrypoint_return_type)
add_output_types(frontend_node, custom_component.get_function_entrypoint_return_type)
logger.debug('Added base classes')
reorder_fields(frontend_node, custom_instance._get_field_order())
return frontend_node.to_dict(add_name=False)
except Exception as exc:
if isinstance(exc, HTTPException):
raise exc
raise HTTPException(
status_code=400,
detail={
'error': ('Invalid type convertion. Please check your code and try again.'),
'traceback': traceback.format_exc(),
},
) from exc
def create_component_template(component):
"""Create a template for a component."""
component_code = component['code']
component_output_types = component['output_types']
component_extractor = CustomComponent(code=component_code)
component_template = build_custom_component_template(component_extractor)
component_template['output_types'] = component_output_types
return component_template
def build_custom_components(components_path: List[str]):
"""Build custom components from the specified paths."""
if not components_path:
return {}
logger.info(f'Building custom components from {components_path}')
custom_components_from_file = {}
processed_paths = set()
for path in components_path:
path_str = str(path)
if path_str in processed_paths:
continue
custom_component_dict = build_custom_component_list_from_path(path_str)
if custom_component_dict:
category = next(iter(custom_component_dict))
logger.info(f'Loading {len(custom_component_dict[category])} component(s) from category {category}')
custom_components_from_file = merge_nested_dicts_with_renaming(
custom_components_from_file, custom_component_dict
)
processed_paths.add(path_str)
return custom_components_from_file
def update_field_dict(field_dict):
"""Update the field dictionary by calling options() or value() if they are callable"""
if 'options' in field_dict and callable(field_dict['options']):
field_dict['options'] = field_dict['options']()
# Also update the "refresh" key
field_dict['refresh'] = True
if 'value' in field_dict and callable(field_dict['value']):
field_dict['value'] = field_dict['value']()
field_dict['refresh'] = True
# Let's check if "range_spec" is a RangeSpec object
if 'rangeSpec' in field_dict and isinstance(field_dict['rangeSpec'], RangeSpec):
field_dict['rangeSpec'] = field_dict['rangeSpec'].model_dump()
def sanitize_field_config(field_config: Dict):
# If any of the already existing keys are in field_config, remove them
for key in [
'name',
'field_type',
'value',
'required',
'placeholder',
'display_name',
'advanced',
'show',
]:
field_config.pop(key, None)
return field_config
def build_component(component):
"""Build a single component."""
logger.debug(f"Building component: {component.get('name'), component.get('output_types')}")
component_name = determine_component_name(component)
component_template = create_component_template(component)
return component_name, component_template
================================================
FILE: src/backend/bisheng/interface/custom_lists.py
================================================
import inspect
from typing import Any
from langchain import llms, memory, text_splitter
from langchain_anthropic import ChatAnthropic
from langchain_community import agent_toolkits, document_loaders, embeddings
from langchain_community.chat_models import ChatVertexAI, MiniMaxChat, ChatTongyi, ChatZhipuAI, \
ChatHunyuan, MoonshotChat
from langchain_community.utilities import requests
from langchain_deepseek import ChatDeepSeek
from langchain_openai import AzureChatOpenAI, ChatOpenAI, OpenAIEmbeddings, AzureOpenAIEmbeddings, OpenAI
from bisheng.interface.agents.custom import CUSTOM_AGENTS
from bisheng.interface.chains.custom import CUSTOM_CHAINS
from bisheng.interface.embeddings.custom import CUSTOM_EMBEDDING
from bisheng.interface.importing.utils import import_class
from bisheng_langchain import chat_models
from bisheng_langchain import document_loaders as contribute_loader
from bisheng_langchain import embeddings as contribute_embeddings
# LLMs
llm_type_to_cls_dict = {}
for k, v in llms.get_type_to_cls_dict().items():
try:
llm_type_to_cls_dict[k] = v()
except Exception:
pass
llm_type_to_cls_dict['ChatAnthropic'] = ChatAnthropic # type: ignore
llm_type_to_cls_dict['AzureChatOpenAI'] = AzureChatOpenAI # type: ignore
llm_type_to_cls_dict['ChatOpenAI'] = ChatOpenAI # type: ignore
llm_type_to_cls_dict['ChatVertexAI'] = ChatVertexAI # type: ignore
llm_type_to_cls_dict['MiniMaxChat'] = MiniMaxChat
llm_type_to_cls_dict['ChatTongyi'] = ChatTongyi
llm_type_to_cls_dict["OpenAI"] = OpenAI
llm_type_to_cls_dict['ChatZhipuAI'] = ChatZhipuAI
llm_type_to_cls_dict['ChatDeepSeek'] = ChatDeepSeek
llm_type_to_cls_dict['ChatHunyuan'] = ChatHunyuan
llm_type_to_cls_dict['MoonshotChat'] = MoonshotChat
# llm contribute
llm_type_to_cls_dict.update({
llm_name: import_class(f'bisheng_langchain.chat_models.{llm_name}')
for llm_name in chat_models.__all__
})
# Toolkits
toolkit_type_to_loader_dict: dict[str, Any] = {
toolkit_name: import_class(f'langchain_community.agent_toolkits.{toolkit_name}')
# if toolkit_name is lower case it is a loader
for toolkit_name in agent_toolkits.__all__ if toolkit_name.islower()
}
toolkit_type_to_cls_dict: dict[str, Any] = {
toolkit_name: import_class(f'langchain_community.agent_toolkits.{toolkit_name}')
# if toolkit_name is not lower case it is a class
for toolkit_name in agent_toolkits.__all__ if not toolkit_name.islower()
}
# Memories
memory_type_to_cls_dict: dict[str, Any] = {}
for memory_name in memory.__all__:
if memory_name.find("ChatMessageHistory") != -1:
memory_type_to_cls_dict[memory_name] = import_class(f"langchain_community.chat_message_histories.{memory_name}")
elif memory_name == "ConversationKGMemory":
memory_type_to_cls_dict[memory_name] = import_class(f"langchain_community.memory.kg.{memory_name}")
elif memory_name == "MotorheadMemory":
memory_type_to_cls_dict[memory_name] = import_class(
f"langchain_community.memory.motorhead_memory.{memory_name}")
elif memory_name == "ZepMemory":
memory_type_to_cls_dict[memory_name] = import_class(f"langchain_community.memory.zep_memory.{memory_name}")
else:
memory_type_to_cls_dict[memory_name] = import_class(f'langchain.memory.{memory_name}')
# Wrappers
wrapper_type_to_cls_dict: dict[str, Any] = {
wrapper.__name__: wrapper
for wrapper in [requests.RequestsWrapper]
}
# Embeddings
embedding_type_to_cls_dict: dict[str, Any] = {
embedding_name: import_class(f'langchain_community.embeddings.{embedding_name}')
for embedding_name in embeddings.__all__
}
embedding_type_to_cls_dict.update({
embedding_name:
import_class(f'bisheng_langchain.embeddings.{embedding_name}')
for embedding_name in contribute_embeddings.__all__
})
embedding_type_to_cls_dict.update({
"OpenAIEmbeddings": OpenAIEmbeddings,
"AzureOpenAIEmbeddings": AzureOpenAIEmbeddings,
})
# Document Loaders
documentloaders_type_to_cls_dict: dict[str, Any] = {
documentloader_name:
import_class(f'langchain_community.document_loaders.{documentloader_name}')
for documentloader_name in document_loaders.__all__
}
# contribute
documentloaders_type_to_cls_dict.update({
loader:
import_class(f'bisheng_langchain.document_loaders.{loader}')
for loader in contribute_loader.__all__
})
# Text Splitters
textsplitter_type_to_cls_dict: dict[str,
Any] = dict(inspect.getmembers(text_splitter, inspect.isclass))
# merge CUSTOM_AGENTS and CUSTOM_CHAINS
CUSTOM_NODES = {
**CUSTOM_AGENTS,
**CUSTOM_CHAINS,
**CUSTOM_EMBEDDING,
} # type: ignore
================================================
FILE: src/backend/bisheng/interface/document_loaders/__init__.py
================================================
================================================
FILE: src/backend/bisheng/interface/document_loaders/base.py
================================================
from typing import Dict, List, Optional, Type
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.custom_lists import documentloaders_type_to_cls_dict
from bisheng.common.services.config_service import settings
from bisheng.template.frontend_node.documentloaders import DocumentLoaderFrontNode
from loguru import logger
from bisheng.utils.util import build_template_from_class
class DocumentLoaderCreator(LangChainTypeCreator):
type_name: str = 'documentloaders'
@property
def frontend_node_class(self) -> Type[DocumentLoaderFrontNode]:
return DocumentLoaderFrontNode
@property
def type_to_loader_dict(self) -> Dict:
return documentloaders_type_to_cls_dict
def get_signature(self, name: str) -> Optional[Dict]:
"""Get the signature of a document loader."""
try:
return build_template_from_class(name, documentloaders_type_to_cls_dict)
except ValueError as exc:
raise ValueError(f'Documment Loader {name} not found') from exc
except AttributeError as exc:
logger.error(f'Documment Loader {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
return [
documentloader.__name__ for documentloader in self.type_to_loader_dict.values()
if documentloader.__name__ in settings.documentloaders or settings.dev
]
documentloader_creator = DocumentLoaderCreator()
================================================
FILE: src/backend/bisheng/interface/embeddings/__init__.py
================================================
================================================
FILE: src/backend/bisheng/interface/embeddings/base.py
================================================
from typing import Dict, List, Optional, Type
from bisheng.custom.customs import get_custom_nodes
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.custom_lists import embedding_type_to_cls_dict
from bisheng.interface.embeddings.custom import CUSTOM_EMBEDDING
from bisheng.common.services.config_service import settings
from bisheng.template.frontend_node.base import FrontendNode
from bisheng.template.frontend_node.embeddings import EmbeddingFrontendNode
from loguru import logger
from bisheng.utils.util import build_template_from_class
class EmbeddingCreator(LangChainTypeCreator):
type_name: str = 'embeddings'
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict = embedding_type_to_cls_dict
for name, embed in CUSTOM_EMBEDDING.items():
# TODO: validate AgentType
self.type_dict[name] = embed # type: ignore
return self.type_dict
@property
def frontend_node_class(self) -> Type[FrontendNode]:
return EmbeddingFrontendNode
def get_signature(self, name: str) -> Optional[Dict]:
"""Get the signature of an embedding."""
try:
if name in get_custom_nodes(self.type_name).keys():
return get_custom_nodes(self.type_name)[name]
else:
return build_template_from_class(name, embedding_type_to_cls_dict)
except ValueError as exc:
raise ValueError(f'Embedding {name} not found') from exc
except AttributeError as exc:
logger.error(f'Embedding {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
names = []
for _, embed in self.type_to_loader_dict.items():
embed_name = (embed.function_name() if hasattr(embed, 'function_name') else embed.__name__)
if embed_name in settings.embeddings or settings.dev:
names.append(embed_name)
return names
embedding_creator = EmbeddingCreator()
================================================
FILE: src/backend/bisheng/interface/embeddings/custom.py
================================================
from typing import List, Optional
from langchain.embeddings.base import Embeddings
from langchain_community.embeddings.dashscope import BATCH_SIZE
from pydantic import Field
from bisheng.llm.domain.llm.embedding import BishengEmbedding
BATCH_SIZE["text-embedding-v4"] = 10 # PengaturanDashScopeThe batch size for is1
class OpenAIProxyEmbedding(Embeddings):
embeddings: Optional[Embeddings] = Field(default=None)
def __init__(self, **kwargs) -> None:
super().__init__(**kwargs)
from bisheng.llm.domain.services import LLMService
self.embeddings = LLMService.get_knowledge_default_embedding(0)
def embed_documents(self, texts: List[str]) -> List[List[float]]:
"""Embed search docs."""
if not texts:
return []
return self.embeddings.embed_documents(texts)
def embed_query(self, text: str) -> List[float]:
"""Embed query text."""
return self.embeddings.embed_query(text)
class FakeEmbedding(Embeddings):
"""To ensure amilvusWait, you can continue to use it when the model is offline"""
def embed_documents(self, texts: List[str]) -> List[List[float]]:
"""embedding"""
return []
def embed_query(self, text: str) -> List[float]:
"""embedding"""
return []
CUSTOM_EMBEDDING = {
'OpenAIProxyEmbedding': OpenAIProxyEmbedding,
'BishengEmbedding': BishengEmbedding,
}
================================================
FILE: src/backend/bisheng/interface/importing/__init__.py
================================================
from bisheng.interface.importing.utils import import_by_type # noqa: F401
# This module is used to import any langchain class by name.
ALL = [
'import_by_type',
]
================================================
FILE: src/backend/bisheng/interface/importing/utils.py
================================================
# This module is used to import any langchain class by name.
import importlib
from typing import Any, ClassVar, Dict, Type
from bisheng.interface.custom import CustomComponent
from bisheng.interface.wrappers.base import wrapper_creator
from bisheng.utils import validate
from langchain.agents import Agent
from langchain.base_language import BaseLanguageModel
from langchain.chains.base import Chain
from langchain.chat_models.base import BaseChatModel
from langchain.prompts import PromptTemplate
from langchain_community.tools import BaseTool
def import_module(module_path: str) -> Any:
"""Import module from module path"""
if 'from' not in module_path:
# Import the module using the module path
return importlib.import_module(module_path)
# Split the module path into its components
_, module_path, _, object_name = module_path.split()
# Import the module using the module path
module = importlib.import_module(module_path)
return getattr(module, object_name)
def import_by_type(_type: str, name: str) -> Any:
from bisheng_langchain import chat_models
"""Import class by type and name"""
if _type is None:
raise ValueError(f'Type cannot be None. Check if {name} is in the config file.')
func_dict: ClassVar[Dict] = {
'agents': import_agent,
'prompts': import_prompt,
'llms': {
'llm': import_llm,
'chat': import_chat_llm,
'contribute': import_chain_contribute_llm
},
'tools': import_tool,
'chains': import_chain,
'toolkits': import_toolkit,
'wrappers': import_wrapper,
'memory': import_memory,
'embeddings': import_embedding,
'vectorstores': import_vectorstore,
'documentloaders': import_documentloader,
'textsplitters': import_textsplitter,
'utilities': import_utility,
'output_parsers': import_output_parser,
'retrievers': import_retriever,
'autogen_roles': import_autogenRoles,
'input_output': import_inputoutput,
'custom_components': import_custom_component,
}
if _type == 'llms':
key = 'contribute' if name in chat_models.__all__ else 'chat' if 'chat' in name.lower(
) else 'llm'
loaded_func = func_dict[_type][key] # type: ignore
else:
loaded_func = func_dict[_type]
return loaded_func(name)
def import_custom_component(custom_component: str) -> CustomComponent:
"""Import custom component from custom component name"""
return import_class('bisheng.interface.custom.custom_component.CustomComponent')
def import_inputoutput(input_output: str) -> Any:
"""Import output parser from output parser name"""
from bisheng.interface.inputoutput.base import input_output_creator
return input_output_creator.type_to_loader_dict[input_output]
def import_output_parser(output_parser: str) -> Any:
"""Import output parser from output parser name"""
from bisheng.interface.output_parsers.base import output_parser_creator
if output_parser in output_parser_creator.type_to_loader_dict:
return output_parser_creator.type_to_loader_dict[output_parser]
return import_module(f'from langchain_community.output_parsers import {output_parser}')
def import_chat_llm(llm: str) -> BaseChatModel:
"""Import chat llm from llm name"""
from bisheng.interface.llms.base import llm_creator
if llm in llm_creator.type_to_loader_dict:
return llm_creator.type_to_loader_dict[llm]
return import_class(f'bisheng_langchain.chat_models.{llm}')
def import_chain_contribute_llm(llm: str) -> BaseChatModel:
"""Import chat llm from llm name"""
from bisheng.interface.llms.base import llm_creator
if llm in llm_creator.type_to_loader_dict:
return llm_creator.type_to_loader_dict[llm]
return import_class(f'bisheng_langchain.chat_models.{llm}')
def import_retriever(retriever: str) -> Any:
"""Import retriever from retriever name"""
from bisheng.interface.retrievers.base import retriever_creator
if retriever in retriever_creator.type_to_loader_dict:
return retriever_creator.type_to_loader_dict[retriever]
return import_module(f'from langchain_community.retrievers import {retriever}')
def import_autogenRoles(autogen: str) -> Any:
return import_module(f'from bisheng_langchain.autogen_role import {autogen}')
def import_memory(memory: str) -> Any:
"""Import memory from memory name"""
from bisheng.interface.memories.base import memory_creator
if memory in memory_creator.type_to_loader_dict:
return memory_creator.type_to_loader_dict[memory]
return import_module(f'from langchain.memory import {memory}')
def import_class(class_path: str) -> Any:
"""Import class from class path"""
module_path, class_name = class_path.rsplit('.', 1)
module = import_module(module_path)
return getattr(module, class_name)
def import_prompt(prompt: str) -> Type[PromptTemplate]:
"""Import prompt from prompt name"""
from bisheng.interface.prompts.base import prompt_creator
if prompt in prompt_creator.type_to_loader_dict:
return prompt_creator.type_to_loader_dict[prompt]
return import_class(f'langchain.prompts.{prompt}')
def import_wrapper(wrapper: str) -> Any:
"""Import wrapper from wrapper name"""
if wrapper in wrapper_creator.type_to_loader_dict:
return wrapper_creator.type_to_loader_dict[wrapper]
def import_toolkit(toolkit: str) -> Any:
"""Import toolkit from toolkit name"""
from bisheng.interface.toolkits.base import toolkits_creator
return toolkits_creator.type_to_loader_dict[toolkit]
def import_agent(agent: str) -> Agent:
"""Import agent from agent name"""
# check for custom agent
from bisheng_langchain import agents
if agent in agents.__all__:
return import_class(f'bisheng_langchain.agents.{agent}')
return import_class(f'langchain.agents.{agent}')
def import_llm(llm: str) -> BaseLanguageModel:
"""Import llm from llm name"""
from bisheng.interface.llms.base import llm_creator
return next(x for x in llm_creator.type_to_loader_dict.values() if x.__name__ == llm)
def import_tool(tool: str) -> BaseTool:
"""Import tool from tool name"""
from bisheng.interface.tools.base import tool_creator
if tool in tool_creator.type_to_loader_dict:
return tool_creator.type_to_loader_dict[tool]['fcn']
return import_class(f'langchain_community.tools.{tool}')
def import_chain(chain: str) -> Type[Chain]:
"""Import chain from chain name"""
from bisheng.interface.chains.base import chain_creator
return next(x for x in chain_creator.type_to_loader_dict.values() if x.__name__ == chain)
def import_embedding(embedding: str) -> Any:
"""Import embedding from embedding name"""
from bisheng.interface.embeddings.base import embedding_creator
return next(x for x in embedding_creator.type_to_loader_dict.values()
if x.__name__ == embedding)
def import_vectorstore(vectorstore: str) -> Any:
"""Import vectorstore from vectorstore name"""
from bisheng_langchain import vectorstores
from bisheng.interface.vector_store.base import vectorstore_creator
if vectorstore_creator.type_to_loader_dict.get(vectorstore) is not None:
return vectorstore_creator.type_to_loader_dict[vectorstore]
if vectorstore in vectorstores.__all__:
return import_class(f'bisheng_langchain.vectorstores.{vectorstore}')
return import_class(f'langchain_community.vectorstores.{vectorstore}')
def import_documentloader(documentloader: str) -> Any:
"""Import documentloader from documentloader name"""
from bisheng_langchain import document_loaders
from bisheng.interface.document_loaders.base import documentloader_creator
if documentloader in document_loaders.__all__:
return import_class(f'bisheng_langchain.document_loaders.{documentloader}')
return next(x for x in documentloader_creator.type_to_loader_dict.values()
if x.__name__ == documentloader)
def import_textsplitter(textsplitter: str) -> Any:
"""Import textsplitter from textsplitter name"""
return import_class(f'langchain.text_splitter.{textsplitter}')
def import_utility(utility: str) -> Any:
"""Import utility from utility name"""
if utility == 'SQLDatabase':
return import_class(f'langchain.sql_database.{utility}')
return import_class(f'langchain_community.utilities.{utility}')
def get_function(code):
"""Get the function"""
function_name = validate.extract_function_name(code)
return validate.create_function(code, function_name)
def eval_custom_component_code(code: str) -> Type[CustomComponent]:
"""Evaluate custom component code"""
class_name = validate.extract_class_name(code)
return validate.create_class(code, class_name)
================================================
FILE: src/backend/bisheng/interface/initialize/__init__.py
================================================
================================================
FILE: src/backend/bisheng/interface/initialize/llm.py
================================================
def initialize_vertexai(class_object, params):
if credentials_path := params.get('credentials'):
from google.oauth2 import service_account # type: ignore
credentials_object = service_account.Credentials.from_service_account_file(
filename=credentials_path
)
params['credentials'] = credentials_object
return class_object(**params)
================================================
FILE: src/backend/bisheng/interface/initialize/loading.py
================================================
import inspect
import json
from typing import TYPE_CHECKING, Any, Callable, Dict, Sequence, Type
import httpx
import openai
from langchain.agents import agent as agent_module
from langchain.agents.agent import AgentExecutor
from langchain.agents.agent_toolkits.base import BaseToolkit
from langchain.agents.tools import BaseTool
from langchain.chains.base import Chain
from langchain.document_loaders.base import BaseLoader
from langchain.vectorstores.base import VectorStore
from langchain_community.utils.openai import is_openai_v1
from loguru import logger
from pydantic import SecretStr, ValidationError, create_model
from pydantic.fields import FieldInfo
from bisheng.common.constants.enums.telemetry import ApplicationTypeEnum
from bisheng.common.services.config_service import settings
from bisheng.core.cache.utils import file_download
from bisheng.interface.agents.base import agent_creator
from bisheng.interface.chains.base import chain_creator
from bisheng.interface.custom_lists import CUSTOM_NODES
from bisheng.interface.embeddings.custom import FakeEmbedding
from bisheng.interface.importing.utils import (eval_custom_component_code, get_function,
import_by_type)
from bisheng.interface.initialize.llm import initialize_vertexai
from bisheng.interface.initialize.utils import (handle_format_kwargs, handle_node_type,
handle_partial_variables, langchain_bug_openv1)
from bisheng.interface.initialize.vector_store import vecstore_initializer
from bisheng.interface.output_parsers.base import output_parser_creator
from bisheng.interface.retrievers.base import retriever_creator
from bisheng.interface.toolkits.base import toolkits_creator
from bisheng.interface.utils import load_file_into_dict
from bisheng.interface.wrappers.base import wrapper_creator
from bisheng.knowledge.domain.models.knowledge import KnowledgeDao
from bisheng.llm.domain import LLMService
from bisheng.utils import validate
from bisheng.utils.constants import NODE_ID_DICT, PRESET_QUESTION
from bisheng_langchain.vectorstores import VectorStoreFilterRetriever
if TYPE_CHECKING:
from bisheng import CustomComponent
def build_vertex_in_params(params: Dict) -> Dict:
from bisheng.graph.vertex.base import Vertex
# If any of the values in params is a Vertex, we will build it
return {
key: value.build() if isinstance(value, Vertex) else value
for key, value in params.items()
}
# from bisheng_langchain.document_loaders.elem_unstrcutured_loader import ElemUnstructuredLoaderV0
async def instantiate_class(node_type: str, base_type: str, params: Dict, user_id=None) -> Any:
"""Instantiate class from module type and key, and params"""
params = convert_params_to_sets(params)
params = convert_kwargs(params)
params_node_id_dict = params.pop(NODE_ID_DICT)
if node_type in CUSTOM_NODES:
if custom_node := CUSTOM_NODES.get(node_type):
if hasattr(custom_node, 'initialize'):
return custom_node.initialize(**params)
return custom_node(**params)
class_object = import_by_type(_type=base_type, name=node_type)
return await instantiate_based_on_type(class_object,
base_type,
node_type,
params,
params_node_id_dict,
user_id=user_id)
def convert_params_to_sets(params):
"""Convert certain params to sets"""
if 'allowed_special' in params:
params['allowed_special'] = set(params['allowed_special'])
if 'disallowed_special' in params:
params['disallowed_special'] = set(params['disallowed_special'])
if 'input_node' in params:
params.pop('input_node')
return params
def convert_kwargs(params):
# if *kwargs are passed as a string, convert to dict
# first find any key that has kwargs or config in it
kwargs_keys = [key for key in params.keys() if 'kwargs' in key or 'config' in key]
for key in kwargs_keys:
if isinstance(params[key], str):
params[key] = json.loads(params[key])
return params
async def instantiate_based_on_type(class_object,
base_type,
node_type,
params,
param_id_dict,
user_id=None):
if base_type == 'agents':
return instantiate_agent(node_type, class_object, params)
elif base_type == 'prompts':
return instantiate_prompt(node_type, class_object, params, param_id_dict)
elif base_type == 'tools':
tool = instantiate_tool(node_type, class_object, params)
if hasattr(tool, 'name') and isinstance(tool, BaseTool):
# tool name shouldn't contain spaces
tool.name = tool.name.replace(' ', '_')
return tool
elif base_type == 'toolkits':
return instantiate_toolkit(node_type, class_object, params)
elif base_type == 'embeddings':
return instantiate_embedding(class_object, params)
elif base_type == 'vectorstores':
return instantiate_vectorstore(node_type, class_object, params)
elif base_type == 'documentloaders':
return instantiate_documentloader(class_object, params)
elif base_type == 'textsplitters':
return instantiate_textsplitter(class_object, params)
elif base_type == 'utilities':
return instantiate_utility(node_type, class_object, params)
elif base_type == 'chains':
return instantiate_chains(node_type, class_object, params, param_id_dict)
elif base_type == 'output_parsers':
return instantiate_output_parser(node_type, class_object, params)
elif base_type == 'llms':
return instantiate_llm(node_type, class_object, params)
elif base_type == 'retrievers':
return instantiate_retriever(node_type, class_object, params)
elif base_type == 'memory':
return instantiate_memory(node_type, class_object, params)
elif base_type == 'custom_components':
return await instantiate_custom_component(node_type, class_object, params, user_id)
elif base_type == 'wrappers':
return instantiate_wrapper(node_type, class_object, params)
elif base_type == 'input_output':
return instantiate_input_output(node_type, class_object, params, param_id_dict)
elif base_type == 'autogen_roles':
return instantiate_autogen_roles(node_type, class_object, params)
else:
return class_object(**params)
async def instantiate_custom_component(node_type, class_object, params, user_id):
params_copy = params.copy()
class_object: 'CustomComponent' = eval_custom_component_code(params_copy.pop('code'))
custom_component = class_object(user_id=user_id)
if 'retriever' in params_copy and hasattr(params_copy['retriever'], 'as_retriever'):
params_copy['retriever'] = params_copy['retriever'].as_retriever()
# Determine if the build method is asynchronous
is_async = inspect.iscoroutinefunction(custom_component.build)
if is_async:
# Await the build method directly if it's async
built_object = await custom_component.build(**params_copy)
else:
# Call the build method directly if it's sync
built_object = custom_component.build(**params_copy)
return built_object, {'repr': custom_component.custom_repr()}
def instantiate_input_output(node_type, class_object, params, id_dict):
if node_type == 'Report':
preset_question = {}
if PRESET_QUESTION in params:
preset_question = params.pop(PRESET_QUESTION)
chains = params.get('chains', [])
chains_idlist = id_dict.get('chains', [])
# Needs to be rightchainsObjects are enriched
chain_list = []
for index, id in enumerate(chains_idlist):
chain_obj = {}
chain_obj['object'] = chains[index]
if id in preset_question:
if isinstance(preset_question[id], list):
for node_id in preset_question[id]:
chain_ = chain_obj.copy()
chain_['node_id'] = node_id[0]
chain_['input'] = {chains[index].input_keys[0]: node_id[1]}
chain_list.append(chain_)
continue
else:
chain_obj['node_id'] = preset_question[id][0]
chain_obj['input'] = {chains[index].input_keys[0]: preset_question[id][1]}
else:
# give a default input
logger.error(f'Report has no question id={id}')
chain_obj['input'] = {chains[index].input_keys[0]: 'start'}
chain_list.append(chain_obj)
params['chains'] = chain_list
# variable
variable = params.get('variables')
variable_node_id = id_dict.get('variables') or []
params['variables'] = []
for index, id in enumerate(variable_node_id):
params['variables'].append({'node_id': id, 'input': variable[index]})
return class_object(**params)
if node_type == 'InputFileNode':
file_path = class_object(**params).text()
if file_path:
file_path, file_name2 = file_download(file_path[0])
return [file_path, file_name2 if file_name2 else file_path[1]]
else:
return ''
if 'file_path' in params:
file_path = params['file_path']
if not file_path:
return ''
if isinstance(file_path, list):
params['file_path'] = file_path[0]
return class_object(**params).text()
def instantiate_autogen_roles(node_type, class_object, params):
return class_object(**params)
def instantiate_wrapper(node_type, class_object, params):
if node_type in wrapper_creator.from_method_nodes:
method = wrapper_creator.from_method_nodes[node_type]
if class_method := getattr(class_object, method, None):
return class_method(**params)
raise ValueError(f'Method {method} not found in {class_object}')
if node_type == 'DallEAPIWrapper' and is_openai_v1():
if 'openai_proxy' in params and params['openai_proxy']:
client_params = langchain_bug_openv1(params)
client_params['http_client'] = httpx.Client(proxies=params.get('openai_proxy'))
params['client'] = openai.OpenAI(**client_params).images
client_params['http_client'] = httpx.AsyncClient(proxies=params.get('openai_proxy'))
params['async_client'] = openai.AsyncOpenAI(**client_params).images
return class_object(**params)
def instantiate_output_parser(node_type, class_object, params):
if node_type in output_parser_creator.from_method_nodes:
method = output_parser_creator.from_method_nodes[node_type]
if class_method := getattr(class_object, method, None):
return class_method(**params)
raise ValueError(f'Method {method} not found in {class_object}')
return class_object(**params)
def instantiate_llm(node_type, class_object, params: Dict, user_llm_request: bool = True):
# This is a workaround so JinaChat works until streaming is implemented
# if "openai_api_base" in params and "jina" in params["openai_api_base"]:
# False if condition is True
params.pop('cache', None)
if is_openai_v1() and params.get('openai_proxy'):
params['http_client'] = httpx.Client(proxies=params.get('openai_proxy'))
params['http_async_client'] = httpx.AsyncClient(proxies=params.get('openai_proxy'))
del params['openai_proxy']
if node_type == '':
anthropic_api_key = params.pop('anthropic_api_key', None)
params['anthropic_api_key'] = SecretStr(anthropic_api_key) if anthropic_api_key else None
if node_type == 'VertexAI':
return initialize_vertexai(class_object=class_object, params=params)
# max_tokens sometimes is a string and should be an int
if 'max_tokens' in params:
if isinstance(params['max_tokens'], str) and params['max_tokens'].isdigit():
params['max_tokens'] = int(params['max_tokens'])
elif not isinstance(params.get('max_tokens'), int):
params.pop('max_tokens', None)
if node_type == 'BishengLLM':
params['app_id'] = "flow(will be deprecated)"
params['app_name'] = "flow(will be deprecated)"
params['app_type'] = ApplicationTypeEnum.SKILL
params['user_id'] = params.get('user_id', 0)
llm = class_object(**params)
llm_config = settings.get_from_db('llm_request')
# Supportrequest_timeout & max_retries
if hasattr(llm, 'request_timeout') and 'request_timeout' in llm_config:
if isinstance(llm_config.get('request_timeout'), str):
llm.request_timeout = int(llm_config.get('request_timeout'))
else:
llm.request_timeout = llm_config.get('request_timeout')
if hasattr(llm, 'max_retries') and 'max_retries' in llm_config:
llm.max_retries = llm_config.get('max_retries')
return llm
def instantiate_memory(node_type, class_object, params):
# process input_key and output_key to remove them if
# they are empty strings
if node_type == 'ConversationEntityMemory':
params.pop('memory_key', None)
for key in ['input_key', 'output_key']:
if key in params and (params[key] == '' or not params[key]):
params.pop(key)
try:
if 'retriever' in params and hasattr(params['retriever'], 'as_retriever'):
params['retriever'] = params['retriever'].as_retriever()
return class_object(**params)
# I want to catch a specific attribute error that happens
# when the object does not have a cursor attribute
except Exception as exc:
if "object has no attribute 'cursor'" in str(exc) or 'object has no field "conn"' in str(
exc):
raise AttributeError(
('Failed to build connection to database.'
f' Please check your connection string and try again. Error: {exc}')) from exc
raise exc
def instantiate_retriever(node_type, class_object, params):
for key, value in params.items():
if 'retriever' in key and hasattr(value, 'as_retriever'):
params[key] = value.as_retriever()
if node_type in retriever_creator.from_method_nodes:
method = retriever_creator.from_method_nodes[node_type]
if class_method := getattr(class_object, method, None):
return class_method(**params)
raise ValueError(f'Method {method} not found in {class_object}')
return class_object(**params)
def instantiate_chains(node_type, class_object: Type[Chain], params: Dict, id_dict: Dict):
if 'retriever' in params:
user_name = params.pop('user_name', '')
if hasattr(params['retriever'], 'as_retriever'):
if settings.get_from_db('file_access'):
# need to verify file access
access_url = settings.get_from_db('file_access') + f'?username={user_name}'
logger.info('file_access_filter url={}', access_url)
vectorstore = VectorStoreFilterRetriever(vectorstore=params['retriever'],
access_url=access_url)
else:
vectorstore = params['retriever'].as_retriever()
params['retriever'] = vectorstore
# sequence chain
if node_type == 'SequentialChain':
# Modifikasisequence Supports customizationchainSequence
params.pop('input_node', '') # sequential Adding input parameters is not supported
try:
chain_order = json.loads(params.pop('chain_order'))
except Exception:
raise Exception('chain_order Not a standard array')
chains_origin = params.get('chains')
chains_dict = {id: index for index, id in enumerate(id_dict.get('chains'))}
params['chains'] = [chains_origin[chains_dict.get(id)] for id in chain_order]
# dict Tukar
if 'headers' in params and isinstance(params['headers'], str):
params['headers'] = json.loads(params['headers'])
if node_type == 'ConversationalRetrievalChain':
params['get_chat_history'] = str
params['combine_docs_chain_kwargs'] = {
'prompt': params.pop('combine_docs_chain_kwargs', None),
'document_prompt': params.pop('document_prompt', None)
}
params['combine_docs_chain_kwargs'] = {
k: v
for k, v in params['combine_docs_chain_kwargs'].items() if v is not None
}
# Manual assemblyMultiPromptChain
if node_type in {'MultiPromptChain', 'MultiRuleChain'}:
destination_chain_name = params['destination_chain_name']
llm_chains = params['LLMChains']
destination_chain = {}
i = 0
for k, name in destination_chain_name.items():
destination_chain[name] = llm_chains[i]
i = i + 1
params.pop('LLMChains')
params.pop('destination_chain_name')
params['destination_chains'] = destination_chain
if node_type in chain_creator.from_method_nodes:
method = chain_creator.from_method_nodes[node_type]
if class_method := getattr(class_object, method, None):
return class_method(**params)
raise ValueError(f'Method {method} not found in {class_object}')
return class_object(**params)
def instantiate_agent(node_type, class_object: Type[agent_module.Agent], params: Dict):
if node_type in agent_creator.from_method_nodes:
method = agent_creator.from_method_nodes[node_type]
if class_method := getattr(class_object, method, None):
agent = class_method(**params)
tools = params.get('tools', [])
return AgentExecutor.from_agent_and_tools(agent=agent,
tools=tools,
handle_parsing_errors=True)
return load_agent_executor(class_object, params)
def instantiate_prompt(node_type, class_object, params: Dict, param_id_dict: Dict):
params, prompt = handle_node_type(node_type, class_object, params)
format_kwargs = handle_format_kwargs(prompt, params)
# Now we'll use partial_format to format the prompt
if format_kwargs:
prompt = handle_partial_variables(prompt, format_kwargs)
no_human_input = set(param_id_dict.keys())
human_input = set(prompt.input_variables).difference(no_human_input)
order_input = list(human_input) + list(set(prompt.input_variables) & no_human_input)
if len(order_input) > 1:
# if node_type == 'ChatPromptTemplate':
if hasattr(prompt, 'prompt') and hasattr(prompt.prompt, 'input_variables'):
prompt.prompt.input_variables = order_input
elif hasattr(prompt, 'input_variables'):
prompt.input_variables = order_input
return prompt, format_kwargs
def instantiate_tool(node_type, class_object: Type[BaseTool], params: Dict):
# build args_schema
args_schema = params.pop('args_schema', '')
if node_type == 'JsonSpec':
if file_dict := load_file_into_dict(params.pop('path')):
params['dict_'] = file_dict
else:
raise ValueError('Invalid file')
return class_object(**params)
elif node_type == 'PythonFunctionTool':
params['func'] = get_function(params.get('code'))
return class_object(**params)
elif node_type == 'PythonFunction':
function_string = params['code']
if isinstance(function_string, str):
return validate.eval_function(function_string)
raise ValueError('Function should be a string')
elif node_type.lower() == 'tool':
tool = class_object(**params)
tool = class_object(**params)
if args_schema and hasattr(tool, 'args_schema'):
fields = {}
for name, prop in args_schema.items():
# evalFunction to execute a string expression and return the result
import typing # noqa
if prop.get('type') == 'string':
field_type = str
else:
field_type = typing.Any
fields[name] = (field_type, FieldInfo(**prop))
tool.args_schema = create_model(name, **fields)
return tool
def instantiate_toolkit(node_type, class_object: Type[BaseToolkit], params: Dict):
loaded_toolkit = class_object(**params)
# Commenting this out for now to use toolkits as normal tools
# if toolkits_creator.has_create_function(node_type):
# return load_toolkits_executor(node_type, loaded_toolkit, params)
if isinstance(loaded_toolkit, BaseToolkit):
return loaded_toolkit.get_tools()
return loaded_toolkit
def instantiate_embedding(class_object, params: Dict):
# params.pop('model', None)
try:
if params.get('openai_proxy'):
params['http_client'] = httpx.Client(proxies=params.get('openai_proxy'))
params['http_async_client'] = httpx.AsyncClient(proxies=params.get('openai_proxy'))
del params['openai_proxy']
if class_object.__name__ == 'OpenAIEmbeddings':
params['check_embedding_ctx_length'] = False
elif class_object.__name__ == 'BishengEmbedding':
params['app_id'] = "flow(will be deprecated)"
params['app_name'] = "flow(will be deprecated)"
params['app_type'] = ApplicationTypeEnum.SKILL
params['user_id'] = params.get('user_id', 0)
return class_object(**params)
except ValidationError:
params = {key: value for key, value in params.items() if key in class_object.__fields__}
return class_object(**params)
def instantiate_vectorstore(node_type: str, class_object: Type[VectorStore], params: Dict):
user_name = params.pop('user_name', '')
search_kwargs = params.pop('search_kwargs', {})
search_type = params.pop('search_type', 'similarity')
if 'documents' not in params:
params['documents'] = []
# Filter knowledge bases for which the user does not have permission
# TODO zgq After the subsequent unified skill execution process, all business-related logic is migrated to before the initial skill object.
if node_type == 'MilvusWithPermissionCheck' or node_type == 'ElasticsearchWithPermissionCheck':
col_name = 'collection_name'
if node_type == 'ElasticsearchWithPermissionCheck':
col_name = 'index_name'
# Get execution users List of knowledge bases with permission to view
knowledge_ids = [one['key'] for one in params[col_name]]
if params.pop('_is_check_auth', True):
knowledge_list = KnowledgeDao.judge_knowledge_permission(user_name, knowledge_ids)
else:
knowledge_list = KnowledgeDao.get_list_by_ids(knowledge_ids)
logger.debug(f'{node_type} after filter, get knowledge_list: {knowledge_list}')
if not knowledge_list:
logger.warning(f'{node_type}: after filter, get zero knowledge')
# Fake if you don't have any knowledge baseembeddingand emptycollection_name
if node_type == 'MilvusWithPermissionCheck':
params[col_name] = []
params['collection_embeddings'] = []
params['partition_keys'] = []
for knowledge in knowledge_list:
params[col_name].append(knowledge.collection_name)
params['collection_embeddings'].append(
LLMService.get_bisheng_knowledge_embedding_sync(0, model_id=int(knowledge.model)))
if knowledge.collection_name.startswith('partition'):
params['partition_keys'].append(knowledge.id)
else:
params['partition_keys'].append(None)
params['embedding'] = params['collection_embeddings'][0] if params[
'collection_embeddings'] else FakeEmbedding()
else:
params[col_name] = [
knowledge.index_name or knowledge.collection_name for knowledge in knowledge_list
]
if initializer := vecstore_initializer.get(class_object.__name__):
vecstore = initializer(class_object, params, search_kwargs)
else:
if 'texts' in params:
params['documents'] = params.pop('texts')
vecstore = class_object.from_documents(**params)
# ! This might not work. Need to test
if search_kwargs and hasattr(vecstore, 'as_retriever'):
if settings.get_from_db('file_access'):
# need to verify file access / For Knowledge Base Only
access_url = settings.get_from_db('file_access') + f'?username={user_name}'
vecstore = VectorStoreFilterRetriever(vectorstore=vecstore,
search_type=search_type,
search_kwargs=search_kwargs,
access_url=access_url)
else:
vecstore = vecstore.as_retriever(search_type=search_type, search_kwargs=search_kwargs)
return vecstore
def instantiate_documentloader(class_object: Type[BaseLoader], params: Dict):
if 'file_filter' in params:
# file_filter will be a string but we need a function
# that will be used to filter the files using file_filter
# like lambda x: x.endswith(".txt") but as we don't know
# anything besides the string, we will simply check if the string is
# in x and if it is, we will return True
file_filter = params.pop('file_filter')
extensions = file_filter.split(',')
params['file_filter'] = lambda x: any(extension.strip() in x for extension in extensions)
if 'file_path' in params:
file_path = params['file_path']
if isinstance(file_path, list):
file_name = file_path[1]
params['file_path'] = file_path[0]
if class_object.__name__ == 'ElemUnstructuredLoaderV0':
params['file_name'] = file_name
metadata = params.pop('metadata', None)
if metadata and isinstance(metadata, str):
try:
metadata = json.loads(metadata)
except json.JSONDecodeError as exc:
raise ValueError('The metadata you provided is not a valid JSON string.') from exc
# make it success when file not present
if 'file_path' in params and not params['file_path']:
return []
docs = class_object(**params).load()
# Now if metadata is an empty dict, we will not add it to the documents
if metadata:
for doc in docs:
# If the document already has metadata, we will not overwrite it
if not doc.metadata:
doc.metadata = metadata
else:
doc.metadata.update(metadata)
return docs
def instantiate_textsplitter(
class_object,
params: Dict,
):
try:
documents = params.pop('documents')
if not documents:
return []
except KeyError as exc:
raise ValueError('The source you provided did not load correctly or was empty.'
'Try changing the chunk_size of the Text Splitter.') from exc
if ('separator_type' in params
and params['separator_type'] == 'Text') or 'separator_type' not in params:
params.pop('separator_type', None)
# separators might come in as an escaped string like \\n
# so we need to convert it to a string
if 'separators' in params:
params['separators'] = (params['separators'].encode().decode('unicode-escape'))
text_splitter = class_object(**params)
else:
from langchain.text_splitter import Language
language = params.pop('separator_type', None)
params['language'] = Language(language)
params.pop('separators', None)
text_splitter = class_object.from_language(**params)
return text_splitter.split_documents(documents)
def instantiate_utility(node_type, class_object, params: Dict):
if node_type == 'SQLDatabase':
return class_object.from_uri(params.pop('uri'))
return class_object(**params)
def replace_zero_shot_prompt_with_prompt_template(nodes):
"""Replace ZeroShotPrompt with PromptTemplate"""
for node in nodes:
if node['data']['type'] == 'ZeroShotPrompt':
# Build Prompt Template
tools = [
tool for tool in nodes if tool['type'] != 'chatOutputNode'
and 'Tool' in tool['data']['node']['base_classes']
]
node['data'] = build_prompt_template(prompt=node['data'], tools=tools)
break
return nodes
def load_agent_executor(agent_class: type[agent_module.Agent], params, **kwargs):
"""Load agent executor from agent class, tools and chain"""
allowed_tools: Sequence[BaseTool] = params.get('allowed_tools', [])
llm_chain = params['llm_chain']
# agent has hidden args for memory. might need to be support
# memory = params["memory"]
# if allowed_tools is not a list or set, make it a list
if not isinstance(allowed_tools, (list, set)) and isinstance(allowed_tools, BaseTool):
allowed_tools = [allowed_tools]
tool_names = [tool.name for tool in allowed_tools]
# Agent class requires an output_parser but Agent classes
# have a default output_parser.
agent = agent_class(allowed_tools=tool_names, llm_chain=llm_chain) # type: ignore
return AgentExecutor.from_agent_and_tools(
agent=agent,
tools=allowed_tools,
handle_parsing_errors=True,
# memory=memory,
**kwargs,
)
def load_toolkits_executor(node_type: str, toolkit: BaseToolkit, params: dict):
create_function: Callable = toolkits_creator.get_create_function(node_type)
if llm := params.get('llm'):
return create_function(llm=llm, toolkit=toolkit)
def build_prompt_template(prompt, tools):
"""Build PromptTemplate from ZeroShotPrompt"""
prefix = prompt['node']['template']['prefix']['value']
suffix = prompt['node']['template']['suffix']['value']
format_instructions = prompt['node']['template']['format_instructions']['value']
tool_strings = '\n'.join([
f"{tool['data']['node']['name']}: {tool['data']['node']['description']}" for tool in tools
])
tool_names = ', '.join([tool['data']['node']['name'] for tool in tools])
format_instructions = format_instructions.format(tool_names=tool_names)
value = '\n\n'.join([prefix, tool_strings, format_instructions, suffix])
prompt['type'] = 'PromptTemplate'
prompt['node'] = {
'template': {
'_type': 'prompt',
'input_variables': {
'type': 'str',
'required': True,
'placeholder': '',
'list': True,
'show': False,
'multiline': False,
},
'output_parser': {
'type': 'BaseOutputParser',
'required': False,
'placeholder': '',
'list': False,
'show': False,
'multline': False,
'value': None,
},
'template': {
'type': 'str',
'required': True,
'placeholder': '',
'list': False,
'show': True,
'multiline': True,
'value': value,
},
'template_format': {
'type': 'str',
'required': False,
'placeholder': '',
'list': False,
'show': False,
'multline': False,
'value': 'f-string',
},
'validate_template': {
'type': 'bool',
'required': False,
'placeholder': '',
'list': False,
'show': False,
'multline': False,
'value': True,
},
},
'description': 'Schema to represent a prompt for an LLM.',
'base_classes': ['BasePromptTemplate'],
}
return prompt
================================================
FILE: src/backend/bisheng/interface/initialize/utils.py
================================================
import contextlib
import json
from typing import Any, Dict, List
import orjson
from langchain.agents import ZeroShotAgent
from langchain.schema import BaseOutputParser, Document
from bisheng.common.utils.util import orjson_dumps
def langchain_bug_openv1(params):
'''Compatible openai v1'''
client_params = {
'api_key': params.get('openai_api_key') or params.get('api_key'),
'organization': params.get('openai_organization'),
'base_url': params.get('openai_api_base'),
'timeout': params.get('request_timeout', 30),
'max_retries': params.get('max_retries', 1),
'default_headers': params.get('default_headers'),
'default_query': params.get('default_query')
}
return client_params
def handle_node_type(node_type, class_object, params: Dict):
if node_type == 'ZeroShotPrompt':
params = check_tools_in_params(params)
prompt = ZeroShotAgent.create_prompt(**params)
elif 'MessagePromptTemplate' in node_type:
prompt = instantiate_from_template(class_object, params)
elif node_type == 'ChatPromptTemplate':
prompt = class_object.from_messages(**params)
elif hasattr(class_object, 'from_template') and params.get('template'):
prompt = class_object.from_template(template=params.pop('template'),
output_parser=params.get('output_parser'))
else:
prompt = class_object(**params)
return params, prompt
def check_tools_in_params(params: Dict):
if 'tools' not in params:
params['tools'] = []
return params
def instantiate_from_template(class_object, params: Dict):
from_template_params = {'template': params.pop('prompt', params.pop('template', ''))}
from_template_params.update(params)
if not from_template_params.get('template'):
raise ValueError('Prompt template is required')
if class_object.__name__ in ('HumanMessagePromptTemplate', 'SystemMessagePromptTemplate'):
from_template_params['template'] = from_template_params['template'][0].template
return class_object.from_template(**from_template_params)
def handle_format_kwargs(prompt, params: Dict):
format_kwargs: Dict[str, Any] = {}
for input_variable in prompt.input_variables:
if input_variable in params:
format_kwargs = handle_variable(params, input_variable, format_kwargs)
return format_kwargs
def handle_partial_variables(prompt, format_kwargs: Dict):
partial_variables = format_kwargs.copy()
partial_variables = {key: value for key, value in partial_variables.items() if value}
# Remove handle_keys otherwise LangChain raises an error
partial_variables.pop('handle_keys', None)
if partial_variables and hasattr(prompt, 'partial'):
return prompt.partial(**partial_variables)
return prompt
def handle_variable(params: Dict, input_variable: str, format_kwargs: Dict):
variable = params[input_variable]
if isinstance(variable, str):
format_kwargs[input_variable] = variable
elif isinstance(variable, dict):
# variable node Special treated
if len(variable) == 0:
format_kwargs[input_variable] = ''
elif len(variable) != 1:
raise ValueError(f'VariableNode contains multi-key {variable.keys()}')
else:
format_kwargs[input_variable] = list(variable.values())[0]
elif isinstance(variable, BaseOutputParser) and hasattr(variable, 'get_format_instructions'):
format_kwargs[input_variable] = variable.get_format_instructions()
elif is_instance_of_list_or_document(variable):
format_kwargs = format_document(variable, input_variable, format_kwargs)
if needs_handle_keys(variable):
format_kwargs = add_handle_keys(input_variable, format_kwargs)
return format_kwargs
def is_instance_of_list_or_document(variable):
return (isinstance(variable, List) and all(isinstance(item, Document) for item in variable)
or isinstance(variable, Document))
def format_document(variable, input_variable: str, format_kwargs: Dict):
variable = variable if isinstance(variable, List) else [variable]
content = format_content(variable)
format_kwargs[input_variable] = content
return format_kwargs
def format_content(variable):
if len(variable) > 1:
return '\n'.join([item.page_content for item in variable if item.page_content])
elif len(variable) == 1:
content = variable[0].page_content
return try_to_load_json(content)
return ''
def try_to_load_json(content):
with contextlib.suppress(json.JSONDecodeError):
content = orjson.loads(content)
if isinstance(content, list):
content = ','.join([str(item) for item in content])
else:
content = orjson_dumps(content)
return content
def needs_handle_keys(variable):
return is_instance_of_list_or_document(variable) or (isinstance(
variable, BaseOutputParser) and hasattr(variable, 'get_format_instructions'))
def add_handle_keys(input_variable: str, format_kwargs: Dict):
if 'handle_keys' not in format_kwargs:
format_kwargs['handle_keys'] = []
format_kwargs['handle_keys'].append(input_variable)
return format_kwargs
================================================
FILE: src/backend/bisheng/interface/initialize/vector_store.py
================================================
import ast
import json
import os
from typing import Any, Callable, Dict, Type
from langchain_community.vectorstores import (FAISS, Chroma, Milvus, MongoDBAtlasVectorSearch,
Pinecone, Qdrant, SupabaseVectorStore, Weaviate)
from loguru import logger
from sqlmodel import select
from bisheng.common.services.config_service import settings
from bisheng.core.database import get_sync_db_session
from bisheng.knowledge.domain.models.knowledge import Knowledge, KnowledgeDao
from bisheng.llm.domain import LLMService
from bisheng_langchain.vectorstores import ElasticKeywordsSearch
def docs_in_params(params: dict) -> bool:
"""Check if params has documents OR texts and one of them is not an empty list,
If any of them is not an empty list, return True, else return False"""
return ('documents' in params and params['documents']) or ('texts' in params
and params['texts'])
def initialize_mongodb(class_object: Type[MongoDBAtlasVectorSearch], params: dict, search: dict):
"""Initialize mongodb and return the class object"""
MONGODB_ATLAS_CLUSTER_URI = params.pop('mongodb_atlas_cluster_uri')
if not MONGODB_ATLAS_CLUSTER_URI:
raise ValueError('Mongodb atlas cluster uri must be provided in the params')
from pymongo import MongoClient
import certifi
client: MongoClient = MongoClient(MONGODB_ATLAS_CLUSTER_URI, tlsCAFile=certifi.where())
db_name = params.pop('db_name', None)
collection_name = params.pop('collection_name', None)
if not db_name or not collection_name:
raise ValueError('db_name and collection_name must be provided in the params')
index_name = params.pop('index_name', None)
if not index_name:
raise ValueError('index_name must be provided in the params')
collection = client[db_name][collection_name]
if not docs_in_params(params):
# __init__ requires collection, embedding and index_name
init_args = {
'collection': collection,
'index_name': index_name,
'embedding': params.get('embedding'),
}
return class_object(**init_args)
if 'texts' in params:
params['documents'] = params.pop('texts')
params['collection'] = collection
params['index_name'] = index_name
return class_object.from_documents(**params)
def initialize_supabase(class_object: Type[SupabaseVectorStore], params: dict, search: dict):
"""Initialize supabase and return the class object"""
from supabase.client import Client, create_client
if 'supabase_url' not in params or 'supabase_service_key' not in params:
raise ValueError('Supabase url and service key must be provided in the params')
if 'texts' in params:
params['documents'] = params.pop('texts')
client_kwargs = {
'supabase_url': params.pop('supabase_url'),
'supabase_key': params.pop('supabase_service_key'),
}
supabase: Client = create_client(**client_kwargs)
if not docs_in_params(params):
params.pop('documents', None)
params.pop('texts', None)
return class_object(client=supabase, **params)
# If there are docs in the params, create a new index
return class_object.from_documents(client=supabase, **params)
def initialize_weaviate(class_object: Type[Weaviate], params: dict, search: dict):
"""Initialize weaviate and return the class object"""
if not docs_in_params(params):
import weaviate # type: ignore
client_kwargs_json = params.get('client_kwargs', '{}')
client_kwargs = json.loads(client_kwargs_json)
client_params = {
'url': params.get('weaviate_url'),
}
client_params.update(client_kwargs)
weaviate_client = weaviate.Client(**client_params)
new_params = {
'client': weaviate_client,
'index_name': params.get('index_name'),
'text_key': params.get('text_key'),
}
return class_object(**new_params)
# If there are docs in the params, create a new index
if 'texts' in params:
params['documents'] = params.pop('texts')
return class_object.from_documents(**params)
def initialize_faiss(class_object: Type[FAISS], params: dict, search: dict):
"""Initialize faiss and return the class object"""
if not docs_in_params(params):
return class_object.load_local
save_local = params.get('save_local')
faiss_index = class_object(**params)
if save_local:
faiss_index.save_local(folder_path=save_local)
return faiss_index
def initialize_pinecone(class_object: Type[Pinecone], params: dict, search: dict):
"""Initialize pinecone and return the class object"""
import pinecone # type: ignore
pinecone_api_key = params.get('pinecone_api_key')
pinecone_env = params.get('pinecone_env')
if pinecone_api_key is None or pinecone_env is None:
if os.getenv('PINECONE_API_KEY') is not None:
pinecone_api_key = os.getenv('PINECONE_API_KEY')
if os.getenv('PINECONE_ENV') is not None:
pinecone_env = os.getenv('PINECONE_ENV')
if pinecone_api_key is None or pinecone_env is None:
raise ValueError('Pinecone API key and environment must be provided in the params')
# initialize pinecone
pinecone.init(
api_key=pinecone_api_key, # find at app.pinecone.io
environment=pinecone_env, # next to api key in console
)
# If there are no docs in the params, return an existing index
# but first remove any texts or docs keys from the params
if not docs_in_params(params):
existing_index_params = {
'embedding': params.pop('embedding'),
}
if 'index_name' in params:
existing_index_params['index_name'] = params.pop('index_name')
if 'namespace' in params:
existing_index_params['namespace'] = params.pop('namespace')
return class_object.from_existing_index(**existing_index_params)
# If there are docs in the params, create a new index
if 'texts' in params:
params['documents'] = params.pop('texts')
return class_object.from_documents(**params)
def initialize_chroma(class_object: Type[Chroma], params: dict, search: dict):
"""Initialize a ChromaDB object from the params"""
persist = params.pop('persist', False)
if not docs_in_params(params):
params.pop('documents', None)
params.pop('texts', None)
params['embedding_function'] = params.pop('embedding')
chromadb = class_object(**params)
else:
if 'texts' in params:
params['documents'] = params.pop('texts')
for doc in params['documents']:
if doc.metadata is None:
doc.metadata = {}
for key, value in doc.metadata.items():
if value is None:
doc.metadata[key] = ''
chromadb = class_object.from_documents(**params)
if persist:
chromadb.persist()
return chromadb
def initialize_qdrant(class_object: Type[Qdrant], params: dict, search: dict):
if not docs_in_params(params):
if 'location' not in params and 'api_key' not in params:
raise ValueError('Location and API key must be provided in the params')
from qdrant_client import QdrantClient
client_params = {
'location': params.pop('location'),
'api_key': params.pop('api_key'),
}
lc_params = {
'collection_name': params.pop('collection_name'),
'embeddings': params.pop('embedding'),
}
client = QdrantClient(**client_params)
return class_object(client=client, **lc_params)
return class_object.from_documents(**params)
def initial_milvus(class_object: Type[Milvus], params: dict, search_kwargs: dict):
if not params.get('connection_args') and settings.get_vectors_conf().milvus.connection_args:
params['connection_args'] = settings.get_vectors_conf().milvus.connection_args
elif isinstance(params.get('connection_args'), str):
print(f"milvus before params={params} type={type(params['connection_args'])}")
params['connection_args'] = json.loads(params.pop('connection_args'))
if 'embedding' not in params:
# Matching Knowledge Base'sembedding
col = params['collection_name']
collection_id = params.pop('collection_id', '')
with get_sync_db_session() as session:
if collection_id:
knowledge = session.get(Knowledge, collection_id)
params['collection_name'] = knowledge.collection_name
else:
knowledge = session.exec(
select(Knowledge).where(Knowledge.collection_name == col)).first()
if not knowledge:
raise ValueError(f'Unable to find knowledge basecollection={col} knowledge_id={collection_id}')
params['embedding'] = LLMService.get_bisheng_knowledge_embedding_sync(0, model_id=int(knowledge.model))
if knowledge.collection_name.startswith('partition'):
search_kwargs.update({'partition_key': knowledge.id})
params['partition_key'] = knowledge.id
logger.info('init_milvus collection_name={} partition={}', params['collection_name'],
search_kwargs)
return class_object.from_documents(**params)
def initial_elastic(class_object: Type[ElasticKeywordsSearch], params: dict, search: dict):
if not params.get('elasticsearch_url') and settings.get_vectors_conf().elasticsearch.elasticsearch_url:
params['elasticsearch_url'] = settings.get_vectors_conf().elasticsearch.elasticsearch_url
if not params.get('ssl_verify') and settings.get_vectors_conf().elasticsearch.ssl_verify:
params['ssl_verify'] = settings.get_vectors_conf().elasticsearch.ssl_verify
if isinstance(params.get('ssl_verify'), str):
params['ssl_verify'] = ast.literal_eval(params['ssl_verify'])
collection_id = params.pop('collection_id', '')
if collection_id:
with get_sync_db_session() as session:
knowledge = session.get(Knowledge, collection_id)
index_name = knowledge.index_name or knowledge.collection_name
params['index_name'] = index_name
params['embedding'] = ''
return class_object.from_documents(**params)
def initial_elastic_vector(class_object: Type[ElasticKeywordsSearch], params: dict, search: dict):
params.update(params.pop('connect_kwargs'))
collection_id = params.pop('collection_id', '')
if collection_id:
knowledge = KnowledgeDao.query_by_id(collection_id)
index_name = knowledge.index_name or knowledge.collection_name
params['index_name'] = index_name
params['embedding'] = LLMService.get_bisheng_knowledge_embedding_sync(0, model_id=int(knowledge.model))
if params['documents']:
return class_object.from_documents(**params)
else:
params.pop('documents')
return class_object(**params)
vecstore_initializer: Dict[str, Callable[[Type[Any], dict], Any]] = {
'Pinecone': initialize_pinecone,
'Chroma': initialize_chroma,
'Qdrant': initialize_qdrant,
'Weaviate': initialize_weaviate,
'FAISS': initialize_faiss,
'Milvus': initial_milvus,
'MilvusWithPermissionCheck': initial_milvus,
'ElasticsearchStore': initial_elastic_vector,
'ElasticKeywordsSearch': initial_elastic,
'ElasticsearchWithPermissionCheck': initial_elastic,
'SupabaseVectorStore': initialize_supabase,
'MongoDBAtlasVectorSearch': initialize_mongodb,
}
================================================
FILE: src/backend/bisheng/interface/inputoutput/__init__.py
================================================
================================================
FILE: src/backend/bisheng/interface/inputoutput/base.py
================================================
from typing import Dict, List, Optional, Type
from bisheng.custom.customs import get_custom_nodes
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.importing.utils import import_class
from bisheng.template.frontend_node.base import FrontendNode
from bisheng.template.frontend_node.input_output import InputOutputNode
from loguru import logger
from bisheng.utils.util import build_template_from_class
from bisheng_langchain import input_output
class IOutputCreator(LangChainTypeCreator):
type_name: str = 'input_output'
@property
def frontend_node_class(self) -> Type[FrontendNode]:
return InputOutputNode
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict = {}
# bisheng-langchain
bisheng = {
node_name: import_class(f'bisheng_langchain.input_output.{node_name}')
for node_name in input_output.__all__
}
self.type_dict.update(bisheng)
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
"""Get the signature of an llm."""
try:
if name in get_custom_nodes(self.type_name).keys():
return get_custom_nodes(self.type_name)[name]
return build_template_from_class(
name,
type_to_cls_dict=self.type_to_loader_dict,
)
except ValueError as exc:
raise ValueError('LLM not found') from exc
except AttributeError as exc:
logger.error(f'LLM {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
return self.type_to_loader_dict.keys()
input_output_creator = IOutputCreator()
================================================
FILE: src/backend/bisheng/interface/listing.py
================================================
from bisheng.utils.lazy_load import LazyLoadDictBase
# from bisheng.interface.agents.base import agent_creator
# from bisheng.interface.autogenRole.base import autogenrole_creator
# from bisheng.interface.chains.base import chain_creator
# from bisheng.interface.document_loaders.base import documentloader_creator
# from bisheng.interface.embeddings.base import embedding_creator
# from bisheng.interface.inputoutput.base import input_output_creator
# from bisheng.interface.llms.base import llm_creator
# from bisheng.interface.memories.base import memory_creator
# from bisheng.interface.output_parsers.base import output_parser_creator
# from bisheng.interface.prompts.base import prompt_creator
# from bisheng.interface.retrievers.base import retriever_creator
# from bisheng.interface.text_splitters.base import textsplitter_creator
# from bisheng.interface.toolkits.base import toolkits_creator
# from bisheng.interface.tools.base import tool_creator
# from bisheng.interface.utilities.base import utility_creator
# from bisheng.interface.vector_store.base import vectorstore_creator
# from bisheng.interface.wrappers.base import wrapper_creator
# def get_type_dict():
# return {
# 'agents': agent_creator.to_list(),
# 'prompts': prompt_creator.to_list(),
# 'llms': llm_creator.to_list(),
# 'tools': tool_creator.to_list(),
# 'chains': chain_creator.to_list(),
# 'memory': memory_creator.to_list(),
# 'toolkits': toolkits_creator.to_list(),
# 'wrappers': wrapper_creator.to_list(),
# 'documentLoaders': documentloader_creator.to_list(),
# 'vectorStore': vectorstore_creator.to_list(),
# 'embeddings': embedding_creator.to_list(),
# 'textSplitters': textsplitter_creator.to_list(),
# 'utilities': utility_creator.to_list(),
# 'outputParsers': output_parser_creator.to_list(),
# 'retrievers': retriever_creator.to_list(),
# 'inputOutput': input_output_creator.to_list(),
# 'autogenRoles': autogenrole_creator.to_list(),
# }
# LANGCHAIN_TYPES_DICT = get_type_dict()
# # Now we'll build a dict with Langchain types and ours
# ALL_TYPES_DICT = {
# **LANGCHAIN_TYPES_DICT,
# 'Custom': ['Custom Tool', 'Python Function'],
# }
class AllTypesDict(LazyLoadDictBase):
def __init__(self):
self._all_types_dict = None
@property
def ALL_TYPES_DICT(self):
return self.all_types_dict
def _build_dict(self):
langchain_types_dict = self.get_type_dict()
return {
**langchain_types_dict,
'Custom': ['Custom Tool', 'Python Function'],
}
def get_type_dict(self):
from bisheng.interface.types import get_all_types_dict
return get_all_types_dict()
lazy_load_dict = AllTypesDict()
================================================
FILE: src/backend/bisheng/interface/llms/__init__.py
================================================
from bisheng.interface.llms.base import LLMCreator
__all__ = ['LLMCreator']
================================================
FILE: src/backend/bisheng/interface/llms/base.py
================================================
from typing import Dict, List, Optional, Type
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.custom_lists import llm_type_to_cls_dict
from bisheng.llm.domain.llm import BishengLLM
from bisheng.common.services.config_service import settings
from bisheng.template.frontend_node.llms import LLMFrontendNode
from loguru import logger
from bisheng.utils.util import build_template_from_class
class LLMCreator(LangChainTypeCreator):
type_name: str = 'llms'
@property
def frontend_node_class(self) -> Type[LLMFrontendNode]:
return LLMFrontendNode
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict = llm_type_to_cls_dict
self.type_dict.update({
'BishengLLM': BishengLLM,
})
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
"""Get the signature of an llm."""
try:
return build_template_from_class(name, llm_type_to_cls_dict)
except ValueError as exc:
raise ValueError('LLM not found') from exc
except AttributeError as exc:
logger.error(f'LLM {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
return [
llm.__name__
for llm in self.type_to_loader_dict.values()
if llm.__name__ in settings.llms or settings.dev
]
llm_creator = LLMCreator()
================================================
FILE: src/backend/bisheng/interface/memories/__init__.py
================================================
from bisheng.interface.memories.base import MemoryCreator
__all__ = ['MemoryCreator']
================================================
FILE: src/backend/bisheng/interface/memories/base.py
================================================
from typing import ClassVar, Dict, List, Optional, Type
from bisheng.custom.customs import get_custom_nodes
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.custom_lists import memory_type_to_cls_dict
from bisheng.common.services.config_service import settings
from bisheng.template.frontend_node.base import FrontendNode
from bisheng.template.frontend_node.memories import MemoryFrontendNode
from loguru import logger
from bisheng.utils.util import build_template_from_class, build_template_from_method
class MemoryCreator(LangChainTypeCreator):
type_name: str = 'memories'
from_method_nodes: ClassVar[Dict] = {
'ZepChatMessageHistory': '__init__',
'SQLiteEntityStore': '__init__',
}
@property
def frontend_node_class(self) -> Type[FrontendNode]:
"""The class type of the FrontendNode created in frontend_node."""
return MemoryFrontendNode
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict = memory_type_to_cls_dict
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
"""Get the signature of a memory."""
try:
if name in get_custom_nodes(self.type_name).keys():
return get_custom_nodes(self.type_name)[name]
elif name in self.from_method_nodes:
return build_template_from_method(
name,
type_to_cls_dict=memory_type_to_cls_dict,
method_name=self.from_method_nodes[name],
)
return build_template_from_class(name, memory_type_to_cls_dict)
except ValueError as exc:
raise ValueError('Memory not found') from exc
except AttributeError as exc:
logger.error(f'Memory {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
return [
memory.__name__ for memory in self.type_to_loader_dict.values()
if memory.__name__ in settings.memories or settings.dev
]
memory_creator = MemoryCreator()
================================================
FILE: src/backend/bisheng/interface/output_parsers/__init__.py
================================================
================================================
FILE: src/backend/bisheng/interface/output_parsers/base.py
================================================
from typing import ClassVar, Dict, List, Optional, Type
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.importing.utils import import_class
from bisheng.common.services.config_service import settings
from bisheng.template.frontend_node.output_parsers import OutputParserFrontendNode
from loguru import logger
from bisheng.utils.util import build_template_from_class, build_template_from_method
from langchain import output_parsers
class OutputParserCreator(LangChainTypeCreator):
type_name: str = 'output_parsers'
from_method_nodes: ClassVar[Dict] = {
'StructuredOutputParser': 'from_response_schemas',
}
@property
def frontend_node_class(self) -> Type[OutputParserFrontendNode]:
return OutputParserFrontendNode
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict = {}
for output_parser_name in output_parsers.__all__:
if not settings.dev and output_parser_name not in settings.output_parsers:
continue
if output_parser_name == "GuardrailsOutputParser":
self.type_dict[output_parser_name] = import_class(f'langchain_community.output_parsers.rail_parser.{output_parser_name}')
else:
self.type_dict[output_parser_name] = import_class(f'langchain.output_parsers.{output_parser_name}')
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
try:
if name in self.from_method_nodes:
return build_template_from_method(
name,
type_to_cls_dict=self.type_to_loader_dict,
method_name=self.from_method_nodes[name],
)
# elif name in get_custom_nodes(self.type_name).keys():
# return get_custom_nodes(self.type_name)[name]
else:
return build_template_from_class(
name,
type_to_cls_dict=self.type_to_loader_dict,
)
except ValueError as exc:
# raise ValueError("OutputParser not found") from exc
logger.error(f'OutputParser {name} not found: {exc}')
except AttributeError as exc:
logger.error(f'OutputParser {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
return list(self.type_to_loader_dict.keys())
output_parser_creator = OutputParserCreator()
================================================
FILE: src/backend/bisheng/interface/prompts/__init__.py
================================================
from bisheng.interface.prompts.base import PromptCreator
__all__ = ['PromptCreator']
================================================
FILE: src/backend/bisheng/interface/prompts/base.py
================================================
from typing import Dict, List, Optional, Type
from bisheng.custom.customs import get_custom_nodes
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.importing.utils import import_class
from bisheng.common.services.config_service import settings
from bisheng.template.frontend_node.prompts import PromptFrontendNode
from loguru import logger
from bisheng.utils.util import build_template_from_class
from langchain import prompts
class PromptCreator(LangChainTypeCreator):
type_name: str = 'prompts'
@property
def frontend_node_class(self) -> Type[PromptFrontendNode]:
return PromptFrontendNode
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict = {
prompt_name: import_class(f'langchain.prompts.{prompt_name}')
# if prompt_name is not lower case it is a class
for prompt_name in prompts.__all__
}
# Merge CUSTOM_PROMPTS into self.type_dict
from bisheng.interface.prompts.custom import CUSTOM_PROMPTS
self.type_dict.update(CUSTOM_PROMPTS)
# Now filter according to settings.prompts
self.type_dict = {
name: prompt
for name, prompt in self.type_dict.items()
if name in settings.prompts or settings.dev
}
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
try:
if name in get_custom_nodes(self.type_name).keys():
return get_custom_nodes(self.type_name)[name]
return build_template_from_class(name, self.type_to_loader_dict)
except ValueError as exc:
# raise ValueError("Prompt not found") from exc
logger.error(f'Prompt {name} not found: {exc}')
except AttributeError as exc:
logger.error(f'Prompt {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
custom_prompts = get_custom_nodes('prompts')
# library_prompts = [
# prompt.__annotations__["return"].__name__
# for prompt in self.type_to_loader_dict.values()
# if prompt.__annotations__["return"].__name__ in settings.prompts
# or settings.dev
# ]
return list(self.type_to_loader_dict.keys()) + list(custom_prompts.keys())
prompt_creator = PromptCreator()
================================================
FILE: src/backend/bisheng/interface/prompts/custom.py
================================================
from typing import Dict, List, Optional, Type
from bisheng.interface.utils import extract_input_variables_from_prompt
from langchain.prompts import PromptTemplate
from pydantic import model_validator
# Steps to create a BaseCustomPrompt:
# 1. Create a prompt template that endes with:
# Current conversation:
# {history}
# Human: {input}
# {ai_prefix}:
# 2. Create a class that inherits from BaseCustomPrompt
# 3. Add the following class attributes:
# template: str = ""
# description: Optional[str]
# ai_prefix: Optional[str] = "{ai_prefix}"
# 3.1. The ai_prefix should be a value in input_variables
# SeriesCharacterPrompt is a working example
# If used in a LLMChain, with a Memory module, it will work as expected
# We should consider creating ConversationalChains that expose custom parameters
# That way it will be easier to create custom prompts
class BaseCustomPrompt(PromptTemplate):
template: str = ''
description: Optional[str]
ai_prefix: Optional[str]
@model_validator(mode='before')
@classmethod
def build_template(cls, values):
format_dict = {}
ai_prefix_format_dict = {}
for key in values.get('input_variables', []):
new_value = values.get(key, f'{{{key}}}')
format_dict[key] = new_value
if key in values['ai_prefix']:
ai_prefix_format_dict[key] = new_value
values['ai_prefix'] = values['ai_prefix'].format(**ai_prefix_format_dict)
values['template'] = values['template'].format(**format_dict)
values['template'] = values['template']
values['input_variables'] = extract_input_variables_from_prompt(
values['template']
)
return values
class SeriesCharacterPrompt(BaseCustomPrompt):
# Add a very descriptive description for the prompt generator
description: Optional[
str
] = 'A prompt that asks the AI to act like a character from a series.'
character: str
series: str
template: str = """I want you to act like {character} from {series}.
I want you to respond and answer like {character}. do not write any explanations. only answer like {character}.
You must know all of the knowledge of {character}.
Current conversation:
{history}
Human: {input}
{character}:"""
ai_prefix: str = '{character}'
input_variables: List[str] = ['character', 'series']
CUSTOM_PROMPTS: Dict[str, Type[BaseCustomPrompt]] = {
'SeriesCharacterPrompt': SeriesCharacterPrompt
}
================================================
FILE: src/backend/bisheng/interface/retrievers/__init__.py
================================================
================================================
FILE: src/backend/bisheng/interface/retrievers/base.py
================================================
from typing import Any, ClassVar, Dict, List, Optional, Type
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.importing.utils import import_class
from bisheng.common.services.config_service import settings
from bisheng.template.frontend_node.retrievers import RetrieverFrontendNode
from loguru import logger
from bisheng.utils.util import build_template_from_class, build_template_from_method
from bisheng_langchain import retrievers as bisheng_retrievers
from langchain.retrievers import MultiQueryRetriever
from langchain_community import retrievers
class RetrieverCreator(LangChainTypeCreator):
type_name: str = 'retrievers'
from_method_nodes: ClassVar[Dict] = {
'MultiQueryRetriever': 'from_llm',
'ZepRetriever': '__init__'
}
@property
def frontend_node_class(self) -> Type[RetrieverFrontendNode]:
return RetrieverFrontendNode
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict: dict[str, Any] = {
retriever_name: import_class(f'langchain_community.retrievers.{retriever_name}')
for retriever_name in retrievers.__all__
}
self.type_dict.update({
retriever_name:
import_class(f'bisheng_langchain.retrievers.{retriever_name}')
for retriever_name in bisheng_retrievers.__all__
})
self.type_dict['MultiQueryRetriever'] = MultiQueryRetriever
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
"""Get the signature of an embedding."""
try:
if name in self.from_method_nodes:
return build_template_from_method(
name,
type_to_cls_dict=self.type_to_loader_dict,
method_name=self.from_method_nodes[name],
)
else:
return build_template_from_class(name, type_to_cls_dict=self.type_to_loader_dict)
except ValueError as exc:
raise ValueError(f'Retriever {name} not found') from exc
except AttributeError as exc:
logger.error(f'Retriever {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
return [
retriever for retriever in self.type_to_loader_dict.keys()
if retriever in settings.retrievers or settings.dev
]
retriever_creator = RetrieverCreator()
================================================
FILE: src/backend/bisheng/interface/run.py
================================================
from typing import Dict, Optional, Tuple, Union
from uuid import UUID
from bisheng.core.cache.utils import memoize_dict
from bisheng.graph import Graph
from loguru import logger
@memoize_dict(maxsize=10)
def build_langchain_object_with_caching(data_graph):
"""
Build langchain object from data_graph.
"""
logger.debug('Building langchain object')
graph = Graph.from_payload(data_graph)
return graph.build()
@memoize_dict(maxsize=10)
def build_sorted_vertices_with_caching(data_graph):
"""
Build langchain object from data_graph.
"""
logger.debug('Building langchain object')
graph = Graph.from_payload(data_graph)
sorted_vertices = graph.topological_sort()
artifacts = {}
for vertex in sorted_vertices:
vertex.build()
if vertex.artifacts:
artifacts.update(vertex.artifacts)
root = graph.build()
langchain_object = next((node for node in root if node.base_type != 'inputOutput'), None)
return langchain_object.build(), artifacts
def build_langchain_object(data_graph):
"""
Build langchain object from data_graph.
"""
logger.debug('Building langchain object')
nodes = data_graph['nodes']
# Add input variables
# nodes = payload.extract_input_variables(nodes)
# Nodes, edges and root node
edges = data_graph['edges']
graph = Graph(nodes, edges)
return graph.build()
async def build_sorted_vertices(data_graph,
user_id: Optional[Union[str, UUID]] = None) -> Tuple[Graph, Dict]:
"""
Build langchain object from data_graph.
"""
logger.debug('Building langchain object')
graph = Graph.from_payload(data_graph)
sorted_vertices = graph.topological_sort()
artifacts = {}
for vertex in sorted_vertices:
vertex.build()
if vertex.artifacts:
artifacts.update(vertex.artifacts)
return graph, artifacts
def get_memory_key(langchain_object):
"""
Given a LangChain object, this function retrieves the current memory key from the object's memory attribute.
It then checks if the key exists in a dictionary of known memory keys and returns the corresponding key,
or None if the current key is not recognized.
"""
mem_key_dict = {
'chat_history': 'history',
'history': 'chat_history',
}
memory_key = langchain_object.memory.memory_key
return mem_key_dict.get(memory_key)
def update_memory_keys(langchain_object, possible_new_mem_key):
"""
Given a LangChain object and a possible new memory key, this function updates the input and output keys in the
object's memory attribute to exclude the current memory key and the possible new key. It then sets the memory key
to the possible new key.
"""
input_key = [
key for key in langchain_object.input_keys
if key not in [langchain_object.memory.memory_key, possible_new_mem_key]
][0]
output_key = [
key for key in langchain_object.output_keys
if key not in [langchain_object.memory.memory_key, possible_new_mem_key]
][0]
keys = [input_key, output_key, possible_new_mem_key]
attrs = ['input_key', 'output_key', 'memory_key']
for key, attr in zip(keys, attrs):
try:
setattr(langchain_object.memory, attr, key)
except ValueError as exc:
logger.debug(f'{langchain_object.memory} has no attribute {attr} ({exc})')
================================================
FILE: src/backend/bisheng/interface/text_splitters/__init__.py
================================================
from bisheng.interface.text_splitters.base import TextSplitterCreator
__all__ = ['TextSplitterCreator']
================================================
FILE: src/backend/bisheng/interface/text_splitters/base.py
================================================
from typing import Dict, List, Optional, Type
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.custom_lists import textsplitter_type_to_cls_dict
from bisheng.common.services.config_service import settings
from bisheng.template.frontend_node.textsplitters import \
TextSplittersFrontendNode
from loguru import logger
from bisheng.utils.util import build_template_from_class
class TextSplitterCreator(LangChainTypeCreator):
type_name: str = 'textsplitters'
@property
def frontend_node_class(self) -> Type[TextSplittersFrontendNode]:
return TextSplittersFrontendNode
@property
def type_to_loader_dict(self) -> Dict:
return textsplitter_type_to_cls_dict
def get_signature(self, name: str) -> Optional[Dict]:
"""Get the signature of a text splitter."""
try:
return build_template_from_class(name, textsplitter_type_to_cls_dict)
except ValueError as exc:
raise ValueError(f'Text Splitter {name} not found') from exc
except AttributeError as exc:
logger.error(f'Text Splitter {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
return [
textsplitter.__name__
for textsplitter in self.type_to_loader_dict.values()
if textsplitter.__name__ in settings.textsplitters or settings.dev
]
textsplitter_creator = TextSplitterCreator()
================================================
FILE: src/backend/bisheng/interface/toolkits/__init__.py
================================================
================================================
FILE: src/backend/bisheng/interface/toolkits/base.py
================================================
from typing import Callable, Dict, List, Optional
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.importing.utils import import_class, import_module
from bisheng.common.services.config_service import settings
from loguru import logger
from bisheng.utils.util import build_template_from_class
from langchain.agents.agent_toolkits.vectorstore.toolkit import (VectorStoreInfo,
VectorStoreRouterToolkit,
VectorStoreToolkit)
from langchain_community import agent_toolkits
class ToolkitCreator(LangChainTypeCreator):
type_name: str = 'toolkits'
all_types: List[str] = agent_toolkits.__all__
create_functions: Dict = {
'JsonToolkit': [],
'SQLDatabaseToolkit': [],
'OpenAPIToolkit': ['create_openapi_agent'],
'VectorStoreToolkit': [
'create_vectorstore_agent',
'create_vectorstore_router_agent',
'VectorStoreInfo',
],
'ZapierToolkit': [],
'PandasToolkit': ['create_pandas_dataframe_agent'],
'CSVToolkit': ['create_csv_agent'],
}
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict = {
toolkit_name: import_class(f'langchain_community.agent_toolkits.{toolkit_name}')
# if toolkit_name is not lower case it is a class
for toolkit_name in agent_toolkits.__all__
if not toolkit_name.islower() and toolkit_name in settings.toolkits
}
self.type_dict.update({
'VectorStoreToolkit': VectorStoreToolkit,
'VectorStoreInfo': VectorStoreInfo,
'VectorStoreRouterToolkit': VectorStoreRouterToolkit
})
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
try:
template = build_template_from_class(name, self.type_to_loader_dict)
# add Tool to base_classes
if 'toolkit' in name.lower() and template:
template['base_classes'].append('Tool')
return template
except ValueError as exc:
raise ValueError('Toolkit not found') from exc
except AttributeError as exc:
logger.error(f'Toolkit {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
return list(self.type_to_loader_dict.keys())
def get_create_function(self, name: str) -> Callable:
if loader_name := self.create_functions.get(name):
return import_module(
f'from langchain_community.agent_toolkits import {loader_name[0]}')
else:
raise ValueError('Toolkit not found')
def has_create_function(self, name: str) -> bool:
# check if the function list is not empty
return bool(self.create_functions.get(name, None))
toolkits_creator = ToolkitCreator()
================================================
FILE: src/backend/bisheng/interface/toolkits/custom.py
================================================
================================================
FILE: src/backend/bisheng/interface/tools/__init__.py
================================================
from bisheng.interface.tools.base import ToolCreator
__all__ = ['ToolCreator']
================================================
FILE: src/backend/bisheng/interface/tools/base.py
================================================
from typing import Dict, List, Optional
from bisheng.custom import customs
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.tools.constants import ALL_TOOLS_NAMES, CUSTOM_TOOLS, FILE_TOOLS, OTHER_TOOLS
from bisheng.interface.tools.util import get_tool_params
from bisheng.common.services.config_service import settings
from bisheng.template.field.base import TemplateField
from bisheng.template.template.base import Template
from bisheng.utils import util
from bisheng.utils.util import build_template_from_class
from langchain_community.agent_toolkits.load_tools import _EXTRA_LLM_TOOLS, _EXTRA_OPTIONAL_TOOLS, _LLM_TOOLS
TOOL_INPUTS = {
'str':
TemplateField(
field_type='str',
required=True,
is_list=False,
show=True,
placeholder='',
value='',
),
'llm':
TemplateField(field_type='BaseLanguageModel', required=True, is_list=False, show=True),
'func':
TemplateField(
field_type='function',
required=True,
is_list=False,
show=True,
multiline=True,
),
'code':
TemplateField(
field_type='str',
required=True,
is_list=False,
show=True,
value='',
multiline=True,
),
'path':
TemplateField(
field_type='file',
required=True,
is_list=False,
show=True,
value='',
suffixes=['.json', '.yaml', '.yml'],
fileTypes=['json', 'yaml', 'yml'],
),
}
class ToolCreator(LangChainTypeCreator):
type_name: str = 'tools'
tools_dict: Optional[Dict] = None
@property
def type_to_loader_dict(self) -> Dict:
if self.tools_dict is None:
all_tools = {}
for tool, tool_fcn in ALL_TOOLS_NAMES.items():
tool_params = get_tool_params(tool_fcn)
tool_name = tool_params.get('name') or tool
if tool_name in settings.tools or settings.dev:
if tool_name == 'JsonSpec':
tool_params['path'] = tool_params.pop('dict_') # type: ignore
all_tools[tool_name] = {
'type': tool,
'params': tool_params,
'fcn': tool_fcn,
}
self.tools_dict = all_tools
return self.tools_dict
def get_signature(self, name: str) -> Optional[Dict]:
"""Get the signature of a tool."""
base_classes = ['Tool', 'BaseTool']
fields = []
params = []
tool_params = {}
# Raise error if name is not in tools
if name not in self.type_to_loader_dict.keys():
raise ValueError('Tool not found')
tool_type: str = self.type_to_loader_dict[name]['type'] # type: ignore
# if tool_type in _BASE_TOOLS.keys():
# params = []
if tool_type in _LLM_TOOLS.keys():
params = ['llm']
elif tool_type in _EXTRA_LLM_TOOLS.keys():
extra_keys = _EXTRA_LLM_TOOLS[tool_type][1]
params = ['llm'] + extra_keys
elif tool_type in _EXTRA_OPTIONAL_TOOLS.keys():
extra_keys = _EXTRA_OPTIONAL_TOOLS[tool_type][1]
params = extra_keys
# elif tool_type == "Tool":
# params = ["name", "description", "func"]
elif tool_type in CUSTOM_TOOLS:
# Get custom tool params
params = self.type_to_loader_dict[name]['params'] # type: ignore
base_classes = ['function']
if node := customs.get_custom_nodes('tools').get(tool_type):
return node
elif tool_type in FILE_TOOLS:
params = self.type_to_loader_dict[name]['params'] # type: ignore
base_classes += [name]
elif tool_type in OTHER_TOOLS:
tool_dict = build_template_from_class(tool_type, OTHER_TOOLS)
fields = tool_dict['template']
# Pop unnecessary fields and add name
fields.pop('_type') # type: ignore
fields.pop('return_direct', None) # type: ignore
fields.pop('verbose', None) # type: ignore
tool_params = {
'name': fields.pop('name')['value'], # type: ignore
'description': fields.pop('description')['value'], # type: ignore
}
fields = [
TemplateField(name=name, field_type=field['type'], **field)
for name, field in fields.items() # type: ignore
]
base_classes += tool_dict['base_classes']
# Copy the field and add the name
for param in params:
field = TOOL_INPUTS.get(param, TOOL_INPUTS['str']).copy()
field.name = param
field.advanced = False
if param == 'aiosession':
field.show = False
field.required = False
fields.append(field)
template = Template(fields=fields, type_name=tool_type)
# add describe
template.add_field(
TemplateField(
field_type='NestedDict',
required=True,
placeholder='',
show=True,
multiline=True,
value='{"arg1": {"type": "string"}}',
name='args_schema',
), )
tool_params = {**tool_params, **self.type_to_loader_dict[name]['params']}
return {
'template': util.format_dict(template.to_dict()),
**tool_params,
'base_classes': base_classes,
}
def to_list(self) -> List[str]:
"""List all load tools"""
return list(self.type_to_loader_dict.keys())
tool_creator = ToolCreator()
================================================
FILE: src/backend/bisheng/interface/tools/constants.py
================================================
from bisheng.interface.importing.utils import import_class
from bisheng.interface.tools.custom import PythonFunction, PythonFunctionTool
from langchain_community.tools import Tool
from langchain_community.agent_toolkits.load_tools import (_BASE_TOOLS, _EXTRA_LLM_TOOLS, _EXTRA_OPTIONAL_TOOLS,
_LLM_TOOLS)
from langchain_community import tools
from langchain_community.tools.json.tool import JsonSpec
from langchain_experimental import tools as python_tools
FILE_TOOLS = {'JsonSpec': JsonSpec}
CUSTOM_TOOLS = {
'Tool': Tool,
'PythonFunctionTool': PythonFunctionTool,
'PythonFunction': PythonFunction,
}
OTHER_TOOLS = {tool: import_class(f'langchain_community.tools.{tool}') for tool in tools.__all__}
PYTHON_TOOLS = {
tool: import_class(f'langchain_experimental.tools.{tool}')
for tool in python_tools.__all__
}
ALL_TOOLS_NAMES = {
**_BASE_TOOLS,
**_LLM_TOOLS, # type: ignore
**{
k: v[0]
for k, v in _EXTRA_LLM_TOOLS.items()
}, # type: ignore
**{
k: v[0]
for k, v in _EXTRA_OPTIONAL_TOOLS.items()
},
**CUSTOM_TOOLS,
**FILE_TOOLS, # type: ignore
**OTHER_TOOLS,
**PYTHON_TOOLS
}
================================================
FILE: src/backend/bisheng/interface/tools/custom.py
================================================
from typing import Callable, Optional
from bisheng.interface.importing.utils import get_function
from bisheng.utils import validate
from langchain_community.tools import Tool
from pydantic import field_validator, BaseModel
class Function(BaseModel):
code: str
function: Optional[Callable] = None
imports: Optional[str] = None
# Eval code and store the function
def __init__(self, **data):
super().__init__(**data)
# Validate the function
@field_validator('code')
@classmethod
def validate_func(cls, v):
try:
validate.eval_function(v)
except Exception as e:
raise e
return v
def get_function(self):
"""Get the function"""
function_name = validate.extract_function_name(self.code)
return validate.create_function(self.code, function_name)
class PythonFunctionTool(Tool):
"""Python function"""
name: str = 'Custom Tool'
description: str
code: str
def ___init__(self, name: str, description: str, code: str):
self.name = name
self.description = description
self.code = code
self.func = get_function(self.code)
super().__init__(name=name, description=description, func=self.func)
class PythonFunction(Function):
"""Python function"""
code: str
================================================
FILE: src/backend/bisheng/interface/tools/util.py
================================================
import ast
import inspect
import textwrap
from typing import Dict, Union
from langchain_community.tools import Tool
def get_func_tool_params(func, **kwargs) -> Union[Dict, None]:
tree = ast.parse(inspect.getsource(func))
# Iterate over the statements in the abstract syntax tree
for node in ast.walk(tree):
# Find the first return statement
if isinstance(node, ast.Return):
tool = node.value
if isinstance(tool, ast.Call):
if isinstance(tool.func, ast.Name) and tool.func.id == 'Tool':
if tool.keywords:
tool_params = {}
for keyword in tool.keywords:
if keyword.arg == 'name':
try:
tool_params['name'] = ast.literal_eval(keyword.value)
except ValueError:
break
elif keyword.arg == 'description':
try:
tool_params['description'] = ast.literal_eval(keyword.value)
except ValueError:
continue
return tool_params
return {
'name': ast.literal_eval(tool.args[0]),
'description': ast.literal_eval(tool.args[2]),
}
#
else:
# get the class object from the return statement
try:
class_obj = eval(compile(ast.Expression(tool), '', 'eval'))
except Exception:
return None
return {
'name': getattr(class_obj, 'name'),
'description': getattr(class_obj, 'description'),
}
# Return None if no return statement was found
return None
def get_class_tool_params(cls, **kwargs) -> Union[Dict, None]:
tree = ast.parse(textwrap.dedent(inspect.getsource(cls)))
tool_params = {}
# Iterate over the statements in the abstract syntax tree
for node in ast.walk(tree):
if isinstance(node, ast.ClassDef):
# Find the class definition and look for methods
for stmt in node.body:
if isinstance(stmt, ast.FunctionDef) and stmt.name == '__init__':
# There is no assignment statements in the __init__ method
# So we need to get the params from the function definition
for arg in stmt.args.args:
if arg.arg == 'name':
# It should be the name of the class
tool_params[arg.arg] = cls.__name__
elif arg.arg == 'self':
continue
# If there is not default value, set it to an empty string
else:
try:
annotation = ast.literal_eval(arg.annotation) # type: ignore
tool_params[arg.arg] = annotation
except ValueError:
tool_params[arg.arg] = ''
# Get the attribute name and the annotation
elif cls != Tool and isinstance(stmt, ast.AnnAssign):
# Get the attribute name and the annotation
tool_params[stmt.target.id] = '' # type: ignore
return tool_params
def get_tool_params(tool, **kwargs) -> Dict:
# Parse the function code into an abstract syntax tree
# Define if it is a function or a class
if inspect.isfunction(tool):
return get_func_tool_params(tool, **kwargs) or {}
elif inspect.isclass(tool):
# Get the parameters necessary to
# instantiate the class
return get_class_tool_params(tool, **kwargs) or {}
else:
raise ValueError('Tool must be a function or class.')
================================================
FILE: src/backend/bisheng/interface/types.py
================================================
from pathlib import Path
from bisheng.interface.agents.base import agent_creator
from bisheng.interface.autogenRole.base import autogenrole_creator
from bisheng.interface.chains.base import chain_creator
from bisheng.interface.custom.directory_reader.utils import merge_nested_dicts_with_renaming
from bisheng.interface.custom.utils import build_custom_components
from bisheng.interface.document_loaders.base import documentloader_creator
from bisheng.interface.embeddings.base import embedding_creator
from bisheng.interface.inputoutput.base import input_output_creator
from bisheng.interface.llms.base import llm_creator
from bisheng.interface.memories.base import memory_creator
from bisheng.interface.output_parsers.base import output_parser_creator
from bisheng.interface.prompts.base import prompt_creator
from bisheng.interface.retrievers.base import retriever_creator
from bisheng.interface.text_splitters.base import textsplitter_creator
from bisheng.interface.toolkits.base import toolkits_creator
from bisheng.interface.tools.base import tool_creator
from bisheng.interface.utilities.base import utility_creator
from bisheng.interface.vector_store.base import vectorstore_creator
from bisheng.interface.wrappers.base import wrapper_creator
from cachetools import LRUCache, cached
def get_type_list():
"""Get a list of all langchain types"""
all_types = build_langchain_types_dict()
# all_types.pop("tools")
for key, value in all_types.items():
all_types[key] = [item['template']['_type'] for item in value.values()]
return all_types
@cached(LRUCache(maxsize=1))
def build_langchain_types_dict(): # sourcery skip: dict-assign-update-to-union
"""Build a dictionary of all langchain types"""
all_types = {}
creators = [
chain_creator,
agent_creator,
prompt_creator,
llm_creator,
memory_creator,
tool_creator,
toolkits_creator,
wrapper_creator,
embedding_creator,
vectorstore_creator,
documentloader_creator,
textsplitter_creator,
utility_creator,
output_parser_creator,
retriever_creator,
input_output_creator,
autogenrole_creator,
]
all_types = {}
for creator in creators:
created_types = creator.to_dict()
if created_types[creator.type_name].values():
all_types.update(created_types)
return all_types
langchain_types_dict = build_langchain_types_dict()
def get_all_types_dict():
"""Get all types dictionary combining native and custom components."""
BASE_COMPONENTS_PATH = str(Path(__file__).parent.parent / 'components')
native_components = build_langchain_types_dict()
custom_components_from_file = build_custom_components([BASE_COMPONENTS_PATH])
return merge_nested_dicts_with_renaming(native_components, custom_components_from_file)
# return native_components
================================================
FILE: src/backend/bisheng/interface/utilities/__init__.py
================================================
================================================
FILE: src/backend/bisheng/interface/utilities/base.py
================================================
from typing import Dict, List, Optional, Type
from bisheng.custom.customs import get_custom_nodes
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.importing.utils import import_class
from bisheng.common.services.config_service import settings
from bisheng.template.frontend_node.utilities import UtilitiesFrontendNode
from loguru import logger
from bisheng.utils.util import build_template_from_class
from langchain_community import utilities
from langchain_community.utilities.sql_database import SQLDatabase
class UtilityCreator(LangChainTypeCreator):
type_name: str = 'utilities'
@property
def frontend_node_class(self) -> Type[UtilitiesFrontendNode]:
return UtilitiesFrontendNode
@property
def type_to_loader_dict(self) -> Dict:
"""
Returns a dictionary mapping utility names to their corresponding loader classes.
If the dictionary has not been created yet, it is created by importing all utility classes
from the langchain.chains module and filtering them according to the settings.utilities list.
"""
if self.type_dict is None:
self.type_dict = {
utility_name: import_class(f'langchain_community.utilities.{utility_name}')
for utility_name in utilities.__all__
}
self.type_dict['SQLDatabase'] = SQLDatabase
# Filter according to settings.utilities
self.type_dict = {
name: utility
for name, utility in self.type_dict.items()
if name in settings.utilities or settings.dev
}
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
"""Get the signature of a utility."""
try:
custom_nodes = get_custom_nodes(self.type_name)
if name in custom_nodes.keys():
return custom_nodes[name]
return build_template_from_class(name, self.type_to_loader_dict)
except ValueError as exc:
raise ValueError(f'Utility {name} not found') from exc
except AttributeError as exc:
logger.error(f'Utility {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
return list(self.type_to_loader_dict.keys())
utility_creator = UtilityCreator()
================================================
FILE: src/backend/bisheng/interface/utils.py
================================================
import base64
import json
import os
import re
from io import BytesIO
import yaml
from PIL.Image import Image
from langchain.base_language import BaseLanguageModel
from bisheng.chat.config import ChatConfig
from bisheng.common.services.config_service import settings
def load_file_into_dict(file_path: str) -> dict:
if not os.path.exists(file_path):
raise FileNotFoundError(f'File not found: {file_path}')
# Files names are UUID, so we can't find the extension
with open(file_path, 'r') as file:
try:
data = json.load(file)
except json.JSONDecodeError:
file.seek(0)
data = yaml.safe_load(file)
except ValueError as exc:
raise ValueError('Invalid file type. Expected .json or .yaml.') from exc
return data
def pil_to_base64(image: Image) -> str:
buffered = BytesIO()
image.save(buffered, format='PNG')
img_str = base64.b64encode(buffered.getvalue())
return img_str.decode('utf-8')
def try_setting_streaming_options(langchain_object, websocket):
# If the LLM type is OpenAI or ChatOpenAI,
# set streaming to True
# First we need to find the LLM
llm = None
if hasattr(langchain_object, 'llm'):
llm = langchain_object.llm
elif hasattr(langchain_object, 'llm_chain') and hasattr(langchain_object.llm_chain, 'llm'):
llm = langchain_object.llm_chain.llm
if isinstance(llm, BaseLanguageModel):
if hasattr(llm, 'streaming') and isinstance(llm.streaming, bool):
llm.streaming = settings.get_from_db('llm_request').get(
'stream') if 'stream' in settings.get_from_db(
'llm_request') else ChatConfig.streaming
elif hasattr(llm, 'stream') and isinstance(llm.stream, bool):
llm.stream = settings.get_from_db('llm_request').get(
'stream') if 'stream' in settings.get_from_db(
'llm_request') else ChatConfig.streaming
return langchain_object
def extract_input_variables_from_prompt(prompt: str) -> list[str]:
"""Extract input variables from prompt."""
return re.findall(r'{(.*?)}', prompt)
================================================
FILE: src/backend/bisheng/interface/vector_store/__init__.py
================================================
================================================
FILE: src/backend/bisheng/interface/vector_store/base.py
================================================
from typing import Any, Dict, List, Optional, Type
from bisheng.interface.base import LangChainTypeCreator
from bisheng.interface.importing.utils import import_class
from bisheng.interface.vector_store.constants import CUSTOM_VECTORSTORE
from bisheng.common.services.config_service import settings
from bisheng.template.frontend_node.vectorstores import VectorStoreFrontendNode
from loguru import logger
from bisheng.utils.util import build_template_from_method
from bisheng_langchain import vectorstores as contribute_vectorstores
from langchain_community import vectorstores
class VectorstoreCreator(LangChainTypeCreator):
type_name: str = 'vectorstores'
@property
def frontend_node_class(self) -> Type[VectorStoreFrontendNode]:
return VectorStoreFrontendNode
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict: dict[str, Any] = {
vectorstore_name:
import_class(f'langchain_community.vectorstores.{vectorstore_name}')
for vectorstore_name in vectorstores.__all__
if vectorstore_name != 'Milvus' # use bisheng_langchain
}
self.type_dict.update({
vectorstore_name:
import_class(f'bisheng_langchain.vectorstores.{vectorstore_name}')
for vectorstore_name in contribute_vectorstores.__all__
})
self.type_dict.update(CUSTOM_VECTORSTORE)
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
"""Get the signature of an embedding."""
try:
return build_template_from_method(
name,
type_to_cls_dict=self.type_to_loader_dict,
method_name='from_texts',
)
except ValueError as exc:
raise ValueError(f'Vector Store {name} not found') from exc
except AttributeError as exc:
logger.error(f'Vector Store {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
return [
vectorstore for vectorstore in self.type_to_loader_dict.keys()
if vectorstore in settings.vectorstores or settings.dev
]
vectorstore_creator = VectorstoreCreator()
================================================
FILE: src/backend/bisheng/interface/vector_store/constants.py
================================================
from bisheng.interface.vector_store.custom import MilvusWithPermissionCheck, ElasticsearchWithPermissionCheck
CUSTOM_VECTORSTORE = {
'MilvusWithPermissionCheck': MilvusWithPermissionCheck,
'ElasticsearchWithPermissionCheck': ElasticsearchWithPermissionCheck
}
================================================
FILE: src/backend/bisheng/interface/vector_store/custom.py
================================================
from abc import ABC
from ast import literal_eval
from typing import TYPE_CHECKING, Any, Callable, Dict, Iterable, List, Optional, Tuple
import jieba
from langchain.chains.llm import LLMChain
from langchain.docstore.document import Document
from langchain.embeddings.base import Embeddings
from langchain.utils import get_from_dict_or_env
from langchain.vectorstores.base import VectorStore
from langchain_community.vectorstores.milvus import Milvus as MilvusLangchain
from langchain_core.language_models import BaseLLM
from langchain_core.prompts import PromptTemplate
from loguru import logger
from bisheng_langchain.vectorstores.elastic_keywords_search import DEFAULT_PROMPT
from bisheng_langchain.vectorstores.milvus import DEFAULT_MILVUS_CONNECTION
if TYPE_CHECKING:
from elasticsearch import Elasticsearch # noqa: F401
class MilvusWithPermissionCheck(MilvusLangchain):
"""
only support multi collection search, but all collection must have same fields
not include create collection
"""
def __init__(self,
embedding_function: Embeddings,
collection_name: list[str] = None,
collection_embeddings: list[Embeddings] = None,
connection_args: Optional[dict[str, Any]] = None,
consistency_level: str = 'Session',
index_params: Optional[dict] = None,
search_params: Optional[dict] = None,
drop_old: Optional[bool] = False,
*,
primary_field: str = 'pk',
text_field: str = 'text',
vector_field: str = 'vector',
partition_field: str = 'knowledge_id',
**kwargs: Any):
"""Initialize the Milvus vector store."""
try:
from pymilvus import Collection, utility
except ImportError:
raise ValueError('Could not import pymilvus python package. '
'Please install it with `pip install pymilvus`.')
# Default search params when one is not provided.
self.default_search_params = {
'IVF_FLAT': {
'metric_type': 'L2',
'params': {
'nprobe': 10
}
},
'IVF_SQ8': {
'metric_type': 'L2',
'params': {
'nprobe': 10
}
},
'IVF_PQ': {
'metric_type': 'L2',
'params': {
'nprobe': 10
}
},
'HNSW': {
'metric_type': 'L2',
'params': {
'ef': 100
}
},
'RHNSW_FLAT': {
'metric_type': 'L2',
'params': {
'ef': 10
}
},
'RHNSW_SQ': {
'metric_type': 'L2',
'params': {
'ef': 10
}
},
'RHNSW_PQ': {
'metric_type': 'L2',
'params': {
'ef': 10
}
},
'IVF_HNSW': {
'metric_type': 'L2',
'params': {
'nprobe': 10,
'ef': 10
}
},
'ANNOY': {
'metric_type': 'L2',
'params': {
'search_k': 10
}
},
'AUTOINDEX': {
'metric_type': 'L2',
'params': {}
},
}
self.embedding_func = embedding_function
self.collection_name = collection_name
self.index_params = index_params
self.search_params = search_params
self.consistency_level = consistency_level
self.connection_args = connection_args
# In order for a collection to be compatible, pk needs to be auto'id and int
self._primary_field = primary_field
# In order for compatiblility, the text field will need to be called "text"
self._text_field = text_field
# In order for compatibility, the vector field needs to be called "vector"
self._vector_field = vector_field
# partion key for multi-tenancy
self._partition_field = partition_field
# if collection_name is None or collection_name.__len__() == 0:
# raise ValueError('collection_name cannot be empty, please provide at least one collection name.')
self.fields: list[str] = []
# Create the connection to the server
if connection_args is None:
connection_args = DEFAULT_MILVUS_CONNECTION
self.alias = self._create_connection_alias(connection_args)
self.col: Optional[List[Collection]] = []
self.col_partition_key: Optional[List[str]] = []
self.collection_embeddings = collection_embeddings
# not used
self.drop_old = drop_old
# Grab the existing collection if it exists
try:
for index, one_collection_name in enumerate(self.collection_name):
if utility.has_collection(one_collection_name, using=self.alias):
self.col.append(Collection(
one_collection_name,
using=self.alias,
))
self.col_partition_key.append(kwargs.get('partition_keys')[index])
except Exception as e:
logger.error(f'milvus operating error={str(e)}')
self.close_connection(self.alias)
raise e
if not self.col:
logger.warning('No collection found, please confirm user have knowledge access')
# raise ValueError("No collection found, please confirm collection name correctly.")
# Initialize the vector store
self._init()
def close_connection(self, using):
logger.warning(f'forbidden close milvus connection: {using}, because alias can be reused by others threads')
def _init(
self,
embeddings: Optional[list] = None,
metadatas: Optional[list[dict]] = None,
partition_names: Optional[list] = None,
replica_number: int = 1,
timeout: Optional[float] = None,
) -> None:
self._extract_fields(col_index=0)
self._create_search_params()
self._load()
def _extract_fields(self, col_index=0) -> None:
"""Grab the existing fields from the Collection"""
from pymilvus import Collection
if not self.col:
return
if isinstance(self.col[col_index], Collection):
schema = self.col[col_index].schema
for x in schema.fields:
self.fields.append(x.name)
# Since primary field is auto-id, no need to track it
self.fields.remove(self._primary_field)
def _create_search_params(self, col_index=0) -> None:
"""Generate search params based on the current index type"""
from pymilvus import Collection
if not self.col:
return
if isinstance(self.col[col_index], Collection) and self.search_params is None:
index = self._get_index(col_index)
if index is not None:
index_type: str = index['index_param']['index_type']
metric_type: str = index['index_param']['metric_type']
self.search_params = self.default_search_params[index_type]
self.search_params['metric_type'] = metric_type
def _get_index(self, col_index=0) -> Optional[dict[str, Any]]:
"""Return the vector index information if it exists"""
from pymilvus import Collection
if not self.col:
return
if isinstance(self.col[col_index], Collection):
for x in self.col[col_index].indexes:
if x.field_name == self._vector_field:
return x.to_dict()
return None
def _load(self) -> None:
"""Load the collection if available."""
from pymilvus import Collection
# Load allcollection
for i, col in enumerate(self.col):
if isinstance(col, Collection) and self._get_index(col_index=i) is not None:
col.load()
@classmethod
def from_texts(
cls,
texts: List[str],
embedding: Embeddings,
metadatas: Optional[List[dict]] = None,
collection_name: list[str] = None,
connection_args: dict[str, Any] = DEFAULT_MILVUS_CONNECTION,
consistency_level: str = 'Session',
index_params: Optional[dict] = None,
search_params: Optional[dict] = None,
drop_old: bool = False,
no_embedding: bool = False,
**kwargs: Any,
):
"""
no insert data into milvus, only search from milvus
"""
vector_db = cls(
embedding_function=embedding,
collection_name=collection_name,
connection_args=connection_args,
consistency_level=consistency_level,
index_params=index_params,
search_params=search_params,
drop_old=drop_old,
**kwargs,
)
return vector_db
def similarity_search(
self,
query: str,
k: int = 4,
param: Optional[dict] = None,
expr: Optional[str] = None,
timeout: Optional[int] = None,
**kwargs: Any,
) -> List[Document]:
"""Perform a similarity search against the query string.
Args:
query (str): The text to search.
k (int, optional): How many results to return. Defaults to 4.
param (dict, optional): The search params for the index type.
Defaults to None.
expr (str, optional): Filtering expression. Defaults to None.
timeout (int, optional): How long to wait before timeout error.
Defaults to None.
kwargs: Collection.search() keyword arguments.
Returns:
List[Document]: Document results for search.
"""
res = self.similarity_search_with_score(query=query,
k=k,
param=param,
expr=expr,
timeout=timeout,
**kwargs)
return [doc for doc, _ in res]
def similarity_search_by_vector(
self,
embedding: List[float],
k: int = 4,
param: Optional[dict] = None,
expr: Optional[str] = None,
timeout: Optional[int] = None,
**kwargs: Any,
) -> List[Document]:
"""Perform a similarity search against the query string.
Args:
embedding (List[float]): The embedding vector to search.
k (int, optional): How many results to return. Defaults to 4.
param (dict, optional): The search params for the index type.
Defaults to None.
expr (str, optional): Filtering expression. Defaults to None.
timeout (int, optional): How long to wait before timeout error.
Defaults to None.
kwargs: Collection.search() keyword arguments.
Returns:
List[Document]: Document results for search.
"""
res = self.similarity_search_with_score_by_vector(embedding=embedding,
k=k,
param=param,
expr=expr,
timeout=timeout,
**kwargs)
return [doc for doc, _ in res]
def similarity_search_with_score(
self,
query: str,
k: int = 4,
param: Optional[dict] = None,
expr: Optional[str] = None,
timeout: Optional[int] = None,
**kwargs: Any,
) -> List[Tuple[Document, float]]:
"""Perform a search on a query string and return results with score.
For more information about the search parameters, take a look at the pymilvus
documentation found here:
https://milvus.io/api-reference/pymilvus/v2.2.6/Collection/search().md
Args:
query (str): The text being searched.
k (int, optional): The amount of results to return. Defaults to 4.
param (dict): The search params for the specified index.
Defaults to None.
expr (str, optional): Filtering expression. Defaults to None.
timeout (int, optional): How long to wait before timeout error.
Defaults to None.
kwargs: Collection.search() keyword arguments.
Returns:
List[float], List[Tuple[Document, any, any]]:
"""
if k == 0:
# pm need to control
return []
if not self.col:
logger.debug('No existing collection to search.')
return []
# Embed the query text.
embedding = self.embedding_func.embed_query(query)
res = self.similarity_search_with_score_by_vector(embedding=embedding,
k=k,
query=query,
param=param,
expr=expr,
timeout=timeout,
**kwargs)
return res
def similarity_search_with_score_by_vector(
self,
embedding: List[float],
k: int = 4,
param: Optional[dict] = None,
query: Optional[str] = None,
expr: Optional[str] = None,
timeout: Optional[int] = None,
**kwargs: Any,
) -> List[Tuple[Document, float]]:
"""Perform a search on a query string and return results with score.
For more information about the search parameters, take a look at the pymilvus
documentation found here:
https://milvus.io/api-reference/pymilvus/v2.2.6/Collection/search().md
Args:
embedding (List[float]): disabled.
k (int, optional): The amount of results to return. Defaults to 4.
param (dict): The search params for the specified index.
Defaults to None.
query (str): The search query, use collection`s embedding calc query embedding.
Defaults to None.
expr (str, optional): Filtering expression. Defaults to None.
timeout (int, optional): How long to wait before timeout error.
Defaults to None.
kwargs: Collection.search() keyword arguments.
Returns:
List[Tuple[Document, float]]: Result doc and score.
"""
if k == 0:
# pm need to control
return []
if not self.col:
logger.debug('No existing collection to search.')
return []
if param is None:
param = self.search_params
# Determine result metadata fields.
output_fields = self.fields[:]
output_fields.remove(self._vector_field)
finally_k = kwargs.pop('k', k)
ret = []
for index, one_col in enumerate(self.col):
search_expr = expr
embedding = self.collection_embeddings[index].embed_query(query)
if self.col_partition_key[index]:
# add parttion
if expr:
search_expr = f"{expr} and {self._partition_field}==\"{self.col_partition_key[index]}\""
else:
search_expr = f"{self._partition_field}==\"{self.col_partition_key[index]}\""
# Perform the search.
res = one_col.search(
data=[embedding],
anns_field=self._vector_field,
param=param,
limit=k,
expr=search_expr,
output_fields=output_fields,
timeout=timeout,
**kwargs,
)
# Organize results.
for result in res[0]:
meta = {x: result.entity.get(x) for x in output_fields}
doc = Document(page_content=meta.pop(self._text_field), metadata=meta)
pair = (doc, result.score)
ret.append(pair)
logger.debug(f'MilvusWithPermissionCheck Search {one_col.name} query: {query} results: {res[0]}')
ret.sort(key=lambda x: x[1])
logger.debug(f'MilvusWithPermissionCheck Search all results: {len(ret)}')
# milvusYes, the smaller the score, the better, so just take the first few digits directly
ret = ret[:finally_k]
logger.debug(f'MilvusWithPermissionCheck Search finally results: {len(ret)}')
return ret
@staticmethod
def _relevance_score_fn(distance: float) -> float:
"""Normalize the distance to a score on a scale [0, 1]."""
return 1 - distance / 2
def _select_relevance_score_fn(self) -> Callable[[float], float]:
return self._relevance_score_fn
class ElasticsearchWithPermissionCheck(VectorStore, ABC):
"""
only search from multi index
"""
def __init__(
self,
elasticsearch_url: str,
index_name: List[str],
drop_old: Optional[bool] = False,
*,
ssl_verify: Optional[Dict[str, Any]] = None,
llm_chain: Optional[LLMChain] = None,
):
"""Initialize with necessary components."""
try:
import elasticsearch
except ImportError:
raise ImportError('Could not import elasticsearch python package. '
'Please install it with `pip install elasticsearch`.')
self.index_name = index_name
self.llm_chain = llm_chain
self.drop_old = drop_old
_ssl_verify = ssl_verify or {}
self.elasticsearch_url = elasticsearch_url
self.ssl_verify = _ssl_verify
try:
self.client = elasticsearch.Elasticsearch(elasticsearch_url, **_ssl_verify)
except ValueError as e:
raise ValueError(f'Your elasticsearch client string is mis-formatted. Got error: {e} ')
def similarity_search(self,
query: str,
k: int = 4,
query_strategy: str = 'match_phrase',
must_or_should: str = 'should',
**kwargs: Any) -> List[Document]:
if k == 0:
# pm need to control
return []
docs_and_scores = self.similarity_search_with_score(query,
k=k,
query_strategy=query_strategy,
must_or_should=must_or_should,
**kwargs)
documents = [d[0] for d in docs_and_scores]
return documents
@staticmethod
def _relevance_score_fn(distance: float) -> float:
"""Normalize the distance to a score on a scale [0, 1]."""
# Todo: normalize the es score on a scale [0, 1]
return distance
def _select_relevance_score_fn(self) -> Callable[[float], float]:
return self._relevance_score_fn
def similarity_search_with_score(self,
query: str,
k: int = 4,
query_strategy: str = 'match_phrase',
must_or_should: str = 'should',
**kwargs: Any) -> List[Tuple[Document, float]]:
if k == 0:
# pm need to control
return []
assert must_or_should in ['must', 'should'], 'only support must and should.'
# llm or jiaba extract keywords
if self.llm_chain:
keywords_str = self.llm_chain.run(query)
logger.debug('elasticsearch llm search keywords:', keywords_str)
try:
keywords = literal_eval(keywords_str)
if not isinstance(keywords, list):
raise ValueError('Keywords extracted by llm is not list.')
except Exception:
keywords = jieba.analyse.extract_tags(query, topK=10, withWeight=False)
else:
keywords = jieba.analyse.extract_tags(query, topK=10, withWeight=False)
keywords = keywords or [query]
logger.debug(f'finally search keywords: {keywords}')
match_query = {'bool': {must_or_should: []}}
for key in keywords:
match_query['bool'][must_or_should].append({query_strategy: {'text': key}})
ret = []
for one_index_name in self.index_name:
response = self.client_search(self.client, one_index_name, match_query, size=k)
hits = [hit for hit in response['hits']['hits']]
for hit in hits:
ret.append((Document(page_content=hit['_source']['text'],
metadata=hit['_source']['metadata']), hit['_score']))
logger.debug(
f'ElasticsearchWithPermissionCheck Search {one_index_name} results: {hits}')
logger.debug(f'ElasticsearchWithPermissionCheck Search all results: {len(ret)}')
finally_k = kwargs.pop('finally_k', k)
ret.sort(key=lambda x: x[1], reverse=True)
ret = ret[:finally_k]
logger.debug(f'ElasticsearchWithPermissionCheck Search finally results: {len(ret)}')
return ret
def add_texts(
self,
texts: Iterable[str],
metadatas: Optional[List[dict]] = None,
ids: Optional[List[str]] = None,
refresh_indices: bool = True,
**kwargs: Any,
) -> List[str]:
pass
@classmethod
def from_texts(
cls,
texts: List[str],
embedding: Embeddings,
metadatas: Optional[List[dict]] = None,
ids: Optional[List[str]] = None,
index_name: Optional[List[str]] = None,
refresh_indices: bool = True,
llm: Optional[BaseLLM] = None,
prompt: Optional[PromptTemplate] = DEFAULT_PROMPT,
drop_old: Optional[bool] = False,
**kwargs: Any,
):
"""Construct ElasticKeywordsSearch wrapper from raw documents.
This is a user-friendly interface that:
1. Embeds documents.
2. Creates a new index for the embeddings in the Elasticsearch instance.
3. Adds the documents to the newly created Elasticsearch index.
This is intended to be a quick way to get started.
Example:
.. code-block:: python
from langchain import ElasticKeywordsSearch
from langchain.embeddings import OpenAIEmbeddings
embeddings = OpenAIEmbeddings()
elastic_vector_search = ElasticKeywordsSearch.from_texts(
texts,
embeddings,
elasticsearch_url="http://localhost:9200"
)
"""
elasticsearch_url = get_from_dict_or_env(kwargs, 'elasticsearch_url', 'ELASTICSEARCH_URL')
if 'elasticsearch_url' in kwargs:
del kwargs['elasticsearch_url']
if llm:
llm_chain = LLMChain(llm=llm, prompt=prompt)
vectorsearch = cls(elasticsearch_url,
index_name,
llm_chain=llm_chain,
drop_old=drop_old,
**kwargs)
else:
vectorsearch = cls(elasticsearch_url, index_name, drop_old=drop_old, **kwargs)
return vectorsearch
def client_search(self, client: Any, index_name: str, script_query: Dict, size: int) -> Any:
version_num = client.info()['version']['number'][0]
version_num = int(version_num)
if version_num >= 8:
response = client.search(index=index_name, query=script_query, size=size)
else:
response = client.search(index=index_name, body={'query': script_query, 'size': size})
return response
def delete(self, **kwargs: Any) -> None:
# TODO: Check if this can be done in bulk
self.client.indices.delete(index=self.index_name)
================================================
FILE: src/backend/bisheng/interface/wrappers/__init__.py
================================================
================================================
FILE: src/backend/bisheng/interface/wrappers/base.py
================================================
from typing import ClassVar, Dict, List, Optional, Type
from bisheng.interface.base import LangChainTypeCreator
from bisheng.template.frontend_node.wrappers import WrappersFrontendNode
from loguru import logger
from bisheng.utils.util import build_template_from_class, build_template_from_method
from bisheng_langchain.utils.azure_dalle_image_generator import AzureDallEWrapper as DallEAPIWrapper
from langchain_community.utilities import requests, sql_database
from sqlmodel import true
class WrapperCreator(LangChainTypeCreator):
type_name: str = 'wrappers'
from_method_nodes: ClassVar[Dict] = {'SQLDatabase': 'from_uri'}
@property
def frontend_node_class(self) -> Type[WrappersFrontendNode]:
"""The class type of the FrontendNode created in frontend_node."""
return WrappersFrontendNode
@property
def type_to_loader_dict(self) -> Dict:
if self.type_dict is None:
self.type_dict = {
wrapper.__name__: wrapper
for wrapper in
[requests.TextRequestsWrapper, sql_database.SQLDatabase, DallEAPIWrapper]
}
return self.type_dict
def get_signature(self, name: str) -> Optional[Dict]:
try:
if name in self.from_method_nodes:
return build_template_from_method(
name,
type_to_cls_dict=self.type_to_loader_dict,
add_function=True,
method_name=self.from_method_nodes[name],
)
return build_template_from_class(name, self.type_to_loader_dict, true)
except ValueError as exc:
raise ValueError('Wrapper not found') from exc
except AttributeError as exc:
logger.error(f'Wrapper {name} not loaded: {exc}')
return None
def to_list(self) -> List[str]:
return list(self.type_to_loader_dict.keys())
wrapper_creator = WrapperCreator()
================================================
FILE: src/backend/bisheng/jcloud.yml
================================================
instance: C4
autoscale_min: 1
================================================
FILE: src/backend/bisheng/knowledge/README.md
================================================
# 知识库模块
## 迁移重构中
================================================
FILE: src/backend/bisheng/knowledge/__init__.py
================================================
================================================
FILE: src/backend/bisheng/knowledge/api/__init__.py
================================================
================================================
FILE: src/backend/bisheng/knowledge/api/dependencies.py
================================================
from fastapi import Depends
from sqlmodel.ext.asyncio.session import AsyncSession
from bisheng.common.dependencies.core_deps import get_db_session
from bisheng.knowledge.domain.repositories.implementations.knowledge_file_repository_impl import \
KnowledgeFileRepositoryImpl
from bisheng.knowledge.domain.repositories.implementations.knowledge_repository_impl import KnowledgeRepositoryImpl
from bisheng.knowledge.domain.repositories.interfaces.knowledge_file_repository import KnowledgeFileRepository
from bisheng.knowledge.domain.repositories.interfaces.knowledge_repository import KnowledgeRepository
from bisheng.knowledge.domain.services.knowledge_file_service import KnowledgeFileService
from bisheng.knowledge.domain.services.knowledge_service import KnowledgeService
async def get_knowledge_repository(
session: AsyncSession = Depends(get_db_session),
) -> KnowledgeRepository:
"""DapatkanKnowledgeRepositoryInstance Dependencies"""
return KnowledgeRepositoryImpl(session)
async def get_knowledge_file_repository(
session: AsyncSession = Depends(get_db_session),
) -> 'KnowledgeFileRepository':
"""DapatkanKnowledgeFileRepositoryInstance Dependencies"""
return KnowledgeFileRepositoryImpl(session)
async def get_knowledge_service(
knowledge_repository: KnowledgeRepository = Depends(get_knowledge_repository),
knowledge_file_repository: KnowledgeFileRepository = Depends(get_knowledge_file_repository),
) -> 'KnowledgeService':
"""DapatkanKnowledgeServiceInstance Dependencies"""
return KnowledgeService(knowledge_repository=knowledge_repository,
knowledge_file_repository=knowledge_file_repository)
async def get_knowledge_file_service(
knowledge_repository: KnowledgeRepository = Depends(get_knowledge_repository),
knowledge_file_repository: KnowledgeFileRepository = Depends(get_knowledge_file_repository),
) -> 'KnowledgeFileService':
"""DapatkanKnowledgeFileServiceInstance Dependencies"""
return KnowledgeFileService(
knowledge_repository=knowledge_repository,
knowledge_file_repository=knowledge_file_repository,
)
================================================
FILE: src/backend/bisheng/knowledge/api/endpoints/__init__.py
================================================
================================================
FILE: src/backend/bisheng/knowledge/api/endpoints/knowledge.py
================================================
import asyncio
import gc
import json
import urllib.parse
from datetime import datetime
from io import BytesIO
from typing import List, Optional, Any, Literal
import pandas as pd
from fastapi import (APIRouter, BackgroundTasks, Body, Depends, File, HTTPException, Query, Request,
UploadFile)
from fastapi.encoders import jsonable_encoder
from loguru import logger
from bisheng.api.services import knowledge_imp
from bisheng.api.services.knowledge import KnowledgeService
from bisheng.api.services.knowledge_imp import add_qa
from bisheng.api.v1.schemas import (KnowledgeFileProcess, UpdatePreviewFileChunk, UploadFileResponse,
UpdateKnowledgeReq, KnowledgeFileReProcess)
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.http_error import UnAuthorizedError, NotFoundError, ServerError
from bisheng.common.errcode.knowledge import KnowledgeCPError, KnowledgeQAError, KnowledgeRebuildingError, \
KnowledgePreviewError, KnowledgeNotQAError, KnowledgeNoEmbeddingError, KnowledgeNotExistError, KnowledgeCPEmptyError
from bisheng.common.errcode.server import NoLlmModelConfigError
from bisheng.common.schemas.api import resp_200, resp_500, UnifiedResponseModel
from bisheng.common.services import telemetry_service
from bisheng.core.cache.redis_manager import get_redis_client
from bisheng.core.cache.utils import save_uploaded_file
from bisheng.core.logger import trace_id_var
from bisheng.database.models.role_access import AccessType
from bisheng.knowledge.api.dependencies import get_knowledge_service, get_knowledge_file_service
from bisheng.knowledge.domain.models.knowledge import (KnowledgeCreate, KnowledgeDao, KnowledgeTypeEnum,
KnowledgeUpdate)
from bisheng.knowledge.domain.models.knowledge import KnowledgeState
from bisheng.knowledge.domain.models.knowledge_file import (KnowledgeFileDao, KnowledgeFileStatus,
QAKnoweldgeDao, QAKnowledgeUpsert, QAStatus)
from bisheng.knowledge.domain.schemas.knowledge_schema import AddKnowledgeMetadataFieldsReq, \
UpdateKnowledgeMetadataFieldsReq, ModifyKnowledgeFileMetaDataReq
from bisheng.llm.domain import LLMService
from bisheng.llm.domain.const import LLMModelType
from bisheng.llm.domain.models import LLMDao
from bisheng.user.domain.models.user import UserDao
from bisheng.utils import generate_uuid, calc_data_sha256
from bisheng.worker.knowledge.qa import insert_qa_celery
# build router
router = APIRouter(prefix='/knowledge', tags=['Knowledge'])
@router.post('/upload')
async def upload_file(*, file: UploadFile = File(...)):
try:
file_name = file.filename
uuid_file_name = await KnowledgeService.save_upload_file_original_name(file_name)
file_path = await save_uploaded_file(file, 'bisheng', uuid_file_name)
if not isinstance(file_path, str):
file_path = str(file_path)
return resp_200(UploadFileResponse(file_path=file_path))
except Exception as e:
logger.error(f'File upload failed: {e}')
raise ServerError(msg=f'File upload failed: {e}')
finally:
await file.close()
@router.post('/upload/{knowledge_id}')
async def upload_knowledge_file(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
knowledge_id: int,
file: UploadFile = File(...)):
""" Knowledge base upload file """
try:
file_name = file.filename
# Save the uploaded file
uuid_file_name = await KnowledgeService.save_upload_file_original_name(file_name)
file_path = await save_uploaded_file(file, 'bisheng', uuid_file_name)
if not isinstance(file_path, str):
file_path = str(file_path)
await file.seek(0)
# Calculate file md5
file_md5 = await asyncio.to_thread(calc_data_sha256, file.file)
# Check for duplicate files
repeat_file = await KnowledgeFileDao.get_repeat_file(
knowledge_id=knowledge_id, file_name=file_name, md5_=file_md5
)
ret = UploadFileResponse(file_path=file_path, file_name=file_name)
if repeat_file:
ret.repeat = True
ret.repeat_update_time = repeat_file.update_time
ret.repeat_file_name = repeat_file.file_name
return resp_200(ret)
except Exception as e:
raise ServerError(msg=f'File upload failed: {e}')
finally:
await file.close()
@router.post('/preview')
async def preview_file_chunk(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
background_tasks: BackgroundTasks,
req_data: KnowledgeFileProcess):
""" Get a chunked preview of a file """
preview_file_id = generate_uuid()
redis_key = f'preview_file:{preview_file_id}'
redis_client = await get_redis_client()
await redis_client.aset(redis_key, {"status": "processing"})
async def exec_task():
try:
parse_type, file_share_url, res, partitions = await KnowledgeService.get_preview_file_chunk(request,
login_user,
req_data)
await redis_client.aset(redis_key, {
"status": "completed",
"data": {
'parse_type': parse_type,
'file_url': file_share_url,
'chunks': [one.model_dump() for one in res],
'partitions': partitions
}
})
except Exception as exc:
logger.exception(f'Preview file chunk error: {exc}')
await redis_client.aset(redis_key, {
"status": "error"
})
background_tasks.add_task(exec_task)
return resp_200(data={'preview_file_id': preview_file_id})
@router.get('/preview/status')
async def get_preview_file_status(
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
preview_file_id: str = Query(..., description='Preview the file returned by the interfaceID')):
redis_key = f'preview_file:{preview_file_id}'
redis_client = await get_redis_client()
file_status = await redis_client.aget(redis_key)
if not file_status:
raise KnowledgePreviewError.http_exception()
if file_status.get('status') == 'error':
raise KnowledgePreviewError.http_exception()
if file_status.get('status') == 'completed':
await redis_client.aexpire_key(redis_key, 10)
return resp_200(data=file_status)
@router.put('/preview')
async def update_preview_file_chunk(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
req_data: UpdatePreviewFileChunk):
""" Updating a chunked preview of a file """
res = await KnowledgeService.update_preview_file_chunk(request, login_user, req_data)
return resp_200(res)
@router.delete('/preview')
async def delete_preview_file_chunk(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
req_data: UpdatePreviewFileChunk):
""" Delete a chunked preview of a file """
res = KnowledgeService.delete_preview_file_chunk(request, login_user, req_data)
return resp_200(res)
@router.post('/process')
def process_knowledge_file(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
background_tasks: BackgroundTasks,
req_data: KnowledgeFileProcess):
""" Uploading Files to the Knowledge Base """
res = KnowledgeService.process_knowledge_file(request, login_user, background_tasks, req_data)
return resp_200(res)
# Modify Segment Reprocessing
@router.post("/process/rebuild")
async def rebuild_knowledge_file(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
req_data: KnowledgeFileReProcess):
""" Reprocessing Knowledge Base Files """
res = await KnowledgeService.rebuild_knowledge_file(request, login_user, req_data)
return resp_200(res)
@router.post('/create')
def create_knowledge(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
knowledge: KnowledgeCreate):
""" Create Knowledge Base. """
db_knowledge = KnowledgeService.create_knowledge(request, login_user, knowledge)
return resp_200(db_knowledge)
@router.post('/copy')
async def copy_knowledge(*,
request: Request,
background_tasks: BackgroundTasks,
login_user: UserPayload = Depends(UserPayload.get_login_user),
knowledge_id: int = Body(..., embed=True),
knowledge_name: str = Body(default=None, embed=True)):
""" Copy Knowledge Base. """
knowledge = await KnowledgeDao.aquery_by_id(knowledge_id)
if not login_user.is_admin and knowledge.user_id != login_user.user_id:
return UnAuthorizedError.return_resp()
knowledge_count = await KnowledgeFileDao.async_count_file_by_filters(
knowledge_id,
status=[KnowledgeFileStatus.PROCESSING.value, KnowledgeFileStatus.WAITING.value],
)
if knowledge.state != KnowledgeState.PUBLISHED.value or knowledge_count > 0:
return KnowledgeCPError.return_resp()
knowledge = await KnowledgeService.copy_knowledge(request, background_tasks, login_user, knowledge, knowledge_name)
return resp_200(knowledge)
@router.post("/qa/copy")
async def copy_qa_knowledge(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
knowledge_id: int = Body(..., embed=True),
knowledge_name: str = Body(default=None, embed=True)):
"""
SalinQAThe knowledge base upon.
:param request:
:param login_user:
:param knowledge_id:
:param knowledge_name: new knowledge name
:return:
"""
qa_knowledge = await KnowledgeDao.aquery_by_id(knowledge_id)
if not login_user.is_admin and qa_knowledge.user_id != login_user.user_id:
return UnAuthorizedError.return_resp()
if qa_knowledge.type != KnowledgeTypeEnum.QA.value:
return KnowledgeNotQAError.return_resp()
qa_knowledge_count = await QAKnoweldgeDao.async_count_by_id(qa_id=qa_knowledge.id)
if qa_knowledge.state != KnowledgeState.PUBLISHED.value:
return KnowledgeCPError.return_resp()
if qa_knowledge_count == 0:
return KnowledgeCPEmptyError.return_resp()
knowledge = await KnowledgeService.copy_qa_knowledge(request, login_user, qa_knowledge, knowledge_name)
return resp_200(knowledge)
@router.get('', status_code=200)
async def get_knowledge(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
name: str = None,
knowledge_type: int = Query(default=KnowledgeTypeEnum.NORMAL.value,
alias='type'),
sort_by: Literal['create_time', 'update_time', 'name'] = Query(default='update_time'),
page_size: Optional[int] = 10,
page_num: Optional[int] = 1):
""" Read all knowledge base information. """
knowledge_type = KnowledgeTypeEnum(knowledge_type)
res, total = await KnowledgeService.get_knowledge(request, login_user, knowledge_type, name,
sort_by,
page_num, page_size)
return resp_200(data={'data': res, 'total': total})
@router.get('/info', status_code=200)
def get_knowledge_info(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
knowledge_id: List[int] = Query(...)):
""" Based on Knowledge BaseIDRead Knowledge Base Information. """
res = KnowledgeService.get_knowledge_info(request, login_user, knowledge_id)
return resp_200(data=res)
@router.put('/', status_code=200)
async def update_knowledge(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
knowledge: KnowledgeUpdate):
res = KnowledgeService.update_knowledge(request, login_user, knowledge)
return resp_200(data=res)
@router.delete('/', status_code=200)
def delete_knowledge(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
knowledge_id: int = Body(..., embed=True)):
""" Delete Knowledge Base Information. """
KnowledgeService.delete_knowledge(request, login_user, knowledge_id)
return resp_200(message='Delete successful')
# Personal Knowledge Base Information Acquisition
@router.get('/personal_knowledge_info', status_code=200)
def get_personal_knowledge_info(
*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user)):
""" Get personal knowledge base information. """
knowledge = KnowledgeDao.get_user_knowledge(login_user.user_id, None,
KnowledgeTypeEnum.PRIVATE)
if not knowledge:
model = LLMService.get_knowledge_llm()
knowledgeCreate = KnowledgeCreate(name='Personal Knowledge Base',
type=KnowledgeTypeEnum.PRIVATE.value,
user_id=login_user.user_id,
model=model.embedding_model_id)
knowledge = [KnowledgeService.create_knowledge(request, login_user, knowledgeCreate)]
return resp_200(data=knowledge)
@router.get('/file_list/{knowledge_id}', status_code=200)
def get_filelist(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
file_name: str = None,
file_ids: List[int] = None,
knowledge_id: int = 0,
page_size: int = 10,
page_num: int = 1,
status: List[int] = Query(default=None)):
""" Get knowledge base file information. """
data, total, flag = KnowledgeService.get_knowledge_files(request, login_user, knowledge_id,
file_name, status, page_num,
page_size, file_ids)
return resp_200({
'data': data,
'total': total,
'writeable': flag,
})
@router.get('/qa/list/{qa_knowledge_id}', status_code=200)
async def get_QA_list(*,
qa_knowledge_id: int,
page_size: int = 10,
page_num: int = 1,
question: Optional[str] = None,
answer: Optional[str] = None,
keyword: Optional[str] = None,
status: Optional[int] = None,
login_user: UserPayload = Depends(UserPayload.get_login_user)):
""" Get knowledge base file information. """
db_knowledge = KnowledgeService.judge_qa_knowledge_write(login_user, qa_knowledge_id)
qa_list, total_count = await knowledge_imp.list_qa_by_knowledge_id(qa_knowledge_id, page_size,
page_num, question, answer,
keyword, status)
user_list = UserDao.get_user_by_ids([qa.user_id for qa in qa_list])
user_map = {user.user_id: user.user_name for user in user_list}
data = [jsonable_encoder(qa) for qa in qa_list]
for qa in data:
qa['questions'] = qa['questions'][0]
qa['answers'] = json.loads(qa['answers'])[0]
qa['user_name'] = user_map.get(qa['user_id'], qa['user_id'])
return resp_200({
'data':
data,
'total':
total_count,
'writeable':
login_user.access_check(db_knowledge.user_id, str(qa_knowledge_id),
AccessType.KNOWLEDGE_WRITE)
})
@router.post('/retry', status_code=200)
def retry(*,
request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
background_tasks: BackgroundTasks,
req_data: dict):
"""Failed Retry"""
KnowledgeService.retry_files(request, login_user, background_tasks, req_data)
return resp_200()
@router.delete('/file/{file_id}', status_code=200)
def delete_knowledge_file(*,
request: Request,
file_id: int,
login_user: UserPayload = Depends(UserPayload.get_login_user)):
""" Delete Knowledge File Information """
KnowledgeService.delete_knowledge_file(request, login_user, [file_id])
return resp_200(message='Delete successful')
@router.get('/chunk', status_code=200)
async def get_knowledge_chunk(request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
knowledge_id: int = Query(..., description='The knowledge base uponID'),
file_ids: List[int] = Query(default=[], description='Doc.ID'),
keyword: str = Query(default='', description='Keywords'),
page: int = Query(default=1, description='Page'),
limit: int = Query(default=10,
description='Number of bars per page Number of bars per page')):
""" Get Knowledge Base Block Content """
# In order to resolvekeywordParameters are sometimes not carried outurldecoderight of privacybug
if keyword.startswith('%'):
keyword = urllib.parse.unquote(keyword)
res, total = KnowledgeService.get_knowledge_chunks(request, login_user, knowledge_id, file_ids,
keyword, page, limit)
return resp_200(data={'data': res, 'total': total})
@router.put('/chunk', status_code=200)
async def update_knowledge_chunk(request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
knowledge_id: int = Body(..., embed=True, description='The knowledge base uponID'),
file_id: int = Body(..., embed=True, description='Doc.ID'),
chunk_index: int = Body(..., embed=True, description='Chunked index number'),
text: str = Body(..., embed=True, description='Chunked content'),
bbox: str = Body(default='', embed=True, description='Block box selection position')):
""" Update Knowledge Base Chunk Content """
KnowledgeService.update_knowledge_chunk(request, login_user, knowledge_id, file_id,
chunk_index, text, bbox)
return resp_200()
@router.delete('/chunk', status_code=200)
async def delete_knowledge_chunk(request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
knowledge_id: int = Body(..., embed=True, description='The knowledge base uponID'),
file_id: int = Body(..., embed=True, description='Doc.ID'),
chunk_index: int = Body(..., embed=True, description='Chunked index number')):
""" Delete Knowledge Base Chunk Content """
KnowledgeService.delete_knowledge_chunk(request, login_user, knowledge_id, file_id,
chunk_index)
return resp_200()
@router.get('/file_share')
async def get_file_share_url(request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
file_id: int = Query(description='File UniqueID')):
original_url, preview_url = KnowledgeService.get_file_share_url(file_id)
return resp_200(data={
'original_url': original_url,
'preview_url': preview_url
})
@router.get('/file_bbox')
async def get_file_bbox(request: Request,
login_user: UserPayload = Depends(UserPayload.get_login_user),
file_id: int = Query(description='File UniqueID')):
res = KnowledgeService.get_file_bbox(request, login_user, file_id)
return resp_200(data=res)
@router.post('/qa/add', status_code=200)
async def qa_add(*, QACreate: QAKnowledgeUpsert,
login_user: UserPayload = Depends(UserPayload.get_login_user)):
""" Add knowledge base information. """
QACreate.user_id = login_user.user_id
db_knowledge = KnowledgeDao.query_by_id(QACreate.knowledge_id)
if db_knowledge.type != KnowledgeTypeEnum.QA.value:
raise NotFoundError()
if not login_user.access_check(
db_knowledge.user_id, str(db_knowledge.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError.http_exception()
db_q = QAKnoweldgeDao.get_qa_knowledge_by_name(QACreate.questions, QACreate.knowledge_id, exclude_id=QACreate.id)
# create repeat question or update
if (db_q and not QACreate.id) or (db_q and QACreate.id and db_q.id != QACreate.id):
raise KnowledgeQAError.http_exception()
add_qa(db_knowledge=db_knowledge, data=QACreate)
return resp_200()
@router.post('/qa/status_switch', status_code=200)
def qa_status_switch(*,
status: int = Body(embed=True),
id: int = Body(embed=True),
login_user: UserPayload = Depends(UserPayload.get_login_user)):
""" Modify Knowledge Base Information. """
qa_db = QAKnoweldgeDao.get_qa_knowledge_by_primary_id(id)
if qa_db.status == status:
return resp_200()
db_knowledge = KnowledgeDao.query_by_id(qa_db.knowledge_id)
if not login_user.access_check(
db_knowledge.user_id, str(db_knowledge.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError.http_exception()
new_qa_db = knowledge_imp.qa_status_change(qa_db, status, db_knowledge)
if not new_qa_db:
return resp_200()
if new_qa_db.status != status:
# Description state switch failed
return resp_500(message=new_qa_db.remark)
return resp_200()
@router.get('/qa/detail', status_code=200)
def qa_detail(*, id: int, login_user: UserPayload = Depends(UserPayload.get_login_user)):
""" Add knowledge base information. """
qa_knowledge = QAKnoweldgeDao.get_qa_knowledge_by_primary_id(id)
qa_knowledge.answers = json.loads(qa_knowledge.answers)[0]
return resp_200(data=qa_knowledge)
@router.post('/qa/append', status_code=200)
def qa_append(
*,
ids: list[int] = Body(..., embed=True),
question: str = Body(..., embed=True),
login_user: UserPayload = Depends(UserPayload.get_login_user),
):
""" Add knowledge base information. """
qa_list = QAKnoweldgeDao.select_list(ids)
knowledge = KnowledgeDao.query_by_id(qa_list[0].knowledge_id)
# check knowledge access
if not login_user.access_check(
knowledge.user_id, str(knowledge.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError.http_exception()
for q in qa_list:
if question in q.questions:
raise KnowledgeQAError.http_exception()
for qa in qa_list:
qa.questions.append(question)
knowledge_imp.add_qa(knowledge, qa)
return resp_200()
@router.delete('/qa/delete', status_code=200)
def qa_delete(*,
ids: list[int] = Body(embed=True),
login_user: UserPayload = Depends(UserPayload.get_login_user)):
""" Delete Knowledge File Information """
qa_list = QAKnoweldgeDao.select_list(ids)
knowledge = KnowledgeDao.query_by_id(qa_list[0].knowledge_id)
if not login_user.access_check(knowledge.user_id, str(knowledge.id),
AccessType.KNOWLEDGE_WRITE):
raise UnAuthorizedError()
if knowledge.type != KnowledgeTypeEnum.QA.value:
raise KnowledgeNotQAError()
knowledge_imp.delete_vector_data(knowledge, ids)
QAKnoweldgeDao.delete_batch(ids)
telemetry_service.log_event_sync(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.DELETE_KNOWLEDGE_FILE,
trace_id=trace_id_var.get())
return resp_200()
@router.post('/qa/auto_question')
def qa_auto_question(
*,
number: int = Body(default=3, embed=True),
ori_question: str = Body(default='', embed=True),
answer: str = Body(default='', embed=True),
login_user: UserPayload = Depends(UserPayload.get_login_user)
):
"""Automatically generate questions from large models"""
questions = knowledge_imp.recommend_question(login_user.user_id, ori_question, number=number, answer=answer)
return resp_200(data={'questions': questions})
@router.get('/qa/export/template', status_code=200)
async def get_export_url():
data = [{"Question": "", "Answer": "", "Similar question 1": "", "Similar question 2": ""}]
df = pd.DataFrame(data)
bio = BytesIO()
with pd.ExcelWriter(bio, engine="openpyxl") as writer:
df.to_excel(writer, sheet_name="Sheet1", index=False)
file_name = f"qa_export_template.xlsx"
bio.seek(0)
file = UploadFile(filename=file_name, file=bio)
file_path = await save_uploaded_file(file, 'bisheng', file_name)
await file.close()
return resp_200({"url": file_path})
@router.get('/qa/export/{qa_knowledge_id}', status_code=200)
async def get_export_url(*,
qa_knowledge_id: int,
question: Optional[str] = None,
answer: Optional[str] = None,
keyword: Optional[str] = None,
status: Optional[int] = None,
max_lines: Optional[int] = 10000,
login_user: UserPayload = Depends(UserPayload.get_login_user)):
# Query the current knowledge base, whether there are write permissions
db_knowledge = KnowledgeService.judge_qa_knowledge_write(login_user, qa_knowledge_id)
if keyword:
question = keyword
page_num = 1
total_num = 0
page_size = max_lines
file_list = []
file_pr = datetime.now().strftime('%Y%m%d%H%M%S')
file_index = 1
while True:
qa_list, total_count = await knowledge_imp.list_qa_by_knowledge_id(qa_knowledge_id, page_size,
page_num, question, answer,
status)
data = [jsonable_encoder(qa) for qa in qa_list]
qa_dict_list = []
all_title = ["Question", "Answer"]
for qa in data:
qa_dict_list.append({
"Question": qa['questions'][0],
"Answer": json.loads(qa['answers'])[0]
})
for index, question in enumerate(qa['questions']):
if index == 0:
continue
key = f"Similar question {index}"
if key not in all_title:
all_title.append(key)
qa_dict_list[-1][key] = question
if len(qa_dict_list) != 0:
df = pd.DataFrame(qa_dict_list)
else:
df = pd.DataFrame([{"Question": "", "Answer": "", "Similar question 1": "", "Similar question 2": ""}])
df = df[all_title]
bio = BytesIO()
with pd.ExcelWriter(bio, engine="openpyxl") as writer:
df.to_excel(writer, sheet_name="Sheet1", index=False)
file_name = f"{file_pr}_{file_index}.xlsx"
file_index = file_index + 1
bio.seek(0)
file_io = UploadFile(filename=file_name, file=bio)
file_path = await save_uploaded_file(file_io, 'bisheng', file_name)
await file_io.close()
file_list.append(file_path)
total_num += len(qa_list)
if len(qa_list) < page_size or total_num >= total_count:
break
return resp_200({"file_list": file_list})
def convert_excel_value(value: Any):
if value is None or value == "":
return ''
if str(value) == 'nan' or str(value) == 'null':
return ''
return str(value)
@router.post('/qa/preview/{qa_knowledge_id}', status_code=200)
def post_import_file(*,
qa_knowledge_id: int,
file_url: str = Body(..., embed=True),
size: Optional[int] = Body(default=0, embed=True),
offset: Optional[int] = Body(default=0, embed=True),
login_user: UserPayload = Depends(UserPayload.get_login_user)):
df = pd.read_excel(file_url)
columns = df.columns.to_list()
if 'Question' not in columns or 'Answer' not in columns:
raise HTTPException(status_code=500, detail='file must have ‘Question’ Or ‘Answer’ column')
data = df.T.to_dict().values()
insert_data = []
for dd in data:
d = QAKnowledgeUpsert(
user_id=login_user.user_id,
knowledge_id=qa_knowledge_id,
answers=[convert_excel_value(dd['Question'])],
questions=[convert_excel_value(dd['Answer'])],
source=4,
create_time=datetime.now(),
update_time=datetime.now())
for key, value in dd.items():
if key.startswith('Similar question') and convert_excel_value(value):
d.questions.append(convert_excel_value(value))
insert_data.append(d)
try:
if size > 0 and offset >= 0:
if offset >= len(insert_data):
insert_data = []
else:
insert_data = insert_data[offset:size]
except Exception as e:
raise HTTPException(status_code=500, detail=e)
return resp_200({"result": insert_data})
@router.post('/qa/import/{qa_knowledge_id}', status_code=200)
def post_import_file(*,
qa_knowledge_id: int,
file_list: list[str] = Body(..., embed=True),
background_tasks: BackgroundTasks,
login_user: UserPayload = Depends(UserPayload.get_login_user)):
# Query the current knowledge base, whether there are write permissions
db_knowledge = KnowledgeService.judge_qa_knowledge_write(login_user, qa_knowledge_id)
insert_result = []
error_result = []
have_question = []
for file_url in file_list:
df = pd.read_excel(file_url)
columns = df.columns.to_list()
if 'Question' not in columns or 'Answer' not in columns:
insert_result.append(0)
continue
data = df.T.to_dict().values()
insert_data = []
have_data = []
all_questions = set()
for index, dd in enumerate(data):
tmp_questions = set()
dd_question = convert_excel_value(dd['Question'])
dd_answer = convert_excel_value(dd['Answer'])
QACreate = QAKnowledgeUpsert(
user_id=login_user.user_id,
knowledge_id=qa_knowledge_id,
answers=[dd_answer],
questions=[dd_question],
source=4,
status=QAStatus.PROCESSING.value)
tmp_questions.add(QACreate.questions[0])
for key, value in dd.items():
if key.startswith('Similar question'):
if tmp_value := convert_excel_value(value):
if tmp_value not in tmp_questions:
QACreate.questions.append(tmp_value)
tmp_questions.add(tmp_value)
db_q = QAKnoweldgeDao.get_qa_knowledge_by_name(QACreate.questions, QACreate.knowledge_id)
if (db_q and not QACreate.id) or len(tmp_questions & all_questions) > 0 or not dd_question or not dd_answer:
have_data.append(index)
else:
insert_data.append(QACreate)
all_questions = all_questions | tmp_questions
result = QAKnoweldgeDao.batch_insert_qa(insert_data)
telemetry_service.log_event_sync(
user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.NEW_KNOWLEDGE_FILE,
trace_id=trace_id_var.get()
)
# async task add qa into milvus and es
for one in result:
insert_qa_celery.delay(one.id)
error_result.append(have_data)
return resp_200({"errors": error_result})
@router.get('/status', status_code=200)
def get_knowledge_status(*, login_user: UserPayload = Depends(UserPayload.get_login_user)):
"""
View Knowledge Base Status Interface
Receive:
1. According UsersidFirst determine if the user has a personal knowledge base and does not return directly200
2. If you have a knowledge base, depending on the useridView the state of your knowledge base, if the state of your personal knowledge base isREBUILDING 3 or FAILED 4 Jam
Return "Personal Knowledge BaseembeddingThe model has been replaced, rebuilding the knowledge base, please try again later" State Code 502
"""
# Query a user's personal knowledge base
user_private_knowledge = KnowledgeDao.get_user_knowledge(
login_user.user_id,
None,
KnowledgeTypeEnum.PRIVATE
)
# If the user does not have a personal knowledge base, go directly back to200
if not user_private_knowledge:
return resp_200({"status": "success"})
# Get the first personal knowledge base (usually users have only one personal knowledge base)
private_knowledge = user_private_knowledge[0]
# Check Knowledge Base Status
if private_knowledge.state == KnowledgeState.REBUILDING.value:
# Return502Status codes and corresponding prompts
raise KnowledgeRebuildingError()
if private_knowledge.state == KnowledgeState.FAILED.value:
# Delay imports to avoid looping imports
from bisheng.worker.knowledge.rebuild_knowledge_worker import rebuild_knowledge_celery
rebuild_knowledge_celery.delay(private_knowledge.id, int(private_knowledge.model), login_user.user_id)
# Return502Status codes and corresponding prompts
raise KnowledgeRebuildingError()
# Knowledge base status is OK, go back200
return resp_200({"status": "success"})
@router.post('/update_knowledge', status_code=200)
def update_knowledge_model(*,
login_user: UserPayload = Depends(UserPayload.get_login_user),
req_data: UpdateKnowledgeReq):
"""
Update Knowledge Base Interface
Update embedding Rebuild Knowledge Base on Model
Receive:
1. According to the incoming from the front-endmodel_id, model_type, let's first determine ifembeddingModels
2. If not, go backresp501("Is notembeddingModels") If so, place theknowledgeAll in the tabletypeare2DatastatusChange to...3，modelChange to Incomingmodel_id
3. everyknowledge_idBoth initiate asynchronous tasks to rebuild the knowledge base
"""
# 1. Verify that isembeddingModels
model_info = LLMDao.get_model_by_id(req_data.model_id)
if not model_info:
return NoLlmModelConfigError.return_resp()
# If the front-end does not passmodel_type, using themodel_type
model_type = req_data.model_type if req_data.model_type else model_info.model_type
if model_type != LLMModelType.EMBEDDING.value:
return KnowledgeNoEmbeddingError.return_resp()
# Process the specified knowledge base
knowledge = KnowledgeDao.query_by_id(req_data.knowledge_id)
if not knowledge:
return KnowledgeNotExistError.return_resp()
if not login_user.access_check(
knowledge.user_id, str(knowledge.id), AccessType.KNOWLEDGE_WRITE
):
return UnAuthorizedError.return_resp()
old_model_id = knowledge.model
# Update Knowledge Base Status and Models
knowledge.model = str(req_data.model_id)
knowledge.name = req_data.knowledge_name
knowledge.description = req_data.description
if int(old_model_id) == int(req_data.model_id):
# If the model does not change, there is no need to rebuild
KnowledgeDao.update_one(knowledge)
return resp_200()
if knowledge.state == KnowledgeState.REBUILDING.value:
return KnowledgeRebuildingError.return_resp()
knowledge.state = KnowledgeState.REBUILDING.value
KnowledgeDao.update_one(knowledge)
# Start asynchronous task
if knowledge.type == KnowledgeTypeEnum.NORMAL.value:
# Delay imports to avoid looping imports
from bisheng.worker.knowledge.rebuild_knowledge_worker import rebuild_knowledge_celery
rebuild_knowledge_celery.delay(knowledge.id, req_data.model_id, login_user.user_id)
elif knowledge.type == KnowledgeTypeEnum.QA.value:
# Delay imports to avoid looping imports
from bisheng.worker.knowledge.qa import rebuild_qa_knowledge_celery
rebuild_qa_knowledge_celery.delay(knowledge.id, req_data.model_id, login_user.user_id)
logger.info(f"Started rebuild task for knowledge_id={knowledge.id} with model_id={req_data.model_id}")
return resp_200()
@router.get("/file/info/{file_id}", description="Get knowledge base file information",
response_model=UnifiedResponseModel)
async def get_knowledge_file_info(*,
login_user: UserPayload = Depends(UserPayload.get_login_user),
file_id: int,
knowledge_file_service=Depends(get_knowledge_file_service)):
"""
Get knowledge base file information
Args:
login_user:
file_id:
knowledge_file_service:
Returns:
"""
knowledge_file_info_res = await knowledge_file_service.get_knowledge_file_info(login_user, file_id)
return resp_200(data=knowledge_file_info_res)
# Adding Metadata Fields to the Knowledge Base
@router.post('/add_metadata_fields', description="Adding Metadata Fields to the Knowledge Base",
response_model=UnifiedResponseModel)
async def add_metadata_fields(*,
login_user: UserPayload = Depends(UserPayload.get_login_user),
req_data: AddKnowledgeMetadataFieldsReq,
knowledge_service=Depends(get_knowledge_service)):
"""
Adding Metadata Fields to the Knowledge Base
"""
knowledge_model = await knowledge_service.add_metadata_fields(login_user, req_data)
return resp_200(data=knowledge_model)
# Modify Knowledge Base Metadata Fields
@router.put('/update_metadata_fields', description="Modify Knowledge Base Metadata Fields",
response_model=UnifiedResponseModel)
async def update_metadata_fields(*,
login_user: UserPayload = Depends(UserPayload.get_login_user),
req_data: UpdateKnowledgeMetadataFieldsReq,
knowledge_service=Depends(get_knowledge_service),
background_tasks: BackgroundTasks
):
"""
Modify Knowledge Base Metadata Fields
Args:
login_user:
req_data:
knowledge_service:
background_tasks:
Returns:
UnifiedResponseModel
"""
knowledge_model = await knowledge_service.update_metadata_fields(login_user, req_data, background_tasks)
return resp_200(data=knowledge_model)
# Delete Knowledge Base Metadata Field
@router.delete('/delete_metadata_fields', description="Delete Knowledge Base Metadata Field",
response_model=UnifiedResponseModel)
async def delete_metadata_fields(*,
login_user: UserPayload = Depends(UserPayload.get_login_user),
knowledge_id: int = Body(..., embed=True, description="The knowledge base uponID"),
field_names: List[str] = Body(..., embed=True,
description="List of field names to delete"),
knowledge_service=Depends(get_knowledge_service),
background_tasks: BackgroundTasks):
"""
Delete Knowledge Base Metadata Field
Args:
login_user:
knowledge_id:
field_names:
knowledge_service:
background_tasks:
Returns:
UnifiedResponseModel
"""
knowledge_model = await knowledge_service.delete_metadata_fields(login_user, knowledge_id, field_names,
background_tasks)
return resp_200(data=knowledge_model)
# Modify Knowledge Base File User Custom Metadata
@router.put('/file/user_metadata', description="Modify Knowledge Base File User Custom Metadata",
response_model=UnifiedResponseModel)
async def modify_file_user_metadata(*,
login_user: UserPayload = Depends(UserPayload.get_login_user),
req_data: ModifyKnowledgeFileMetaDataReq,
knowledge_file_service=Depends(get_knowledge_file_service)):
"""
Modify Knowledge Base File User Custom Metadata
Args:
login_user:
req_data:
knowledge_file_service:
Returns:
UnifiedResponseModel
"""
knowledge_file_model = await knowledge_file_service.modify_file_user_metadata(login_user, req_data)
return resp_200(data=knowledge_file_model)
================================================
FILE: src/backend/bisheng/knowledge/api/endpoints/qa.py
================================================
import asyncio
import json
from typing import Annotated
from fastapi import APIRouter, Body
from sqlmodel import select
from bisheng.api.services.knowledge import KnowledgeService
from bisheng.api.v1.schemas import resp_200
from bisheng.common.errcode.qa import BackendProcessingError
from bisheng.core.database import get_sync_db_session
from bisheng.database.models.recall_chunk import RecallChunk
from bisheng.knowledge.domain.models.knowledge_file import KnowledgeFileDao
# build router
router = APIRouter(prefix='/qa', tags=['QA'])
# Some interfaces of the traceability module
@router.get('/keyword')
async def get_answer_keyword(message_id: int):
# Get Hitskey
conter = 3
while True:
with get_sync_db_session() as session:
chunks = session.exec(
select(RecallChunk).where(RecallChunk.message_id == message_id)).first()
# keywords
if chunks:
keywords = chunks.keywords
return resp_200(json.loads(keywords))
else:
# Delay Loop
if conter <= 0:
break
await asyncio.sleep(1)
conter -= 1
raise BackendProcessingError()
@router.post('/chunk', status_code=200)
def get_original_file(message_id: Annotated[int, Body(embed=True)],
keys: Annotated[str, Body(embed=True)]):
# Get Hitskey
with get_sync_db_session() as session:
chunks = session.exec(
select(RecallChunk).where(RecallChunk.message_id == message_id)).all()
if not chunks:
return resp_200(message='No bulkpost found in Trashchunks')
# chunk care of all offile
file_ids = {chunk.file_id for chunk in chunks}
db_knowledge_files = KnowledgeFileDao.get_file_by_ids(list(file_ids))
id2file = {file.id: file for file in db_knowledge_files}
# keywords
keywords = keys.split(';') if keys else []
result = []
for index, chunk in enumerate(chunks):
file = id2file.get(chunk.file_id)
chunk_res = json.loads(json.loads(chunk.meta_data).get('bbox'))
file_access = json.loads(chunk.meta_data).get('right', True)
chunk_res['right'] = file_access
if file_access and file:
# Preview filesurl
original_url, preview_url = KnowledgeService.get_file_share_url(file.id)
chunk_res['source_url'] = preview_url
chunk_res['original_url'] = original_url
chunk_res['source'] = file.file_name
else:
chunk_res['source_url'] = ''
chunk_res['original_url'] = ''
chunk_res['source'] = ''
chunk_res['score'] = round(match_score(chunk.chunk, keywords),
2) if len(keywords) > 0 else 0
chunk_res['file_id'] = chunk.file_id
chunk_res['parse_type'] = file.parse_type if file else ''
result.append(chunk_res)
# sort_and_filter_all_chunks(keywords, all_chunk)
return resp_200(result)
def find_lcsubstr(s1, s2):
m = [[0 for i in range(len(s2) + 1)] for j in range(len(s1) + 1)]
mmax = 0
p = 0
for i in range(len(s1)):
for j in range(len(s2)):
if s1[i] == s2[j]:
m[i + 1][j + 1] = m[i][j] + 1
if m[i + 1][j + 1] > mmax:
mmax = m[i + 1][j + 1]
p = i + 1
return s1[p - mmax:p], mmax
def match_score(chunk, keywords):
"""
After deduplicationkeywordsBlanketchunkWhat percentage is covered?
"""
hit_num = 0
# # Exact match
# for keyword in keywords:
# if keyword in chunk:
# hit_num += 1
# fuzzy matching, keywords2/3The above is included
for keyword in keywords:
res = find_lcsubstr(keyword, chunk)
if res[1] >= 2 / 3 * len(keyword):
hit_num += 1
return hit_num / len(keywords)
def sort_and_filter_all_chunks(keywords, all_chunks, thr=0.0):
"""
1. answerExtract and deduplicate keywords
2. Calculated keywords arechunkPercentage of coverage (=matched_key_num / all_key_num), calculate each in turnchunk
3. From high to low in terms of coverage ratio, yeschunkSort
4. Filter coverage ratio is less than thresholdThrright of privacychunkwhile retaining at least onechunk(To prevent the threshold from being too high, put all thechunkhave been filtered out)
"""
keywords = set(keywords)
chunk_match_score = []
for index, chunk in enumerate(all_chunks):
chunk_match_score.append(match_score(chunk, keywords))
sorted_res = sorted(enumerate(chunk_match_score), key=lambda x: -x[1])
remain_chunks = [all_chunks[elem[0]] for elem in sorted_res if elem[1] >= thr]
if not remain_chunks:
remain_chunks = [all_chunks[sorted_res[0][0]]]
return remain_chunks
================================================
FILE: src/backend/bisheng/knowledge/api/router.py
================================================
from .endpoints.knowledge import router as knowledge_router
from .endpoints.qa import router as qa_router
__all__ = ['knowledge_router', 'qa_router']
================================================
FILE: src/backend/bisheng/knowledge/domain/__init__.py
================================================
================================================
FILE: src/backend/bisheng/knowledge/domain/knowledge_rag.py
================================================
from typing import Dict
from langchain_core.embeddings import Embeddings
from langchain_core.vectorstores import VectorStore
from langchain_elasticsearch import AsyncElasticsearchStore, ElasticsearchStore
from langchain_milvus import Milvus
from bisheng.common.errcode.http_error import NotFoundError
from bisheng.knowledge.domain.models.knowledge import Knowledge, KnowledgeDao
from bisheng.knowledge.rag.elasticsearch_factory import ElasticsearchFactory
from bisheng.knowledge.rag.milvus_factory import MilvusFactory
from bisheng.llm.domain import LLMService
class KnowledgeRag:
""" initialize knowledge rag components """
@classmethod
async def _get_knowledge(cls, knowledge: Knowledge = None, knowledge_id: int = None) -> Knowledge:
if not knowledge:
knowledge = await KnowledgeDao.aquery_by_id(knowledge_id)
if not knowledge:
raise NotFoundError.http_exception()
return knowledge
@classmethod
def _get_knowledge_sync(cls, knowledge: Knowledge = None, knowledge_id: int = None):
if not knowledge:
knowledge = KnowledgeDao.query_by_id(knowledge_id)
if not knowledge:
raise NotFoundError.http_exception()
return knowledge
@classmethod
async def init_knowledge_milvus_vectorstore(cls, invoke_user_id: int, knowledge: Knowledge = None,
knowledge_id: int = None, **kwargs) -> Milvus:
knowledge = await cls._get_knowledge(knowledge, knowledge_id)
embedding = await LLMService.get_bisheng_knowledge_embedding(model_id=int(knowledge.model),
invoke_user_id=invoke_user_id)
return cls.init_milvus_vectorstore(knowledge.collection_name, embedding, **kwargs)
@classmethod
def init_knowledge_milvus_vectorstore_sync(cls, invoke_user_id: int, knowledge: Knowledge = None,
knowledge_id: int = None, **kwargs) -> Milvus:
knowledge = cls._get_knowledge_sync(knowledge, knowledge_id)
embedding = LLMService.get_bisheng_knowledge_embedding_sync(model_id=int(knowledge.model),
invoke_user_id=invoke_user_id)
return cls.init_milvus_vectorstore(knowledge.collection_name, embedding, **kwargs)
@classmethod
async def init_knowledge_es_vectorstore(cls, knowledge: Knowledge = None, knowledge_id: int = None,
**kwargs) -> AsyncElasticsearchStore:
knowledge = await cls._get_knowledge(knowledge, knowledge_id)
return cls.init_es_vectorstore(knowledge.index_name, **kwargs)
@classmethod
def init_knowledge_es_vectorstore_sync(cls, knowledge: Knowledge = None, knowledge_id: int = None,
**kwargs) -> ElasticsearchStore:
knowledge = cls._get_knowledge_sync(knowledge, knowledge_id)
return cls.init_es_vectorstore_sync(knowledge.index_name, **kwargs)
@classmethod
def get_multi_knowledge_vectorstore_sync(cls, invoke_user_id: int, knowledge_ids: list[int], user_name: str = None,
check_auth: bool = True, include_es: bool = True,
include_milvus: bool = True) \
-> Dict[int, Dict[str, VectorStore | Knowledge]]:
""" get multiple knowledge vectorstore, including milvus and es
return: {
knowledge_id: {
"knowledge": Knowledge
"milvus": Milvus,
"es": ElasticsearchStore,
},
}
"""
if not include_es and not include_milvus:
raise RuntimeError('at least one of include_es and include_milvus must be True')
if check_auth:
if not user_name:
raise RuntimeError('knowledge check auth user_name must be provided')
knowledge_list = KnowledgeDao.judge_knowledge_permission(user_name, knowledge_ids)
else:
knowledge_list = KnowledgeDao.get_list_by_ids(knowledge_ids)
ret = {}
for knowledge in knowledge_list:
ret[knowledge.id] = {
"knowledge": knowledge,
"milvus": None,
"es": None,
}
if include_milvus:
vectorstore = cls.init_knowledge_milvus_vectorstore_sync(invoke_user_id, knowledge)
ret[knowledge.id]["milvus"] = vectorstore
if include_es:
es_vectorstore = cls.init_knowledge_es_vectorstore_sync(knowledge)
ret[knowledge.id]["es"] = es_vectorstore
return ret
@classmethod
async def get_multi_knowledge_vectorstore(cls, invoke_user_id: int, knowledge_ids: list[int], user_name: str = None,
check_auth: bool = True, include_es: bool = True,
include_milvus: bool = True) \
-> Dict[int, Dict[str, VectorStore | Knowledge]]:
""" get multiple knowledge vectorstore, including milvus and es
return: {
knowledge_id: {
"knowledge": Knowledge
"milvus": Milvus,
"es": ElasticsearchStore,
},
}
"""
if knowledge_ids is None or len(knowledge_ids) == 0:
return {}
if not include_es and not include_milvus:
raise RuntimeError('at least one of include_es and include_milvus must be True')
if check_auth:
if not user_name:
raise RuntimeError('knowledge check auth user_name must be provided')
knowledge_list = await KnowledgeDao.ajudge_knowledge_permission(user_name, knowledge_ids)
else:
knowledge_list = await KnowledgeDao.aget_list_by_ids(knowledge_ids)
ret = {}
for knowledge in knowledge_list:
ret[knowledge.id] = {
"knowledge": knowledge,
"milvus": None,
"es": None,
}
if include_milvus:
vectorstore = await cls.init_knowledge_milvus_vectorstore(invoke_user_id, knowledge)
ret[knowledge.id]["milvus"] = vectorstore
if include_es:
es_vectorstore = await cls.init_knowledge_es_vectorstore(knowledge)
ret[knowledge.id]["es"] = es_vectorstore
return ret
@classmethod
def init_milvus_vectorstore(cls, collection_name: str, embeddings: Embeddings, **kwargs) -> Milvus:
""" init milvus vectorstore by collection name and model id """
return MilvusFactory.init_vectorstore(collection_name, embeddings, **kwargs)
@classmethod
def init_es_vectorstore(cls, index_name: str, **kwargs) -> AsyncElasticsearchStore:
""" init es vectorstore by index name """
return ElasticsearchFactory.init_vectorstore(index_name, **kwargs)
@classmethod
def init_es_vectorstore_sync(cls, index_name: str, **kwargs) -> ElasticsearchStore:
""" init es vectorstore by index name """
return ElasticsearchFactory.init_vectorstore_sync(index_name, **kwargs)
================================================
FILE: src/backend/bisheng/knowledge/domain/models/__init__.py
================================================
================================================
FILE: src/backend/bisheng/knowledge/domain/models/knowledge.py
================================================
from datetime import datetime
from enum import Enum
from typing import Any, List, Optional, Union, Dict
from pydantic import BaseModel, field_validator
from sqlalchemy import JSON
from sqlmodel import Column, DateTime, Field, delete, func, or_, select, text, update
from sqlmodel.sql.expression import Select, SelectOfScalar, col
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session, get_async_db_session
from bisheng.database.models.role_access import AccessType, RoleAccessDao
from bisheng.knowledge.domain.models.knowledge_file import KnowledgeFile, KnowledgeFileDao
from bisheng.user.domain.models.user import UserDao
from bisheng.user.domain.models.user_role import UserRoleDao
class KnowledgeTypeEnum(Enum):
QA = 1 # QAThe knowledge base upon
NORMAL = 0 # Docly Knowledge Base
PRIVATE = 2 # Workbench Personal Knowledge Base
class KnowledgeState(Enum):
UNPUBLISHED = 0
PUBLISHED = 1 # Document Knowledge Base Success Status
COPYING = 2
REBUILDING = 3 # Status in Document Knowledge Base Reconstruction
FAILED = 4 # Status of Documentation Knowledge Base Reconstruction Failure
class MetadataFieldType(str, Enum):
""" Metadata field type"""
STRING = "string"
NUMBER = "number"
TIME = "time"
# Case-insensitive enumeration matching
@classmethod
def _missing_(cls, value: Any) -> Optional["MetadataFieldType"]:
if isinstance(value, str):
for member in cls:
if member.value.lower() == value.lower():
return member
return None
class KnowledgeBase(SQLModelSerializable):
user_id: Optional[int] = Field(default=None, index=True)
name: str = Field(index=True, min_length=1, max_length=200,
description='Knowledge Base Name')
type: int = Field(index=False, default=0,
description='Knowledge Base Type, value from KnowledgeTypeEnum')
description: Optional[str] = Field(default=None, index=True)
model: Optional[str] = Field(default=None, index=False)
collection_name: Optional[str] = Field(default=None, index=False)
index_name: Optional[str] = Field(default=None, index=False)
state: Optional[int] = Field(index=False, default=KnowledgeState.PUBLISHED.value,
description='value from KnowledgeState')
metadata_fields: Optional[List[Dict]] = Field(default=None, sa_column=Column(JSON, nullable=True),
description="Metadata Field Configuration for Knowledge Base")
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
@field_validator('model', mode='before')
@classmethod
def convert_model(cls, v: Any) -> str:
if isinstance(v, int):
v = str(v)
return v
class Knowledge(KnowledgeBase, table=True):
id: Optional[int] = Field(default=None, primary_key=True)
class KnowledgeRead(KnowledgeBase):
id: int
user_name: Optional[str] = None
copiable: Optional[bool] = None
class KnowledgeUpdate(BaseModel):
knowledge_id: int
name: Optional[str] = None
description: Optional[str] = None
class KnowledgeCreate(KnowledgeBase):
is_partition: Optional[bool] = None
class KnowledgeDao(KnowledgeBase):
@classmethod
def insert_one(cls, data: Knowledge) -> Knowledge:
with get_sync_db_session() as session:
session.add(data)
session.commit()
session.refresh(data)
return data
@classmethod
async def async_insert_one(cls, data: Knowledge) -> Knowledge:
async with get_async_db_session() as session:
session.add(data)
await session.commit()
await session.refresh(data)
return data
@classmethod
def update_one(cls, data: Knowledge) -> Knowledge:
with get_sync_db_session() as session:
session.add(data)
session.commit()
session.refresh(data)
return data
@classmethod
async def aupdate_one(cls, data: Knowledge) -> Knowledge:
async with get_async_db_session() as session:
session.add(data)
await session.commit()
await session.refresh(data)
return data
@classmethod
async def async_update_state(cls, knowledge_id: int, state: KnowledgeState, update_time: Optional[datetime] = None):
async with get_async_db_session() as session:
statement = update(Knowledge).where(col(Knowledge.id) == knowledge_id)
statement = statement.values(state=state.value,
update_time=update_time or datetime.now())
await session.exec(statement)
await session.commit()
@classmethod
def update_state(cls, knowledge_id: int, state: KnowledgeState, update_time: Optional[datetime] = None):
with get_sync_db_session() as session:
statement = update(Knowledge).where(col(Knowledge.id) == knowledge_id)
statement = statement.values(state=state.value,
update_time=update_time or datetime.now())
session.exec(statement)
session.commit()
@classmethod
def update_knowledge_update_time(cls, knowledge: Knowledge):
statement = update(Knowledge).where(Knowledge.id == knowledge.id).values(
update_time=text('NOW()'))
with get_sync_db_session() as session:
session.exec(statement)
session.commit()
@classmethod
def query_by_id(cls, knowledge_id: int) -> Knowledge:
with get_sync_db_session() as session:
return session.get(Knowledge, knowledge_id)
@classmethod
async def aquery_by_id(cls, knowledge_id: int) -> Knowledge:
async with get_async_db_session() as session:
return await session.get(Knowledge, knowledge_id)
@classmethod
async def async_query_by_id(cls, knowledge_id: int) -> Knowledge:
async with get_async_db_session() as session:
result = await session.execute(select(Knowledge).where(Knowledge.id == knowledge_id))
return result.scalars().first()
@classmethod
def get_list_by_ids(cls, ids: List[int]) -> List[Knowledge]:
with get_sync_db_session() as session:
return session.exec(select(Knowledge).where(Knowledge.id.in_(ids))).all()
@classmethod
async def aget_list_by_ids(cls, ids: List[int]) -> List[Knowledge]:
async with get_async_db_session() as session:
result = await session.exec(select(Knowledge).where(col(Knowledge.id).in_(ids)))
return result.all()
@classmethod
def _user_knowledge_filters(
cls,
statement: Any,
user_id: int,
knowledge_id_extra: List[int] = None,
knowledge_type: KnowledgeTypeEnum = None,
name: str = None,
page: int = 0,
limit: int = 0,
filter_knowledge: List[int] = None) -> Union[Select, SelectOfScalar]:
if knowledge_id_extra:
statement = statement.where(
or_(Knowledge.id.in_(knowledge_id_extra), Knowledge.user_id == user_id))
else:
statement = statement.where(Knowledge.user_id == user_id)
if filter_knowledge:
statement = statement.where(Knowledge.id.in_(filter_knowledge))
if knowledge_type:
statement = statement.where(Knowledge.type == knowledge_type.value)
if name:
conditions = [col(Knowledge.name).like(f'%{name}%'), col(Knowledge.description).like(f'%{name}%')]
file_knowledge_ids = KnowledgeFileDao.get_knowledge_ids_by_name(name)
if file_knowledge_ids:
conditions.append(Knowledge.id.in_(file_knowledge_ids))
if conditions:
statement = statement.where(or_(*conditions))
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
return statement
@classmethod
def get_user_knowledge(cls,
user_id: int,
knowledge_id_extra: List[int] = None,
knowledge_type: KnowledgeTypeEnum = None,
name: str = None,
page: int = 0,
limit: int = 10,
filter_knowledge: List[int] = None) -> List[Knowledge]:
statement = select(Knowledge)
statement = cls._user_knowledge_filters(statement, user_id, knowledge_id_extra,
knowledge_type, name, page, limit,
filter_knowledge)
statement = statement.order_by(Knowledge.update_time.desc())
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
async def aget_user_knowledge(cls,
user_id: int,
knowledge_id_extra: List[int] = None,
knowledge_type: KnowledgeTypeEnum = None,
name: str = None,
sort_by: str = "update_time",
page: int = 0,
limit: int = 10,
filter_knowledge: List[int] = None) -> List[Knowledge]:
statement = select(Knowledge)
statement = cls._user_knowledge_filters(statement, user_id, knowledge_id_extra,
knowledge_type, name, page, limit,
filter_knowledge)
if sort_by == "create_time":
statement = statement.order_by(Knowledge.create_time.desc())
elif sort_by == "update_time":
statement = statement.order_by(Knowledge.update_time.desc())
elif sort_by == "name":
statement = statement.order_by(text('CONVERT(name USING gbk) ASC'))
async with get_async_db_session() as session:
return (await session.exec(statement)).all()
@classmethod
def count_user_knowledge(cls,
user_id: int,
knowledge_id_extra: List[int] = None,
knowledge_type: KnowledgeTypeEnum = None,
name: str = None) -> int:
statement = select(func.count(Knowledge.id))
statement = cls._user_knowledge_filters(statement, user_id, knowledge_id_extra,
knowledge_type, name)
with get_sync_db_session() as session:
return session.scalar(statement)
@classmethod
async def acount_user_knowledge(cls,
user_id: int,
knowledge_id_extra: List[int] = None,
knowledge_type: KnowledgeTypeEnum = None,
name: str = None) -> int:
statement = select(func.count(Knowledge.id))
statement = cls._user_knowledge_filters(statement, user_id, knowledge_id_extra,
knowledge_type, name)
async with get_async_db_session() as session:
return await session.scalar(statement)
@classmethod
def count_by_filter(cls, filters: List[Any]) -> int:
with get_sync_db_session() as session:
return session.scalar(select(Knowledge.id).where(*filters))
@classmethod
def judge_knowledge_permission(cls, user_name: str,
knowledge_ids: List[int]) -> List[Knowledge]:
"""
Based on username and knowledge baseIDList to get a list of knowledge bases that the user has permission to view
:param user_name: Username
:param knowledge_ids: The knowledge base uponIDVertical
:return: Returns a list of knowledge bases that the user has permissions
"""
# get user info
user_info = UserDao.get_user_by_username(user_name)
if not user_info:
return []
# Query the role the user belongs to
role_list = UserRoleDao.get_user_roles(user_info.user_id)
if not role_list:
return []
role_id_list = []
is_admin = False
for role in role_list:
role_id_list.append(role.role_id)
if role.role_id == 1:
is_admin = True
# admin User has all knowledge base permissions
if is_admin:
return KnowledgeDao.get_list_by_ids(knowledge_ids)
# query role List of knowledge bases with permissions
role_access_list = RoleAccessDao.find_role_access(role_id_list, [str(one) for one in knowledge_ids],
AccessType.KNOWLEDGE)
user_knowledge_list = cls.get_user_knowledge(user_info.user_id,
knowledge_id_extra=[int(access.third_id) for access in
role_access_list],
filter_knowledge=knowledge_ids)
return user_knowledge_list
@classmethod
async def ajudge_knowledge_permission(cls, user_name: str,
knowledge_ids: List[int]) -> List[Knowledge]:
"""
By Username and Knowledge BaseIDlist, asynchronously get a list of knowledge bases that the user has permission to view
Args:
user_name:
knowledge_ids:
Returns:
"""
# get user info
user_info = await UserDao.aget_user_by_username(user_name)
if not user_info:
return []
# Query the role the user belongs to
role_list = await UserRoleDao.aget_user_roles(user_info.user_id)
if not role_list:
return []
role_id_list = []
is_admin = False
for role in role_list:
role_id_list.append(role.role_id)
if role.role_id == 1:
is_admin = True
# admin User has all knowledge base permissions
if is_admin:
return await cls.aget_list_by_ids(knowledge_ids)
# query role List of knowledge bases with permissions
role_access_list = await RoleAccessDao.afind_role_access(role_id_list, [str(one) for one in knowledge_ids],
AccessType.KNOWLEDGE)
# Query whether the knowledge base created by the user is included
user_knowledge_list = await cls.aget_user_knowledge(user_info.user_id,
knowledge_id_extra=[int(access.third_id) for access in
role_access_list],
filter_knowledge=knowledge_ids)
return user_knowledge_list
@classmethod
def filter_knowledge_by_ids(cls,
knowledge_ids: List[int],
keyword: str = None,
page: int = 0,
limit: int = 0) -> (List[Knowledge], int):
"""
Based on keywords and knowledge baseidFilter out the corresponding knowledge base
"""
statement = select(Knowledge)
count_statement = select(func.count(Knowledge.id))
if knowledge_ids:
statement = statement.where(Knowledge.id.in_(knowledge_ids))
count_statement = count_statement.where(Knowledge.id.in_(knowledge_ids))
if keyword:
statement = statement.where(
or_(Knowledge.name.like('%' + keyword + '%'),
Knowledge.description.like('%' + keyword + '%')))
count_statement = count_statement.where(
or_(Knowledge.name.like('%' + keyword + '%'),
Knowledge.description.like('%' + keyword + '%')))
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
statement = statement.order_by(Knowledge.update_time.desc())
with get_sync_db_session() as session:
return session.exec(statement).all(), session.scalar(count_statement)
@classmethod
def generate_all_knowledge_filter(cls,
statement,
name: str = None,
knowledge_type: KnowledgeTypeEnum = None):
if knowledge_type:
statement = statement.where(Knowledge.type == knowledge_type.value)
if name:
conditions = [col(Knowledge.name).like(f'%{name}%'), col(Knowledge.description).like(f'%{name}%')]
file_knowledge_ids = KnowledgeFileDao.get_knowledge_ids_by_name(name)
if file_knowledge_ids:
conditions.append(Knowledge.id.in_(file_knowledge_ids))
if conditions:
statement = statement.where(or_(*conditions))
return statement
@classmethod
def get_all_knowledge(cls,
name: str = None,
knowledge_type: KnowledgeTypeEnum = None,
page: int = 0,
limit: int = 0) -> List[Knowledge]:
statement = select(Knowledge)
statement = cls.generate_all_knowledge_filter(statement,
name=name,
knowledge_type=knowledge_type)
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
statement = statement.order_by(Knowledge.update_time.desc())
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
async def aget_all_knowledge(cls,
name: str = None,
knowledge_type: KnowledgeTypeEnum = None,
sort_by: str = "update_time",
page: int = 0,
limit: int = 0) -> List[Knowledge]:
statement = select(Knowledge)
statement = cls.generate_all_knowledge_filter(statement,
name=name,
knowledge_type=knowledge_type)
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
if sort_by == "create_time":
statement = statement.order_by(Knowledge.create_time.desc())
elif sort_by == "update_time":
statement = statement.order_by(Knowledge.update_time.desc())
elif sort_by == "name":
statement = statement.order_by(text('CONVERT(name USING gbk) ASC'))
async with get_async_db_session() as session:
return (await session.exec(statement)).all()
@classmethod
def count_all_knowledge(cls,
name: str = None,
knowledge_type: KnowledgeTypeEnum = None) -> int:
statement = select(func.count(Knowledge.id))
statement = cls.generate_all_knowledge_filter(statement,
name=name,
knowledge_type=knowledge_type)
with get_sync_db_session() as session:
return session.scalar(statement)
@classmethod
async def acount_all_knowledge(cls,
name: str = None,
knowledge_type: KnowledgeTypeEnum = None) -> int:
statement = select(func.count(Knowledge.id))
statement = cls.generate_all_knowledge_filter(statement,
name=name,
knowledge_type=knowledge_type)
async with get_async_db_session() as session:
return await session.scalar(statement)
@classmethod
def update_knowledge_list(cls, knowledge_list: List[Knowledge]):
with get_sync_db_session() as session:
for knowledge in knowledge_list:
session.add(knowledge)
session.commit()
@classmethod
def get_knowledge_by_name(cls, name: str, user_id: int = 0) -> Knowledge:
""" Get Knowledge Base Details by Knowledge Base Name """
statement = select(Knowledge).where(Knowledge.name == name)
if user_id:
statement = statement.where(Knowledge.user_id == user_id)
with get_sync_db_session() as session:
return session.exec(statement).first()
@classmethod
def delete_knowledge(cls, knowledge_id: int, only_clear: bool = False):
"""
Delete or empty the knowledge base
"""
# Medical Treatment:knowledge file
with get_sync_db_session() as session:
session.exec(delete(KnowledgeFile).where(KnowledgeFile.knowledge_id == knowledge_id))
# Do not delete knowledge base records when clearing the knowledge base
if not only_clear:
session.exec(delete(Knowledge).where(Knowledge.id == knowledge_id))
session.commit()
@classmethod
def get_knowledge_by_time_range(cls, start_time: datetime, end_time: datetime, page: int = 0,
page_size: int = 0) -> List[Knowledge]:
""" Get a list of knowledge bases based on the creation timeframe """
statement = select(Knowledge).where(
Knowledge.create_time >= start_time,
Knowledge.create_time < end_time
)
if page and page_size:
statement = statement.offset((page - 1) * page_size).limit(page_size)
statement = statement.order_by(col(Knowledge.id).asc())
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
def get_first_knowledge(cls) -> Optional[Knowledge]:
""" Get the first knowledge base """
statement = select(Knowledge).order_by(col(Knowledge.id).asc()).limit(1)
with get_sync_db_session() as session:
return session.exec(statement).first()
================================================
FILE: src/backend/bisheng/knowledge/domain/models/knowledge_file.py
================================================
import json
from datetime import datetime
from enum import Enum
from typing import List, Optional, Dict, Any, Literal
# if TYPE_CHECKING:
from pydantic import field_validator
from sqlalchemy import JSON, Column, DateTime, String, or_, text, Text
from sqlmodel import Field, delete, func, select, update, col
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_async_db_session, get_sync_db_session
from bisheng.database.base import async_get_count, get_count
class KnowledgeFileStatus(Enum):
PROCESSING = 1 # Sedang diproses
SUCCESS = 2 # Berhasil
FAILED = 3 # Parse Failure
REBUILDING = 4 # Rebuilding
WAITING = 5 # In queue:
TIMEOUT = 6 # Super24Hour not parsed, parsing timeout
class QAStatus(Enum):
DISABLED = 0 # User manually closedQA
ENABLED = 1 # Enabled
PROCESSING = 2 # Sedang diproses
FAILED = 3 # QAFailed to insert vector library
class ParseType(Enum):
LOCAL = 'local' # Local mode resolution
UNS = 'uns' # unsService resolution, all converted topdfDoc.
# 1.3.0After the enumeration, the previous belongs to the file parsed on the version
ETL4LM = 'etl4lm' # etl4lmService Insights, includingpdfLayout Analysis for
UN_ETL4LM = 'un_etl4lm' # Nonetl4lmService parsing, nobboxContent, only source files andmdDoc.
class KnowledgeFileBase(SQLModelSerializable):
user_id: Optional[int] = Field(default=None, index=True)
user_name: Optional[str] = Field(default=None, index=True)
knowledge_id: int = Field(index=True)
file_name: str = Field(max_length=200, index=True)
file_size: Optional[int] = Field(default=None, index=False, description='File size inbytes')
md5: Optional[str] = Field(default=None, index=False)
parse_type: Optional[str] = Field(default=ParseType.LOCAL.value,
index=False,
description='Files parsed in what mode')
split_rule: Optional[str] = Field(default=None, sa_column=Column(Text), description='Files parsed in what mode')
bbox_object_name: Optional[str] = Field(default='', description='bboxFiles inminioStored object name')
status: Optional[int] = Field(default=KnowledgeFileStatus.WAITING.value,
index=False,
description='1: Parsing;2: Resolved successfully;3: Parse Failure')
object_name: Optional[str] = Field(default=None, index=False, description='Files inminioStored object name')
user_metadata: Optional[Dict[str, Any]] = Field(default_factory=dict, sa_column=Column(JSON, nullable=True),
description='User-defined metadata')
remark: Optional[str] = Field(default='', sa_column=Column(String(length=4096)))
updater_id: Optional[int] = Field(default=None, index=True, description='Last updated by userID')
updater_name: Optional[str] = Field(default=None, index=True)
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class QAKnowledgeBase(SQLModelSerializable):
user_id: Optional[int] = Field(default=None, index=True)
knowledge_id: int = Field(index=True)
questions: List[str] = Field(index=False)
answers: str = Field(index=False)
source: Optional[int] = Field(default=0, index=False, description='0: Unknown 1: Manual2: Audit, 3: api, 4: Batch import')
status: Optional[int] = Field(default=1, index=False,
description='1: Activate0: Close, the user manually closes;2: Sedang diproses3Failed to insert')
extra_meta: Optional[str] = Field(default=None, index=False)
remark: Optional[str] = Field(default='', sa_column=Column(String(length=4096)))
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
@field_validator('questions')
@classmethod
def validate_json(cls, v):
# dict_keys(['description', 'name', 'id', 'data'])
if not v:
return v
if not isinstance(v, List):
raise ValueError('question must be a valid JSON')
return v
@field_validator('answers')
@classmethod
def validate_answer(cls, v):
# dict_keys(['description', 'name', 'id', 'data'])
if not v:
return v
if isinstance(v, List):
return json.dumps(v, ensure_ascii=False)
return v
class KnowledgeFile(KnowledgeFileBase, table=True):
id: Optional[int] = Field(default=None, primary_key=True)
class QAKnowledge(QAKnowledgeBase, table=True):
id: Optional[int] = Field(default=None, primary_key=True)
questions: Optional[List[str]] = Field(default=None, sa_column=Column(JSON))
answers: Optional[str] = Field(default=None, sa_column=Column(Text))
class KnowledgeFileRead(KnowledgeFileBase):
id: int
class KnowledgeFileCreate(KnowledgeFileBase):
pass
class QAKnowledgeUpsert(QAKnowledgeBase):
"""Support modification"""
id: Optional[int] = None
answers: Optional[List[str] | str] = None
class KnowledgeFileDao(KnowledgeFileBase):
@classmethod
async def query_by_id(cls, file_id: int) -> Optional[KnowledgeFile]:
async with get_async_db_session() as session:
result = await session.execute(select(KnowledgeFile).where(KnowledgeFile.id == file_id))
return result.scalars().first()
@classmethod
def query_by_id_sync(cls, file_id: int) -> Optional[KnowledgeFile]:
with get_sync_db_session() as session:
return session.exec(select(KnowledgeFile).where(KnowledgeFile.id == file_id)).first()
@classmethod
def get_file_simple_by_knowledge_id(cls, knowledge_id: int, page: int, page_size: int):
offset = (page - 1) * page_size
with get_sync_db_session() as session:
return session.query(KnowledgeFile.id, KnowledgeFile.object_name).filter(
KnowledgeFile.knowledge_id == knowledge_id).order_by(
KnowledgeFile.id.asc()).offset(offset).limit(page_size).all()
@classmethod
def count_file_by_knowledge_id(cls, knowledge_id: int):
with get_sync_db_session() as session:
return session.query(func.count(
KnowledgeFile.id)).filter(KnowledgeFile.knowledge_id == knowledge_id).scalar()
@classmethod
def delete_batch(cls, file_ids: List[int]) -> bool:
with get_sync_db_session() as session:
session.exec(delete(KnowledgeFile).where(KnowledgeFile.id.in_(file_ids)))
session.commit()
return True
@classmethod
def add_file(cls, knowledge_file: KnowledgeFile) -> KnowledgeFile:
with get_sync_db_session() as session:
session.add(knowledge_file)
session.commit()
session.refresh(knowledge_file)
return knowledge_file
@classmethod
def update(cls, knowledge_file):
with get_sync_db_session() as session:
session.add(knowledge_file)
session.commit()
session.refresh(knowledge_file)
return knowledge_file
@classmethod
async def async_update(cls, knowledge_file):
async with get_async_db_session() as session:
session.add(knowledge_file)
await session.commit()
await session.refresh(knowledge_file)
return knowledge_file
@classmethod
def update_file_status(cls, file_ids: list[int], status: KnowledgeFileStatus, reason: str = None):
""" Batch update file status """
statement = update(KnowledgeFile).where(KnowledgeFile.id.in_(file_ids)).values(status=status.value,
remark=reason)
with get_sync_db_session() as session:
session.exec(statement)
session.commit()
@classmethod
def get_file_by_condition(cls, knowledge_id: int, md5_: str = None, file_name: str = None):
with get_sync_db_session() as session:
sql = select(KnowledgeFile).where(KnowledgeFile.knowledge_id == knowledge_id)
if md5_:
sql = sql.where(KnowledgeFile.md5 == md5_)
if file_name:
sql = sql.where(KnowledgeFile.file_name == file_name)
return session.exec(sql).all()
@classmethod
async def get_repeat_file(cls, knowledge_id: int, md5_: str = None, file_name: str = None):
sql = select(KnowledgeFile).where(KnowledgeFile.knowledge_id == knowledge_id)
if md5_ and file_name:
sql = sql.where(
or_(
KnowledgeFile.md5 == md5_,
KnowledgeFile.file_name == file_name
)
)
elif md5_:
sql = sql.where(KnowledgeFile.md5 == md5_)
elif file_name:
sql = sql.where(KnowledgeFile.file_name == file_name)
async with get_async_db_session() as session:
result = await session.exec(sql)
return result.first()
@classmethod
def select_list(cls, file_ids: List[int]):
if not file_ids:
return []
with get_sync_db_session() as session:
knowledge_files = session.exec(
select(KnowledgeFile).where(KnowledgeFile.id.in_(file_ids))).all()
if not knowledge_files:
raise ValueError('Doc.IDDoes not exist')
return knowledge_files
@classmethod
def get_file_by_ids(cls, file_ids: List[int]) -> List[KnowledgeFile]:
if not file_ids:
return []
with get_sync_db_session() as session:
return session.exec(select(KnowledgeFile).where(KnowledgeFile.id.in_(file_ids))).all()
@classmethod
def get_file_by_filters(cls,
knowledge_id: int,
file_name: str = None,
status: List[int] = None,
page: int = 0,
page_size: int = 0,
file_ids: List[int] = None) -> List[KnowledgeFile]:
statement = select(KnowledgeFile).where(KnowledgeFile.knowledge_id == knowledge_id)
if file_name:
statement = statement.where(KnowledgeFile.file_name.like(f'%{file_name}%'))
if status:
statement = statement.where(KnowledgeFile.status.in_(status))
if file_ids:
statement = statement.where(KnowledgeFile.id.in_(file_ids))
if page and page_size:
statement = statement.offset((page - 1) * page_size).limit(page_size)
statement = statement.order_by(KnowledgeFile.update_time.desc())
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
def get_files_by_multiple_status(cls, knowledge_id: int, status_list: List[int]) -> List[KnowledgeFile]:
"""
Based on Knowledge BaseIDand status list query file
Args:
knowledge_id: The knowledge base uponID
status_list: List of status values
Returns:
List[KnowledgeFile]: Matching Files List
"""
statement = select(KnowledgeFile).where(
KnowledgeFile.knowledge_id == knowledge_id,
KnowledgeFile.status.in_(status_list)
)
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
def count_file_by_filters(cls,
knowledge_id: int,
file_name: str = None,
status: List[int] = None,
file_ids: List[int] = None) -> int:
statement = select(func.count(
KnowledgeFile.id)).where(KnowledgeFile.knowledge_id == knowledge_id)
if file_name:
statement = statement.where(KnowledgeFile.file_name.like(f'%{file_name}%'))
if status:
statement = statement.where(KnowledgeFile.status.in_(status))
if file_ids:
statement = statement.where(KnowledgeFile.id.in_(file_ids))
with get_sync_db_session() as session:
return session.scalar(statement)
@classmethod
async def async_count_file_by_filters(cls,
knowledge_id: int,
file_name: str = None,
status: List[int] = None,
file_ids: List[int] = None) -> int:
statement = select(func.count(
KnowledgeFile.id)).where(KnowledgeFile.knowledge_id == knowledge_id)
if file_name:
statement = statement.where(KnowledgeFile.file_name.like(f'%{file_name}%'))
if status:
statement = statement.where(KnowledgeFile.status.in_(status))
if file_ids:
statement = statement.where(KnowledgeFile.id.in_(file_ids))
async with get_async_db_session() as session:
result = await session.execute(statement)
return result.scalar_one()
@classmethod
def get_knowledge_ids_by_name(cls, file_name: str) -> List[int]:
statement = select(KnowledgeFile.knowledge_id).where(KnowledgeFile.file_name.like(f'%{file_name}%')).group_by(
KnowledgeFile.knowledge_id)
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
def update_status_bulk(cls, file_ids: List[int], status: KnowledgeFileStatus, remark: str = "") -> None:
"""
Batch update file status
Args:
file_ids: Doc.IDVertical
status: New status value
Returns:
None
"""
if not file_ids:
return
statement = (
update(KnowledgeFile)
.where(col(KnowledgeFile.id).in_(file_ids))
)
statement = statement.values(status=status.value)
if remark:
statement = statement.values(remark=remark)
with get_sync_db_session() as session:
session.exec(statement)
session.commit()
@classmethod
def filter_file_by_metadata_fields(cls, knowledge_id: int, logical: Literal["and", "or"],
metadata_filters: List[Dict[str, Dict[str, Any]]]) -> List[int]:
"""
Filter knowledge files based on user-defined metadata fields
:param knowledge_id: The knowledge base uponID
:param logical: Logical operators, supporting "AND" OR "OR"
:param metadata_filters: User-defined metadata fields and their corresponding values
[{
field_a: {
'comparison': '=',
'value': 'some_value',
'extra_filter': [
{
'comparison': '!=',
'value': 'other_value'
}
]
}
}]
:return: Eligible Knowledge FilesIDVertical
"""
statement = "select id from knowledgefile where knowledge_id = :knowledge_id and "
params = {"knowledge_id": knowledge_id}
params_index = 1
field_statement = []
for metadata_filter in metadata_filters:
for key, key_info in metadata_filter.items():
key_comparison = key_info['comparison']
key_value = key_info['value']
extra_filter = key_info.get('extra_filter')
if key_value is not None:
params_key = f"tmp_params_{params_index}"
params[params_key] = key_value
sub_statement = f"{key} {key_comparison} :{params_key}"
else:
sub_statement = f"{key} {key_comparison}"
if extra_filter:
for sub_info in extra_filter:
params_index += 1
params_key = f"tmp_params_{params_index}"
params[params_key] = sub_info['value']
sub_statement += f" AND {key} {sub_info['comparison']} :{params_key}"
sub_statement = f"({sub_statement})"
field_statement.append(sub_statement)
params_index += 1
field_statement = f" {logical} ".join(field_statement)
statement += f"({field_statement})"
with get_sync_db_session() as session:
file_ids = []
result = session.execute(text(statement), params)
for one in result:
file_ids.append(one[0])
return file_ids
@classmethod
def update_file_updater(cls, file_id: int, updater_id: int, updater_name: str) -> None:
"""
Update Knowledge File Updater Information
:param file_id: Knowledge DocumentsID
:param updater_id: User who updated ID
:param updater_name: Updated By Username
:return: None
"""
statement = update(KnowledgeFile).where(col(KnowledgeFile.id) == file_id)
statement = statement.values(updater_id=updater_id, updater_name=updater_name)
with get_sync_db_session() as session:
session.exec(statement)
session.commit()
class QAKnoweldgeDao(QAKnowledgeBase):
@classmethod
def get_qa_knowledge_by_knowledge_id(cls, knowledge_id: int, page: int, page_size: int):
offset = (page - 1) * page_size
state = select(QAKnowledge).where(
QAKnowledge.knowledge_id == knowledge_id, ).offset(offset).limit(page_size)
with get_sync_db_session() as session:
return session.exec(state).all()
@classmethod
def get_qa_knowledge_by_knowledge_ids(cls, knowledge_ids: List[int]) -> List[QAKnowledge]:
with get_sync_db_session() as session:
return session.exec(
select(QAKnowledge).where(QAKnowledge.knowledge_id.in_(knowledge_ids))).all()
@classmethod
async def aget_qa_knowledge_by_knowledge_ids(cls, knowledge_ids: List[int]) -> List[QAKnowledge]:
async with get_async_db_session() as session:
result = await session.exec(
select(QAKnowledge).where(col(QAKnowledge.knowledge_id).in_(knowledge_ids)))
return result.all()
@classmethod
def get_qa_knowledge_by_primary_id(cls, qa_id: int) -> QAKnowledge:
with get_sync_db_session() as session:
return session.exec(select(QAKnowledge).where(QAKnowledge.id == qa_id)).first()
@classmethod
def get_qa_knowledge_by_name(cls, question: List[str], knowledge_id: int, exclude_id: int = None) -> QAKnowledge:
with get_sync_db_session() as session:
group_filters = []
for one in question:
group_filters.append(func.json_contains(QAKnowledge.questions, json.dumps(one)))
statement = select(QAKnowledge).where(
or_(*group_filters)).where(QAKnowledge.knowledge_id == knowledge_id)
if exclude_id:
statement = statement.where(QAKnowledge.id != exclude_id)
return session.exec(statement).first()
@classmethod
def update(cls, qa_knowledge: QAKnowledge):
if qa_knowledge.id is None:
raise ValueError('idTidak boleh kosong.')
with get_sync_db_session() as session:
session.add(qa_knowledge)
session.commit()
session.refresh(qa_knowledge)
return qa_knowledge
@classmethod
def delete_batch(cls, qa_ids: List[int]) -> bool:
with get_sync_db_session() as session:
session.exec(delete(QAKnowledge).where(QAKnowledge.id.in_(qa_ids)))
session.commit()
return True
@classmethod
def select_list(cls, ids: List[int]) -> List[QAKnowledge]:
with get_sync_db_session() as session:
QAKnowledges = session.exec(select(QAKnowledge).where(QAKnowledge.id.in_(ids))).all()
if not QAKnowledges:
raise ValueError('Knowledge base does not exist')
return QAKnowledges
@classmethod
def insert_qa(cls, qa_knowledge: QAKnowledgeUpsert):
with get_sync_db_session() as session:
qa = QAKnowledge.model_validate(qa_knowledge)
session.add(qa)
session.commit()
session.refresh(qa)
return qa
@classmethod
def batch_insert_qa(cls, qa_knowledges: List[QAKnowledgeUpsert]) -> List[QAKnowledge]:
with get_sync_db_session() as session:
qas = []
for qa_knowledge in qa_knowledges:
qa = QAKnowledge.model_validate(qa_knowledge)
qas.append(qa)
session.add_all(qas)
session.commit()
for qa in qas:
session.refresh(qa)
return qas
@classmethod
async def total_count(cls, sql):
async with get_async_db_session() as session:
return await async_get_count(session, sql)
@classmethod
async def query_by_condition(cls, sql):
async with get_async_db_session() as session:
result = await session.exec(sql)
return result.all()
@classmethod
def query_by_condition_v1(cls, source: List[int], create_start: str, create_end: str):
with get_sync_db_session() as session:
sql = select(QAKnowledge).where(QAKnowledge.source.in_(source)).where(
QAKnowledge.create_time.between(create_start, create_end))
return session.exec(sql).all()
# accordingqa_idTotal Fetched
@classmethod
async def async_count_by_id(cls, qa_id: int) -> int:
async with get_async_db_session() as session:
statement = select(func.count(QAKnowledge.id)).where(QAKnowledge.knowledge_id == qa_id)
return await async_get_count(session, statement)
@classmethod
def count_by_id(cls, qa_id: int) -> int:
with get_sync_db_session() as session:
statement = select(func.count(QAKnowledge.id)).where(QAKnowledge.knowledge_id == qa_id)
return get_count(session, statement)
@classmethod
def batch_update_status_by_ids(cls, qa_ids: List[int],
status: QAStatus,
remark: str = "") -> None:
"""
accordingQAkey learning pointsIDBulk Update Status
:param qa_ids: QAkey learning pointsIDVertical
:param status: Status
:param remark: Remark
:return:
"""
statement = (
update(QAKnowledge).where(col(QAKnowledge.id).in_(qa_ids))
)
statement = statement.values(status=status.value).values(remark=remark)
with get_sync_db_session() as session:
session.exec(statement)
session.commit()
# according knowledge_id Update status
@classmethod
def update_status_by_knowledge_id(cls, knowledge_id: int, status: QAStatus, remark: str = "") -> None:
"""
according knowledge_id Update status
:param knowledge_id: The knowledge base uponID
:param status: Status
:param remark: Remark
:return:
"""
statement = (
update(QAKnowledge).where(col(QAKnowledge.knowledge_id) == knowledge_id)
)
statement = statement.values(status=status.value).values(remark=remark)
with get_sync_db_session() as session:
session.exec(statement)
session.commit()
================================================
FILE: src/backend/bisheng/knowledge/domain/repositories/__init__.py
================================================
================================================
FILE: src/backend/bisheng/knowledge/domain/repositories/implementations/__init__.py
================================================
================================================
FILE: src/backend/bisheng/knowledge/domain/repositories/implementations/knowledge_file_repository_impl.py
================================================
from typing import Any, Coroutine
from sqlmodel import select, col
from sqlmodel.ext.asyncio.session import AsyncSession
from bisheng.common.repositories.implementations.base_repository_impl import BaseRepositoryImpl
from bisheng.knowledge.domain.models.knowledge_file import KnowledgeFile
from bisheng.knowledge.domain.repositories.interfaces.knowledge_file_repository import KnowledgeFileRepository
class KnowledgeFileRepositoryImpl(BaseRepositoryImpl[KnowledgeFile, int], KnowledgeFileRepository):
"""Knowledge Base Repository Implementation Class"""
def __init__(self, session: AsyncSession):
super().__init__(session, KnowledgeFile)
# according knowledge_idAndknowledge_file_ids Dapatkanuser_metadata Data field
async def get_user_metadata_by_knowledge_file_ids(self, knowledge_id: int, knowledge_file_ids: list[int]) -> dict[
int | None, list[dict[str, Any]] | None]:
query = select(KnowledgeFile).where(
KnowledgeFile.knowledge_id == knowledge_id,
col(KnowledgeFile.id).in_(knowledge_file_ids)
)
result = await self.session.exec(query)
knowledge_files = result.all()
user_metadata_dict = {}
for knowledge_file in knowledge_files:
if knowledge_file.user_metadata:
# Sort by newness
sorted_user_metadata = dict(sorted(knowledge_file.user_metadata.items(), key=lambda item: item[1].get("updated_at", 0), reverse=False))
user_metadata_dict[knowledge_file.id] = sorted_user_metadata
else:
user_metadata_dict[knowledge_file.id] = {}
return user_metadata_dict
================================================
FILE: src/backend/bisheng/knowledge/domain/repositories/implementations/knowledge_repository_impl.py
================================================
from typing import Union
from sqlmodel import Session
from sqlmodel.ext.asyncio.session import AsyncSession
from bisheng.common.repositories.implementations.base_repository_impl import BaseRepositoryImpl
from bisheng.knowledge.domain.models.knowledge import Knowledge
from bisheng.knowledge.domain.repositories.interfaces.knowledge_repository import KnowledgeRepository
class KnowledgeRepositoryImpl(BaseRepositoryImpl[Knowledge, int], KnowledgeRepository):
"""Knowledge Base Repository Implementation Class"""
def __init__(self, session: Union[AsyncSession, Session]):
super().__init__(session, Knowledge)
================================================
FILE: src/backend/bisheng/knowledge/domain/repositories/interfaces/__init__.py
================================================
================================================
FILE: src/backend/bisheng/knowledge/domain/repositories/interfaces/knowledge_file_repository.py
================================================
from abc import ABC
from typing import Any
from bisheng.common.repositories.interfaces.base_repository import BaseRepository
from bisheng.knowledge.domain.models.knowledge_file import KnowledgeFile
class KnowledgeFileRepository(BaseRepository[KnowledgeFile, int], ABC):
"""Knowledge Base File Repository Interface Class"""
async def get_user_metadata_by_knowledge_file_ids(self, knowledge_id: int,
knowledge_file_ids: list[int]) ->dict[
int | None, list[dict[str, Any]] | None]:
"""according knowledge_idAndknowledge_file_ids Dapatkanuser_metadata Data field"""
pass
================================================
FILE: src/backend/bisheng/knowledge/domain/repositories/interfaces/knowledge_repository.py
================================================
from abc import ABC
from bisheng.common.repositories.interfaces.base_repository import BaseRepository
from bisheng.knowledge.domain.models.knowledge import Knowledge
class KnowledgeRepository(BaseRepository[Knowledge, int], ABC):
"""Knowledge Base Repository Interface"""
pass
================================================
FILE: src/backend/bisheng/knowledge/domain/schemas/__init__.py
================================================
================================================
FILE: src/backend/bisheng/knowledge/domain/schemas/knowledge_file_schema.py
================================================
from typing import Optional, Type, Any, Dict
from pydantic import Field
from bisheng.knowledge.domain.models.knowledge_file import KnowledgeFileRead, KnowledgeFile
class KnowledgeFileInfoRes(KnowledgeFileRead):
"""Knowledge Base File Information Response Body"""
creat_user: Optional[str] = Field(default=None, description='Create user name')
update_user: Optional[str] = Field(default=None, description='Update user name')
@classmethod
def from_orm_extra(cls, model: KnowledgeFile, extra: Dict[str, Any]) -> "KnowledgeFileInfoRes":
"""FROMORMModel and Additional Information Create Response Body Instance"""
knowledge_file_info = cls.model_validate(model)
knowledge_file_info.creat_user = extra.get('creat_user')
knowledge_file_info.update_user = extra.get('update_user')
return knowledge_file_info
================================================
FILE: src/backend/bisheng/knowledge/domain/schemas/knowledge_rag_schema.py
================================================
from typing import Optional
from pydantic import BaseModel, Field
class Metadata(BaseModel):
""" Metadata model for knowledge retrieval-augmented generation (RAG) documents."""
document_id: Optional[int] = Field(default=None, description="Document ID")
document_name: Optional[str] = Field(default=None, description="Document Name")
abstract: Optional[str] = Field(default=None, description="Abstract of the document")
chunk_index: Optional[int] = Field(default=None, description="Chunk Index")
bbox: Optional[str] = Field(default=None, description="Bounding Box information")
page: Optional[int] = Field(default=None, description="Page number")
knowledge_id: Optional[int] = Field(default=None, description="Knowledge ID")
upload_time: Optional[int] = Field(default=None, description="Upload timestamp")
update_time: Optional[int] = Field(default=None, description="Update timestamp")
uploader: Optional[str] = Field(default=None, description="Uploader's name")
updater: Optional[str] = Field(default=None, description="Updater's name")
user_metadata: Optional[dict] = Field(default=None, description="User-defined metadata")
================================================
FILE: src/backend/bisheng/knowledge/domain/schemas/knowledge_schema.py
================================================
import re
from datetime import datetime
from typing import Optional, List, Any
from pydantic import BaseModel, Field, field_validator
from bisheng.knowledge.domain.models.knowledge import MetadataFieldType
class UpdatePreviewFileChunk(BaseModel):
knowledge_id: int = Field(..., description='The knowledge base uponID')
file_path: str = Field(..., description='FilePath')
text: str = Field(..., description='Text block Content')
chunk_index: int = Field(..., description='Text block index, Insidemetadatamile')
bbox: Optional[str] = Field(default='', description='Text blocksbboxMessage')
class MetadataField(BaseModel):
"""Metadata Field Model"""
field_name: str = Field(..., max_length=255, description='Metadata field name')
field_type: MetadataFieldType = Field(..., description='Metadata field type')
updated_at: int = Field(default_factory=lambda: int(datetime.now().timestamp()),
description='Metadata field update timestamp')
@field_validator('field_name')
@classmethod
def validate_field_name(cls, v: str) -> str:
# Must consist of lowercase letters, numbers, underscores, and must begin with a lowercase letter
pattern = r'^[a-z][a-z0-9_]*$'
if not re.match(pattern, v):
raise ValueError(
'field_name must start with a lowercase letter and contain only lowercase letters, numbers, and underscores, current value: {v}')
return v
class AddKnowledgeMetadataFieldsReq(BaseModel):
"""Add Knowledge Base Metadata Field Request Body"""
knowledge_id: int = Field(..., description='The knowledge base uponID')
metadata_fields: List[MetadataField] = Field(..., description='List of metadata fields to add')
class UpdateKnowledgeMetadataFieldsReq(BaseModel):
"""Update Knowledge Base Metadata Field Request Body"""
class UpdateMetadataFieldName(BaseModel):
"""Update Metadata Field Name Model"""
old_field_name: str = Field(..., description='Old Metadata Field Name')
new_field_name: str = Field(..., max_length=255, description='New Metadata Field Name')
@field_validator('new_field_name')
@classmethod
def validate_new_field_name(cls, v: str) -> str:
# Must consist of lowercase letters, numbers, underscores, and must begin with a lowercase letter
pattern = r'^[a-z][a-z0-9_]*$'
if not re.match(pattern, v):
raise ValueError(
f"new_field_name must start with a lowercase letter and contain only lowercase letters, numbers, and underscores, current value: {v}")
return v
@field_validator('old_field_name')
@classmethod
def validate_old_field_name(cls, v: str) -> str:
# Tidak boleh kosong.
if not v:
raise ValueError("old_field_name cannot be empty")
return v
knowledge_id: int = Field(..., description='The knowledge base uponID')
metadata_fields: List[UpdateMetadataFieldName] = Field(..., description='List of metadata fields to update')
class FileUserMetaDataInfo(BaseModel):
field_name: str = Field(..., description='Metadata field name')
field_value: Optional[Any] = Field(default=None, description='Metadata field value')
updated_at: int = Field(default_factory=lambda: int(datetime.now().timestamp()),
description='Metadata field update timestamp')
class ModifyKnowledgeFileMetaDataReq(BaseModel):
"""Change Knowledge Base File Metadata Request Body"""
knowledge_file_id: int = Field(..., description='Knowledge Base FilesID')
user_metadata_list: List[FileUserMetaDataInfo] = Field(..., description='List of file metadata to add')
================================================
FILE: src/backend/bisheng/knowledge/domain/services/__init__.py
================================================
================================================
FILE: src/backend/bisheng/knowledge/domain/services/knowledge_file_service.py
================================================
import copy
from typing import List
from loguru import logger
from bisheng.common.constants.vectorstore_metadata import KNOWLEDGE_RAG_METADATA_SCHEMA
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.http_error import UnAuthorizedError
from bisheng.common.errcode.knowledge import KnowledgeFileNotExistError, KnowledgeMetadataFieldNotExistError, \
KnowledgeMetadataFieldExistError, KnowledgeMetadataValueTypeConvertError
from bisheng.database.models.role_access import AccessType
from bisheng.knowledge.domain import utils
from bisheng.knowledge.domain.knowledge_rag import KnowledgeRag
from bisheng.knowledge.domain.repositories.interfaces.knowledge_file_repository import KnowledgeFileRepository
from bisheng.knowledge.domain.repositories.interfaces.knowledge_repository import KnowledgeRepository
from bisheng.knowledge.domain.schemas.knowledge_file_schema import KnowledgeFileInfoRes
from bisheng.knowledge.domain.schemas.knowledge_schema import ModifyKnowledgeFileMetaDataReq, MetadataField
from bisheng.open_endpoints.domain.schemas.knowledge import DeleteUserMetadataReq
from bisheng.user.domain.models.user import UserDao
class KnowledgeFileService:
"""Knowledge File Service Class"""
def __init__(self, knowledge_file_repository: 'KnowledgeFileRepository',
knowledge_repository: 'KnowledgeRepository'):
self.knowledge_file_repository = knowledge_file_repository
self.knowledge_repository = knowledge_repository
async def get_knowledge_file_info(self, login_user: 'UserPayload', knowledge_file_id: int):
"""Get Knowledge File Information"""
knowledge_file_model = await self.knowledge_file_repository.find_by_id(
entity_id=knowledge_file_id)
if not knowledge_file_model:
raise KnowledgeFileNotExistError()
knowledge_model = await self.knowledge_repository.find_by_id(
entity_id=knowledge_file_model.knowledge_id)
# Permission check
if not await login_user.async_access_check(
knowledge_model.user_id, str(knowledge_file_model.knowledge_id), AccessType.KNOWLEDGE
):
raise UnAuthorizedError()
create_user = await UserDao.aget_user(user_id=knowledge_file_model.user_id)
update_user = await UserDao.aget_user(user_id=knowledge_file_model.updater_id)
knowledge_file_info_res = KnowledgeFileInfoRes.from_orm_extra(model=knowledge_file_model,
extra={
'creat_user': create_user.user_name if create_user else '',
'update_user': update_user.user_name if update_user else create_user.user_name
})
if not knowledge_file_info_res.user_metadata:
metadata_field_dict = {item['field_name']: MetadataField(**item) for item in
knowledge_model.metadata_fields or []}
for key, item in knowledge_file_info_res.user_metadata.items():
if key in metadata_field_dict:
item['field_type'] = metadata_field_dict[key].field_type
return knowledge_file_info_res
@staticmethod
async def modify_milvus_file_user_metadata(invoke_user_id: int, knowledge_model, knowledge_file_id,
user_metadata: dict):
"""Change Milvus User metadata for files in"""
vector_client = await KnowledgeRag.init_knowledge_milvus_vectorstore(invoke_user_id, knowledge=knowledge_model,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
# Find all the data first
search_result = await vector_client.aclient.query(collection_name=knowledge_model.collection_name,
filter=f"document_id == {knowledge_file_id}", limit=10000)
# Modify User Metadata
for item in search_result:
item["user_metadata"] = user_metadata
# Bulk Update Data
await vector_client.aclient.upsert(collection_name=vector_client.collection_name,
data=search_result)
@staticmethod
async def modify_elasticsearch_file_user_metadata(knowledge_model, knowledge_file_id, user_metadata: dict):
"""Change Elasticsearch User metadata for files in"""
es_client = await KnowledgeRag.init_knowledge_es_vectorstore(knowledge=knowledge_model,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
# Use update_by_query to update eligible documents
res = await es_client.client.update_by_query(
index=knowledge_model.index_name,
body={
"script": {
"source": "ctx._source.metadata.user_metadata = params.user_metadata",
"lang": "painless",
"params": {"user_metadata": user_metadata}
},
"query": {
"term": {"metadata.document_id": knowledge_file_id}
}
}
)
logger.info(f"Elasticsearch update_by_query result: {res}")
async def modify_file_user_metadata(self, login_user: 'UserPayload',
modify_file_metadata_req: 'ModifyKnowledgeFileMetaDataReq'):
"""Add Knowledge File Metadata"""
knowledge_file_model = await self.knowledge_file_repository.find_by_id(
entity_id=modify_file_metadata_req.knowledge_file_id)
if not knowledge_file_model:
raise KnowledgeFileNotExistError()
knowledge_model = await self.knowledge_repository.find_by_id(
entity_id=knowledge_file_model.knowledge_id)
# Permission check
if not await login_user.async_access_check(
knowledge_model.user_id, str(knowledge_file_model.knowledge_id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError()
metadata_field_dict = {item['field_name']: MetadataField(**item) for item in
knowledge_model.metadata_fields or []}
# Initialize metadata if it's None
if knowledge_file_model.user_metadata is None:
knowledge_file_model.user_metadata = {}
# Create a new dictionary to store updated metadata
new_current_user_metadata = {}
for item in modify_file_metadata_req.user_metadata_list:
if item.field_name in metadata_field_dict.keys():
item_dict = item.model_dump()
# Data type conversion
try:
field_type = metadata_field_dict[item.field_name].field_type
field_value = utils.metadata_value_type_convert(
value=item_dict['field_value'], target_type=field_type)
item_dict['field_value'] = field_value
except Exception as e:
logger.error(f"Metadata value type conversion error: {e}")
continue
item_dict['field_type'] = metadata_field_dict[item.field_name].field_type
item_dict.pop('field_name')
new_current_user_metadata[item.field_name] = item_dict
# Update user metadata for knowledge files
knowledge_file_model.user_metadata = new_current_user_metadata
knowledge_file_model.updater_id = login_user.user_id
knowledge_file_model = await self.knowledge_file_repository.update(knowledge_file_model)
user_metadata = {key: value.get('field_value') for key, value in knowledge_file_model.user_metadata.items()}
# Change Milvus, Elasticsearch Corresponding metadata in
await self.modify_milvus_file_user_metadata(
login_user.user_id,
knowledge_model=knowledge_model,
knowledge_file_id=knowledge_file_model.id,
user_metadata=user_metadata
)
await self.modify_elasticsearch_file_user_metadata(
knowledge_model=knowledge_model,
knowledge_file_id=knowledge_file_model.id,
user_metadata=user_metadata
)
return knowledge_file_model
async def add_file_user_metadata(self, login_user: 'UserPayload', knowledge_id: int,
add_file_metadata_req: 'List[ModifyKnowledgeFileMetaDataReq]'):
"""Add Knowledge File Metadata"""
knowledge_model = await self.knowledge_repository.find_by_id(
entity_id=knowledge_id)
if not knowledge_model:
raise KnowledgeFileNotExistError()
# Permission check
if not await login_user.async_access_check(
knowledge_model.user_id, str(knowledge_model.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError()
metadata_field_dict = {item['field_name']: MetadataField(**item) for item in
knowledge_model.metadata_fields or []}
existing_files = await self.knowledge_file_repository.find_by_ids(
[req.knowledge_file_id for req in add_file_metadata_req])
existing_files_dict = {file.id: file for file in existing_files}
updated_knowledge_files = []
for modify_file_metadata_req in add_file_metadata_req:
knowledge_file_model = existing_files_dict.get(modify_file_metadata_req.knowledge_file_id)
# Check if knowledge file exists
if not knowledge_file_model:
raise KnowledgeFileNotExistError(
msg=f"Knowledge Base FilesID:{modify_file_metadata_req.knowledge_file_id} Does not exist")
# Initialize metadata if it's None
if knowledge_file_model.user_metadata is None:
knowledge_file_model.user_metadata = {}
current_user_metadata = copy.deepcopy(knowledge_file_model.user_metadata)
for item in modify_file_metadata_req.user_metadata_list:
if item.field_name not in metadata_field_dict.keys():
raise KnowledgeMetadataFieldNotExistError(field_name=item.field_name)
elif item.field_name in current_user_metadata.keys():
raise KnowledgeMetadataFieldExistError(
field_name=item.field_name,
msg=f"Knowledge Base FilesID:{modify_file_metadata_req.knowledge_file_id} Metadata field already exists:{item.field_name}"
)
item_dict = item.model_dump()
# Data type conversion
try:
field_type = metadata_field_dict[item.field_name].field_type
field_value = utils.metadata_value_type_convert(
value=item_dict['field_value'], target_type=field_type)
item_dict['field_value'] = field_value
except Exception as e:
raise KnowledgeMetadataValueTypeConvertError(
msg=f"Meta data fields {item.field_name} Value type conversion error: {e}")
item_dict['field_type'] = metadata_field_dict[item.field_name].field_type
item_dict.pop('field_name')
current_user_metadata[item.field_name] = item_dict
# Update user metadata for knowledge files
knowledge_file_model.user_metadata = current_user_metadata
knowledge_file_model.updater_id = login_user.user_id
updated_knowledge_files.append(knowledge_file_model)
# Bulk Update Knowledge Files
for knowledge_file_model in updated_knowledge_files:
knowledge_file_model = await self.knowledge_file_repository.update(knowledge_file_model)
user_metadata = {key: value.get('field_value') for key, value in knowledge_file_model.user_metadata.items()}
# Change Milvus, Elasticsearch Corresponding metadata in
await self.modify_milvus_file_user_metadata(
login_user.user_id,
knowledge_model=knowledge_model,
knowledge_file_id=knowledge_file_model.id,
user_metadata=user_metadata
)
await self.modify_elasticsearch_file_user_metadata(
knowledge_model=knowledge_model,
knowledge_file_id=knowledge_file_model.id,
user_metadata=user_metadata
)
return updated_knowledge_files
# Batch modify file user metadata
async def batch_modify_file_user_metadata(self, login_user: 'UserPayload',
knowledge_id: int,
modify_file_metadata_reqs: 'List[ModifyKnowledgeFileMetaDataReq]'):
"""Batch Modify Knowledge File Metadata"""
knowledge_model = await self.knowledge_repository.find_by_id(
entity_id=knowledge_id)
if not knowledge_model:
raise KnowledgeFileNotExistError()
# Permission check
if not await login_user.async_access_check(
knowledge_model.user_id, str(knowledge_model.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError()
metadata_field_dict = {item['field_name']: MetadataField(**item) for item in
knowledge_model.metadata_fields or []}
existing_files = await self.knowledge_file_repository.find_by_ids(
[req.knowledge_file_id for req in modify_file_metadata_reqs])
existing_files_dict = {file.id: file for file in existing_files}
updated_knowledge_files = []
for modify_file_metadata_req in modify_file_metadata_reqs:
knowledge_file_model = existing_files_dict.get(modify_file_metadata_req.knowledge_file_id)
if not knowledge_file_model:
raise KnowledgeFileNotExistError(
msg=f"Knowledge Base FilesID:{modify_file_metadata_req.knowledge_file_id} Does not exist")
# Initialize metadata if it's None
if knowledge_file_model.user_metadata is None:
knowledge_file_model.user_metadata = {}
current_user_metadata = copy.deepcopy(knowledge_file_model.user_metadata)
# Update user metadata for knowledge files
for item in modify_file_metadata_req.user_metadata_list:
if item.field_name not in metadata_field_dict.keys():
raise KnowledgeMetadataFieldNotExistError(field_name=item.field_name)
if item.field_name not in current_user_metadata.keys():
raise KnowledgeMetadataFieldNotExistError(
field_name=item.field_name,
msg=f"Knowledge Base FilesID:{modify_file_metadata_req.knowledge_file_id} No metadata fields exist:{item.field_name}"
)
existing_item = current_user_metadata.get(item.field_name)
try:
# Data Type
field_type = metadata_field_dict[item.field_name].field_type
# Update values and update time for existing fields
field_value = utils.metadata_value_type_convert(
value=item.field_value, target_type=field_type)
existing_item['field_value'] = field_value
current_user_metadata[item.field_name] = existing_item
except Exception as e:
raise KnowledgeMetadataValueTypeConvertError(
msg=f"Meta data fields {item.field_name} Value type conversion error: {e}")
knowledge_file_model.user_metadata = current_user_metadata
knowledge_file_model.updater_id = login_user.user_id
updated_knowledge_files.append(knowledge_file_model)
# Bulk Update Knowledge Files
for knowledge_file_model in updated_knowledge_files:
knowledge_file_model = await self.knowledge_file_repository.update(knowledge_file_model)
user_metadata = {key: value.get('field_value') for key, value in knowledge_file_model.user_metadata.items()}
# Change Milvus, Elasticsearch Corresponding metadata in
await self.modify_milvus_file_user_metadata(
login_user.user_id,
knowledge_model=knowledge_model,
knowledge_file_id=knowledge_file_model.id,
user_metadata=user_metadata
)
await self.modify_elasticsearch_file_user_metadata(
knowledge_model=knowledge_model,
knowledge_file_id=knowledge_file_model.id,
user_metadata=user_metadata
)
return updated_knowledge_files
async def batch_delete_file_user_metadata(self, login_user: 'UserPayload',
knowledge_id: int,
delete_user_metadata_req: 'List[DeleteUserMetadataReq]'):
"""
Bulk Delete Knowledge File Metadata
Args:
login_user:
knowledge_id:
delete_user_metadata_req:
Returns:
"""
knowledge_model = await self.knowledge_repository.find_by_id(
entity_id=knowledge_id)
if not knowledge_model:
raise KnowledgeFileNotExistError()
# Permission check
if not await login_user.async_access_check(
knowledge_model.user_id, str(knowledge_model.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError()
existing_files = await self.knowledge_file_repository.find_by_ids(
[req.knowledge_file_id for req in delete_user_metadata_req])
existing_files_dict = {file.id: file for file in existing_files}
updated_knowledge_files = []
for delete_metadata_req in delete_user_metadata_req:
knowledge_file_model = existing_files_dict.get(delete_metadata_req.knowledge_file_id)
if not knowledge_file_model:
raise KnowledgeFileNotExistError(
msg=f"Knowledge Base FilesID:{delete_metadata_req.knowledge_file_id} Does not exist")
current_user_metadata = copy.deepcopy(knowledge_file_model.user_metadata) or {}
# Delete the specified metadata field
for field_name in delete_metadata_req.field_names:
if field_name not in current_user_metadata.keys():
raise KnowledgeMetadataFieldNotExistError(
field_name=field_name,
msg=f"Knowledge Base FilesID:{delete_metadata_req.knowledge_file_id} No metadata fields exist:{field_name}"
)
current_user_metadata.pop(field_name)
knowledge_file_model.user_metadata = current_user_metadata
knowledge_file_model.updater_id = login_user.user_id
updated_knowledge_files.append(knowledge_file_model)
# Bulk Update Knowledge Files
for knowledge_file_model in updated_knowledge_files:
knowledge_file_model = await self.knowledge_file_repository.update(knowledge_file_model)
user_metadata = {key: value.get('field_value') for key, value in knowledge_file_model.user_metadata.items()}
# Change Milvus, Elasticsearch Corresponding metadata in
await self.modify_milvus_file_user_metadata(
login_user.user_id,
knowledge_model=knowledge_model,
knowledge_file_id=knowledge_file_model.id,
user_metadata=user_metadata
)
await self.modify_elasticsearch_file_user_metadata(
knowledge_model=knowledge_model,
knowledge_file_id=knowledge_file_model.id,
user_metadata=user_metadata
)
return updated_knowledge_files
async def list_knowledge_file_user_metadata(self, login_user: 'UserPayload',
knowledge_id: int,
knowledge_file_ids: List[int]):
"""
List user metadata for knowledge files
Args:
login_user:
knowledge_id:
knowledge_file_ids:
Returns:
"""
knowledge_model = await self.knowledge_repository.find_by_id(
entity_id=knowledge_id)
if not knowledge_model:
raise KnowledgeFileNotExistError()
# Permission check
if not await login_user.async_access_check(
knowledge_model.user_id, str(knowledge_model.id), AccessType.KNOWLEDGE
):
raise UnAuthorizedError()
user_metadata_dict = await self.knowledge_file_repository.get_user_metadata_by_knowledge_file_ids(
knowledge_id=knowledge_id,
knowledge_file_ids=knowledge_file_ids)
return user_metadata_dict
================================================
FILE: src/backend/bisheng/knowledge/domain/services/knowledge_service.py
================================================
import copy
from datetime import datetime
from fastapi import BackgroundTasks
from bisheng.common.constants.vectorstore_metadata import KNOWLEDGE_RAG_METADATA_SCHEMA
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.http_error import UnAuthorizedError
from bisheng.common.errcode.knowledge import KnowledgeNotExistError, KnowledgeMetadataFieldConflictError, \
KnowledgeMetadataFieldExistError, KnowledgeMetadataFieldNotExistError, KnowledgeMetadataFieldImmutableError
from bisheng.database.models.role_access import AccessType
from bisheng.knowledge.domain.knowledge_rag import KnowledgeRag
from bisheng.knowledge.domain.models.knowledge import KnowledgeDao, Knowledge
from bisheng.knowledge.domain.repositories.interfaces.knowledge_file_repository import KnowledgeFileRepository
from bisheng.knowledge.domain.repositories.interfaces.knowledge_repository import KnowledgeRepository
from bisheng.knowledge.domain.schemas.knowledge_schema import AddKnowledgeMetadataFieldsReq, \
UpdateKnowledgeMetadataFieldsReq
from bisheng.utils.util import retry_async
class KnowledgeService:
"""Service class for managing knowledge domain operations."""
def __init__(self, knowledge_repository: 'KnowledgeRepository',
knowledge_file_repository: 'KnowledgeFileRepository'):
self.knowledge_repository = knowledge_repository
self.knowledge_file_repository = knowledge_file_repository
async def add_metadata_fields(self, login_user: UserPayload, add_metadata_fields: AddKnowledgeMetadataFieldsReq):
"""Add metadata fields to a knowledge entity."""
knowledge_model = await self.knowledge_repository.find_by_id(entity_id=add_metadata_fields.knowledge_id)
if not knowledge_model:
raise KnowledgeNotExistError()
# Permission check
if not await login_user.async_access_check(
knowledge_model.user_id, str(knowledge_model.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError()
# Initialize metadata_fields if it's None
if knowledge_model.metadata_fields is None:
knowledge_model.metadata_fields = []
existing_field_names = {field["field_name"] for field in knowledge_model.metadata_fields}
# Built field names
built_field_names = [item.field_name for item in KNOWLEDGE_RAG_METADATA_SCHEMA]
# Determine if the added field conflicts with an existing field
for field in add_metadata_fields.metadata_fields:
if field.field_name in existing_field_names:
raise KnowledgeMetadataFieldExistError(field_name=field.field_name)
elif field.field_name in built_field_names:
raise KnowledgeMetadataFieldConflictError(field_name=field.field_name)
metadata_fields = copy.deepcopy(knowledge_model.metadata_fields)
# Add new metadata fields, avoiding duplicates
for field in add_metadata_fields.metadata_fields:
if field.field_name not in existing_field_names:
metadata_fields.append(field.model_dump())
knowledge_model.metadata_fields = metadata_fields
knowledge_model = await self.knowledge_repository.update(knowledge_model)
return knowledge_model
# Update Milvus and Elasticsearch metadata field names
async def update_vectorstore_metadata_field_names(self, invoke_user_id: int, knowledge_model, field_name_map):
"""Update metadata field names in Milvus and Elasticsearch vector stores."""
# Update Milvus metadata field names
# milvus_vectorstore = KnowledgeRag.init_knowledge_milvus_vectorstore_sync(knowledge=knowledge_model)
# Implement Milvus metadata field name update logic here
# Update Elasticsearch metadata field names
# es_vectorstore = KnowledgeRag.init_knowledge_es_vectorstore_sync(knowledge=knowledge_model)
# Implement Elasticsearch metadata field name update logic here
knowledge_model_files = await self.knowledge_file_repository.find_all(knowledge_id=knowledge_model.id)
vector_client = await KnowledgeRag.init_knowledge_milvus_vectorstore(invoke_user_id, knowledge=knowledge_model,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
es_client = await KnowledgeRag.init_knowledge_es_vectorstore(knowledge=knowledge_model,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
# Requestmilvus
@retry_async(delay=3)
async def request_milvus(new_data):
# Bulk Update Data
await vector_client.aclient.upsert(collection_name=vector_client.collection_name, data=new_data)
# Requestes
@retry_async(delay=3)
async def request_es(request_body):
await es_client.client.update_by_query(
index=knowledge_model.index_name,
body=request_body
)
for knowledge_file in knowledge_model_files:
# Update Milvus metadata
# Implement Milvus metadata field name update logic for each knowledge file here
# Query all vectors in this knowledge file in Milvus.
search_result = await vector_client.aclient.query(collection_name=knowledge_model.collection_name,
filter=f"document_id == {knowledge_file.id}", limit=10000)
# Modify User Metadata Field Name
for item in search_result:
for old_field_name, new_field_name in field_name_map.items():
if old_field_name in item["user_metadata"]:
item["user_metadata"][new_field_name] = item["user_metadata"].pop(old_field_name)
# Bulk Update Data
await request_milvus(search_result)
# Update Elasticsearch metadata
# Implement Elasticsearch metadata field name update logic for each knowledge file here
# Use update_by_query to update eligible documents
script_lines = []
for old_field_name, new_field_name in field_name_map.items():
script_lines.append(
f"if (ctx._source.metadata.user_metadata.containsKey('{old_field_name}')) " +
"{ ctx._source.metadata.user_metadata['" + new_field_name + "'] = " +
"ctx._source.metadata.user_metadata.remove('" + old_field_name + "'); }"
)
script_source = " ".join(script_lines)
body = {
"script": {
"source": script_source,
"lang": "painless"
},
"query": {
"term": {"metadata.document_id": knowledge_file.id}
}
}
# Update es
await request_es(body)
# Update knowledge file's user_metadata field
user_metadata_dict = copy.deepcopy(knowledge_file.user_metadata)
for old_field_name, new_field_name in field_name_map.items():
user_metadata = user_metadata_dict.pop(old_field_name, None)
if user_metadata is not None:
user_metadata["updated_at"] = int(datetime.now().timestamp())
user_metadata[new_field_name] = user_metadata
knowledge_file.user_metadata = user_metadata_dict
await self.knowledge_file_repository.update(knowledge_file)
async def update_metadata_fields(self, login_user: UserPayload,
update_metadata_fields: UpdateKnowledgeMetadataFieldsReq,
background_tasks: BackgroundTasks):
"""Update metadata field names in a knowledge entity."""
knowledge_model = await self.knowledge_repository.find_by_id(entity_id=update_metadata_fields.knowledge_id)
if not knowledge_model:
raise KnowledgeNotExistError()
# Permission check
if not await login_user.async_access_check(
knowledge_model.user_id, str(knowledge_model.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError()
if knowledge_model.metadata_fields is None:
return knowledge_model # No metadata fields to update
# Built field names
built_field_names = [item.field_name for item in KNOWLEDGE_RAG_METADATA_SCHEMA]
existing_field_names = {field["field_name"] for field in knowledge_model.metadata_fields}
for field in update_metadata_fields.metadata_fields:
if field.old_field_name in built_field_names:
raise KnowledgeMetadataFieldImmutableError(field_name=field.old_field_name)
elif field.new_field_name in built_field_names:
raise KnowledgeMetadataFieldConflictError(field_name=field.new_field_name)
elif field.new_field_name in existing_field_names:
raise KnowledgeMetadataFieldExistError(field_name=field.new_field_name)
elif field.old_field_name not in existing_field_names:
raise KnowledgeMetadataFieldNotExistError(field_name=field.old_field_name)
field_name_map = {
field_update.old_field_name: field_update.new_field_name
for field_update in update_metadata_fields.metadata_fields
}
# Check if all old field names exist and new field names do not exist
for old_field_name in field_name_map.keys():
if old_field_name not in existing_field_names or field_name_map[old_field_name] in existing_field_names:
return knowledge_model
metadata_fields = copy.deepcopy(knowledge_model.metadata_fields)
# Update metadata field names
for i, field in enumerate(metadata_fields):
if field["field_name"] in field_name_map:
metadata_fields[i]["field_name"] = field_name_map[field["field_name"]]
metadata_fields[i]["updated_at"] = int(datetime.now().timestamp())
knowledge_model.metadata_fields = metadata_fields
knowledge_model = await self.knowledge_repository.update(knowledge_model)
# Milvus and ES metadata field name update logic
background_tasks.add_task(
self.update_vectorstore_metadata_field_names,
login_user.user_id,
knowledge_model,
field_name_map
)
return knowledge_model
async def delete_vectorstore_metadata_fields(self, invoke_user_id: int, knowledge_model, field_names: list[str]):
"""Delete metadata fields in Milvus and Elasticsearch vector stores."""
# Delete Milvus metadata fields
# milvus_vectorstore = KnowledgeRag.init_knowledge_milvus_vectorstore_sync(knowledge=knowledge_model)
# Implement Milvus metadata field deletion logic here
# Delete Elasticsearch metadata fields
# es_vectorstore = KnowledgeRag.init_knowledge_es_vectorstore_sync(knowledge=knowledge_model)
# Implement Elasticsearch metadata field deletion logic here
knowledge_model_files = await self.knowledge_file_repository.find_all(knowledge_id=knowledge_model.id)
vector_client = await KnowledgeRag.init_knowledge_milvus_vectorstore(invoke_user_id, knowledge=knowledge_model,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
es_client = await KnowledgeRag.init_knowledge_es_vectorstore(knowledge=knowledge_model,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
# Requestmilvus
@retry_async(delay=3)
async def request_milvus(new_data):
# Bulk Update Data
await vector_client.aclient.upsert(collection_name=vector_client.collection_name, data=new_data)
# Requestes
@retry_async(delay=3)
async def request_es(request_body):
await es_client.client.update_by_query(
index=knowledge_model.index_name,
body=request_body
)
for knowledge_file in knowledge_model_files:
# Delete Milvus metadata fields
# Implement Milvus metadata field deletion logic for each knowledge file here
# Query all vectors in this knowledge file in Milvus.
search_result = await vector_client.aclient.query(collection_name=knowledge_model.collection_name,
filter=f"document_id == {knowledge_file.id}", limit=10000)
# Delete the specified metadata field
for item in search_result:
for field_name in field_names:
if field_name in item["user_metadata"]:
del item["user_metadata"][field_name]
# Bulk Update Data
await request_milvus(search_result)
# Delete Elasticsearch metadata fields
# Implement Elasticsearch metadata field deletion logic for each knowledge file here
# Use update_by_query to update eligible documents
script_lines = []
for field_name in field_names:
script_lines.append(
f"ctx._source.metadata.user_metadata.remove('{field_name}');"
)
script_source = " ".join(script_lines)
body = {
"script": {
"source": script_source,
"lang": "painless"
},
"query": {
"term": {"metadata.document_id": knowledge_file.id}
}
}
# Update es
await request_es(body)
# Update knowledge file's user_metadata field
knowledge_file.user_metadata = {
key: value for key, value in knowledge_file.user_metadata.items()
if key not in field_names
}
await self.knowledge_file_repository.update(knowledge_file)
async def delete_metadata_fields(self, login_user: UserPayload, knowledge_id: int, field_names: list[str],
background_tasks: BackgroundTasks):
"""Delete metadata fields from a knowledge entity."""
knowledge_model = await self.knowledge_repository.find_by_id(entity_id=knowledge_id)
if not knowledge_model:
raise KnowledgeNotExistError()
# Permission check
if not await login_user.async_access_check(
knowledge_model.user_id, str(knowledge_model.id), AccessType.KNOWLEDGE_WRITE
):
raise UnAuthorizedError()
# Initialize metadata_fields if it's None
if knowledge_model.metadata_fields is None:
knowledge_model.metadata_fields = []
existing_field_names = [field["field_name"] for field in knowledge_model.metadata_fields]
# Check if all field names to be deleted exist
for field_name in field_names:
if field_name not in existing_field_names:
raise KnowledgeMetadataFieldNotExistError(field_name=field_name)
# Filter out metadata fields to be deleted
metadata_fields = [
field for field in knowledge_model.metadata_fields
if field['field_name'] not in field_names
]
knowledge_model.metadata_fields = metadata_fields
knowledge_model = await self.knowledge_repository.update(knowledge_model)
# Milvus and ES metadata field deletion logic
background_tasks.add_task(
self.delete_vectorstore_metadata_fields,
login_user.user_id,
knowledge_model,
field_names
)
return knowledge_model
async def list_metadata_fields(self, default_user, knowledge_id):
"""
List metadata fields of a knowledge entity.
Args:
default_user:
knowledge_id:
Returns:
"""
knowledge_model = await self.knowledge_repository.find_by_id(entity_id=knowledge_id)
if not knowledge_model:
raise KnowledgeNotExistError()
# Permission check
if not await default_user.async_access_check(
knowledge_model.user_id, str(knowledge_model.id), AccessType.KNOWLEDGE
):
raise UnAuthorizedError()
if knowledge_model.metadata_fields is None:
knowledge_model.metadata_fields = []
# Sortmetadata_fields by updated_at desc
knowledge_model.metadata_fields.sort(key=lambda x: x.get("updated_at", 0), reverse=True)
return {
"knowledge_id": knowledge_model.id,
"metadata_fields": knowledge_model.metadata_fields or []
}
@classmethod
def get_all_knowledge_by_time_range(cls, start_data: datetime, end_data: datetime, page: int = 1,
page_size: int = 10):
"""Get all the knowledge bases created in a certain timeframe"""
return KnowledgeDao.get_knowledge_by_time_range(start_data, end_data, page, page_size)
@classmethod
def get_first_knowledge(cls) -> Knowledge | None:
return KnowledgeDao.get_first_knowledge()
================================================
FILE: src/backend/bisheng/knowledge/domain/utils.py
================================================
from datetime import datetime
from bisheng.knowledge.domain.models.knowledge import MetadataFieldType
# Convert Time String to Timestamp
def time_str_to_timestamp(time_str):
try:
return int(time_str)
except:
pass
try:
dt = datetime.strptime(time_str, "%Y-%m-%d %H:%M:%S")
return int(dt.timestamp())
except:
pass
try:
dt = datetime.strptime(time_str, "%Y-%m-%dT%H:%M:%S")
return int(dt.timestamp())
except:
pass
try:
dt = datetime.strptime(time_str, "%Y/%m/%d %H:%M:%S")
return int(dt.timestamp())
except:
pass
try:
dt = datetime.strptime(time_str, "%Y.%m.%d %H:%M:%S")
return int(dt.timestamp())
except:
pass
try:
dt = datetime.strptime(time_str, "%Y-%m-%dT%H:%M:%S.%fZ")
return int(dt.timestamp())
except:
pass
try:
dt = datetime.strptime(time_str, "%Y-%m-%dT%H:%M:%S.%f")
return int(dt.timestamp())
except:
pass
try:
dt = datetime.strptime(time_str, "%Yyear%mMonth%dHarian %H:%M:%S")
return int(dt.timestamp())
except:
raise ValueError("Unsupported time format")
# metadata Data Format Conversion
def metadata_value_type_convert(value, target_type: MetadataFieldType):
try:
if target_type == MetadataFieldType.NUMBER:
if not value:
return 0
return int(value)
elif target_type == MetadataFieldType.STRING:
if not value:
return None
return str(value)
elif target_type == MetadataFieldType.TIME:
if not value:
return None
timestamp = time_str_to_timestamp(value)
return timestamp
else:
raise ValueError("Unsupported target type")
except Exception as e:
raise ValueError(f"Failed to convert value '{value}' to type '{target_type}': {e}")
================================================
FILE: src/backend/bisheng/knowledge/rag/__init__.py
================================================
================================================
FILE: src/backend/bisheng/knowledge/rag/elasticsearch_factory.py
================================================
from typing import Optional, List, Dict
from elasticsearch import Elasticsearch, AsyncElasticsearch
from bisheng.common.schemas.rag_schema import RagMetadataFieldSchema
from bisheng.common.services.config_service import settings
from bisheng.core.vectorstore import ElasticsearchStore, AsyncElasticsearchStore, BM25Strategy, AsyncBM25Strategy
def generate_metadata_mappings(metadata_schemas: Optional[List[RagMetadataFieldSchema]]):
"""
Generate Elasticsearch metadata mappings from RagMetadataFieldSchema list.
Args:
metadata_schemas:
Returns:
"""
metadata_mappings: Optional[Dict[str, any]] = None
for schema in metadata_schemas or []:
if metadata_mappings is None:
metadata_mappings = {}
if schema.field_type == 'text':
metadata_mappings[schema.field_name] = {'type': 'text',
'fields': {'keyword': {'type': 'keyword', 'ignore_above': 256}}}
elif schema.field_type == 'boolean':
metadata_mappings[schema.field_name] = {'type': 'boolean'}
elif schema.field_type == 'int8' or schema.field_type == 'int16':
metadata_mappings[schema.field_name] = {'type': 'short'}
elif schema.field_type == 'int32':
metadata_mappings[schema.field_name] = {'type': 'integer'}
elif schema.field_type == 'int64':
metadata_mappings[schema.field_name] = {'type': 'long'}
elif schema.field_type == 'float':
metadata_mappings[schema.field_name] = {'type': 'float'}
elif schema.field_type == 'double':
metadata_mappings[schema.field_name] = {'type': 'double'}
elif schema.field_type == 'json':
metadata_mappings[schema.field_name] = {'type': 'flattened'}
return metadata_mappings
class ElasticsearchFactory:
@staticmethod
def init_vectorstore_sync(index_name: str, **kwargs) -> ElasticsearchStore:
"""Initialize an ElasticsearchStore vectorstore for keywords."""
es_conf = settings.get_vectors_conf().elasticsearch
metadata_schemas: Optional[List[RagMetadataFieldSchema]] = kwargs.pop('metadata_schemas', None)
metadata_mappings = generate_metadata_mappings(metadata_schemas)
es_client = ElasticsearchStore(
index_name=index_name,
strategy=BM25Strategy(),
es_connection=Elasticsearch(hosts=es_conf.elasticsearch_url, **es_conf.ssl_verify),
metadata_mappings=metadata_mappings,
**kwargs
)
return es_client
@staticmethod
def init_vectorstore(index_name: str, **kwargs) -> AsyncElasticsearchStore:
"""Asynchronously initialize an ElasticsearchStore vectorstore for keywords."""
es_conf = settings.get_vectors_conf().elasticsearch
metadata_schemas: Optional[List[RagMetadataFieldSchema]] = kwargs.pop('metadata_schemas', None)
metadata_mappings = generate_metadata_mappings(metadata_schemas)
es_client = AsyncElasticsearchStore(
index_name=index_name,
strategy=AsyncBM25Strategy(),
es_connection=AsyncElasticsearch(hosts=es_conf.elasticsearch_url, **es_conf.ssl_verify),
metadata_mappings=metadata_mappings,
**kwargs
)
return es_client
================================================
FILE: src/backend/bisheng/knowledge/rag/milvus_factory.py
================================================
from typing import Optional, List, Dict
from langchain_core.embeddings import Embeddings
from pymilvus import DataType
from bisheng.common.schemas.rag_schema import RagMetadataFieldSchema
from bisheng.common.services.config_service import settings
from bisheng.core.vectorstore import Milvus
_default_index_params = {"index_type": "HNSW", "metric_type": "L2", "params": {"M": 8, "efConstruction": 64}}
class MilvusFactory:
@staticmethod
def init_vectorstore(collection_name: str, embedding_function: Embeddings, **kwargs) -> Milvus:
conf = settings.get_vectors_conf().milvus
connection_args = conf.connection_args.copy()
if connection_args.get('host') and connection_args.get('port'):
uri = f"http://{connection_args.pop('host')}:{connection_args.pop('port')}"
connection_args['uri'] = uri
metadata_schemas: Optional[List[RagMetadataFieldSchema]] = kwargs.pop('metadata_schemas', None)
milvus_metadata_schema: Optional[Dict[str, any]] = None
for schema in metadata_schemas or []:
if milvus_metadata_schema is None:
milvus_metadata_schema = {}
schema_kwargs = schema.kwargs or {}
if schema.field_type == 'text':
milvus_metadata_schema[schema.field_name] = {'dtype': DataType.VARCHAR,
"kwargs": schema_kwargs}
elif schema.field_type == 'int8':
milvus_metadata_schema[schema.field_name] = {'dtype': DataType.INT8, "kwargs": schema_kwargs}
elif schema.field_type == 'int16':
milvus_metadata_schema[schema.field_name] = {'dtype': DataType.INT16, "kwargs": schema_kwargs}
elif schema.field_type == 'int32':
milvus_metadata_schema[schema.field_name] = {'dtype': DataType.INT32, "kwargs": schema_kwargs}
elif schema.field_type == 'int64':
milvus_metadata_schema[schema.field_name] = {'dtype': DataType.INT64, "kwargs": schema_kwargs}
elif schema.field_type == 'float':
milvus_metadata_schema[schema.field_name] = {'dtype': DataType.FLOAT, "kwargs": schema_kwargs}
elif schema.field_type == 'double':
milvus_metadata_schema[schema.field_name] = {'dtype': DataType.DOUBLE, "kwargs": schema_kwargs}
elif schema.field_type == 'json':
milvus_metadata_schema[schema.field_name] = {'dtype': DataType.JSON, "kwargs": schema_kwargs}
elif schema.field_type == 'boolean':
milvus_metadata_schema[schema.field_name] = {'dtype': DataType.BOOL, "kwargs": schema_kwargs}
return Milvus(
embedding_function=embedding_function,
collection_name=collection_name,
connection_args=connection_args,
auto_id=True,
index_params=_default_index_params,
metadata_schema=milvus_metadata_schema,
**kwargs
)
================================================
FILE: src/backend/bisheng/linsight/__init__.py
================================================
================================================
FILE: src/backend/bisheng/linsight/api/__init__.py
================================================
================================================
FILE: src/backend/bisheng/linsight/api/endpoints/__init__.py
================================================
================================================
FILE: src/backend/bisheng/linsight/api/endpoints/linsight.py
================================================
import asyncio
import json
import os
import time
from datetime import datetime
from typing import List, Literal, Optional, Union
from urllib import parse
from fastapi import APIRouter, Depends, Body, Query, UploadFile, File, BackgroundTasks, Request, Form
from loguru import logger
from pydantic import BaseModel, ValidationError
from sse_starlette import EventSourceResponse
from starlette.responses import StreamingResponse
from starlette.websockets import WebSocket
from bisheng.api.services.invite_code.invite_code import InviteCodeService
from bisheng.api.services.knowledge import KnowledgeService
from bisheng.linsight.domain.services.message_stream_handle import MessageStreamHandle
from bisheng.linsight.domain.services.sop_manage import SOPManageService
from bisheng.linsight.domain.services.workbench_impl import LinsightWorkbenchImpl
from bisheng.api.v1.schema.base_schema import PageList
from bisheng.linsight.domain.schemas.inspiration_schema import SOPManagementSchema, SOPManagementUpdateSchema
from bisheng.linsight.domain.schemas.linsight_schema import LinsightQuestionSubmitSchema, DownloadFilesSchema, \
SubmitFileSchema, LinsightToolSchema, ToolChildrenSchema
from bisheng.api.v1.schemas import UnifiedResponseModel, resp_200
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum, ApplicationTypeEnum
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.http_error import UnAuthorizedError, NotFoundError
from bisheng.common.errcode.linsight import LinsightQuestionError, LinsightUseUpError, LinsightModifySopError, \
LinsightStartTaskError, LinsightSessionVersionRunningError, LinsightQueueStatusError, FileUploadError, \
SopShowcaseError
from bisheng.common.errcode.server import InvalidOperationError, ResourceDownloadError
from bisheng.common.schemas.telemetry.event_data_schema import ApplicationAliveEventData, ApplicationProcessEventData
from bisheng.common.services import telemetry_service
from bisheng.common.services.config_service import settings
from bisheng.core.cache.redis_manager import get_redis_client
from bisheng.core.logger import trace_id_var
from bisheng.core.storage.minio.minio_manager import get_minio_storage
from bisheng.linsight.domain.models.linsight_session_version import LinsightSessionVersionDao, SessionVersionStatusEnum, \
LinsightSessionVersion
from bisheng.linsight.domain.models.linsight_sop import LinsightSOPDao, LinsightSOPRecord
from bisheng.knowledge.domain.models.knowledge import KnowledgeTypeEnum, KnowledgeDao
from bisheng.linsight.domain.services.state_message_manager import LinsightStateMessageManager, MessageData, \
MessageEventType
from bisheng.share_link.api.dependencies import header_share_token_parser
from bisheng.share_link.domain.models.share_link import ShareLink
from bisheng.utils import util
router = APIRouter()
# Inspiration Upload File
@router.post("/workbench/upload-file", summary="Inspiration Upload File", response_model=UnifiedResponseModel)
async def upload_file(
background_tasks: BackgroundTasks,
file: UploadFile = File(...),
login_user: UserPayload = Depends(UserPayload.get_login_user)) -> UnifiedResponseModel:
"""
Inspiration Upload File
:param background_tasks:
:param file: files uploaded
:param login_user: Logged in user information
:return: Upload results
"""
try:
# Call the implementation class to process file uploads
upload_result = await LinsightWorkbenchImpl.upload_file(file)
background_tasks.add_task(LinsightWorkbenchImpl.parse_file, upload_result, login_user.user_id)
result = {
"file_id": upload_result.get("file_id"),
"file_name": upload_result.get("original_filename"),
"parsing_status": upload_result.get("parsing_status"),
}
except Exception as e:
logger.error(f"Upload Failed: {str(e)}")
return FileUploadError.return_resp()
finally:
await file.close()
# Back to upload results
return resp_200(data=result,
message="Key file uploaded successfully! and start parsing. Please check the resolution status later.")
# Get file resolution status
@router.post("/workbench/file-parsing-status", summary="Get file resolution status",
response_model=UnifiedResponseModel)
async def get_file_parsing_status(
file_ids: List[str] = Body(..., description="Doc.IDVertical", embed=True),
login_user: UserPayload = Depends(UserPayload.get_login_user)) -> UnifiedResponseModel:
"""
Get file resolution status
:param file_ids:
:param login_user:
:return:
"""
# Call the implementation class to get the file parsing state
key_prefix = LinsightWorkbenchImpl.FILE_INFO_REDIS_KEY_PREFIX
file_ids = [f"{key_prefix}{file_id}" for file_id in file_ids]
redis_client = await get_redis_client()
# Use Redis right of privacy amget Method Get file parsing status in batches
parsing_status = await redis_client.amget(file_ids)
return resp_200(data=parsing_status, message="File parsing status retrieved successfully")
@router.post("/workbench/file_download", summary="Inspiration File Download", response_model=UnifiedResponseModel)
async def linsight_file_download(
file_url: str = Body(..., embed=True),
session_version_id: str = Body(..., embed=True),
login_user: UserPayload = Depends(UserPayload.get_login_user),
share_link: Union['ShareLink', None] = Depends(header_share_token_parser)) -> UnifiedResponseModel:
session_version_model = await LinsightSessionVersionDao.get_by_id(session_version_id)
if not session_version_model:
raise NotFoundError()
# judge permission
if session_version_model.user_id != login_user.user_id and not login_user.is_admin():
# Access by sharing a link
if (share_link is None or
share_link.meta_data is None or
share_link.meta_data.get("versionId") != session_version_id):
raise UnAuthorizedError()
minio_client = await get_minio_storage()
file_url = file_url.lstrip("/")
if file_url.startswith(minio_client.bucket):
file_url = file_url[len(minio_client.bucket) + 1:]
file_share_url = await minio_client.get_share_link(file_url)
return resp_200(data={
"file_path": file_share_url
})
# Submit an Idea User Issue Request
@router.post("/workbench/submit", summary="Submit an Idea User Issue Request")
async def submit_linsight_workbench(
submit_obj: LinsightQuestionSubmitSchema = Body(..., description="Idea User Issue Submitter"),
login_user: UserPayload = Depends(UserPayload.get_login_user)) -> EventSourceResponse:
"""
Submit an Idea User Issue Request
:param submit_obj:
:param login_user:
:return:
"""
logger.info(f"Users {login_user.user_id} Submit an Idea Question: {submit_obj.question}")
async def event_generator():
"""
Event generator for generatingSSE events
"""
try:
system_config = await settings.aget_all_config()
# DapatkanLinsight_invitation_code
linsight_invitation_code = system_config.get("linsight_invitation_code", False)
if linsight_invitation_code:
if await InviteCodeService.use_invite_code(user_id=login_user.user_id) is False:
yield LinsightUseUpError().to_sse_event_instance()
return
message_session_model, linsight_session_version_model = await LinsightWorkbenchImpl.submit_user_question(
submit_obj,
login_user)
response_data = {
"message_session": message_session_model.model_dump(),
"linsight_session_version": linsight_session_version_model.model_dump()
}
except Exception as e:
yield LinsightQuestionError(exception=e).to_sse_event_instance()
return
yield {
"event": "linsight_workbench_submit",
"data": json.dumps(response_data)
}
# Task Title Generation
title_data = await LinsightWorkbenchImpl.task_title_generate(question=submit_obj.question,
chat_id=message_session_model.chat_id,
login_user=login_user)
linsight_session_version_model.title = title_data.get("task_title")
await LinsightSessionVersionDao.insert_one(linsight_session_version_model)
yield {
"event": "linsight_workbench_title_generate",
"data": json.dumps(title_data)
}
return EventSourceResponse(event_generator())
# workbench Generate and Reimagine IdeasSOP
@router.post("/workbench/generate-sop", summary="Generate and Reimagine IdeasSOP", response_model=UnifiedResponseModel)
async def generate_sop(
request: Request,
linsight_session_version_id: str = Body(..., description="Inspiration Conversation VersionID"),
previous_session_version_id: str = Body(None, description="Previous Invisible Conversation VersionID"),
feedback_content: str = Body(None, description="User feedback content"),
reexecute: bool = Body(False, description="Whether to rerun the buildSOP"),
sop_id: int = Body(None, description="Featured Cases'ID"),
example_session_version_id: str = Body(default=None, description="Reference Cases'linsight_version_id"),
login_user: UserPayload = Depends(UserPayload.get_login_user)) -> EventSourceResponse:
"""
Generate and Reimagine IdeasSOP
:param previous_session_version_id:
:param reexecute:
:param linsight_session_version_id:
:param feedback_content:
:param sop_id:
:param login_user:
:return:
"""
logger.info(
f"Start Generating and Redesigning IdeasSOP, Inscription Conversation VersionID: {linsight_session_version_id} ")
start_time = time.time()
session_version = await LinsightSessionVersionDao.get_by_id(linsight_session_version_id)
# Get a list of knowledge bases with permissions
if not session_version:
raise NotFoundError.http_exception()
example_sop = None
if sop_id:
sop_db = await SOPManageService.get_sop_by_id(sop_id)
example_sop = sop_db.content if sop_db else None
elif example_session_version_id:
example_session_version = await LinsightSessionVersionDao.get_by_id(example_session_version_id)
example_sop = example_session_version.sop if example_session_version else None
res = []
linsight_conf = settings.get_linsight_conf()
if session_version.org_knowledge_enabled and linsight_conf.max_knowledge_num > 0:
res, _ = await KnowledgeService.get_knowledge(request, login_user, KnowledgeTypeEnum.NORMAL, None, 1,
linsight_conf.max_knowledge_num)
if session_version.personal_knowledge_enabled:
knowledge = await KnowledgeDao.aget_user_knowledge(login_user.user_id, None,
KnowledgeTypeEnum.PRIVATE)
if knowledge:
res.extend(knowledge)
async def event_generator():
"""
Event generator for generatingSSE events
"""
# BuatSOP
sop_generate = LinsightWorkbenchImpl.generate_sop(
linsight_session_version_id=linsight_session_version_id,
previous_session_version_id=previous_session_version_id,
feedback_content=feedback_content,
reexecute=reexecute,
login_user=login_user,
knowledge_list=res,
example_sop=example_sop
)
async for event in sop_generate:
yield event
# End
yield {
"event": "sop_generate_complete",
"data": json.dumps({"message": "SOPGeneration and re-planning complete"})
}
try:
return EventSourceResponse(event_generator())
finally:
end_time = time.time()
await telemetry_service.log_event(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.APPLICATION_ALIVE,
trace_id=trace_id_var.get(),
event_data=ApplicationAliveEventData(
app_id=ApplicationTypeEnum.LINSIGHT.value,
app_name=ApplicationTypeEnum.LINSIGHT.value,
app_type=ApplicationTypeEnum.LINSIGHT,
chat_id=session_version.session_id,
start_time=int(start_time),
end_time=int(end_time)
))
await telemetry_service.log_event(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.APPLICATION_PROCESS,
trace_id=trace_id_var.get(),
event_data=ApplicationProcessEventData(
app_id=ApplicationTypeEnum.LINSIGHT.value,
app_name=ApplicationTypeEnum.LINSIGHT.value,
app_type=ApplicationTypeEnum.LINSIGHT,
chat_id=session_version.session_id,
start_time=int(start_time),
end_time=int(end_time),
process_time=int((end_time - start_time) * 1000)
))
# workbench Changesop
@router.post("/workbench/sop-modify", summary="Modify InspirationSOP", response_model=UnifiedResponseModel)
async def modify_sop(
sop_content: str = Body(..., description="SOPContents"),
linsight_session_version_id: str = Body(..., description="Inspiration Conversation VersionID"),
login_user: UserPayload = Depends(UserPayload.get_login_user)) -> UnifiedResponseModel:
"""
Modify InspirationSOP
:param sop_content:
:param linsight_session_version_id:
:param login_user:
:return:
"""
session_version_model = await LinsightSessionVersionDao.get_by_id(
linsight_session_version_id=linsight_session_version_id)
if not session_version_model:
return NotFoundError.return_resp()
if login_user.user_id != session_version_model.user_id:
return UnAuthorizedError.return_resp()
try:
modify_res = await LinsightWorkbenchImpl.modify_sop(linsight_session_version_id=linsight_session_version_id,
sop_content=sop_content)
except Exception as e:
return LinsightModifySopError.return_resp(data=str(e))
return resp_200(modify_res)
# workbench to process
@router.post("/workbench/start-execute", summary="Start Executing Reims", response_model=UnifiedResponseModel)
async def start_execute_sop(
background_tasks: BackgroundTasks,
linsight_session_version_id: str = Body(..., description="Inspiration Conversation VersionID", embed=True),
login_user: UserPayload = Depends(UserPayload.get_login_user)) -> UnifiedResponseModel:
"""
Start Executing ReimsSOP
:param linsight_session_version_id:
:param login_user:
:return:
"""
session_version_model = await LinsightSessionVersionDao.get_by_id(
linsight_session_version_id=linsight_session_version_id)
if not session_version_model:
return NotFoundError.return_resp()
if login_user.user_id != session_version_model.user_id:
return UnAuthorizedError.return_resp()
if session_version_model.status in [SessionVersionStatusEnum.COMPLETED, SessionVersionStatusEnum.TERMINATED,
SessionVersionStatusEnum.IN_PROGRESS]:
# The Inspiration session version has been completed or is being executed and cannot be executed again
return LinsightSessionVersionRunningError.return_resp()
from bisheng.linsight.worker import LinsightQueue
try:
redis_client = await get_redis_client()
queue = LinsightQueue('queue', namespace="linsight", redis=redis_client)
await queue.put(data=linsight_session_version_id)
# will besopWrite to record table
background_tasks.add_task(SOPManageService.add_sop_record, LinsightSOPRecord(
name=session_version_model.title,
description=None,
user_id=login_user.user_id,
content=session_version_model.sop,
linsight_version_id=session_version_model.id,
create_time=session_version_model.create_time,
))
except Exception as e:
logger.error(f"Failed to start the Ideas task: {str(e)}")
await InviteCodeService.revoke_invite_code(user_id=login_user.user_id)
return LinsightStartTaskError.return_resp(data=str(e))
return resp_200(data=True,
message="Ideas execution task has started, execution results will be returned via message flow")
# workbench User input
@router.post("/workbench/user-input", summary="User input Ideas", response_model=UnifiedResponseModel)
async def user_input(
session_version_id: str = Body(..., description="Inspiration Conversation VersionID"),
linsight_execute_task_id: str = Body(..., description="Inspiration Task ExecutionID"),
input_content: str = Body(..., description="User input"),
files: Optional[List[SubmitFileSchema]] = Body(None, description="User-uploaded files"),
login_user: UserPayload = Depends(UserPayload.get_login_user)) -> UnifiedResponseModel:
"""
User input
:param files:
:param session_version_id:
:param input_content:
:param linsight_execute_task_id:
:param login_user:
:return:
"""
session_version_model = await LinsightSessionVersionDao.get_by_id(
linsight_session_version_id=session_version_id)
if not session_version_model:
return NotFoundError.return_resp()
if login_user.user_id != session_version_model.user_id:
return UnAuthorizedError.return_resp()
state_message_manager = LinsightStateMessageManager(session_version_id=session_version_id)
# If there are documents Process files first
processed_files = await LinsightWorkbenchImpl.human_participate_add_file(session_version_model, files=files)
await state_message_manager.set_user_input(task_id=linsight_execute_task_id, user_input=input_content,
files=processed_files)
return resp_200(data=True, message="User input submitted")
# workbench Submitting Execution Result Feedback
@router.post("/workbench/submit-feedback", summary="Submitting Execution Result Feedback",
response_model=UnifiedResponseModel)
async def submit_feedback(
background_tasks: BackgroundTasks,
linsight_session_version_id: str = Body(..., description="Inspiration Conversation VersionID"),
feedback: str = Body(None, description="User feedback"),
score: int = Body(0, ge=0, le=5, description="Users rating1-5cent"),
is_reexecute: bool = Body(False, description="Whether to re-execute"),
cancel_feedback: bool = Body(False, description="Cancel feedback"),
login_user: UserPayload = Depends(UserPayload.get_login_user)) -> UnifiedResponseModel:
"""
Submitting Execution Result Feedback
:param background_tasks:
:param cancel_feedback:
:param linsight_session_version_id:
:param feedback:
:param score:
:param is_reexecute:
:param login_user:
:return:
"""
session_version_model = await LinsightSessionVersionDao.get_by_id(
linsight_session_version_id=linsight_session_version_id)
if not session_version_model:
return NotFoundError.return_resp()
if login_user.user_id != session_version_model.user_id:
return UnAuthorizedError.return_resp()
if score is not None and 0 < score <= 5:
session_version_model.score = score
await SOPManageService.update_sop_record_score(session_version_model.id, score)
if feedback is not None:
session_version_model.execute_feedback = feedback
else:
session_version_model.execute_feedback = "User did not provide feedback"
# If the feedback is canceled
if cancel_feedback:
session_version_model.execute_feedback = "User canceled feedback"
await LinsightSessionVersionDao.insert_one(session_version_model)
return resp_200(data=True, message="Submit successful.")
session_version_model = await LinsightSessionVersionDao.insert_one(session_version_model)
if is_reexecute:
# Re-implementing the Logic of Ideas
system_config = await settings.aget_all_config()
# DapatkanLinsight_invitation_code
linsight_invitation_code = system_config.get("linsight_invitation_code", False)
if linsight_invitation_code:
if await InviteCodeService.use_invite_code(user_id=login_user.user_id) is False:
return LinsightUseUpError.return_resp()
# Inspiration Conversation Version
linsight_session_version_model = LinsightSessionVersion(
session_id=session_version_model.session_id,
user_id=login_user.user_id,
question=session_version_model.question,
tools=session_version_model.tools,
org_knowledge_enabled=session_version_model.org_knowledge_enabled,
personal_knowledge_enabled=session_version_model.personal_knowledge_enabled,
files=session_version_model.files,
title=session_version_model.title
)
linsight_session_version_model = await LinsightSessionVersionDao.insert_one(linsight_session_version_model)
return resp_200(data=linsight_session_version_model.model_dump(),
message="The submission successfully succeeded.")
else:
if feedback is not None and feedback.strip() != "":
await SOPManageService.update_sop_record_feedback(session_version_model.id, feedback)
return resp_200(data=True, message="Submit successful.")
# workbench Termination
@router.post("/workbench/terminate-execute", summary="Termination of execution of Ideas",
response_model=UnifiedResponseModel)
async def terminate_execute(
linsight_session_version_id: str = Body(..., description="Inspiration Conversation VersionID", embed=True),
login_user: UserPayload = Depends(UserPayload.get_login_user)) -> UnifiedResponseModel:
"""
Termination of execution of Ideas
:param linsight_session_version_id:
:param login_user:
:return:
"""
# The logic of executing Spirituality is now terminated
session_version_model = await LinsightSessionVersionDao.get_by_id(
linsight_session_version_id=linsight_session_version_id)
if not session_version_model:
return NotFoundError.return_resp()
if login_user.user_id != session_version_model.user_id:
return UnAuthorizedError.return_resp()
if session_version_model.status == SessionVersionStatusEnum.COMPLETED:
# return resp_500(code=400, message="Execution cannot be terminated because the Inspiration session version has been completed")
return InvalidOperationError.return_resp()
if session_version_model.status == SessionVersionStatusEnum.TERMINATED:
# return resp_500(code=400, message="Execution terminated for Inspiration session version")
return InvalidOperationError.return_resp()
from bisheng.linsight.worker import LinsightQueue
redis_client = await get_redis_client()
queue = LinsightQueue('queue', namespace="linsight", redis=redis_client)
try:
# Remove task from queue
await queue.remove(linsight_session_version_id)
except Exception as e:
logger.error(f"Failed to delete queue task: {str(e)}")
# Update status is terminated
session_version_model.status = SessionVersionStatusEnum.TERMINATED
state_message_manager = LinsightStateMessageManager(session_version_id=linsight_session_version_id)
await state_message_manager.set_session_version_info(session_version_model)
state_message_manager = LinsightStateMessageManager(session_version_id=session_version_model.id)
# Push termination message
await state_message_manager.push_message(
MessageData(
event_type=MessageEventType.TASK_TERMINATED,
data={
"message": "Task has been actively stopped by the user",
"session_id": session_version_model.id,
"terminated_at": datetime.now().isoformat()
}
)
)
return resp_200(data=True, message="Idea Execution Terminated")
# Get all the Inspiration information for the current session
@router.get("/workbench/session-version-list", summary="Get all the Inspiration information for the current session",
response_model=UnifiedResponseModel)
async def get_linsight_session_version_list(
session_id: str = Query(..., description="SessionsID"),
login_user: UserPayload = Depends(UserPayload.get_login_user),
share_link: Union['ShareLink', None] = Depends(header_share_token_parser)
) -> UnifiedResponseModel:
"""
Get all the Inspiration information for the current session
:param share_link:
:param session_id:
:param login_user:
:return:
"""
linsight_session_version_models = await LinsightWorkbenchImpl.get_linsight_session_version_list(session_id)
if linsight_session_version_models and login_user.user_id != linsight_session_version_models[0].user_id:
# Access by sharing a link
session_version_ids = [model.id for model in linsight_session_version_models]
# Access by sharing a link
if (share_link is None or
share_link.meta_data is None or
share_link.meta_data.get("versionId") not in session_version_ids):
return UnAuthorizedError.return_resp()
# Only return to the shared version of the Inspiration session
linsight_session_version_models = [
model for model in linsight_session_version_models if model.id == share_link.meta_data.get("versionId")
]
return resp_200([model.model_dump() for model in linsight_session_version_models])
# Get task execution details
@router.get("/workbench/execute-task-detail", summary="Get task execution details", response_model=UnifiedResponseModel)
async def get_execute_task_detail(
session_version_id: str = Query(..., description="Inspiration Conversation VersionID"),
login_user: UserPayload = Depends(UserPayload.get_login_user),
share_link: Union['ShareLink', None] = Depends(header_share_token_parser)) -> UnifiedResponseModel:
"""
Get task execution details
:param share_link:
:param session_version_id:
:param login_user:
:return:
"""
execute_task_models = await LinsightWorkbenchImpl.get_execute_task_detail(session_version_id)
if not execute_task_models:
return resp_200([])
linsight_session_version_model = await LinsightSessionVersionDao.get_by_id(session_version_id)
if login_user.user_id != linsight_session_version_model.user_id:
# Access by sharing a link
if (share_link is None or
share_link.meta_data is None or
share_link.meta_data.get("versionId") != session_version_id):
return UnAuthorizedError.return_resp()
return resp_200(execute_task_models)
# Creating an Idea Task Message Flow websocket
@router.websocket("/workbench/task-message-stream", name="task_message_stream")
async def task_message_stream(
websocket: WebSocket,
session_version_id: str = Query(..., description="Inspiration Conversation VersionID"),
login_user: UserPayload = Depends(UserPayload.get_login_user_from_ws)):
"""
Creating an Idea Task Message Flow websocket
:param Authorize:
:param websocket:
:param session_version_id:
:return:
"""
start_time = time.time()
try:
message_handler = MessageStreamHandle(websocket=websocket, session_version_id=session_version_id)
await message_handler.connect()
except Exception as e:
await websocket.close(code=1000, reason=str(e))
return
finally:
end_time = time.time()
session_version_info = await LinsightSessionVersionDao.get_by_id(session_version_id)
await telemetry_service.log_event(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.APPLICATION_ALIVE,
trace_id=trace_id_var.get(),
event_data=ApplicationAliveEventData(
app_id=ApplicationTypeEnum.LINSIGHT.value,
app_name=ApplicationTypeEnum.LINSIGHT.value,
app_type=ApplicationTypeEnum.LINSIGHT,
chat_id=session_version_info.session_id if session_version_info else "",
start_time=int(start_time),
end_time=int(end_time)
))
await telemetry_service.log_event(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.APPLICATION_PROCESS,
trace_id=trace_id_var.get(),
event_data=ApplicationProcessEventData(
app_id=ApplicationTypeEnum.LINSIGHT.value,
app_name=ApplicationTypeEnum.LINSIGHT.value,
app_type=ApplicationTypeEnum.LINSIGHT,
chat_id=session_version_info.session_id if session_version_info else "",
start_time=int(start_time),
end_time=int(end_time),
process_time=int((end_time - start_time) * 1000)
))
# Batch Download Task Files
@router.post("/workbench/batch-download-files", summary="Batch Download Task Files")
async def batch_download_files(
zip_name: str = Body(..., description="Package name"),
file_info_list: List[DownloadFilesSchema] = Body(..., description="File Information List"),
login_user: UserPayload = Depends(UserPayload.get_login_user)):
"""
Batch Download Task Files
:param zip_name:
:param file_info_list:
:param login_user:
:return:
"""
try:
# Call to implement class processing batch download
zip_bytes = await LinsightWorkbenchImpl.batch_download_files(file_info_list)
zip_name = zip_name if os.path.splitext(zip_name)[-1] == ".zip" else f"{zip_name}.zip"
# Convert to unicode String
zip_name = parse.quote(zip_name)
return StreamingResponse(
iter([zip_bytes]),
media_type="application/zip",
headers={
"Content-Disposition": f"attachment; filename={zip_name}"
}
)
except Exception as e:
logger.error(f"Failed to download file in bulk: {str(e)}")
return ResourceDownloadError.return_resp(data=str(e))
# Get Queue Queue Status
@router.get("/workbench/queue-status", summary="Get Ideas Queue Queue Status", response_model=UnifiedResponseModel)
async def get_queue_status(
session_version_id: str = Query(..., description="Inspiration Conversation VersionID"),
login_user: UserPayload = Depends(UserPayload.get_login_user)) -> UnifiedResponseModel:
"""
Get Ideas Queue Queue Status
:param session_version_id:
:param login_user:
:return:
"""
from bisheng.linsight.worker import LinsightQueue
redis_client = await get_redis_client()
queue = LinsightQueue('queue', namespace="linsight", redis=redis_client)
try:
index = await queue.index(session_version_id)
return resp_200(data={"index": index}, message="Get Ideas queue queue status successfully")
except Exception as e:
logger.error(f"Failed to get Ideas queue queue status: {str(e)}")
return LinsightQueueStatusError.return_resp(data=str(e))
# InspirationmdTransferpdf or docx Mengunduh
@router.post("/workbench/download-md-to-pdf-or-docx", summary="InspirationmdTransferpdf or docx Mengunduh")
async def download_md_to_pdf_or_docx(
file_info: DownloadFilesSchema = Body(..., description="File information"),
to_type: Literal["pdf", "docx"] = Body(..., description="the target file type of the conversion,pdfORdocx"),
login_user: UserPayload = Depends(UserPayload.get_login_user)):
"""
InspirationmdTransferpdf or docx Mengunduh
:param file_info:
:param to_type:
:param login_user:
:return:
"""
try:
# Call the implementation class to process the file download
file_name, file_bytes = await LinsightWorkbenchImpl.download_file(file_info)
md_str = file_bytes.decode('utf-8')
# Filename Removal Extension
file_name = os.path.splitext(file_name)[0]
if to_type == "pdf":
from bisheng.common.utils.markdown_cmpnt.md_to_pdf import md_to_pdf_bytes
converted_bytes = await util.sync_func_to_async(md_to_pdf_bytes)(md_str)
file_name = f"{file_name}.pdf"
content_type = "application/pdf"
else:
from bisheng.common.utils.markdown_cmpnt.md_to_docx.markdocx import MarkDocx
mark_docx = MarkDocx()
converted_bytes, _ = await util.sync_func_to_async(mark_docx)(md_str)
file_name = f"{file_name}.docx"
content_type = "application/vnd.openxmlformats-officedocument.wordprocessingml.document"
# Convert to unicode String
file_name = parse.quote(file_name)
return StreamingResponse(
iter([converted_bytes]),
media_type=content_type,
headers={
"Content-Disposition": f"attachment; filename={file_name}"
}
)
except Exception as e:
logger.error(f"This content failed to load: {str(e)}")
return ResourceDownloadError.return_resp(data=str(e))
@router.post("/sop/add", summary="Add InspirationSOP", response_model=UnifiedResponseModel)
async def add_sop(
sop_obj: SOPManagementSchema = Body(..., description="SOPObjects"),
login_user: UserPayload = Depends(UserPayload.get_login_user)) -> UnifiedResponseModel:
"""
Add InspirationSOP
:return:
"""
if not login_user.is_admin():
return UnAuthorizedError.return_resp()
return await SOPManageService.add_sop(sop_obj, user_id=login_user.user_id)
@router.post("/sop/update", summary="Update IdeasSOP", response_model=UnifiedResponseModel)
async def update_sop(
sop_obj: SOPManagementUpdateSchema = Body(..., description="SOPObjects"),
login_user: UserPayload = Depends(UserPayload.get_admin_user)) -> UnifiedResponseModel:
"""
Update IdeasSOP
:return:
"""
sop_obj.user_id = login_user.user_id
return await SOPManageService.update_sop(sop_obj, update_version_id=False)
@router.get("/sop/list", summary="Get IdeasSOPVertical", response_model=UnifiedResponseModel)
async def get_sop_list(
keywords: str = Query(None, description="Keywords Search"),
showcase: bool = Query(None, description="Get featured cases only?"),
page: int = Query(1, ge=1, description="Page"),
page_size: int = Query(10, ge=1, le=100, description="Items per page"),
sort: Literal["asc", "desc"] = Query("desc", description="Sort ByascORdesc"),
login_user: UserPayload = Depends(UserPayload.get_admin_user)) -> UnifiedResponseModel:
"""
Get IdeasSOPVertical
:return:
"""
sop_pages = await SOPManageService.get_sop_list(keywords=keywords, showcase=showcase, page=page,
page_size=page_size,
sort=sort)
return resp_200(data=sop_pages)
@router.get("/sop/record", summary="Get IdeasSOPRecord", response_model=UnifiedResponseModel)
async def get_sop_record(login_user: UserPayload = Depends(UserPayload.get_admin_user),
keyword: str = Query(None, description="Search keyword ..."),
sort: str = Query(default='desc', description="Sort ByascORdesc"),
page: int = Query(1, ge=1, description="Page"),
page_size: int = Query(10, ge=1, le=100, description="Items per page")):
res, count = await SOPManageService.get_sop_record(keyword, sort, page, page_size)
return resp_200(PageList(total=count, list=res))
@router.post("/sop/record/sync", summary="SynchronoussopRecord toSOPGallery", response_model=UnifiedResponseModel)
async def sync_sop_record(
login_user: UserPayload = Depends(UserPayload.get_admin_user),
record_ids: list[int] = Body(..., description="sopThe only one in the record sheetid"),
override: Optional[bool] = Body(default=False,
description="Force override or not"),
save_new: Optional[bool] = Body(default=False,
description="Do you want to save as newsop")) -> UnifiedResponseModel:
"""
SynchronousSOP"Log to"SOPGallery
"""
repeat_name = await SOPManageService.sync_sop_record(record_ids, override, save_new)
return resp_200(data={
"repeat_name": repeat_name,
}, message="success")
@router.post("/sop/upload", summary="Batch importSOPWarehousing", response_model=UnifiedResponseModel)
async def upload_sop_file(
file: UploadFile = File(..., description="Uploaded bySOPDoc."),
override: Optional[bool] = Body(default=False, description="Force override or not"),
save_new: Optional[bool] = Body(default=False, description="Do you want to save as newsop"),
ignore_error: Optional[bool] = Body(default=False,
description="Whether to ignore the file and find the wrong record"),
login_user: UserPayload = Depends(UserPayload.get_admin_user)) -> UnifiedResponseModel:
"""
Batch importSOPWarehousing
"""
try:
success_rows, error_rows, repeat_rows = await SOPManageService.upload_sop_file(login_user, file, ignore_error,
override, save_new)
return resp_200(data={
"success_rows": success_rows,
"error_rows": error_rows,
"repeat_rows": repeat_rows,
})
except Exception as e:
raise e
finally:
await file.close()
@router.delete("/sop/remove", summary="Delete IdeasSOP", response_model=UnifiedResponseModel)
async def remove_sop(
sop_ids: List[int] = Body(..., description="SOPUniqueness quantificationIDVertical", embed=True),
login_user: UserPayload = Depends(UserPayload.get_admin_user)) -> UnifiedResponseModel:
"""
Delete IdeasSOP
:return:
"""
return await SOPManageService.remove_sop(sop_ids, login_user)
@router.get("/sop/showcase", summary="InspirationsopLibrary's Featured Cases", response_model=UnifiedResponseModel)
async def get_sop_banner(
page: int = Query(1, ge=1, description="Page"),
page_size: int = Query(10, ge=1, le=100, description="Items per page"),
sort: Literal["asc", "desc"] = Query("desc", description="Sort ByascORdesc"),
login_user: UserPayload = Depends(UserPayload.get_login_user)) -> UnifiedResponseModel:
"""
Set or cancel IdeasSOPLibrary's Featured Cases
:return:
"""
sop_pages = await SOPManageService.get_sop_list(showcase=True, page=page, page_size=page_size, sort=sort)
return resp_200(data=sop_pages)
@router.post("/sop/showcase", summary="Set or unset a featured case for Inspirations",
response_model=UnifiedResponseModel)
async def set_sop_banner(
sop_id: int = Body(..., description="SOPUniqueness quantificationID"),
showcase: bool = Body(..., description="Set as featured case or not"),
login_user: UserPayload = Depends(UserPayload.get_admin_user)) -> UnifiedResponseModel:
"""
Set or cancel IdeasSOPLibrary's Featured Cases
:return:
"""
# CorrectionSOPpresence or does it
existing_sop = await LinsightSOPDao.get_sops_by_ids([sop_id])
if not existing_sop:
raise NotFoundError.http_exception(msg="sop not found")
if showcase:
# Setting as featured case requires checking for run results
existing_sop = existing_sop[0]
if not existing_sop.linsight_version_id:
raise SopShowcaseError.http_exception()
execute_task_models = await LinsightWorkbenchImpl.get_execute_task_detail(existing_sop.linsight_version_id)
if not execute_task_models:
raise SopShowcaseError.http_exception()
await LinsightSOPDao.set_sop_showcase(sop_id, showcase)
return resp_200()
@router.get("/sop/showcase/result", summary="Obtain the results of the execution of the selected cases of Lingsi",
response_model=UnifiedResponseModel)
async def get_sop_showcase_result(
sop_id: int = Query(None, description="SOPUniqueness quantificationID"),
linsight_version_id: str = Query(None,
description="Inspiration Conversation VersionID, use this parameter first"),
login_user: UserPayload = Depends(UserPayload.get_login_user)) -> UnifiedResponseModel:
if not linsight_version_id:
# CorrectionSOPpresence or does it
existing_sop = await LinsightSOPDao.get_sops_by_ids([sop_id])
if not existing_sop:
raise NotFoundError.http_exception(msg="sop not found")
linsight_version_id = existing_sop[0].linsight_version_id
if not linsight_version_id:
return resp_200(data={"version_info": None, "execute_tasks": []})
version_info = await LinsightSessionVersionDao.get_by_id(linsight_version_id)
# Outstanding sessions do not return execution results
if not version_info or version_info.status != SessionVersionStatusEnum.COMPLETED:
return resp_200(data={
"version_info": None,
"execute_tasks": []
})
execute_task_models = await LinsightWorkbenchImpl.get_execute_task_detail(linsight_version_id)
return resp_200(data={
"version_info": version_info,
"execute_tasks": execute_task_models
})
class IntegratedExecuteRequestBody(BaseModel):
query: Optional[str] = Body(None, description="User Submitted Questions")
sop_content: Optional[str] = Body(None, description="User SubmittedSOPContents")
tool_ids: List[int] = Body(None, description="Selected ToolsIDVertical")
org_knowledge_enabled: bool = Body(False, description="Whether to enable organization knowledge base")
personal_knowledge_enabled: bool = Body(False, description="Whether or not to enable Personal Knowledge Base")
# Generate Inspiration OnlySOPNo
only_generate_sop: bool = Body(False, description="Whether to generate onlySOPNo")
# Lingsi Integrated Execution Interface
@router.post("/integrated-execute", summary="Lingsi Integrated Execution Interface")
async def integrated_execute(
request: Request,
body_param: str = Form(..., description="Request Body Parameters,JSONString",
example='{"query": "Please write one for mePythonfunction that calculates the sum of two numbers.", "tool_ids": [1, 2], "org_knowledge_enabled": true, "personal_knowledge_enabled": false}'),
files: List[UploadFile] = File(None, description="Uploaded files list:"),
login_user: UserPayload = Depends(UserPayload.get_login_user)
) -> EventSourceResponse:
"""
Lingsi Integrated Execution Interface
:param body_param: Request Body Parameters,JSONString
:param request: Request object
:param files: Uploaded files list:
:param login_user: Logged in user information
:return: SSEEvent Flow Response
"""
# ======================== Parameter Validation ========================
try:
body_param = IntegratedExecuteRequestBody.model_validate_json(body_param)
if not body_param.query and not body_param.sop_content:
logger.error(
f"Users {login_user.user_id} Bad request body parameters: queryAndsop_contentCannot be empty at the same time")
return EventSourceResponse(iter([{
"event": "error",
"data": json.dumps({
"error": "Bad request body parameters",
"message": "queryAndsop_contentCannot be empty at the same time",
"code": "PARAM_ERROR"
})
}]))
except ValidationError as e:
logger.error(f"Users {login_user.user_id} Request body parameter parsing failed: {str(e)}")
return EventSourceResponse(iter([{
"event": "error",
"data": json.dumps({
"error": "Request body parameter parsing failed",
"message": str(e),
"code": "PARAM_VALIDATION_ERROR"
})
}]))
except Exception as e:
logger.error(f"Users {login_user.user_id} Parameter parsing exception: {str(e)}")
return EventSourceResponse(iter([{
"event": "error",
"data": json.dumps({
"error": "Parameter parsing exception",
"message": str(e),
"code": "PARAM_PARSE_ERROR"
})
}]))
logger.info(f"Users {login_user.user_id} Submit an Idea Question: {body_param.query}")
# ======================== Upload file and parse ========================
upload_file_results = []
if files:
logger.info(f"Users {login_user.user_id} Start Upload {len(files)} files")
for idx, file in enumerate(files):
try:
# File size and type validation
if file.size and file.size > 100 * 1024 * 1024: # 100MBLimit
raise ValueError(f"Doc. {file.filename} Size exceeds limit(100MB)")
if not file.filename:
raise ValueError(f"Doc. {idx + 1} The uploaded file has no filename")
logger.debug(f"Starting file upload: {file.filename}")
upload_result = await LinsightWorkbenchImpl.upload_file(file)
if not upload_result:
raise ValueError(f"Doc. {file.filename} Upload failed, return result is empty")
# Parse files asynchronously to increase timeout control
parse_result = await asyncio.wait_for(
LinsightWorkbenchImpl.parse_file(upload_result, login_user.user_id),
timeout=300 # 5Minute Timeout
)
if not parse_result:
raise ValueError(f"Doc. {file.filename} Parsing failed, returned empty result")
upload_file_results.append(parse_result)
logger.debug(f"Doc. {file.filename} Upload parsing complete")
except asyncio.TimeoutError:
error_msg = f"Doc. {file.filename} Resolve Timeout"
logger.error(f"Users {login_user.user_id} {error_msg}")
return EventSourceResponse(iter([{
"event": "error",
"data": json.dumps({
"error": "File parsing timeout",
"message": error_msg,
"code": "FILE_PARSE_TIMEOUT"
})
}]))
except Exception as e:
error_msg = f"Doc. {getattr(file, 'filename', f'{idx + 1} No filename')} Upload parsing error: {str(e)}"
logger.error(f"Users {login_user.user_id} {error_msg}")
return EventSourceResponse(iter([{
"event": "error",
"data": json.dumps({
"error": "Upload Failed",
"message": error_msg,
"code": "FILE_UPLOAD_ERROR"
})
}]))
# Check file parsing results
if upload_file_results:
failed_files = [
f.get("original_filename", "Unknown file")
for f in upload_file_results
if f.get("parsing_status") == "failed"
]
if failed_files:
error_msg = f"The following files failed to be parsed: {', '.join(failed_files)}"
logger.error(f"Users {login_user.user_id} {error_msg}")
return EventSourceResponse(iter([{
"event": "error",
"data": json.dumps({
"error": "File parsing failed",
"message": error_msg,
"code": "FILE_PARSE_FAILED",
"failed_files": failed_files
})
}]))
async def event_generator():
"""
Event generator for generatingSSE events
"""
linsight_session_version_model = None
state_message_manager = None
try:
# ======================== Submit an Idea Question ========================
try:
submit_files = []
if upload_file_results:
for f in upload_file_results:
if not f.get("file_id"):
logger.warning(f"Doc. {f.get('original_filename', 'Unknown')} missing? file_id")
continue
submit_files.append(SubmitFileSchema(
file_id=f.get("file_id"),
file_name=f.get("original_filename"),
parsing_status=f.get("parsing_status")
))
submit_tools = None
if body_param.tool_ids:
try:
submit_tools = [LinsightToolSchema(
id="1",
is_preset=1,
children=[
ToolChildrenSchema(id=int(tool_id))
for tool_id in body_param.tool_ids
]
)]
except (ValueError, TypeError) as e:
logger.error(f"Users {login_user.user_id} ToolsIDfailed to transform: {str(e)}")
yield {
"event": "error",
"data": json.dumps({
"error": "ToolsIDFormat salah.",
"message": f"ToolsIDMust be numeric: {str(e)}",
"code": "INVALID_TOOL_ID"
})
}
return
submit_obj = LinsightQuestionSubmitSchema(
question=body_param.query if body_param.query else "User did not provide a question",
org_knowledge_enabled=body_param.org_knowledge_enabled,
personal_knowledge_enabled=body_param.personal_knowledge_enabled,
files=submit_files,
tools=submit_tools
)
_, linsight_session_version_model = await LinsightWorkbenchImpl.submit_user_question(
submit_obj, login_user
)
if not linsight_session_version_model:
raise ValueError("Failed to submit the idea question, the return result is empty")
yield {
"event": "linsight_workbench_submit",
"data": linsight_session_version_model.model_dump_json()
}
except Exception as e:
error_msg = f"Failed to submit Idea Question: {str(e)}"
logger.error(f"Users {login_user.user_id} {error_msg}")
yield {
"event": "error",
"data": json.dumps({
"error": "Failed to submit question",
"message": error_msg,
"code": "SUBMIT_QUESTION_ERROR"
})
}
return
# ======================== BuatSOP ========================
if body_param.sop_content:
# User Submitted DirectlySOPcontent, skipping generationSOPStep
await LinsightSessionVersionDao.modify_sop_content(
linsight_session_version_id=linsight_session_version_model.id,
sop_content=body_param.sop_content
)
else:
try:
knowledge_res = []
linsight_conf = settings.get_linsight_conf()
# Get the organization's knowledge base
if (linsight_session_version_model.org_knowledge_enabled and
linsight_conf and linsight_conf.max_knowledge_num > 0):
try:
org_knowledge, _ = await KnowledgeService.get_knowledge(
request, login_user, KnowledgeTypeEnum.NORMAL, None, 1,
linsight_conf.max_knowledge_num
)
if org_knowledge:
knowledge_res.extend(org_knowledge)
logger.debug(f"Get {len(org_knowledge)} organization knowledge")
except Exception as e:
logger.warning(
f"Users {login_user.user_id} Failed to get organization knowledge base: {str(e)}")
# Proceed without interrupting the process
# Get your own knowledge base
if linsight_session_version_model.personal_knowledge_enabled:
try:
personal_knowledge = await KnowledgeDao.aget_user_knowledge(
login_user.user_id, None, KnowledgeTypeEnum.PRIVATE
)
if personal_knowledge:
knowledge_res.extend(personal_knowledge)
logger.debug(f"Get {len(personal_knowledge)} personal knowledge")
except Exception as e:
logger.warning(
f"Users {login_user.user_id} Failed to get personal knowledge base: {str(e)}")
# Proceed without interrupting the process
# BuatSOP
sop_generate = LinsightWorkbenchImpl.generate_sop(
linsight_session_version_id=linsight_session_version_model.id,
previous_session_version_id=None,
feedback_content=None,
reexecute=False,
login_user=login_user,
knowledge_list=knowledge_res
)
async for event in sop_generate:
if event.get("event") == "error":
yield event
return
yield event
yield {
"event": "sop_generate_complete",
"data": json.dumps({"message": "SOPBuild Complete"})
}
except Exception as e:
error_msg = f"SOPGeneration Failed: {str(e)}"
logger.error(f"Users {login_user.user_id} {error_msg}")
yield {
"event": "error",
"data": json.dumps({
"error": "SOPGeneration Failed",
"message": error_msg,
"code": "SOP_GENERATE_ERROR"
})
}
return
if body_param.only_generate_sop:
return
# ======================== to process ========================
try:
from bisheng.linsight.worker import LinsightQueue
redis_client = await get_redis_client()
queue = LinsightQueue('queue', namespace="linsight", redis=redis_client)
await queue.put(data=linsight_session_version_model.id)
yield {
"event": "linsight_execute_submitted",
"data": json.dumps({"message": "Idea Execution Task Submitted"})
}
except Exception as e:
error_msg = f"Failed to submit execution task: {str(e)}"
logger.error(f"Users {login_user.user_id} {error_msg}")
yield {
"event": "error",
"data": json.dumps({
"error": "Failed to submit execution task",
"message": error_msg,
"code": "SUBMIT_TASK_ERROR"
})
}
return
# ======================== Consumer Message Flow ========================
try:
state_message_manager = LinsightStateMessageManager(
session_version_id=linsight_session_version_model.id
)
final_result_message = None
message_count = 0
max_messages = 10000 # Prevents infinite loops
max_wait_time = 60 * 60 # Maximum Wait1Jam
start_time = time.time()
while message_count < max_messages:
# Check timeout
if time.time() - start_time > max_wait_time:
logger.warning(f"Users {login_user.user_id} Message consumption timed out")
yield {
"event": "warning",
"data": json.dumps({
"message": "The execution time is longer and may take more time to complete",
"code": "EXECUTION_TIMEOUT_WARNING"
})
}
break
try:
message = await asyncio.wait_for(
state_message_manager.pop_message(),
timeout=10.0 # 10seconds timeout
)
except asyncio.TimeoutError:
linsight_session_version_model = await state_message_manager.get_session_version_info()
if linsight_session_version_model.status in [
SessionVersionStatusEnum.COMPLETED,
SessionVersionStatusEnum.TERMINATED,
SessionVersionStatusEnum.FAILED
]:
message = MessageData(
event_type=MessageEventType.FINAL_RESULT if linsight_session_version_model.status == SessionVersionStatusEnum.COMPLETED else MessageEventType.TASK_TERMINATED,
data=linsight_session_version_model.model_dump()
)
if message.event_type == MessageEventType.FINAL_RESULT:
final_result_message = message
yield {
"event": "linsight_execute_message",
"data": message.model_dump_json()
}
logger.info(f"Users {login_user.user_id} Idea execution has ended, stop getting messages")
break
# Timeout to continue waiting
continue
except Exception as e:
logger.error(f"Users {login_user.user_id} Failed to fetch messages: {str(e)}")
break
if message:
message_count += 1
yield {
"event": "linsight_execute_message",
"data": message.model_dump_json()
}
# Save final result message
if message.event_type == MessageEventType.FINAL_RESULT:
final_result_message = message
# Check termination conditions
if message.event_type in [
MessageEventType.ERROR_MESSAGE,
MessageEventType.TASK_TERMINATED,
MessageEventType.FINAL_RESULT
]:
break
else:
await asyncio.sleep(1)
# Process final result message
final_files = []
all_from_session_files = []
if final_result_message and final_result_message.data:
try:
session_version_model = LinsightSessionVersion.model_validate(final_result_message.data)
if session_version_model.output_result:
final_files = session_version_model.output_result.get("final_files", [])
all_from_session_files = session_version_model.output_result.get("all_from_session_files",
[])
minio_client = await get_minio_storage()
# Generate file sharing link
for final_file in final_files:
if final_file.get("url"):
try:
final_file["url"] = await minio_client.get_share_link(final_file["url"],
clear_host=False)
except Exception as e:
logger.warning(f"Failed to generate final file share link: {str(e)}")
for session_file in all_from_session_files:
if session_file.get("url"):
try:
session_file["url"] = await minio_client.get_share_link(session_file["url"],
clear_host=False)
except Exception as e:
logger.warning(f"Failed to generate session file share link: {str(e)}")
except Exception as e:
logger.error(f"Users {login_user.user_id} Failed to process final result: {str(e)}")
yield {
"event": "final_result_files",
"data": json.dumps({
"final_files": final_files,
"all_from_session_files": all_from_session_files
})
}
except Exception as e:
error_msg = f"Message consumption failed: {str(e)}"
logger.error(f"Users {login_user.user_id} {error_msg}")
yield {
"event": "error",
"data": json.dumps({
"error": "Message consumption failed",
"message": error_msg,
"code": "MESSAGE_CONSUME_ERROR"
})
}
return
except Exception as e:
# Catch all unhandled exceptions
error_msg = f"Interface Execution Exception: {str(e)}"
logger.error(f"Users {login_user.user_id} {error_msg}", exc_info=True)
yield {
"event": "error",
"data": json.dumps({
"error": "System Exception",
"message": error_msg,
"code": "SYSTEM_ERROR"
})
}
finally:
pass
return EventSourceResponse(event_generator())
================================================
FILE: src/backend/bisheng/linsight/api/router.py
================================================
from fastapi import APIRouter
from .endpoints.linsight import router as linsight_router
router = APIRouter(prefix="/linsight", tags=["Inspiration"])
router.include_router(linsight_router)
__all__ = ["router"]
================================================
FILE: src/backend/bisheng/linsight/domain/__init__.py
================================================
================================================
FILE: src/backend/bisheng/linsight/domain/models/__init__.py
================================================
================================================
FILE: src/backend/bisheng/linsight/domain/models/linsight_execute_task.py
================================================
from datetime import datetime
from enum import Enum
from typing import Optional, Dict, List
from sqlalchemy import Enum as SQLEnum, Column, JSON, DateTime, text, CHAR, ForeignKey, update
from sqlmodel import Field, select, col
from bisheng.core.database import get_async_db_session
from bisheng.database.base import uuid_hex
from bisheng.common.models.base import SQLModelSerializable
class ExecuteTaskTypeEnum(str, Enum):
"""
Idea execution task type enumeration
"""
# Single Task
SINGLE = "single"
# Has subtasks
COMPOSITE = "composite"
class ExecuteTaskStatusEnum(str, Enum):
"""
Idea Execution Task Status Enumeration
"""
# Not Started
NOT_STARTED = "not_started"
# Sedang berlangsung
IN_PROGRESS = "in_progress"
# Berhasil
SUCCESS = "success"
# Waiting for user input
WAITING_FOR_USER_INPUT = "waiting_for_user_input"
# User input complete
USER_INPUT_COMPLETED = "user_input_completed"
# Kalah
FAILED = "failed"
# TERMINATION
TERMINATED = "terminated"
class LinsightExecuteTaskBase(SQLModelSerializable):
"""
Idea Execution Task Model Base Class
"""
session_version_id: str = Field(..., description='Session VersionID',
sa_column=Column(CHAR(36), ForeignKey("linsight_session_version.id"),
nullable=False))
parent_task_id: Optional[str] = Field(None, description='Parent Task:ID',
sa_column=Column(CHAR(36), ForeignKey("linsight_execute_task.id"),
nullable=True))
previous_task_id: Optional[str] = Field(None, description='Previous TaskID',
sa_column=Column(CHAR(36),
nullable=True))
next_task_id: Optional[str] = Field(None, description='[patterns/patterns_ParallelJoin.xml?ROU_NEXT_TASK] Next TaskID',
sa_column=Column(CHAR(36),
nullable=True))
task_type: ExecuteTaskTypeEnum = Field(..., description='Task type',
sa_column=Column(SQLEnum(ExecuteTaskTypeEnum), nullable=False))
task_data: Optional[dict] = Field(None, description='Task Data', sa_type=JSON, nullable=True)
# input_prompt: Optional[str] = Field(None, description='Enter a prompt', sa_type=Text, nullable=True)
# user_input: Optional[str] = Field(None, description='User input', sa_type=Text, nullable=True)
history: Optional[List[Dict]] = Field(None, description='Execute Step Record', sa_type=JSON, nullable=True)
status: ExecuteTaskStatusEnum = Field(ExecuteTaskStatusEnum.NOT_STARTED, description="Status Misi",
sa_column=Column(SQLEnum(ExecuteTaskStatusEnum), nullable=False))
result: Optional[Dict] = Field(None, description='Result of Task', sa_type=JSON, nullable=True)
class LinsightExecuteTask(LinsightExecuteTaskBase, table=True):
"""
Ideas Execution Task Model, sopThe library will also reference the data here
"""
id: str = Field(default_factory=uuid_hex, description='TaskID',
sa_column=Column(CHAR(36), unique=True, nullable=False, primary_key=True))
create_time: datetime = Field(default_factory=datetime.now, description='Creation Time',
sa_column=Column(DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=True, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
__tablename__ = "linsight_execute_task"
class LinsightExecuteTaskDao(object):
"""
Ideas Execution Task Data Access Objects
"""
@classmethod
async def get_by_id(cls, task_id: str) -> Optional[LinsightExecuteTask]:
"""
By TaskIDGet Tasks
:param task_id: TaskID
:return: Task Objects
"""
async with get_async_db_session() as session:
statement = select(LinsightExecuteTask).where(LinsightExecuteTask.id == str(task_id))
task = await session.exec(statement)
return task.first()
@classmethod
async def get_by_session_version_id(cls, session_version_id: str, is_parent_task: bool = False) -> List[
LinsightExecuteTask]:
"""
Based on session versionIDGet all tasks
:param is_parent_task:
:param session_version_id: Session VersionID
:return: Task list
"""
async with get_async_db_session() as session:
statement = select(LinsightExecuteTask).where(
LinsightExecuteTask.session_version_id == str(session_version_id))
if is_parent_task:
statement = statement.where(col(LinsightExecuteTask.parent_task_id).is_(None))
tasks = await session.exec(statement)
return tasks.all()
@classmethod
async def batch_create_tasks(cls, tasks: List[LinsightExecuteTask]) -> List[LinsightExecuteTask]:
"""
Batch Create Tasks
:param tasks: Task list
:return: Post-Created Task List
"""
async with get_async_db_session() as session:
session.add_all(tasks)
await session.commit()
return tasks
@classmethod
async def update_by_id(cls, task_id: str, **kwargs) -> Optional[LinsightExecuteTask]:
"""
By TaskIDUpdate Details
:param task_id: TaskID
:param kwargs: Update fields
:return: Updated task object
"""
async with get_async_db_session() as session:
statement = select(LinsightExecuteTask).where(LinsightExecuteTask.id == task_id)
task = await session.exec(statement)
task = task.first()
if not task:
return None
for key, value in kwargs.items():
setattr(task, key, value)
session.add(task)
await session.commit()
await session.refresh(task)
return task
# accordingsession_version_idBulk update task status
@classmethod
async def batch_update_status_by_session_version_id(cls, session_version_ids: List[str],
status: ExecuteTaskStatusEnum,
where) -> None:
"""
Based on session versionIDBulk update task status
:param session_version_ids:
:param status:
:param where:
:return:
"""
async with get_async_db_session() as session:
statement = (
update(LinsightExecuteTask)
.where(col(LinsightExecuteTask.session_version_id).in_(session_version_ids)) # Explicit Transfer str
)
if where:
statement = statement.where(*where)
statement = statement.values(status=status)
await session.exec(statement)
await session.commit()
================================================
FILE: src/backend/bisheng/linsight/domain/models/linsight_session_version.py
================================================
import logging
from datetime import datetime
from enum import Enum
from typing import List, Dict, Optional
from sqlalchemy import Column, Text, JSON, Boolean, Enum as SQLEnum, DateTime, text, ForeignKey, CHAR, func
from sqlmodel import Field, select, col, update
from bisheng.core.database import get_async_db_session
from bisheng.database.base import uuid_hex
from bisheng.common.models.base import SQLModelSerializable
logger = logging.getLogger(__name__)
class SessionVersionStatusEnum(str, Enum):
"""
Ideas Session Version Status Enumeration
"""
# not implemented
NOT_STARTED = "not_started"
# Sedang berlangsung
IN_PROGRESS = "in_progress"
# Run Completed
COMPLETED = "completed"
# has failed to run...
FAILED = "failed"
# SOP Generation Failed
SOP_GENERATION_FAILED = "sop_generation_failed"
# TERMINATION
TERMINATED = "terminated"
class LinsightSessionVersionBase(SQLModelSerializable):
"""
Inspiration Conversation Version Model Base Class
"""
session_id: str = Field(..., description='SessionsID', sa_column=Column(CHAR(36),
ForeignKey("message_session.chat_id"),
nullable=False,
index=True))
user_id: int = Field(..., description='UsersID', foreign_key="user.user_id", nullable=False)
question: str = Field(..., description='User Questions', sa_type=Text, nullable=False)
title: Optional[str] = Field(None, description='Session title', sa_type=Text, nullable=True)
tools: Optional[List[Dict]] = Field(None, description='List of available tools', sa_type=JSON, nullable=True)
# Personal Knowledge Base
personal_knowledge_enabled: bool = Field(False, description='Whether or not to enable Personal Knowledge Base', sa_type=Boolean)
# Organization Knowledge Base
org_knowledge_enabled: bool = Field(False, description='Whether to enable organization knowledge base', sa_type=Boolean)
files: Optional[List[Dict]] = Field(None, description='Uploaded files list:', sa_type=JSON, nullable=True)
sop: Optional[str] = Field(None, description='SOPContents', sa_type=Text, nullable=True)
output_result: Optional[Dict] = Field(None, description='Output Results', sa_type=JSON, nullable=True)
status: SessionVersionStatusEnum = Field(default=SessionVersionStatusEnum.NOT_STARTED, description='Session Version Status',
sa_column=Column(SQLEnum(SessionVersionStatusEnum), nullable=False))
score: Optional[int] = Field(None, description='Session Score', ge=1, le=5, nullable=True)
# Execution Result Feedback Information
execute_feedback: Optional[str] = Field(None, description='Execution Result Feedback Information', sa_type=Text, nullable=True)
# Is there a re-execution
has_reexecute: bool = Field(default=False, description='Is there a re-execution', sa_type=Boolean, nullable=False)
# Version
version: datetime = Field(default_factory=datetime.now, description='Session Version Created Time', sa_type=DateTime)
class LinsightSessionVersion(LinsightSessionVersionBase, table=True):
"""
Inspiration Conversation Version Model
"""
id: str = Field(default_factory=uuid_hex, description='Session VersionID',
sa_column=Column(CHAR(36), unique=True, nullable=False, primary_key=True))
create_time: datetime = Field(default_factory=datetime.now, description='Creation Time',
sa_column=Column(DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=True, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
__tablename__ = "linsight_session_version"
class LinsightSessionVersionDao(object):
"""
Inspiration Session Version Data Access Objects
"""
@staticmethod
async def insert_one(session_version: LinsightSessionVersion) -> LinsightSessionVersion:
"""
Insert an Invisible Sessions version record
:param session_version: Inspiration Conversation Version Object
:return: Invisible Conversation Version Object Created
"""
async with get_async_db_session() as session:
session.add(session_version)
await session.commit()
await session.refresh(session_version)
return session_version
@staticmethod
async def get_by_id(linsight_session_version_id: str) -> Optional[LinsightSessionVersion]:
"""
According to the Inspiration Conversation versionIDGet Ideas Conversation Version
:param linsight_session_version_id: Inspiration Conversation VersionID
:return: Inspiration Conversation Version Object
"""
async with get_async_db_session() as session:
statement = select(LinsightSessionVersion).where(
LinsightSessionVersion.id == str(linsight_session_version_id)) # Explicit Transfer str
result = await session.exec(statement)
return result.first()
@staticmethod
async def get_session_versions_by_session_id(session_id: str) -> List[LinsightSessionVersion]:
"""
By ConversationIDGet all Ideas Conversation versions
:param session_id: SessionsID
:return: Inspiration Session Version List
"""
async with get_async_db_session() as session:
statement = select(LinsightSessionVersion).where(
LinsightSessionVersion.session_id == str(session_id)).order_by(
col(LinsightSessionVersion.version).desc())
return (await session.exec(statement)).all()
@staticmethod
async def modify_sop_content(linsight_session_version_id: str, sop_content: str):
"""
Modify Inspiration Conversation Version ofSOPContents
:param linsight_session_version_id:
:param sop_content:
:return:
"""
async with get_async_db_session() as session:
stmt = (
update(LinsightSessionVersion)
.where(col(LinsightSessionVersion.id) == str(linsight_session_version_id)) # Explicit Transfer str
.values(sop=sop_content)
)
result = await session.exec(stmt)
if result.rowcount == 0:
logger.warning(f"No session version found with ID: {linsight_session_version_id}")
await session.commit()
@staticmethod
async def get_session_version_by_file_id(file_id: str) -> Optional[LinsightSessionVersion]:
"""
According to DOCUMENTIDGet Ideas Conversation Version
:param file_id: Doc.ID
:return: Inspiration Conversation Version Object
"""
async with get_async_db_session() as session:
statement = select(LinsightSessionVersion).where(
func.json_search(LinsightSessionVersion.files, 'all', file_id)
)
result = await session.exec(statement)
return result.first()
# Get a list of Ideas session versions based on task status
@staticmethod
async def get_session_versions_by_status(status: SessionVersionStatusEnum) -> List[LinsightSessionVersion]:
"""
Get a list of Ideas session versions based on task status
:param status: Session Version Status
:return: Inspiration Session Version List
"""
async with get_async_db_session() as session:
statement = select(LinsightSessionVersion).where(
LinsightSessionVersion.status == status
)
result = await session.exec(statement)
return result.all()
# Bulk Update Ideas Session Version Status
@staticmethod
async def batch_update_session_versions_status(session_version_ids: List[str], status: SessionVersionStatusEnum,
**kwargs) -> None:
"""
Bulk Update Ideas Session Version Status
:param session_version_ids: Session VersionIDVertical
:param status: New Session Version Status
"""
async with get_async_db_session() as session:
stmt = (
update(LinsightSessionVersion)
.where(col(LinsightSessionVersion.id).in_(session_version_ids))
.values(status=status, **kwargs) # Support for additional field updates
)
await session.exec(stmt)
await session.commit()
================================================
FILE: src/backend/bisheng/linsight/domain/models/linsight_sop.py
================================================
from datetime import datetime
from typing import Optional, Dict, Any, List, Literal
from loguru import logger
from sqlalchemy import update
from sqlalchemy.dialects.mysql import LONGTEXT
from sqlmodel import Field, select, delete, col, or_, func, Column, Text, DateTime, text, CHAR
from bisheng.linsight.domain.schemas.inspiration_schema import SOPManagementUpdateSchema
from bisheng.core.database import get_async_db_session
from bisheng.database.base import async_get_count
from bisheng.common.models.base import SQLModelSerializable
class LinsightSOPBase(SQLModelSerializable):
"""
Inspiration SOPModel Base Class
"""
name: str = Field(..., description='SOPPart Name', sa_column=Column(Text, nullable=False))
description: Optional[str] = Field(default=None, description='SOPDescription', sa_column=Column(Text))
user_id: int = Field(..., description='UsersID', foreign_key="user.user_id", nullable=False)
content: str = Field(..., description='SOPContents',
sa_column=Column(LONGTEXT, nullable=False, comment="SOPContents"))
rating: Optional[int] = Field(default=0, ge=0, le=5, description='SOPScore, Range0-5')
showcase: Optional[bool] = Field(default=False, index=True, description='Whether to display it on the homepage as a featured case')
vector_store_id: Optional[str] = Field(..., description='Vector StorageID',
sa_column=Column(CHAR(36), nullable=False, comment="Vector StorageID"))
linsight_version_id: Optional[str] = Field(default=None,
description='Inspiration Conversation VersionID, used to query the running results of a featured case',
sa_column=Column(CHAR(36), nullable=True))
create_time: datetime = Field(default_factory=datetime.now, description='Creation Time',
sa_column=Column(DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=True, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class LinsightSOP(LinsightSOPBase, table=True):
"""
Inspiration SOPModels
"""
__tablename__ = "linsight_sop"
id: Optional[int] = Field(default=None, primary_key=True, description='SOPUniqueness quantificationID')
class LinsightSOPRecord(SQLModelSerializable, table=True):
"""
InspirationSOPRun the record sheet to record the results generated during the execution of the Ideasop
"""
__tablename__ = "linsight_sop_record"
id: Optional[int] = Field(default=None, primary_key=True, description='SOPRecord UniqueID')
name: str = Field(..., description='SOPPart Name', sa_column=Column(Text, nullable=False))
description: Optional[str] = Field(default=None, description='SOPDescription', sa_column=Column(Text))
user_id: int = Field(..., description='UsersID', foreign_key="user.user_id", nullable=False)
content: str = Field(..., description='SOPContents',
sa_column=Column(LONGTEXT, nullable=False, comment="SOPContents"))
rating: Optional[int] = Field(default=0, ge=0, le=5, description='SOPScore, Range0-5')
execute_feedback: Optional[str] = Field(None, description='Execution Result Feedback Information', sa_type=Text, nullable=True)
linsight_version_id: Optional[str] = Field(default=None, description='Inspiration Conversation Versionid, sync ratings')
create_time: datetime = Field(default_factory=datetime.now, description='Creation Time',
sa_column=Column(DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=True, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class LinsightSOPDao(LinsightSOPBase):
"""
Inspiration SOPData Access Objects
"""
@classmethod
async def create_sop(cls, sop: LinsightSOP) -> LinsightSOP:
async with get_async_db_session() as session:
session.add(sop)
await session.commit()
await session.refresh(sop)
return sop
@classmethod
async def update_sop(cls, sop_obj: SOPManagementUpdateSchema) -> LinsightSOP:
async with get_async_db_session() as session:
# UseUpdateStatement UpdatesSOP
statement = select(LinsightSOP).where(LinsightSOP.id == sop_obj.id)
result = await session.exec(statement)
sop = result.first()
if not sop:
raise ValueError("SOP not found")
# will besop_objField values for were updated tosopin the example
for key, value in sop_obj.model_dump().items():
if hasattr(sop, key) and value is not None:
setattr(sop, key, value)
sop.update_time = datetime.now() # Update modification time
session.add(sop)
await session.commit()
await session.refresh(sop)
return sop
@classmethod
async def get_sop_page(cls, keywords: Optional[str] = None, showcase: bool = None,
sort: Literal["asc", "desc"] = "desc", page: int = 1,
page_size: int = 10) -> Dict[str, Any]:
"""
DapatkanSOPPagination List
"""
statement = select(LinsightSOP)
if keywords:
statement = statement.where(
LinsightSOP.name.ilike(f'%{keywords}%') |
LinsightSOP.description.ilike(f'%{keywords}%') |
LinsightSOP.content.ilike(f'%{keywords}%')
)
# according rating And create_time Sort
if sort == "asc":
statement = statement.order_by(col(LinsightSOP.rating).asc(), col(LinsightSOP.update_time).asc())
else:
statement = statement.order_by(col(LinsightSOP.rating).desc(), col(LinsightSOP.update_time).desc())
if showcase is not None:
statement = statement.where(LinsightSOP.showcase == showcase)
async with get_async_db_session() as session:
total_count = await async_get_count(session, statement)
statement = statement.offset((page - 1) * page_size).limit(page_size)
result = (await session.exec(statement)).all()
return {
"total": total_count,
"current_page": page,
"page_size": page_size,
"items": [result.model_dump() for result in result]
}
@classmethod
async def get_sops_by_ids(cls, sop_ids: List[int]) -> List[LinsightSOP]:
"""
accordingSOP IDGet listSOPObjects
"""
async with get_async_db_session() as session:
statement = select(LinsightSOP).where(col(LinsightSOP.id).in_(sop_ids))
result = await session.exec(statement)
sop_list = result.all()
return sop_list
@classmethod
async def get_sops_by_names(cls, names: list[str]) -> List[LinsightSOP]:
"""
accordingSOPGet Name ListSOPObjects
"""
statement = select(LinsightSOP).where(col(LinsightSOP.name).in_(names))
async with get_async_db_session() as session:
result = await session.exec(statement)
sop_list = result.all()
return sop_list
@classmethod
async def remove_sop(cls, sop_ids: List[int]) -> bool:
"""
DeleteSOP
"""
async with get_async_db_session() as session:
delete_statement = delete(LinsightSOP).where(col(LinsightSOP.id).in_(sop_ids))
result = await session.exec(delete_statement)
await session.commit()
logger.info(f"Deleted {result.rowcount} SOP(s) with IDs: {sop_ids}")
return True
@classmethod
async def get_sop_by_session_id(cls, session_id: str) -> Optional[LinsightSOP]:
"""
According to the Inspiration ConversationID Dapatkan OP
"""
async with get_async_db_session() as session:
statement = select(LinsightSOP).where(LinsightSOP.linsight_session_id == session_id)
result = await session.exec(statement)
sop = result.first()
return sop if sop else None
@classmethod
async def get_sop_by_vector_store_ids(cls, vector_store_ids: List[str]) -> List[LinsightSOP]:
"""
According to vector storageIDGet listSOPObjects
"""
async with get_async_db_session() as session:
statement = select(LinsightSOP).where(col(LinsightSOP.vector_store_id).in_(vector_store_ids))
result = await session.exec(statement)
sop_list = result.all()
return sop_list
@classmethod
async def get_all_sops(cls) -> List[LinsightSOP]:
"""
Fetch allSOP
"""
async with get_async_db_session() as session:
statement = select(LinsightSOP)
result = await session.exec(statement)
sop_list = result.all()
return sop_list
@classmethod
async def create_sop_record(cls, sop_record: LinsightSOPRecord) -> LinsightSOPRecord:
"""
Insert an entrySOPRecord
"""
async with get_async_db_session() as session:
session.add(sop_record)
await session.commit()
await session.refresh(sop_record)
return sop_record
@classmethod
async def _filter_sop_record_statement(cls, statement, keywords: str = None, user_ids: list[int] = None) -> select:
"""
Build.SOPRecorded Query Statements
"""
or_params = []
if keywords:
or_params.extend([
LinsightSOPRecord.name.like(f'%{keywords}%'),
LinsightSOPRecord.description.like(f'%{keywords}%'),
LinsightSOPRecord.content.like(f'%{keywords}%')
])
if user_ids:
or_params.append(LinsightSOPRecord.user_id.in_(user_ids))
if or_params:
statement = statement.where(or_(*or_params))
return statement
@classmethod
async def filter_sop_record(cls, keywords: str = None, user_ids: list[int] = None, page: int = None,
page_size: int = None, sort: str = None) -> List[LinsightSOPRecord]:
"""
Fetch allSOPRecord, Key Matchname、description、content。user_idsIs UserIDList. Between filters areorRelationship
"""
statement = select(LinsightSOPRecord)
statement = await cls._filter_sop_record_statement(statement, keywords, user_ids)
if page and page_size:
statement = statement.offset((page - 1) * page_size).limit(page_size)
if sort == "asc":
statement = statement.order_by(col(LinsightSOPRecord.create_time).asc())
else:
statement = statement.order_by(col(LinsightSOPRecord.create_time).desc())
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
async def count_sop_record(cls, keywords: str = None, user_ids: list[int] = None) -> int:
"""
StatisticsSOPNumber of Records
"""
statement = select(func.count(LinsightSOPRecord.id))
statement = await cls._filter_sop_record_statement(statement, keywords, user_ids)
async with get_async_db_session() as session:
return await session.scalar(statement)
@classmethod
async def get_sop_record_by_ids(cls, ids: list[int]) -> List[LinsightSOPRecord]:
"""
accordingSOPRecordIDGet listSOPRecord Objects
"""
statement = select(LinsightSOPRecord).where(col(LinsightSOPRecord.id).in_(ids))
async with get_async_db_session() as session:
result = await session.exec(statement)
sop_record_list = result.all()
return sop_record_list
@classmethod
async def update_sop_record_score(cls, linsight_version_id: str, rating: int) -> bool:
"""
UpdateSOPRecord Score
"""
statement = update(LinsightSOPRecord).where(
col(LinsightSOPRecord.linsight_version_id) == linsight_version_id).values(rating=rating)
async with get_async_db_session() as session:
await session.exec(statement)
await session.commit()
return True
@classmethod
async def update_sop_record_feedback(cls, linsight_version_id: str, execute_feedback: str) -> bool:
"""
UpdateSOPRecord Execution Result Feedback Information
"""
statement = update(LinsightSOPRecord).where(
col(LinsightSOPRecord.linsight_version_id) == linsight_version_id).values(execute_feedback=execute_feedback)
async with get_async_db_session() as session:
await session.exec(statement)
await session.commit()
return True
@classmethod
async def set_sop_showcase(cls, sop_id: int, showcase: bool) -> bool:
"""
PengaturanSOPWhether to display it on the homepage as a featured case
"""
statement = update(LinsightSOP).where(
col(LinsightSOP.id) == sop_id).values(showcase=showcase)
async with get_async_db_session() as session:
await session.exec(statement)
await session.commit()
return True
================================================
FILE: src/backend/bisheng/linsight/domain/schemas/__init__.py
================================================
================================================
FILE: src/backend/bisheng/linsight/domain/schemas/inspiration_schema.py
================================================
from typing import Optional
from pydantic import Field, BaseModel, field_validator
# SOPManaging Schema
class SOPManagementSchema(BaseModel):
"""SOPManaging Schema"""
name: str = Field(..., description="SOPPart Name")
description: str = Field(None, description="SOPDescription")
content: str = Field(..., description="SOPContents")
rating: int = Field(0, ge=0, le=5, description="SOPScore, Range0-5")
linsight_version_id: Optional[str] = Field(default=None, description="LinsightSession Version ofID")
@field_validator("name", mode="before")
def validate_name(cls, v):
# LimitSOPName length does not exceed500characters
return v[:500]
class SOPManagementUpdateSchema(SOPManagementSchema):
"""SOPManage updates Schema"""
id: int = Field(..., description="SOPUniqueness quantificationID")
user_id: Optional[int] = Field(None, description="UsersID")
showcase: Optional[bool] = Field(None, description="Whether to display")
================================================
FILE: src/backend/bisheng/linsight/domain/schemas/linsight_schema.py
================================================
from typing import List, Dict, Optional
from pydantic import BaseModel, Field, field_validator
from bisheng.linsight.domain.models.linsight_sop import LinsightSOPRecord
from bisheng_langchain.linsight.event import NeedUserInput
class ToolChildrenSchema(BaseModel):
id: int = Field(..., description="Toolsid")
name: Optional[str] = Field(None, description="Tool name")
tool_key: Optional[str] = Field(None, description="Toolskey")
desc: Optional[str] = Field(None, description="Tools Description")
# Opt-IntoolSchema
class LinsightToolSchema(BaseModel):
id: int = Field(..., description="Tool LevelID")
name: Optional[str] = Field(None, description="Tool name")
is_preset: int = Field(1, description="Whether or not it is a preset tool")
desc: Optional[str] = Field(None, description="Tools Description")
# childTools List
children: Optional[List[ToolChildrenSchema]] = Field(..., description="Subtools List")
class SubmitFileSchema(BaseModel):
file_id: str = Field(..., description="File UniqueID")
file_name: str = Field(..., description="File Name")
parsing_status: str = Field(..., description="File parsing status")
# Submit a problemSchema
class LinsightQuestionSubmitSchema(BaseModel):
question: str = Field(..., description="User Submitted Questions")
org_knowledge_enabled: bool = Field(False, description="Whether to enable organization knowledge base")
personal_knowledge_enabled: bool = Field(False, description="Whether or not to enable Personal Knowledge Base")
files: Optional[List[SubmitFileSchema]] = Field(None, description="Uploaded files list:")
tools: Optional[List[LinsightToolSchema]] = Field(None, description="List of available tools")
@field_validator("tools")
@classmethod
def validate_tools(cls, v: List[LinsightToolSchema]) -> List[Dict]:
if not v:
return []
# Convert tool to dictionary format
return [tool.model_dump() for tool in v]
class DownloadFilesSchema(BaseModel):
file_name: str = Field(..., description="File Name")
file_url: str = Field(..., description="File download link")
class SopRecordRead(LinsightSOPRecord, table=False):
user_name: Optional[str] = Field(default=None, description="Client Name")
class UserInputEventSchema(NeedUserInput):
files: Optional[List[Dict[str, str]]] = Field(None, description="Uploaded files list:")
user_input: Optional[str] = Field(None, description="User input")
is_completed: bool = Field(False, description="Is it completed")
================================================
FILE: src/backend/bisheng/linsight/domain/services/__init__.py
================================================
================================================
FILE: src/backend/bisheng/linsight/domain/services/message_stream_handle.py
================================================
from starlette.websockets import WebSocket
from bisheng.linsight.domain.services.state_message_manager import LinsightStateMessageManager, MessageData, MessageEventType
class MessageStreamHandle(object):
def __init__(self, websocket: 'WebSocket', session_version_id: str):
"""
Inisialisasi MessageStreamHandle
:param websocket:
"""
self._websocket = websocket
self.session_version_id = session_version_id
self._state_message_manager: LinsightStateMessageManager = LinsightStateMessageManager(
session_version_id=session_version_id)
async def send_message(self, message_data: str) -> None:
"""
Send Message To WebSocket
:param message_data: Message to send
"""
await self._websocket.send_text(message_data)
async def receive_message(self) -> str:
"""
Received from WebSocket Message
:return:
"""
return await self._websocket.receive_text()
async def send_json(self, json_data: dict) -> None:
"""
Send JSON Data to WebSocket
:param json_data: To be sent JSON DATA
"""
await self._websocket.send_json(json_data)
async def receive_json(self) -> dict:
"""
Received from WebSocket right of privacy JSON DATA
:return:
"""
return await self._websocket.receive_json()
# Medical Treatment: WebSocket Connected Lifecycle Events
async def connect(self) -> None:
"""
Connecting to devices WebSocket
"""
await self._websocket.accept()
while True:
try:
message = await self._state_message_manager.pop_message()
if message:
await self.send_json(message.model_dump())
if message.event_type in [MessageEventType.ERROR_MESSAGE, MessageEventType.TASK_TERMINATED,
MessageEventType.FINAL_RESULT]:
await self._websocket.close(code=1000, reason="Session finished or error occurred")
break
except Exception as e:
await self.send_json(
MessageData(event_type=MessageEventType.ERROR_MESSAGE, data={"error": str(e)}).model_dump())
await self._websocket.close(code=1000, reason=f"Error: {str(e)}")
break
async def disconnect(self) -> None:
"""
detach WebSocket CONNECT
"""
await self._websocket.close(code=1000, reason="Client disconnected")
================================================
FILE: src/backend/bisheng/linsight/domain/services/sop_manage.py
================================================
import io
import json
import uuid
from typing import List, Dict, Literal
import openpyxl
from fastapi import UploadFile
from langchain_core.documents import Document
from langchain_core.embeddings import Embeddings
from langchain_core.language_models import BaseChatModel
from langchain_text_splitters import RecursiveCharacterTextSplitter
from loguru import logger
from bisheng.api.services.knowledge_imp import decide_vectorstores, extract_code_blocks
from bisheng.linsight.domain.schemas.inspiration_schema import SOPManagementSchema, SOPManagementUpdateSchema
from bisheng.linsight.domain.schemas.linsight_schema import SopRecordRead
from bisheng.api.v1.schemas import UnifiedResponseModel, resp_200
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode import BaseErrorCode
from bisheng.common.errcode.http_error import NotFoundError
from bisheng.common.errcode.linsight import (
LinsightAddSopError, LinsightUpdateSopError, LinsightDeleteSopError,
LinsightVectorModelError, LinsightDocSearchError, LinsightDocNotFoundError, SopContentOverLimitError
)
from bisheng.common.errcode.server import (
NoEmbeddingModelError, EmbeddingModelNotExistError, EmbeddingModelTypeError, UploadFileEmptyError
)
from bisheng.common.services.config_service import settings
from bisheng.core.prompts.manager import get_prompt_manager
from bisheng.linsight.domain.models.linsight_sop import LinsightSOP, LinsightSOPDao, LinsightSOPRecord
from bisheng.interface.embeddings.custom import FakeEmbedding
from bisheng.llm.domain.const import LLMModelType
from bisheng.llm.domain.models import LLMDao
from bisheng.llm.domain.services import LLMService
from bisheng.user.domain.models.user import UserDao
from bisheng.utils import util
from bisheng_langchain.rag.init_retrievers import KeywordRetriever, BaselineVectorRetriever
from bisheng_langchain.retrievers import EnsembleRetriever
from bisheng_langchain.vectorstores import ElasticKeywordsSearch, Milvus
class SOPManageService:
__doc__ = "InspirationSOPManage Services"
collection_name = "col_linsight_sop"
@staticmethod
async def generate_sop_summary(invoke_user_id: int, sop_content: str, llm: BaseChatModel = None) -> Dict[str, str]:
"""BuatSOPAbstract"""
default_summary = {"sop_title": "SOP Title", "sop_description": "SOP Description"}
try:
if llm is None:
workbench_conf = await LLMService.get_workbench_llm()
linsight_conf = settings.get_linsight_conf()
llm = await LLMService.get_bisheng_linsight_llm(invoke_user_id=invoke_user_id,
model_id=workbench_conf.task_model.id,
temperature=linsight_conf.default_temperature)
prompt_service = await get_prompt_manager()
prompt_obj = prompt_service.render_prompt(
namespace="sop",
prompt_name="gen_sop_summary",
sop_detail=sop_content
)
prompt = [
("system", prompt_obj.prompt.system),
("user", prompt_obj.prompt.user)
]
response = await llm.ainvoke(prompt)
if not response.content:
return default_summary
code_ret = extract_code_blocks(response.content)
if code_ret:
return json.loads(code_ret[0])
return json.loads(response.content)
except Exception as e:
logger.exception(f"BuatSOPSummary failed: {e}")
return default_summary
@staticmethod
async def add_sop_record(sop_record: LinsightSOPRecord) -> LinsightSOPRecord:
"""
TambahSOPRecord
"""
if not sop_record.description:
sop_summary = await SOPManageService.generate_sop_summary(sop_record.user_id, sop_record.content, None)
sop_record.description = sop_summary["sop_description"]
return await LinsightSOPDao.create_sop_record(sop_record)
@staticmethod
async def get_sop_record(keyword: str = None, sort: str = None, page: int = 1, page_size: int = 10) -> \
(List[SopRecordRead], int):
"""
Query by keywordSOPRecord
"""
user_ids = []
if keyword:
# If there are keywords, get the user firstIDVertical
user_ids = await UserDao.afilter_users(user_ids=[], keyword=keyword)
user_ids = [one.user_id for one in user_ids]
res = await LinsightSOPDao.filter_sop_record(keyword, user_ids, page, page_size, sort)
count = await LinsightSOPDao.count_sop_record(keyword, user_ids)
if not res:
return [], 0
all_users = await UserDao.afilter_users(user_ids=[one.user_id for one in res])
all_users = {
one.user_id: one.user_name for one in all_users
}
result = []
for one in res:
new_one = SopRecordRead.model_validate(one)
new_one.user_name = all_users.get(one.user_id, str(one.user_id))
result.append(new_one)
return result, count
@staticmethod
async def update_sop_record_score(session_version_id: str, score: int) -> None:
await LinsightSOPDao.update_sop_record_score(session_version_id, score)
@staticmethod
async def update_sop_record_feedback(session_version_id: str, feedback: str) -> None:
await LinsightSOPDao.update_sop_record_feedback(session_version_id, feedback)
@classmethod
async def sync_sop_record(cls, record_ids: list[int], override: bool = False, save_new: bool = False) \
-> list[str] | None:
"""
SynchronousSOPRecord
:param record_ids: SOPRecordIDVertical
:param override: Do you want to overwrite the existingSOP
:param save_new: Do you want to save the newSOP
:return: If there is a duplicateSOPrecords, return a list of duplicate record names, otherwise returnNone
"""
sop_records = await LinsightSOPDao.get_sop_record_by_ids(record_ids)
return await cls._sync_sop_record(sop_records, override, save_new)
@staticmethod
async def _sync_sop_record(sop_records: list[LinsightSOPRecord], override: bool = False, save_new: bool = False) \
-> list[str] | None:
"""
If there is a duplicateSOPrecords, returning a list of duplicate record names
"""
records_name_dict = {}
repeat_names = set()
name_set = set()
sop_list = []
oversize_records = []
new_records = []
for one in sop_records:
if len(one.content) > 50000:
oversize_records.append(one.name)
continue
new_records.append(one)
if one.name not in name_set:
records_name_dict[one.name] = one
name_set.add(one.name)
sop_records = new_records
if not sop_records and oversize_records:
raise SopContentOverLimitError(data={"sop_name": "、".join(oversize_records)})
if name_set:
sop_list = await LinsightSOPDao.get_sops_by_names(list(name_set))
for one in sop_list:
repeat_names.add(one.name)
if override:
# Update existing firstsopGallery
override_name_dict = {}
for one in sop_list:
if one_record := records_name_dict.get(one.name):
await SOPManageService.update_sop(SOPManagementUpdateSchema(
id=one.id,
name=one.name,
description=one_record.description,
content=one_record.content,
rating=one_record.rating,
linsight_version_id=one_record.linsight_version_id,
showcase=False,
user_id=one_record.user_id,
))
override_name_dict[one.name] = True
# Add the restsopRecord
for one in records_name_dict.values():
if one.name in override_name_dict:
continue
await SOPManageService.add_sop(SOPManagementSchema(
name=one.name,
description=one.description,
content=one.content,
rating=one.rating,
linsight_version_id=one.linsight_version_id,
), one.user_id)
elif save_new:
for one in sop_records:
new_name = one.name
if new_name in repeat_names:
# Add suffix if there are duplicate records, Limit Length500characters
new_name = f"{one.name}Dungeon"
await SOPManageService.add_sop(SOPManagementSchema(
name=new_name,
description=one.description,
content=one.content,
rating=one.rating,
linsight_version_id=one.linsight_version_id,
), one.user_id)
else:
# Explain that there is a duplicate record, which needs to be confirmed by the user
if sop_list:
return list(repeat_names)
# Insert a record into the database
for one in sop_records:
await SOPManageService.add_sop(SOPManagementSchema(
name=one.name,
description=one.description,
content=one.content,
rating=one.rating,
linsight_version_id=one.linsight_version_id,
), one.user_id)
if oversize_records:
raise SopContentOverLimitError(data={"sop_name": "、".join(oversize_records)})
return None
@classmethod
async def parse_sop_file(cls, file: UploadFile) -> (list, list):
"""
analyzingSOPDoc.
:param file: FilePath
"""
if not file.size:
raise UploadFileEmptyError()
error_rows = []
success_rows = []
wb = None
try:
wb = openpyxl.load_workbook(io.BytesIO(file.file.read()), read_only=True, data_only=True)
sheet = wb.active
max_rows = sheet.max_row
for i in range(2, max_rows + 1):
name = sheet.cell(row=i, column=1).value
description = sheet.cell(row=i, column=2).value
content = sheet.cell(row=i, column=3).value
error_msg = []
if not name:
error_msg.append("name_empty")
if not content:
error_msg.append("description_empty")
if len(str(name)) >= 500:
error_msg.append("name_over_size")
if len(str(content)) >= 50000:
error_msg.append("content_over_size")
if description and len(str(description)) >= 1000:
error_msg.append("description_over_size")
if error_msg:
error_rows.append({
"index": i,
"error_msg": error_msg
})
else:
success_rows.append({
"name": str(name),
"description": str(description) if description is not None else "",
"content": str(content),
})
finally:
if wb:
wb.close()
return success_rows, error_rows
@classmethod
async def upload_sop_file(cls, login_user: UserPayload, file: UploadFile, ignore_error: bool, override: bool,
save_new: bool) \
-> (List[Dict], List[Dict], List[str]):
"""
Upload itSOPDoc.
:param login_user: Logged in user information
:param file: FilePath
:param ignore_error: Do you want to ignore the error
:param override: Do you want to overwrite the existingSOP
:param save_new: Do you want to save the newSOP
:return: Upload results, success_rows, error_rows, repeat_names
"""
success_rows, error_rows = await cls.parse_sop_file(file)
if (error_rows or len(success_rows) == 0) and not ignore_error:
return success_rows, error_rows
if not success_rows:
return [], [], []
records = [LinsightSOPRecord(**one, user_id=login_user.user_id) for one in success_rows]
repeat_name_list = await cls._sync_sop_record(records, override=override, save_new=save_new)
return [], [], repeat_name_list
@classmethod
async def get_sop_list(cls, keywords: str = None, sort: Literal["asc", "desc"] = "desc", showcase: bool = False,
page: int = 1, page_size: int = 10) -> dict:
"""
DapatkanSOPVertical
:param keywords: Keyword
:param sort: Sort By
:param page: Page
:param page_size: Items per page
:param showcase: Whether to show only the selected cases ofSOP
:return: SOPLists and totals
"""
sop_pages = await LinsightSOPDao.get_sop_page(keywords=keywords, showcase=showcase, page=page,
page_size=page_size,
sort=sort)
user_ids = list(set([one["user_id"] for one in sop_pages["items"]]))
user_map = UserDao.aget_user_by_ids(user_ids=user_ids)
user_map = {one.user_id: one.user_name for one in await user_map}
for one in sop_pages["items"]:
one["user_name"] = user_map.get(one["user_id"], str(one["user_id"]))
return sop_pages
@staticmethod
async def add_sop(sop_obj: SOPManagementSchema, user_id: int) -> UnifiedResponseModel | None:
"""
Add NewSOP
:param user_id:
:param sop_obj:
:return: Added bySOPObjects
"""
# Get the current global configuration ofembeddingModels
workbench_conf = await LLMService.get_workbench_llm()
try:
emb_model_id = workbench_conf.embedding_model.id
if not emb_model_id:
return NoEmbeddingModelError.return_resp()
except AttributeError:
return NoEmbeddingModelError.return_resp()
# CorrectionembeddingModels
embed_info = LLMDao.get_model_by_id(int(emb_model_id))
if not embed_info:
return EmbeddingModelNotExistError.return_resp()
if embed_info.model_type != LLMModelType.EMBEDDING.value:
return EmbeddingModelTypeError.return_resp()
vector_store_id = uuid.uuid4().hex
embeddings = await LLMService.get_bisheng_linsight_embedding(model_id=embed_info.id,
invoke_user_id=user_id)
try:
vector_client: Milvus = decide_vectorstores(
SOPManageService.collection_name, "Milvus", embeddings
)
es_client: ElasticKeywordsSearch = decide_vectorstores(
SOPManageService.collection_name, "ElasticKeywordsSearch", FakeEmbedding()
)
metadatas = [{"vector_store_id": vector_store_id}]
vector_client.add_texts([sop_obj.content[0:10000]], metadatas=metadatas)
es_client.add_texts([sop_obj.content], ids=[vector_store_id], metadatas=metadatas)
except Exception as e:
return LinsightAddSopError.return_resp(data=str(e))
sop_dict = sop_obj.model_dump(exclude_unset=True)
sop_dict["vector_store_id"] = vector_store_id # Set Vector StorageID
# Here you can add a database operation that willsop_objSave to Database
sop_model = LinsightSOP(**sop_dict)
sop_model.user_id = user_id
sop_model = await LinsightSOPDao.create_sop(sop_model)
return resp_200(data=sop_model)
@staticmethod
async def update_sop(sop_obj: SOPManagementUpdateSchema,
update_version_id: bool = True) -> UnifiedResponseModel | None:
"""
UpdateSOP
:param sop_obj:
:param update_version_id: Do you want to update the versionID
:return: Post UpdateSOPObjects
"""
# CorrectionSOPpresence or does it
existing_sop = await LinsightSOPDao.get_sops_by_ids([sop_obj.id])
if not existing_sop:
return NotFoundError.return_resp()
if not update_version_id:
sop_obj.linsight_version_id = existing_sop[0].linsight_version_id
if sop_obj.content != existing_sop[0].content:
# Get the current global configuration ofembeddingModels
workbench_conf = await LLMService.get_workbench_llm()
try:
emb_model_id = workbench_conf.embedding_model.id
if not emb_model_id:
return NoEmbeddingModelError.return_resp()
except AttributeError:
return NoEmbeddingModelError.return_resp()
vector_store_id = existing_sop[0].vector_store_id
embeddings = await LLMService.get_bisheng_linsight_embedding(invoke_user_id=sop_obj.user_id,
model_id=int(emb_model_id))
# Update Vector Store
try:
vector_client: Milvus = decide_vectorstores(
SOPManageService.collection_name, "Milvus", embeddings
)
es_client: ElasticKeywordsSearch = decide_vectorstores(
SOPManageService.collection_name, "ElasticKeywordsSearch", FakeEmbedding()
)
vector_client.delete(expr=f"vector_store_id == '{vector_store_id}'")
es_client.delete([vector_store_id])
metadatas = [{"vector_store_id": vector_store_id}]
vector_client.add_texts([sop_obj.content[0:10000]], metadatas=metadatas)
es_client.add_texts([sop_obj.content], ids=[vector_store_id], metadatas=metadatas)
except Exception as e:
return LinsightUpdateSopError.return_resp(data=str(e))
# Update databaseSOP
sop_model = await LinsightSOPDao.update_sop(sop_obj)
return resp_200(data=sop_model)
@staticmethod
async def remove_sop(sop_ids: list[int], login_user: UserPayload) -> UnifiedResponseModel | None:
"""
DeleteSOP
:param login_user:
:param sop_ids: SOPUniqueness quantificationIDVertical
:return: Remove result
"""
# CorrectionSOPpresence or does it
existing_sops = await LinsightSOPDao.get_sops_by_ids(sop_ids)
if not existing_sops:
return NotFoundError.return_resp()
# Delete data in vector store
try:
vector_store_ids = [sop.vector_store_id for sop in existing_sops]
vector_client: Milvus = decide_vectorstores(
SOPManageService.collection_name, "Milvus", FakeEmbedding()
)
es_client: ElasticKeywordsSearch = decide_vectorstores(
SOPManageService.collection_name, "ElasticKeywordsSearch", FakeEmbedding()
)
vector_client.delete(expr=f"vector_store_id in {vector_store_ids}")
es_client.delete(vector_store_ids)
except Exception as e:
return LinsightDeleteSopError.return_resp(data=str(e))
# Delete from databaseSOP
await LinsightSOPDao.remove_sop(sop_ids=sop_ids)
return resp_200(data=True)
@classmethod
async def get_sop_by_id(cls, sop_id: int) -> LinsightSOP | None:
"""
accordingIDDapatkanSOP
:param sop_id: SOPUniqueness quantificationID
:return: SOPObjects
"""
sop_models = await LinsightSOPDao.get_sops_by_ids([sop_id])
if not sop_models:
return None
return sop_models[0]
# sop Library Retrieval
@classmethod
async def search_sop(cls, invoke_user_id: int, query: str, k: int = 3) -> (List[Document], BaseErrorCode | None):
"""
CariSOP
:param k:
:param query: Keywords Search
:return: search results
"""
# Get the current global configuration ofembeddingModels
try:
vector_search = True
es_search = True
error_msg = None
workbench_conf = await LLMService.get_workbench_llm()
if workbench_conf.embedding_model is None or not workbench_conf.embedding_model.id:
vector_search = False
error_msg = LinsightVectorModelError
else:
try:
emb_model_id = workbench_conf.embedding_model.id
embeddings = await LLMService.get_bisheng_linsight_embedding(invoke_user_id=invoke_user_id,
model_id=int(emb_model_id))
await embeddings.aembed_query("test")
except Exception as e:
logger.error(f"Vector retrieval model initialization failed: {str(e)}")
vector_search = False
error_msg = LinsightVectorModelError
# Create Text Splitter
text_splitter = RecursiveCharacterTextSplitter()
retrievers = []
if vector_search and es_search:
emb_model_id = workbench_conf.embedding_model.id
embeddings = await LLMService.get_bisheng_linsight_embedding(invoke_user_id=invoke_user_id,
model_id=int(emb_model_id))
vector_client: Milvus = decide_vectorstores(
SOPManageService.collection_name, "Milvus", embeddings
)
es_client: ElasticKeywordsSearch = decide_vectorstores(
SOPManageService.collection_name, "ElasticKeywordsSearch", FakeEmbedding()
)
keyword_retriever = KeywordRetriever(keyword_store=es_client, search_kwargs={"k": 100},
text_splitter=text_splitter)
baseline_vector_retriever = BaselineVectorRetriever(vector_store=vector_client,
search_kwargs={"k": 100},
text_splitter=text_splitter)
retrievers = [keyword_retriever, baseline_vector_retriever]
elif es_search and not vector_search:
# Search with keywords only
es_client: ElasticKeywordsSearch = decide_vectorstores(
SOPManageService.collection_name, "ElasticKeywordsSearch", FakeEmbedding()
)
keyword_retriever = KeywordRetriever(keyword_store=es_client, search_kwargs={"k": 100},
text_splitter=text_splitter)
retrievers = [keyword_retriever]
elif vector_search and not es_search:
# Use vector retrieval only
emb_model_id = workbench_conf.embedding_model.id
embeddings = await LLMService.get_bisheng_linsight_embedding(invoke_user_id=invoke_user_id,
model_id=int(emb_model_id))
vector_client: Milvus = decide_vectorstores(
SOPManageService.collection_name, "Milvus", embeddings
)
baseline_vector_retriever = BaselineVectorRetriever(vector_store=vector_client,
search_kwargs={"k": 100},
text_splitter=text_splitter)
retrievers = [baseline_vector_retriever]
else:
error_msg = LinsightDocSearchError
return [], error_msg
retriever = EnsembleRetriever(retrievers=retrievers, weights=[0.5, 0.5] if len(retrievers) > 1 else [1.0])
# Perform Retrieval
results = await retriever.ainvoke(input=query)
if not results:
return [], error_msg
vector_store_ids = [doc.metadata.get("vector_store_id") for doc in results if
doc.metadata.get("vector_store_id")]
# accordingvector_store_idsQuerying repositoriessop
sop_models = await LinsightSOPDao.get_sop_by_vector_store_ids(vector_store_ids)
sop_model_vector_store_ids = [sop.vector_store_id for sop in sop_models]
# Filter results to ensure that only those that exist in the database are returnedSOP
results = [doc for doc in results if doc.metadata.get("vector_store_id") in sop_model_vector_store_ids]
# Before filtering and fetchingk result(s) found
results = results[:k]
return results, error_msg
except Exception as e:
logger.error(f"Failed to search the instruction manual: {str(e)}")
return [], LinsightDocNotFoundError
# RebuildSOP VectorStore
@classmethod
async def rebuild_sop_vector_store_task(cls, embeddings: Embeddings):
"""
RebuildSOPVector Storage
:return: Reconstruction Results
"""
try:
# Fetch allSOP
all_sops = await LinsightSOPDao.get_all_sops()
if not all_sops:
logger.info("NoSOPData needs to be reconstructed for vector storage")
return None
# Wrapper synchronization function is asynchronous function
def sync_func(sops, emb):
"""
Synchronization function for rebuildingSOPVector Storage
:param emb:
:param sops:
:return:
"""
vector_client: Milvus = decide_vectorstores(
SOPManageService.collection_name, "Milvus", emb
)
# Delete existing vector storecollection
if vector_client.col is not None:
logger.info("Delete existingSOPVector Storagecollection")
vector_client.col.drop()
vector_client.col = None
vector_client.fields = []
metadatas = [{"vector_store_id": sop.vector_store_id} for sop in sops]
contents = [sop.content for sop in sops]
batch_size = 16
for i in range(0, len(contents), batch_size):
batch_contents = contents[i:i + batch_size]
batch_metadatas = metadatas[i:i + batch_size]
# Add NewSOPData to vector storage
vector_client.add_texts(batch_contents, metadatas=batch_metadatas)
logger.info("SOPVector store rebuild completed: {}".format(len(sops)))
# Userun_asyncRun Synchronization Function
await util.sync_func_to_async(sync_func)(all_sops, embeddings)
return None
except Exception as e:
logger.exception(f"RebuildSOPVector store failed: {str(e)}")
return None
# if __name__ == '__main__':
# # test code
# results, error_msg = asyncio.run(SOPManageService.search_sop(query="Guidelines for the preparation of bidding documents", k=3))
#
# print(results)
# print(error_msg)
================================================
FILE: src/backend/bisheng/linsight/domain/services/state_message_manager.py
================================================
import asyncio
import pickle
from enum import Enum
from typing import List, Dict, Any, Optional
from loguru import logger
from pydantic import BaseModel, Field
from bisheng.linsight.domain.schemas.linsight_schema import UserInputEventSchema
from bisheng.common.errcode.http_error import ServerError
from bisheng.core.cache.redis_manager import get_redis_client_sync, get_redis_client
from bisheng.linsight.domain.models.linsight_execute_task import ExecuteTaskStatusEnum, LinsightExecuteTaskDao, \
LinsightExecuteTask
from bisheng.linsight.domain.models.linsight_session_version import LinsightSessionVersion, LinsightSessionVersionDao
from bisheng.utils.util import retry_async
from bisheng_langchain.linsight.event import BaseEvent
class MessageEventType(str, Enum):
"""
Message event type enumeration
"""
# tasks starting
TASK_START = "task_start"
# Generate Tasks
TASK_GENERATE = "task_generate"
# Task status update
TASK_STATUS_UPDATE = "task_status_update"
# User input
USER_INPUT = "user_input"
# User input complete
USER_INPUT_COMPLETED = "user_input_completed"
# Task Execution Steps
TASK_EXECUTE_STEP = "task_execute_step"
# Mission-End
TASK_END = "task_end"
# Error Message
ERROR_MESSAGE = "error_message"
# Final Result
FINAL_RESULT = "final_result"
# Mission terminated
TASK_TERMINATED = "task_terminated"
class MessageData(BaseModel):
"""Message Data Model"""
event_type: MessageEventType
data: Dict[str, Any]
timestamp: Optional[float] = Field(default_factory=lambda: asyncio.get_event_loop().time())
class LinsightStateMessageManager:
"""Idea State and Message Manager"""
# Class Constant
DEFAULT_EXPIRATION = 3600
DEFAULT_RETRY_ATTEMPTS = 3
DEFAULT_RETRY_DELAY = 1
KEY_PREFIX = "linsight_tasks:"
def __init__(self, session_version_id: str):
"""
Initializing the Inspiration State and Message Manager
Args:
session_version_id: Session VersionID
"""
self._session_version_id = session_version_id
self._redis_client = get_redis_client_sync()
self._logger = logger
# Redis keyManaging
self._key_prefix = f"{self.KEY_PREFIX}{session_version_id}:"
self._keys = {
'session_version_info': f"{self._key_prefix}session_version_info",
'messages': f"{self._key_prefix}messages",
'execution_tasks': f"{self._key_prefix}execution_tasks:"
}
async def _handle_redis_operation(self, operation, *args, **kwargs):
"""
ImpuestoRedisOperation error handling
Args:
operation: RedisAction Function
*args: Position JSON
**kwargs: Keyword Parameters
Returns:
Operating result
Raises:
Exception: RedisException thrown when operation failed
"""
try:
return await operation(*args, **kwargs)
except Exception as e:
self._logger.error(f"Redis operation failed: {e}")
raise
@retry_async(num_retries=DEFAULT_RETRY_ATTEMPTS, delay=DEFAULT_RETRY_DELAY)
async def push_message(self, message: MessageData) -> None:
"""
Push messages toRedisin the list
Args:
message: Message Model
"""
self._logger.info(f"Pushing message: {message.event_type}")
await self._handle_redis_operation(
self._redis_client.arpush,
self._keys['messages'],
message.model_dump()
)
self._logger.info(f"Message pushed: {message.event_type}")
@retry_async(num_retries=DEFAULT_RETRY_ATTEMPTS, delay=DEFAULT_RETRY_DELAY)
async def pop_message(self) -> Optional[MessageData]:
"""
FROMRedisA message pops up in the list
Returns:
Message Model orNone
"""
try:
message_data = await self._redis_client.ablpop(self._keys['messages'])
if message_data:
return MessageData.model_validate(message_data)
return None
except Exception as e:
self._logger.error(f"Failed to pop message: {e}")
raise e
@retry_async(num_retries=DEFAULT_RETRY_ATTEMPTS, delay=DEFAULT_RETRY_DELAY)
async def set_session_version_info(self, session_version_model) -> None:
"""
Set session version information
Args:
session_version_model: Session Version Model
"""
# Using Transactions to Ensure Data Consistency
async with self._redis_client.async_pipeline() as pipe:
try:
# Write database first
await LinsightSessionVersionDao.insert_one(session_version_model)
# Write AgainRedis
await pipe.set(
self._keys['session_version_info'],
pickle.dumps(session_version_model.model_dump()),
ex=self.DEFAULT_EXPIRATION
)
await pipe.execute()
self._logger.info(f"Session version info set: {self._session_version_id}")
except Exception as e:
self._logger.error(f"Failed to set session version info: {e}")
raise
@retry_async(num_retries=DEFAULT_RETRY_ATTEMPTS, delay=DEFAULT_RETRY_DELAY)
async def get_session_version_info(self) -> Optional[LinsightSessionVersion]:
"""
Get session version information
Returns:
Session Version Information Model orNone
"""
try:
info = await self._handle_redis_operation(
self._redis_client.aget,
self._keys['session_version_info']
)
if not info:
self._logger.warning(f"No session version info found for {self._session_version_id}")
session_version_model = await LinsightSessionVersionDao.get_by_id(self._session_version_id)
await self.set_session_version_info(session_version_model)
return session_version_model
return LinsightSessionVersion.model_validate(info)
except Exception as e:
self._logger.error(f"Failed to get session version info: {e}")
session_version_model = await LinsightSessionVersionDao.get_by_id(self._session_version_id)
await self.set_session_version_info(session_version_model)
return session_version_model
@retry_async(num_retries=DEFAULT_RETRY_ATTEMPTS, delay=DEFAULT_RETRY_DELAY)
async def set_execution_tasks(self, tasks: List[LinsightExecuteTask]) -> None:
"""
Set Execution Information
Args:
tasks: Execute Task List
"""
if not tasks:
self._logger.warning("No tasks provided to set_execution_tasks")
return
try:
# Batch WriteRedis
tasks_mapping = {
f"{self._keys['execution_tasks']}{task.id}": task.model_dump()
for task in tasks
}
await self._redis_client.amset(tasks_mapping, expiration=self.DEFAULT_EXPIRATION)
except Exception as e:
self._logger.error(f"Failed to set execution tasks: {e}")
raise
@retry_async(num_retries=DEFAULT_RETRY_ATTEMPTS, delay=DEFAULT_RETRY_DELAY)
async def update_execution_task_status(
self,
task_id: str,
status: ExecuteTaskStatusEnum,
**kwargs
) -> Dict[str, Any]:
"""
Update Execution Status
Args:
task_id: TaskID
status: New status
**kwargs: Other update fields
Returns:
Updated task data
"""
try:
# Update database first
task_model = await LinsightExecuteTaskDao.update_by_id(
task_id,
status=status,
**kwargs
)
# Update againRedis
task_key = f"{self._keys['execution_tasks']}{task_id}"
task_data = task_model.model_dump()
await self._redis_client.aset(
task_key,
task_data,
expiration=self.DEFAULT_EXPIRATION
)
self._logger.info(f"Updated task {task_id} status to {status}")
return task_data
except Exception as e:
self._logger.error(f"Failed to update task {task_id} status: {e}")
raise
@retry_async(num_retries=DEFAULT_RETRY_ATTEMPTS, delay=DEFAULT_RETRY_DELAY)
async def set_user_input(self, task_id: str, user_input: str, files: List[Dict[str, str]] = None) -> None:
"""
Set User Input
Args:
task_id: TaskID
user_input: User input
files: Related Documents List
"""
task_key = f"{self._keys['execution_tasks']}{task_id}"
try:
task_model = await self.get_execution_task(task_id)
if not task_model:
raise ValueError(f"Task with ID {task_id} not found in Redis or database.")
user_input_event = task_model.history[-1] if task_model.history else None
if user_input_event is None or user_input_event.get("step_type") != "call_user_input":
raise ValueError(f"Task with ID {task_id} does not support user input.")
user_input_event = UserInputEventSchema.model_validate(user_input_event)
user_input_event.user_input = user_input
user_input_event.files = files
user_input_event.is_completed = True
task_model.history[-1] = user_input_event.model_dump()
task_model.status = ExecuteTaskStatusEnum.USER_INPUT_COMPLETED
# Using Transactions to Ensure Data Consistency
async with self._redis_client.async_pipeline() as pipe:
await pipe.set(task_key, pickle.dumps(task_model.model_dump()), ex=self.DEFAULT_EXPIRATION)
await pipe.execute()
# Database updating
await LinsightExecuteTaskDao.update_by_id(
task_id,
status=ExecuteTaskStatusEnum.USER_INPUT_COMPLETED,
history=task_model.history
)
self._logger.info(f"Set user input for task {task_id}")
except Exception as e:
self._logger.error(f"Failed to set user input for task {task_id}: {e}")
raise ServerError.http_exception()
@retry_async(num_retries=DEFAULT_RETRY_ATTEMPTS, delay=DEFAULT_RETRY_DELAY)
async def get_execution_task(self, task_id: str) -> Optional[LinsightExecuteTask]:
"""
Get task execution information
Args:
task_id: TaskID
Returns:
Execute Task Model orNone
"""
task_key = f"{self._keys['execution_tasks']}{task_id}"
try:
task_data = await self._redis_client.aget(task_key)
if task_data:
return LinsightExecuteTask.model_validate(task_data)
# Automatically close purchase order afterRedisNo data in, fetching from database
task_model = await LinsightExecuteTaskDao.get_by_id(task_id)
await self.set_execution_tasks([task_model])
return task_model
except Exception as e:
self._logger.error(f"Failed to get execution task {task_id}: {e}")
return None
@retry_async(num_retries=DEFAULT_RETRY_ATTEMPTS, delay=DEFAULT_RETRY_DELAY)
async def add_execution_task_step(self, task_id: str, step: BaseEvent) -> None:
"""
Add Execute Task Step
Args:
task_id: TaskID
step: Execution Steps
"""
task_key = f"{self._keys['execution_tasks']}{task_id}"
try:
task_data = await self.get_execution_task(task_id)
if not task_data:
raise ValueError(f"Task with ID {task_id} not found in Redis.")
task_model = LinsightExecuteTask.model_validate(task_data)
# Initialization History
if task_model.history is None:
task_model.history = []
# Adding new Steps
task_model.history.append(step.model_dump())
# Update Redis and database
await self._redis_client.aset(
task_key,
task_model.model_dump(),
expiration=self.DEFAULT_EXPIRATION
)
await LinsightExecuteTaskDao.update_by_id(
task_id,
history=task_model.history
)
self._logger.info(f"Added step to task {task_id}")
except Exception as e:
self._logger.error(f"Failed to add step to task {task_id}: {e}")
raise
async def get_execution_tasks(self):
"""
Get All Execute Tasks
Returns:
Execute Task List
"""
try:
pattern = f"{self._keys['execution_tasks']}*"
task_keys = await self._redis_client.akeys(pattern)
if not task_keys:
return []
tasks_data = await self._redis_client.amget(task_keys)
tasks = [LinsightExecuteTask.model_validate(task) for task in tasks_data if task]
if not tasks:
tasks = await LinsightExecuteTaskDao.get_by_session_version_id(
session_version_id=self._session_version_id)
return tasks
except Exception as e:
self._logger.error(f"Failed to get execution tasks: {e}")
return []
async def cleanup_session_data(self) -> None:
"""
Cleanup Session RelatedRedisDATA
"""
try:
pattern = f"{self._key_prefix}*"
keys = await self._redis_client.akeys(pattern)
if keys:
await self._redis_client.adelete(*keys)
self._logger.info(f"Cleaned up {len(keys)} keys for session {self._session_version_id}")
except Exception as e:
self._logger.error(f"Failed to cleanup session data: {e}")
raise
async def get_session_stats(self) -> Dict[str, Any]:
"""
Get session statistics
Returns:
Dictionary containing session statistics
"""
try:
stats = {
'session_version_id': self._session_version_id,
'message_count': await self._redis_client.allen(self._keys['messages']),
'has_session_info': await self._redis_client.exists(self._keys['session_version_info']),
'task_count': 0
}
# Calculate number of tasks
pattern = f"{self._keys['execution_tasks']}*"
task_keys = await self._redis_client.akeys(pattern)
stats['task_count'] = len(task_keys)
return stats
except Exception as e:
self._logger.error(f"Failed to get session stats: {e}")
return {'error': str(e)}
# Clean up all session-relatedRedisDATA
@classmethod
async def cleanup_all_sessions(cls) -> None:
"""
Clean up all session-relatedRedisDATA
"""
try:
redis_client = await get_redis_client()
pattern = f"{cls.KEY_PREFIX}*"
keys = await redis_client.async_connection.keys(pattern)
if keys:
await redis_client.async_connection.delete(*keys)
logger.info(f"Cleaned up {len(keys)} keys for all sessions")
except Exception as e:
logger.error(f"Failed to cleanup all session data: {e}")
return
================================================
FILE: src/backend/bisheng/linsight/domain/services/workbench_impl.py
================================================
import asyncio
import json
import os
import uuid
from dataclasses import dataclass
from io import BytesIO
from typing import Dict, List, Optional, AsyncGenerator, Tuple, Any
from urllib.parse import unquote
from e2b.sandbox.filesystem.filesystem import WriteEntry
from fastapi import UploadFile
from langchain_core.tools import BaseTool
from loguru import logger
from bisheng.api.services.knowledge_imp import decide_vectorstores, async_read_chunk_text
from bisheng.linsight.domain.services.sop_manage import SOPManageService
from bisheng.api.services.workstation import WorkStationService
from bisheng.linsight.domain.schemas.linsight_schema import LinsightQuestionSubmitSchema, DownloadFilesSchema, \
SubmitFileSchema
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum, ApplicationTypeEnum
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode import BaseErrorCode
from bisheng.common.errcode.http_error import UnAuthorizedError
from bisheng.common.errcode.linsight import LinsightToolInitError, LinsightBishengLLMError, LinsightGenerateSopError
from bisheng.common.schemas.telemetry.event_data_schema import NewMessageSessionEventData
from bisheng.common.services import telemetry_service
from bisheng.common.services.config_service import settings
from bisheng.core.cache.redis_manager import get_redis_client
from bisheng.core.cache.utils import save_file_to_folder, CACHE_DIR
from bisheng.core.logger import trace_id_var
from bisheng.core.prompts.manager import get_prompt_manager
from bisheng.core.storage.minio.minio_manager import get_minio_storage
from bisheng.database.models.flow import FlowType
from bisheng.linsight.domain.models.linsight_execute_task import LinsightExecuteTaskDao
from bisheng.linsight.domain.models.linsight_session_version import LinsightSessionVersionDao, SessionVersionStatusEnum, \
LinsightSessionVersion
from bisheng.linsight.domain.models.linsight_sop import LinsightSOPRecord
from bisheng.database.models.session import MessageSessionDao, MessageSession
from bisheng.interface.embeddings.custom import FakeEmbedding
from bisheng.knowledge.domain.models.knowledge import KnowledgeRead, KnowledgeTypeEnum
from bisheng.llm.domain.llm import BishengLLM
from bisheng.llm.domain.services import LLMService
from bisheng.tool.domain.models.gpts_tools import GptsToolsDao
from bisheng.tool.domain.services.executor import ToolExecutor
from bisheng.tool.domain.services.tool import ToolServices
from bisheng.utils import util
from bisheng.utils.util import async_calculate_md5
from bisheng_langchain.linsight.const import ExecConfig
@dataclass
class TaskNode:
"""Task node to build the task tree"""
task: Any # LinsightExecuteTask Objects
children: List['TaskNode'] = None
def __post_init__(self):
if self.children is None:
self.children = []
def to_dict(self) -> Dict:
"""Convert task nodes to dictionary format"""
task_dict = self.task.model_dump()
task_dict['children'] = [child.to_dict() for child in self.children]
return task_dict
class LinsightWorkbenchImpl:
"""LinsightWorkbench Implementation Class"""
# Class Constant
COLLECTION_NAME_PREFIX = "col_linsight_file_"
FILE_INFO_REDIS_KEY_PREFIX = "linsight_file:"
CACHE_EXPIRATION_HOURS = 24
class LinsightError(Exception):
"""LinsightRelated Errors"""
pass
class SearchSOPError(Exception):
"""SOPRetrieve Error"""
def __init__(self, error_class: BaseErrorCode):
super().__init__(error_class.Msg)
self.error_class = error_class
class ToolsInitializationError(Exception):
"""Tool initialization error"""
class BishengLLMError(Exception):
"""Bisheng LLMRelated Errors"""
@classmethod
async def _get_llm(cls, invoke_user_id: int) -> (BishengLLM, Any):
# Get and validate the workbench configuration
workbench_conf = await cls._get_workbench_config()
# BuatLLMInstances
linsight_conf = settings.get_linsight_conf()
llm = await LLMService.get_bisheng_linsight_llm(invoke_user_id=invoke_user_id,
model_id=workbench_conf.task_model.id,
temperature=linsight_conf.default_temperature)
return llm, workbench_conf
@classmethod
async def human_participate_add_file(cls, linsight_session_version: LinsightSessionVersion,
files: List[SubmitFileSchema]) -> Optional[List]:
"""
Adding Files When Manually Involved
:param linsight_session_version:
:param files:
:return:
"""
if not files:
return None
processed_files = await cls._process_submitted_files(files, linsight_session_version.session_id)
if linsight_session_version.files:
linsight_session_version.files.extend(processed_files)
else:
linsight_session_version.files = processed_files
await LinsightSessionVersionDao.insert_one(linsight_session_version)
return processed_files
@classmethod
async def submit_user_question(cls, submit_obj: LinsightQuestionSubmitSchema,
login_user: UserPayload) -> tuple[MessageSession, LinsightSessionVersion]:
"""
Submit user issue and create session
Args:
submit_obj: Submitted Question Objects
login_user: Logged in user information
Returns:
tuple: (Message Session Model, Inspiration Conversation Version Model)
Raises:
LinsightError: When creating a session fails
"""
try:
# Generate Unique SessionsID
chat_id = uuid.uuid4().hex
# Process files (if present)
processed_files = await cls._process_submitted_files(submit_obj.files, chat_id)
# Create a message session
message_session = MessageSession(
chat_id=chat_id,
flow_id=ApplicationTypeEnum.LINSIGHT.value,
flow_name='New Chat',
flow_type=FlowType.LINSIGHT.value,
user_id=login_user.user_id
)
message_session = await MessageSessionDao.async_insert_one(message_session)
# RecordTelemetryJournal
await telemetry_service.log_event(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.NEW_MESSAGE_SESSION,
trace_id=trace_id_var.get(),
event_data=NewMessageSessionEventData(
session_id=message_session.chat_id,
app_id=ApplicationTypeEnum.LINSIGHT.value,
source="platform",
app_name=ApplicationTypeEnum.LINSIGHT.value,
app_type=ApplicationTypeEnum.LINSIGHT
)
)
# Create Ideas Conversation Version
linsight_session_version = LinsightSessionVersion(
session_id=chat_id,
user_id=login_user.user_id,
question=submit_obj.question,
tools=submit_obj.tools,
org_knowledge_enabled=submit_obj.org_knowledge_enabled,
personal_knowledge_enabled=submit_obj.personal_knowledge_enabled,
files=processed_files
)
linsight_session_version = await LinsightSessionVersionDao.insert_one(linsight_session_version)
return message_session, linsight_session_version
except Exception as e:
logger.error(f"Failed to submit user question: {str(e)}")
raise cls.LinsightError(f"Failed to submit user question: {str(e)}")
@classmethod
async def _process_submitted_files(cls, files: Optional[List[SubmitFileSchema]], chat_id: str) -> Optional[List]:
"""
Process Submitted Files
Args:
files: List of files
chat_id: SessionsID
Returns:
List of processed files
"""
if not files:
return None
file_ids = []
for file in files:
if file.parsing_status != "completed":
raise cls.LinsightError(f"file {file.file_name} status is error: {file.parsing_status}")
file_ids.append(file.file_id)
redis_keys = [f"{cls.FILE_INFO_REDIS_KEY_PREFIX}{file_id}" for file_id in file_ids]
redis_client = await get_redis_client()
processed_files = await redis_client.amget(redis_keys)
for file_info in processed_files:
if file_info:
await cls._copy_file_to_session_storage(file_info, chat_id)
return processed_files
@classmethod
async def _copy_file_to_session_storage(cls, file_info: Dict, chat_id: str) -> None:
"""
Copy file to session store
Args:
file_info: File information
chat_id: SessionsID
"""
source_object_name = file_info.get("markdown_file_path")
if source_object_name:
original_filename = file_info.get("original_filename")
markdown_filename = f"{original_filename.rsplit('.', 1)[0]}.md"
new_object_name = f"linsight/{chat_id}/{source_object_name}"
minio_client = await get_minio_storage()
await minio_client.copy_object(
source_object=source_object_name,
dest_object=new_object_name,
source_bucket=minio_client.tmp_bucket,
dest_bucket=minio_client.bucket
)
file_info["markdown_file_path"] = new_object_name
file_info["markdown_filename"] = markdown_filename
@classmethod
async def task_title_generate(cls, question: str, chat_id: str,
login_user: UserPayload) -> Dict:
"""
Generate task title
Args:
question: User Questions
chat_id: SessionsID
login_user: Logged in user information
Returns:
Dictionary with task title
"""
try:
llm, _ = await cls._get_llm(login_user.user_id)
# Buatprompt
prompt = await cls._generate_title_prompt(question)
# Generate task title
task_title = await llm.ainvoke(prompt)
if not task_title.content:
raise ValueError("Failed to generate task title, please check the model configuration or input")
# Update session title
await cls._update_session_title(chat_id, task_title.content)
return {
"task_title": task_title.content,
"chat_id": chat_id,
"error_message": None
}
except Exception as e:
logger.error(f"Failed to generate task title: {str(e)}")
return {
"task_title": "New Chat",
"chat_id": chat_id,
"error_message": str(e)
}
@classmethod
async def _get_workbench_config(cls):
"""Get and validate the workbench configuration"""
workbench_conf = await LLMService.get_workbench_llm()
if not workbench_conf or not workbench_conf.task_model:
raise cls.BishengLLMError("The task has been terminated, please contact the administrator to check the status of the Ideas task execution model")
return workbench_conf
@classmethod
async def _generate_title_prompt(cls, question: str) -> List[Tuple[str, str]]:
"""Generate Title Generateprompt"""
prompt_service = await get_prompt_manager()
prompt_obj = prompt_service.render_prompt(
namespace="gen_title",
prompt_name="linsight",
USER_GOAL=question
)
return [
("system", prompt_obj.prompt.system),
("user", prompt_obj.prompt.user)
]
@classmethod
async def _update_session_title(cls, chat_id: str, title: str) -> None:
"""Update session title"""
session = await MessageSessionDao.async_get_one(chat_id)
if session:
session.flow_name = title
await MessageSessionDao.async_insert_one(session)
@classmethod
async def get_linsight_session_version_list(cls, session_id: str) -> List[LinsightSessionVersion]:
"""
Get a list of Invisible Conversation Versions
Args:
session_id: SessionsID
Returns:
Inspiration Session Version List
"""
return await LinsightSessionVersionDao.get_session_versions_by_session_id(session_id)
@classmethod
async def modify_sop(cls, linsight_session_version_id: str, sop_content: str) -> Dict:
"""
Modify Inspiration Conversation Version ofSOPContents
Args:
linsight_session_version_id: Session VersionID
sop_content: SOPContents
Returns:
Operating result
"""
try:
await LinsightSessionVersionDao.modify_sop_content(
linsight_session_version_id=linsight_session_version_id,
sop_content=sop_content
)
return {"success": True, "message": "modify sop content successfully"}
except Exception as e:
logger.error(f"ChangeSOPContent failed: {str(e)}")
raise cls.LinsightError(str(e))
@classmethod
async def generate_sop(cls, linsight_session_version_id: str,
previous_session_version_id: Optional[str] = None,
feedback_content: Optional[str] = None,
reexecute: bool = False,
login_user: Optional[UserPayload] = None,
knowledge_list: List[KnowledgeRead] = None,
example_sop: Optional[str] = None) -> AsyncGenerator[Dict, None]:
"""
BuatSOPContents
Args:
linsight_session_version_id: Current Session VersionID
previous_session_version_id: Previous Session VersionID
feedback_content: Content of feedback
reexecute: Whether to re-execute
login_user: Logged in user information
knowledge_list: Knowledge Base Columns
example_sop: Ref.sop, incoming when doing the same modelsopContents
Yields:
Date GeneratedSOPContent Events
"""
error_message = None
try:
# Get workbench configuration and session version
session_version = await cls._get_session_version(linsight_session_version_id)
if login_user.user_id != session_version.user_id:
yield UnAuthorizedError().to_sse_event_instance()
return
try:
# BuatLLMand tools
llm, workbench_conf = await cls._get_llm(session_version.user_id)
except Exception as e:
logger.error(f"BuatSOPContent failed: session_version_id={linsight_session_version_id}, error={str(e)}")
raise cls.BishengLLMError(str(e))
tools = await cls._prepare_tools(session_version, llm)
# Preparation History Summary
history_summary = await cls._prepare_history_summary(
reexecute, previous_session_version_id
)
# Create proxy and generateSOP
agent = await cls._create_linsight_agent(session_version, llm, tools, workbench_conf)
if previous_session_version_id:
session_version = await LinsightSessionVersionDao.get_by_id(previous_session_version_id)
content = ""
async for res in cls._generate_sop_content(
agent, session_version, feedback_content, history_summary, knowledge_list, example_sop=example_sop,
login_user=login_user
):
if isinstance(res, cls.SearchSOPError):
yield res.error_class.to_sse_event(event="search_sop_error")
continue
content += res.content
yield {
"event": "generate_sop_content",
"data": res.model_dump_json()
}
# Update session version status and sop content
await LinsightSessionVersionDao.modify_sop_content(
linsight_session_version_id=linsight_session_version_id,
sop_content=content
)
logger.info(f"BuatSOPContent Success: session_version_id={linsight_session_version_id}")
except cls.ToolsInitializationError as e:
logger.exception(
f"Failed to initialize the Inspiration Workbench tool: session_version_id={linsight_session_version_id}, error={str(e)}")
error_message = LinsightToolInitError(exception=e)
except cls.BishengLLMError as e:
logger.exception(f"Bisheng LLMError-free: session_version_id={linsight_session_version_id}, error={str(e)}")
error_message = LinsightBishengLLMError(exception=e)
except Exception as e:
logger.exception(f"BuatSOPContent failed: session_version_id={linsight_session_version_id}, error={str(e)}")
error_message = LinsightGenerateSopError(exception=e)
finally:
if error_message:
session_version = await LinsightSessionVersionDao.get_by_id(linsight_session_version_id)
if session_version:
session_version.sop = f"{error_message.Msg}: {str(error_message.exception)}"
session_version.status = SessionVersionStatusEnum.SOP_GENERATION_FAILED
await LinsightSessionVersionDao.insert_one(session_version)
yield error_message.to_sse_event_instance()
@classmethod
async def _get_session_version(cls, session_version_id: str) -> LinsightSessionVersion:
"""Get session version"""
session_version = await LinsightSessionVersionDao.get_by_id(session_version_id)
if not session_version:
raise cls.LinsightError("Inspiration session version does not exist")
return session_version
@classmethod
async def _prepare_tools(cls, session_version: LinsightSessionVersion,
llm: BishengLLM) -> List[BaseTool]:
"""Preparation Tools List"""
try:
tools = await cls.init_linsight_config_tools(session_version, llm)
root_path = os.path.join(CACHE_DIR, "linsight", session_version.id)
os.makedirs(root_path, exist_ok=True)
linsight_tools = await ToolServices.init_linsight_tools(root_path=root_path)
tools.extend(linsight_tools)
return tools
except Exception as e:
raise cls.ToolsInitializationError(str(e))
@classmethod
async def prepare_file_list(cls, session_version: LinsightSessionVersion) -> List[str]:
"""Prepare File List"""
file_list = []
template_str = """@{filename}File Storage Information:{{'Files stored in a semantic repositoryid':'{file_id}','File storage address':'{markdown}'}}@"""
if not session_version.files:
return file_list
for file in session_version.files:
file_list.append(template_str.format(filename=file['original_filename'],
file_id=file['file_id'],
markdown=f"./{file['markdown_filename']}"))
return file_list
@classmethod
async def prepare_knowledge_list(cls, knowledge_list: list[KnowledgeRead]) -> List[str]:
res = []
if not knowledge_list:
return res
# Check if there is a personal knowledge base
template_str = """@{name}Stored information for:{{'The knowledge base is stored in a semantic repositoryid':'{id}'}}@"""
for one in knowledge_list:
if one.type == KnowledgeTypeEnum.PRIVATE.value:
res.append(template_str.format(name="Personal Knowledge Base", id=one.id))
else:
knowledge_str = template_str.format(name=one.name, id=one.id)
if one.description:
knowledge_str += f"，{one.name}is described as{one.description}"
res.append(knowledge_str)
return res
@classmethod
async def _prepare_history_summary(cls, reexecute: bool,
previous_session_version_id: str) -> List[str]:
"""Preparation History Summary"""
history_summary = []
if reexecute and previous_session_version_id:
execute_tasks = await LinsightExecuteTaskDao.get_by_session_version_id(previous_session_version_id)
for task in execute_tasks:
if task.result:
answer = task.result.get("answer", "")
if answer:
history_summary.append(answer)
return history_summary
@classmethod
async def _create_linsight_agent(cls, session_version: LinsightSessionVersion,
llm: BishengLLM, tools: List[BaseTool],
workbench_conf):
"""BuatLinsightAgent"""
from bisheng_langchain.linsight.agent import LinsightAgent
root_path = os.path.join(CACHE_DIR, "linsight", session_version.id[:8])
linsight_conf = settings.get_linsight_conf()
exec_config = ExecConfig(**linsight_conf.model_dump(), debug_id=session_version.id)
return LinsightAgent(
file_dir=root_path,
query=session_version.question,
llm=llm,
tools=tools,
task_mode=workbench_conf.linsight_executor_mode,
exec_config=exec_config,
)
@classmethod
async def _generate_sop_content(cls, agent, session_version: LinsightSessionVersion,
feedback_content: Optional[str],
history_summary: List[str],
knowledge_list: List[KnowledgeRead] = None,
example_sop: str = None,
login_user: Optional[UserPayload] = None
) -> AsyncGenerator:
"""BuatSOPContents"""
file_list = await cls.prepare_file_list(session_version)
knowledge_list = await cls.prepare_knowledge_list(knowledge_list)
if example_sop:
async for res in agent.generate_sop(sop=example_sop, file_list=file_list, knowledge_list=knowledge_list):
yield res
elif feedback_content is None:
# RetrieveSOPTemplates
sop_template, search_sop_error = await SOPManageService.search_sop(
invoke_user_id=login_user.user_id,
query=session_version.question, k=3
)
if search_sop_error:
search_sop_error: BaseErrorCode
logger.error(f"RetrieveSOPTemplate failed: {search_sop_error.Msg}")
yield cls.SearchSOPError(error_class=search_sop_error)
sop_template = "\n\n".join([
f"Examples:\n\n{sop.page_content}"
for sop in sop_template if sop.page_content
])
async for res in agent.generate_sop(sop=sop_template, file_list=file_list, knowledge_list=knowledge_list):
yield res
else:
sop_template = session_version.sop if session_version.sop else ""
async for res in agent.feedback_sop(
sop=sop_template,
feedback=feedback_content,
history_summary=history_summary if history_summary else None,
file_list=file_list,
knowledge_list=knowledge_list
):
yield res
@classmethod
async def get_execute_task_detail(cls, session_version_id: str,
login_user: Optional[UserPayload] = None):
"""
Get task execution details
Args:
session_version_id: Inspiration Conversation VersionID
login_user: Logged in user information
Returns:
Execute Task Detail List
"""
execute_tasks = await LinsightExecuteTaskDao.get_by_session_version_id(session_version_id)
if not execute_tasks:
return []
# 1. Get Level 1 Tasks parent_task_id Yes None Task
root_tasks = [task for task in execute_tasks if task.parent_task_id is None]
# 2. accordingprevious_task_idAND:next_task_idSort first level tasks
def sort_tasks_by_chain(tasks: List[Any]) -> List[Any]:
"""
Sort task list by task chain
previous_task_idYes Noneis the first task,next_task_idYes Noneis the last task.
"""
if not tasks:
return []
# Create a task dictionary for quick lookups
task_dict = {task.id: task for task in tasks}
# Find the start node of the chain (previous_task_id are None）
start_tasks = [task for task in tasks if task.previous_task_id is None]
sorted_tasks = []
for start_task in start_tasks:
# Build task chains from each start node
current_task = start_task
chain = []
while current_task is not None:
chain.append(current_task)
# Setujunext_task_idFind next task
next_task_id = current_task.next_task_id
current_task = task_dict.get(next_task_id) if next_task_id else None
sorted_tasks.extend(chain)
# Dealing with possible orphaned tasks (neitherpreviousNothing, either!nextpointing to them)
processed_ids = {task.id for task in sorted_tasks}
orphan_tasks = [task for task in tasks if task.id not in processed_ids]
sorted_tasks.extend(orphan_tasks)
return sorted_tasks
# Sort first level tasks
sorted_root_tasks = sort_tasks_by_chain(root_tasks)
# 3. Build task tree Use parent_task_id Associate subtasks with parent tasks
def build_task_tree(parent_tasks: List[Any], all_tasks: List[Any]) -> List[TaskNode]:
"""
Build task tree
"""
# Create task mapping
task_map = {task.id: task for task in all_tasks}
# By Parent TaskIDGroup subtasks
children_map = {}
for task in all_tasks:
if task.parent_task_id:
if task.parent_task_id not in children_map:
children_map[task.parent_task_id] = []
children_map[task.parent_task_id].append(task)
def build_node(task: Any) -> TaskNode:
"""Recursively build task nodes"""
node = TaskNode(task=task)
# Get subtasks
child_tasks = children_map.get(task.id, [])
# Sort subtasks
sorted_child_tasks = sort_tasks_by_chain(child_tasks)
# Recursively build child nodes
for child_task in sorted_child_tasks:
child_node = build_node(child_task)
node.children.append(child_node)
return node
# Build Root Node List
root_nodes = []
for parent_task in parent_tasks:
root_node = build_node(parent_task)
root_nodes.append(root_node)
return root_nodes
# Build task tree
task_tree = build_task_tree(sorted_root_tasks, execute_tasks)
# 4. Returns the root node list of the task tree
result = [node.to_dict() for node in task_tree]
return result
@classmethod
async def upload_file(cls, file: UploadFile) -> Dict:
"""
Upload files to the Inspiration Workbench
Args:
file: files uploaded
Returns:
File Information Dictionary
"""
# Generate file information
file_id = uuid.uuid4().hex[:8] # Buat8Bit Unique FileID
# url Code decode The file name
original_filename = unquote(file.filename)
file_extension = original_filename.split('.')[-1] if '.' in original_filename else ''
unique_filename = f"{file_id}.{file_extension}"
# Save file
file_path = await save_file_to_folder(file, 'linsight', unique_filename)
upload_result = {
"file_id": file_id,
"filename": unique_filename,
"original_filename": original_filename,
"file_path": file_path,
"parsing_status": "running",
}
# Cache Result
await cls._cache_parse_result(file_id, upload_result)
return upload_result
@classmethod
async def parse_file(cls, upload_result: Dict, invoke_user_id: int) -> Dict:
"""
Parsing uploaded files
Args:
upload_result: Upload results
invoke_user_id: Call UserID
Returns:
Parsing results
"""
logger.info(f"Start parsing files: {upload_result}")
file_id = upload_result["file_id"]
original_filename = upload_result["original_filename"]
file_path = upload_result["file_path"]
try:
# Get workbench configuration
workbench_conf = await cls._get_workbench_config()
collection_name = f"{cls.COLLECTION_NAME_PREFIX}{workbench_conf.embedding_model.id}"
# Asynchronous execution of file parsing
parse_result = await cls._parse_file(invoke_user_id, file_id, file_path, original_filename,
collection_name, workbench_conf)
# Cache Result
await cls._cache_parse_result(file_id, parse_result)
logger.info(f"File analysis complete: {parse_result}")
except Exception as e:
logger.error(f"File parsing failed: file_id={file_id}, error={str(e)}")
parse_result = {
"file_id": file_id,
"original_filename": original_filename,
"parsing_status": "failed",
"error_message": str(e)
}
await cls._cache_parse_result(file_id, parse_result)
return parse_result
@classmethod
async def _parse_file(cls, invoke_user_id: int, file_id: str, file_path: str, original_filename: str,
collection_name: str, workbench_conf) -> Dict:
"""
Synchronize parsed files
Args:
file_id: Doc.ID
file_path: FilePath
original_filename: Original Filename
collection_name: Set Name
workbench_conf: Workbench configuration
Returns:
Parsing results
"""
# Read file contents
try:
texts, _, parse_type, _ = await async_read_chunk_text(
invoke_user_id=invoke_user_id,
input_file=file_path,
file_name=original_filename,
separator=['\n\n', '\n'],
separator_rule=['after', 'after'],
chunk_size=1000,
chunk_overlap=100,
no_summary=True
)
# BuatmarkdownContents
markdown_content = "\n".join(texts)
markdown_bytes = markdown_content.encode('utf-8')
# SAVINGmarkdownDoc.
markdown_filename = f"{file_id}.md"
minio_client = await get_minio_storage()
await minio_client.put_object_tmp(markdown_filename, markdown_bytes)
markdown_md5 = await async_calculate_md5(markdown_bytes)
# Process vector storage
await cls._process_vector_storage(invoke_user_id, texts, file_id, collection_name, workbench_conf)
return {
"file_id": file_id,
"original_filename": original_filename,
"parsing_status": "completed",
"parse_type": parse_type,
"markdown_filename": markdown_filename,
"markdown_file_path": markdown_filename,
"markdown_file_md5": markdown_md5,
"embedding_model_id": workbench_conf.embedding_model.id,
"collection_name": collection_name
}
except Exception as e:
logger.error(f"File parsing failed: file_id={file_id}, error={str(e)}")
return {
"file_id": file_id,
"original_filename": original_filename,
"parsing_status": "failed",
"error_message": str(e)
}
@classmethod
async def _process_vector_storage(cls, invoke_user_id: int, texts: List[str], file_id: str,
collection_name: str, workbench_conf) -> None:
"""Process vector storage"""
# Buatembeddings
embeddings = await LLMService.get_bisheng_linsight_embedding(model_id=workbench_conf.embedding_model.id,
invoke_user_id=invoke_user_id)
# Create Vector Store
vector_client = decide_vectorstores(collection_name, "Milvus", embeddings)
es_client = decide_vectorstores(collection_name, "ElasticKeywordsSearch", FakeEmbedding())
# Adding Text to Vector Storage
metadatas = [{"file_id": file_id} for _ in texts]
await vector_client.aadd_texts(texts, metadatas=metadatas)
await es_client.aadd_texts(texts, metadatas=metadatas)
@classmethod
async def _cache_parse_result(cls, file_id: str, parse_result: Dict) -> None:
"""Cache Result"""
redis_client = await get_redis_client()
key = f"{cls.FILE_INFO_REDIS_KEY_PREFIX}{file_id}"
await redis_client.aset(
key=key,
value=parse_result,
expiration=60 * 60 * cls.CACHE_EXPIRATION_HOURS
)
@classmethod
async def _init_bisheng_code_tool(cls, config_tool_ids: List[int], file_dir: str, user_id: int) -> List[BaseTool]:
"""
Code interpreter tool for special handling initialization Bi Lift
"""
tools = []
bisheng_code_tool = await GptsToolsDao.aget_tool_by_tool_key(tool_key='bisheng_code_interpreter')
if not bisheng_code_tool or bisheng_code_tool.id not in config_tool_ids:
return tools
# Individual initialization code interpreter tool
config_tool_ids.remove(bisheng_code_tool.id)
code_config = json.loads(bisheng_code_tool.extra) if bisheng_code_tool.extra else {}
if "config" not in code_config:
code_config["config"] = {}
if "local" not in code_config["config"]:
code_config["config"]["local"] = {}
code_config["config"]["local"]["local_sync_path"] = file_dir
if "e2b" not in code_config["config"]:
code_config["config"]["e2b"] = {}
code_config["config"]["e2b"]["local_sync_path"] = file_dir
# Default60Validity period of minutes
code_config["config"]["e2b"]["timeout"] = 3600
code_config["config"]["e2b"]["keep_sandbox"] = True
file_list = []
for root, dirs, files in os.walk(file_dir):
for file in files:
file_path = os.path.join(root, file)
file_list.append(WriteEntry(data=file_path, path=file_path.replace(file_dir, ".")))
code_config["config"]["e2b"]["file_list"] = file_list
bisheng_code_tool.extra = code_config
tools = await ToolExecutor.init_by_tool_id(tool=bisheng_code_tool, app_id=ApplicationTypeEnum.LINSIGHT.value,
app_name=ApplicationTypeEnum.LINSIGHT.value,
app_type=ApplicationTypeEnum.LINSIGHT,
user_id=user_id)
return [tools]
@classmethod
async def init_linsight_config_tools(cls, session_version: LinsightSessionVersion,
llm: BishengLLM, need_upload: bool = False, file_dir: str = None) -> List[
BaseTool]:
"""
Tools for initializing Invis configurations
Args:
session_version: Session Version Model
llm: LLMInstances
need_upload: Do I need to bind a user-uploaded file to the code interpreter?
file_dir: Root directory for user uploaded files
Returns:
Tools List
"""
tools = []
if not session_version.tools:
return tools
# &Extraction toolID
tool_ids = cls._extract_tool_ids(session_version.tools)
# Tools to get workbench configurationsID
ws_config = await WorkStationService.aget_config()
config_tool_ids = cls._extract_tool_ids(ws_config.linsightConfig.tools or [])
# todo Better tool initialization scheme
if need_upload and file_dir:
try:
bisheng_code_tool = await cls._init_bisheng_code_tool(config_tool_ids, file_dir,
user_id=session_version.user_id)
tools.extend(bisheng_code_tool)
except Exception as e:
logger.error(f"Failed to initialize BiSheng code interpreter tool: session_version_id={session_version.id}, error={str(e)}")
# Filter Effective ToolsID
valid_tool_ids = [tid for tid in tool_ids if tid in config_tool_ids]
# Initialization Tools
if valid_tool_ids:
tools.extend(await ToolExecutor.init_by_tool_ids(valid_tool_ids, app_id=ApplicationTypeEnum.LINSIGHT.value,
app_name=ApplicationTypeEnum.LINSIGHT.value,
app_type=ApplicationTypeEnum.LINSIGHT,
user_id=session_version.user_id))
return tools
@classmethod
def _extract_tool_ids(cls, tools: List[Dict]) -> List[int]:
"""
Extract Tools from Tool ConfigurationID
Args:
tools: Tool Configuration List
Returns:
ToolsIDVertical
"""
tool_ids = []
for tool in tools:
if tool.get("children"):
tool_ids.extend(int(child.get("id")) for child in tool["children"] if child.get("id"))
return tool_ids
@classmethod
async def feedback_regenerate_sop_task(cls, session_version_model: LinsightSessionVersion,
feedback: str) -> None:
"""
Regenerate from feedbackSOPTask
Args:
session_version_model: Inspiration Conversation Version Model
feedback: Content of feedback
"""
try:
file_list = await cls.prepare_file_list(session_version_model)
# BuatLLMand tools
llm, workbench_conf = await cls._get_llm(session_version_model.user_id)
tools = await cls._prepare_tools(session_version_model, llm)
# Get a summary of your history
history_summary = await cls._get_history_summary(session_version_model.id)
# Create proxy and generateSOP
agent = await cls._create_linsight_agent(session_version_model, llm, tools, workbench_conf)
sop_content = ""
sop_template = session_version_model.sop or ''
async for res in agent.feedback_sop(
sop=sop_template,
feedback=feedback,
history_summary=history_summary if history_summary else None,
file_list=file_list
):
sop_content += res.content
# sopWrite it down on the record sheet, thissopThere is no need to associate sessions because there is no need to update scores
await SOPManageService.add_sop_record(LinsightSOPRecord(
name=session_version_model.title,
description=None,
user_id=session_version_model.user_id,
content=sop_content,
))
except cls.ToolsInitializationError as e:
logger.exception(f"Failed to initialize the Inspiration Workbench tool: session_version_id={session_version_model.id}, error={str(e)}")
except Exception as e:
logger.exception(f"Feedback regenerationSOPMisi Gagal: session_version_id={session_version_model.id}, error={str(e)}")
@classmethod
async def _get_history_summary(cls, session_version_id: str) -> List[str]:
"""Get a summary of your history"""
history_summary = []
execute_tasks = await LinsightExecuteTaskDao.get_by_session_version_id(session_version_id)
for task in execute_tasks:
if task.result:
answer = task.result.get("answer", "")
if answer:
history_summary.append(answer)
return history_summary
@classmethod
async def download_file(cls, file_info: DownloadFilesSchema) -> Tuple[str, bytes]:
"""Download individual files"""
minio_client = await get_minio_storage()
object_name = file_info.file_url
object_name = object_name.replace(f"/{minio_client.bucket}/", "")
try:
bytes_io = BytesIO()
file_byte = await minio_client.get_object(bucket_name=minio_client.bucket,
object_name=object_name)
bytes_io.write(file_byte)
bytes_io.seek(0)
return file_info.file_name, bytes_io.getvalue()
except Exception as e:
logger.error(f"Download failed {object_name}: {e}")
return object_name, b''
@classmethod
async def batch_download_files(cls, file_info_list: List[DownloadFilesSchema]) -> bytes:
"""
Batch Download Files
Args:
file_info_list: File Information List
Returns:
List containing file download information
"""
# Batch Download Files
download_tasks = [cls.download_file(file_info) for file_info in file_info_list]
results = await asyncio.gather(*download_tasks)
# Filter download failed files
successful_files = [res for res in results if res[1]]
if not successful_files:
raise ValueError("File not downloaded successfully, could not be generatedZIP")
zip_bytes = util.bytes_to_zip(successful_files)
return zip_bytes
================================================
FILE: src/backend/bisheng/linsight/domain/task_exec.py
================================================
import asyncio
import os
import shutil
import traceback
from contextlib import asynccontextmanager
from datetime import datetime
from typing import Optional, List, Dict, Callable
from langchain_core.language_models import BaseChatModel
from loguru import logger
from bisheng.api.services.invite_code.invite_code import InviteCodeService
from bisheng.linsight.domain.services.workbench_impl import LinsightWorkbenchImpl
from bisheng.linsight.domain.schemas.linsight_schema import UserInputEventSchema
from bisheng.common.services.config_service import settings
from bisheng.core.cache.utils import create_cache_folder_async, CACHE_DIR
from bisheng.core.external.http_client.http_client_manager import get_http_client
from bisheng.core.logger import trace_id_var
from bisheng.core.storage.minio.minio_manager import get_minio_storage
from bisheng.linsight.domain.models.linsight_execute_task import LinsightExecuteTaskDao, ExecuteTaskStatusEnum, \
ExecuteTaskTypeEnum, LinsightExecuteTask
from bisheng.linsight.domain.models.linsight_session_version import LinsightSessionVersionDao, SessionVersionStatusEnum, \
LinsightSessionVersion
from bisheng.linsight.domain import utils as linsight_execute_utils
from bisheng.linsight.domain.services.state_message_manager import LinsightStateMessageManager, MessageData, MessageEventType
from bisheng.llm.domain.services import LLMService
from bisheng.tool.domain.services.tool import ToolServices
from bisheng_langchain.linsight.agent import LinsightAgent
from bisheng_langchain.linsight.const import TaskStatus, ExecConfig
from bisheng_langchain.linsight.event import NeedUserInput, GenerateSubTask, ExecStep, TaskStart, TaskEnd, BaseEvent
class TaskExecutionError(Exception):
"""Task Execution Exception"""
pass
class UserTerminationError(Exception):
"""User proactively terminates exceptions"""
pass
# Task already in progress Exception
class TaskAlreadyInProgressError(Exception):
"""Task already in progress Exception"""
pass
class LinsightWorkflowTask:
"""Workflow Task Executor - Responsible for managing the entire mission lifecycle"""
USER_TERMINATION_CHECK_INTERVAL = 2
def __init__(self):
self._state_manager: Optional[LinsightStateMessageManager] = None
self._is_terminated = False
self._termination_task: Optional[asyncio.Task] = None
self._final_result: Optional[TaskEnd] = None
self.file_dir: Optional[str] = None
self.session_version_id: Optional[str] = None
self.step_event_extra_files: List[Dict] = [] # File information for storing additional processing of step events
self.llm: Optional[BaseChatModel] = None # For storageLLMInstances
# ==================== Resource Management ====================
@asynccontextmanager
async def _managed_execution(self):
"""Context manager for managing execution resources"""
self._state_manager = LinsightStateMessageManager(self.session_version_id)
session_model = await self._get_session_model(self.session_version_id)
# Check session status
if await self._is_session_in_progress(session_model):
raise TaskAlreadyInProgressError("Task already in progress")
try:
# Start Termination Monitoring
await self._start_termination_monitor(session_model)
# Initialization file directory
self.file_dir = await self._init_file_directory(session_model)
yield session_model
finally:
await self._cleanup_resources()
async def _cleanup_resources(self):
"""Clean up resources"""
try:
# Stop Terminating Monitoring
await self._stop_termination_monitor()
# Clean File Directory
if self.file_dir and os.path.exists(self.file_dir):
shutil.rmtree(self.file_dir, ignore_errors=True)
except Exception as e:
logger.error(f"Resource cleanup failed: {e}")
# ==================== Core Execution Logic ====================
async def async_run(self, session_version_id: str) -> None:
"""Asynchronous Task Execution Entry"""
self.session_version_id = session_version_id
trace_id_var.set(self.session_version_id)
logger.info(f"Start the task: session_version_id={self.session_version_id}")
try:
async with self._managed_execution() as session_model:
await self._execute_workflow(session_model)
except UserTerminationError:
logger.info(f"Task was actively terminated by the user: session_version_id={self.session_version_id}")
except TaskAlreadyInProgressError:
logger.warning(f"Task already in progress: session_version_id={self.session_version_id}")
except TaskExecutionError as e:
logger.error(f"Task execution failed: : session_version_id={self.session_version_id}")
await self._handle_execution_error(e)
except Exception as e:
logger.error(f"Unknown error: session_version_id={self.session_version_id}, error={e}")
await self._handle_execution_error(e)
async def _execute_workflow(self, session_model: LinsightSessionVersion):
"""Execute the core logic of the workflow"""
# Update session status to in progress
await self._update_session_status(session_model, SessionVersionStatusEnum.IN_PROGRESS)
# Initialization Execution Component
self.llm = await self._get_llm(invoke_user_id=session_model.user_id)
tools = await self._generate_tools(session_model)
try:
# Build Tool List
linsight_tools = await ToolServices.init_linsight_tools(root_path=self.file_dir)
tools.extend(linsight_tools)
# Create agent
agent = await self._create_agent(session_model, tools)
# Check if terminated during initialization
self._check_termination()
# Generate and save tasks
task_info = await agent.generate_task(session_model.sop)
await self._save_task_info(session_model, task_info)
# Do Task
success = await self._execute_agent_tasks(agent, task_info, session_model)
finally:
# Clean the sandbox of the code interpreter
for one in tools:
if one.name == "bisheng_code_interpreter":
one.close()
break
if success:
await self._handle_task_completion(session_model)
else:
await self._handle_user_termination(session_model)
raise UserTerminationError("Task terminated by user")
# ==================== Session and State Management ====================
async def _get_session_model(self, session_version_id: str) -> LinsightSessionVersion:
"""Get session model"""
try:
return await LinsightSessionVersionDao.get_by_id(session_version_id)
except Exception as e:
raise TaskExecutionError(f"Failed to get session model: {e}")
async def _is_session_in_progress(self, session_model: LinsightSessionVersion) -> bool:
"""Check if the session is already in progress"""
if session_model.status == SessionVersionStatusEnum.IN_PROGRESS:
logger.info(f"Sessions {session_model.id} Already in progress")
return True
return False
async def _update_session_status(self, session_model: LinsightSessionVersion, status: SessionVersionStatusEnum):
"""Update session status"""
session_model.status = status
await self._state_manager.set_session_version_info(session_model)
# ==================== Component Initialization ====================
async def _get_llm(self, invoke_user_id: int) -> BaseChatModel:
"""DapatkanLLMInstances"""
try:
workbench_conf = await LLMService.get_workbench_llm()
linsight_conf = settings.get_linsight_conf()
return await LLMService.get_bisheng_linsight_llm(invoke_user_id=invoke_user_id,
model_id=workbench_conf.task_model.id,
temperature=linsight_conf.default_temperature)
except Exception as e:
raise TaskExecutionError("The task has been terminated, please contact the administrator to check the status of the Ideas task execution model")
@create_cache_folder_async
async def _init_file_directory(self, session_model: LinsightSessionVersion) -> str:
"""Initialization file directory"""
file_dir = os.path.join(CACHE_DIR, "linsight", session_model.id[:8])
file_dir = os.path.normpath(file_dir)
os.makedirs(file_dir, exist_ok=True)
if not session_model.files:
return file_dir
# Concurrent downloads
download_tasks = [
self._download_file(file_info, file_dir)
for file_info in session_model.files
]
results = await asyncio.gather(*download_tasks, return_exceptions=True)
# Record Download Failed Files
for i, result in enumerate(results):
if isinstance(result, Exception):
file_name = os.path.basename(session_model.files[i]["markdown_file_path"])
logger.error(f"This content failed to load {file_name}: {result}")
return file_dir
async def _download_file(self, file_info: dict, target_dir: str) -> str:
"""Download individual files"""
object_name = file_info["markdown_file_path"]
file_name = file_info.get("markdown_filename", os.path.basename(object_name))
file_path = os.path.join(target_dir, file_name)
minio_client = await get_minio_storage()
try:
file_url = await minio_client.get_share_link(object_name, clear_host=False)
http_client = await get_http_client()
with open(file_path, "wb") as f:
async for chunk in http_client.stream(method="GET", url=str(file_url)):
f.write(chunk)
if not os.path.exists(file_path) or os.path.getsize(file_path) == 0:
raise ValueError(f"File download failed or empty: {object_name}")
return file_path
except Exception as e:
logger.error(f"Download failed {object_name}: {e}")
raise
async def _generate_tools(self, session_model: LinsightSessionVersion) -> List:
"""Build Tool List"""
if not session_model.tools:
return []
return await LinsightWorkbenchImpl.init_linsight_config_tools(session_version=session_model, llm=self.llm,
need_upload=True, file_dir=self.file_dir)
async def _create_agent(self, session_model: LinsightSessionVersion, tools: List) -> LinsightAgent:
workbench_conf = await LLMService.get_workbench_llm()
linsight_conf = settings.get_linsight_conf()
exec_config = ExecConfig(**linsight_conf.model_dump(), debug_id=session_model.id)
"""Create agent"""
return LinsightAgent(
llm=self.llm,
query=session_model.question,
tools=tools,
file_dir=self.file_dir,
task_mode=workbench_conf.linsight_executor_mode,
exec_config=exec_config,
)
# ==================== Mission Execution ====================
async def _save_task_info(self, session_model: LinsightSessionVersion, task_info: List[dict]):
"""Save Task Information"""
try:
tasks = []
# step_idIt doesn't have to be regular.step_int, FROMagentGot ittask_infoThe order is the execution order
sorted_data = task_info
for index, task_info in enumerate(sorted_data):
previous_task_id = sorted_data[index - 1]["id"] if index > 0 else None
next_task_id = sorted_data[index + 1]["id"] if index < len(sorted_data) - 1 else None
task = LinsightExecuteTask(
id=task_info["id"],
parent_task_id=task_info.get("parent_id"),
session_version_id=session_model.id,
previous_task_id=previous_task_id,
next_task_id=next_task_id,
task_type=ExecuteTaskTypeEnum.COMPOSITE if task_info.get(
"node_loop") else ExecuteTaskTypeEnum.SINGLE,
task_data=task_info
)
tasks.append(task)
await LinsightExecuteTaskDao.batch_create_tasks(tasks)
await self._state_manager.set_execution_tasks(tasks)
# Push Generate Task Message
await self._state_manager.push_message(MessageData(
event_type=MessageEventType.TASK_GENERATE,
data={"tasks": [task.model_dump() for task in tasks]}
))
logger.info(f"Set {len(tasks)} execution tasks")
except Exception as e:
logger.error(f"Failed to save task information: {e}")
raise TaskExecutionError(f"Failed to save task information: {e}")
async def _execute_agent_tasks(self, agent, task_info: List[dict],
session_model) -> bool:
"""Perform agent tasks - Modified version supports user termination"""
async def agent_execution():
"""Agent performs a task"""
file_list = await LinsightWorkbenchImpl.prepare_file_list(session_model)
async for event in agent.ainvoke(task_info, session_model.sop, file_list=file_list):
await self._handle_event(agent, event, session_model)
return True
async def termination_monitor():
"""End monitoring task"""
while True:
await asyncio.sleep(0.5) # Qn,0.5Seconds checked once
self._check_termination()
try:
# Create two concurrent tasks
# Preparing user-uploaded files
agent_task = asyncio.create_task(agent_execution())
monitor_task = asyncio.create_task(termination_monitor())
# Waiting for any one task to be completed
done, pending = await asyncio.wait(
[agent_task, monitor_task],
return_when=asyncio.FIRST_COMPLETED
)
# Cancel Incomplete Task
for task in pending:
task.cancel()
try:
await task
except asyncio.CancelledError:
logger.debug("Successfully unsuspended task")
pass
# Review completed task results
for task in done:
if task.exception():
if isinstance(task.exception(), UserTerminationError):
logger.info("Agent task terminated by user")
return False
else:
raise task.exception()
else:
# If the agent task is completed normally
if task == agent_task:
logger.info("The agent task is completed normally")
return True
return True
except UserTerminationError:
logger.info("Agent task terminated by user")
return False
except Exception as e:
logger.error(f"task_exec_error {traceback.format_exc()}")
raise TaskExecutionError(f"Agent task execution failed: {e}")
# ==================== Event processing ====================
async def _handle_event(self, agent: LinsightAgent, event: BaseEvent, session_model: LinsightSessionVersion):
"""handle incidents"""
event_handlers: Dict[type[BaseEvent], Callable] = {
GenerateSubTask: self._handle_generate_subtask,
TaskStart: self._handle_task_start,
TaskEnd: self._handle_task_end,
NeedUserInput: self._handle_need_user_input,
ExecStep: self._handle_exec_step,
}
handler = event_handlers.get(type(event))
if handler:
await handler(agent, event, session_model)
else:
logger.warning(f"Unknown event type: {type(event)}")
async def _handle_generate_subtask(self, agent: LinsightAgent, event: GenerateSubTask,
session_model: LinsightSessionVersion):
"""Handle build subtask events"""
await self._save_task_info(session_model, event.subtask)
logger.debug(f"Generate subtasks: {event}")
async def _handle_task_start(self, agent: LinsightAgent, event: TaskStart, session_model: LinsightSessionVersion):
"""Handle task start events"""
task_data = await self._state_manager.update_execution_task_status(
task_id=event.task_id,
status=ExecuteTaskStatusEnum.IN_PROGRESS
)
await self._state_manager.push_message(
MessageData(event_type=MessageEventType.TASK_START, data=task_data)
)
async def _handle_task_end(self, agent: LinsightAgent, event: TaskEnd, session_model: LinsightSessionVersion):
"""Handle task end events"""
status = ExecuteTaskStatusEnum.SUCCESS if event.status == TaskStatus.SUCCESS.value else ExecuteTaskStatusEnum.FAILED
task_data = await self._state_manager.update_execution_task_status(
task_id=event.task_id,
status=status,
result={"answer": event.answer},
task_data=event.data
)
await self._state_manager.push_message(
MessageData(event_type=MessageEventType.TASK_END, data=task_data)
)
# Save Final Result
self._final_result = event
async def _handle_need_user_input(self, agent: LinsightAgent, event: NeedUserInput,
session_model: LinsightSessionVersion):
"""Handle events that require user input"""
asyncio.create_task(self._wait_for_user_input(agent, event))
async def _handle_exec_step(self, agent: LinsightAgent, event: ExecStep, session_model: LinsightSessionVersion):
"""Handle execution step events"""
# Additional Processing Step Events
event = await linsight_execute_utils.handle_step_event_extra(event, self)
await self._state_manager.add_execution_task_step(event.task_id, step=event)
await self._state_manager.push_message(
MessageData(event_type=MessageEventType.TASK_EXECUTE_STEP, data=event.model_dump())
)
async def _wait_for_user_input(self, agent: LinsightAgent, event: NeedUserInput):
"""Waiting for user input"""
try:
await self._state_manager.add_execution_task_step(event.task_id, step=event)
# Update status to wait for user input
await self._state_manager.update_execution_task_status(
event.task_id,
status=ExecuteTaskStatusEnum.WAITING_FOR_USER_INPUT,
)
# Push user input events
await self._state_manager.push_message(
MessageData(event_type=MessageEventType.USER_INPUT, data=event.model_dump())
)
# Wait for user input to complete
task_model = await self._wait_for_input_completion(event.task_id)
user_input_event = task_model.history[-1] if task_model.history else None
if user_input_event is None or user_input_event.get("step_type") != "call_user_input":
raise TaskExecutionError(f"Task {event.task_id} No user input events found while waiting for user input")
user_input_event = UserInputEventSchema.model_validate(user_input_event)
# Check if the user uploaded a file
if user_input_event.files:
# Concurrent downloads
download_tasks = [
self._download_file(file_info, self.file_dir)
for file_info in user_input_event.files
]
results = await asyncio.gather(*download_tasks, return_exceptions=True)
# Record Download Failed Files
for i, result in enumerate(results):
if isinstance(result, Exception):
file_name = os.path.basename(user_input_event.files[i]["markdown_file_path"])
logger.error(f"This content failed to load {file_name}: {result}")
user_input_event.user_input += f"\n\nUploaded File:\n" + \
"\n".join([f"- {file}" for file in results if
not isinstance(file, Exception)])
# Push Input Completion Event
await self._state_manager.push_message(
MessageData(event_type=MessageEventType.USER_INPUT_COMPLETED, data=task_model.model_dump())
)
# Resume Task
await agent.continue_task(event.task_id, user_input_event.user_input)
except Exception as e:
raise TaskExecutionError(f"Failed to wait for user input task_id={event.task_id}: {e}")
async def _wait_for_input_completion(self, task_id: str) -> Optional[LinsightExecuteTask]:
"""Wait for user input to complete"""
while True:
self._check_termination()
task_model = await self._state_manager.get_execution_task(task_id)
if task_model is None:
raise ValueError(f"Task {task_id} Does not exist")
if task_model.status == ExecuteTaskStatusEnum.USER_INPUT_COMPLETED:
return task_model
await asyncio.sleep(1)
# ==================== Terminate Inspection ====================
def _check_termination(self):
"""Check if terminated"""
if self._is_terminated:
logger.info("Termination signal detected, ready to terminate agent task")
raise UserTerminationError("Task terminated by user")
async def _start_termination_monitor(self, session_model: LinsightSessionVersion):
"""Start Termination Monitoring"""
async def monitor():
while not self._is_terminated:
try:
if await self._check_user_termination():
self._is_terminated = True
break
await asyncio.sleep(self.USER_TERMINATION_CHECK_INTERVAL)
except Exception as e:
logger.error(f"Terminate Monitoring Exception: {e}")
await asyncio.sleep(self.USER_TERMINATION_CHECK_INTERVAL)
self._termination_task = asyncio.create_task(monitor())
async def _stop_termination_monitor(self):
"""Stop Terminating Monitoring"""
if self._termination_task and not self._termination_task.done():
self._termination_task.cancel()
try:
await self._termination_task
except asyncio.CancelledError:
pass
async def _check_user_termination(self) -> bool:
"""Check if the user is actively terminating"""
if self._is_terminated:
return True
try:
current_session = await self._state_manager.get_session_version_info()
return (current_session and
current_session.status == SessionVersionStatusEnum.TERMINATED)
except Exception as e:
logger.error(f"Failed to check user termination status: {e}")
return False
async def _handle_user_termination(self, session_model: LinsightSessionVersion):
"""Handling user unsolicited termination"""
logger.info(f"Handling User Termination {session_model.id}")
session_model.status = SessionVersionStatusEnum.TERMINATED
session_model.output_result = {"answer": "Task has been actively stopped by the user"}
await self._state_manager.set_session_version_info(session_model)
# Set all tasks to failed
await self._set_tasks_failed()
# Push termination message
await self._state_manager.push_message(
MessageData(
event_type=MessageEventType.TASK_TERMINATED,
data={
"message": "Task has been actively stopped by the user",
"session_id": session_model.id,
"terminated_at": datetime.now().isoformat()
}
)
)
# ==================== Task Completion Processing ====================
async def _handle_task_completion(self, session_model: LinsightSessionVersion):
"""Processing Task Completion"""
if not self._final_result:
logger.error("No final task results found")
return
if self._final_result.status == TaskStatus.SUCCESS.value:
await self._handle_task_success(session_model)
else:
await self._handle_task_failure(session_model, "Task execution failed: ")
async def _handle_task_success(self, session_model: LinsightSessionVersion):
"""Processing task successful"""
try:
# Read File Directory File Details
file_details = await linsight_execute_utils.read_file_directory(self.file_dir)
logger.debug(f"Read File Directory File Details: {file_details}")
final_result_files = await linsight_execute_utils.get_final_result_file(
session_model=session_model,
file_details=file_details,
answer=self._final_result.answer
)
execution_tasks = await self._state_manager.get_execution_tasks()
all_from_session_files = await linsight_execute_utils.get_all_files_from_session(
execution_tasks=execution_tasks, file_details=file_details)
# Update session status
session_model.status = SessionVersionStatusEnum.COMPLETED
session_model.output_result = {
"answer": self._final_result.answer,
"final_files": final_result_files,
"all_from_session_files": all_from_session_files
}
# Save session information and push messages
await self._state_manager.set_session_version_info(session_model)
await self._state_manager.push_message(
MessageData(
event_type=MessageEventType.FINAL_RESULT,
data=session_model.model_dump()
)
)
logger.info(f"Task completed successfully, processed {len(final_result_files)} files")
except Exception as e:
logger.error(f"An error occurred while processing the task successfully: {e}")
raise TaskExecutionError(f"An error occurred while processing the task successfully: {e}")
# Modify All Task Failure Processing Logic
async def _set_tasks_failed(self):
"""Set all tasks to failed"""
try:
# Get All Execute Tasks
execution_tasks = await self._state_manager.get_execution_tasks()
for task in execution_tasks:
# Update each task status to Terminated
if task.status not in [ExecuteTaskStatusEnum.TERMINATED, ExecuteTaskStatusEnum.SUCCESS,
ExecuteTaskStatusEnum.FAILED]:
await self._state_manager.update_execution_task_status(task_id=task.id,
status=ExecuteTaskStatusEnum.TERMINATED)
except Exception as e:
logger.warning(f"Error setting task failed: {e}")
async def _handle_task_failure(self, session_model: LinsightSessionVersion, error_msg: str):
"""Processing task failed"""
session_model.status = SessionVersionStatusEnum.FAILED
session_model.output_result = {"error_message": error_msg}
await self._state_manager.set_session_version_info(session_model)
# Set all tasks to failed
await self._set_tasks_failed()
await self._state_manager.push_message(
MessageData(event_type=MessageEventType.ERROR_MESSAGE, data={"error": error_msg})
)
system_config = await settings.aget_all_config()
# DapatkanLinsight_invitation_code
linsight_invitation_code = system_config.get("linsight_invitation_code", False)
if linsight_invitation_code:
await InviteCodeService.revoke_invite_code(user_id=session_model.user_id)
async def _handle_execution_error(self, error: Exception):
"""Processing execution error"""
try:
session_model = await LinsightSessionVersionDao.get_by_id(self.session_version_id)
await self._handle_task_failure(session_model, str(error))
except Exception as e:
logger.error(f"Processing execution error failed: session_version_id={self.session_version_id}, error={e}")
================================================
FILE: src/backend/bisheng/linsight/domain/utils.py
================================================
import asyncio
import os
import uuid
from typing import List, Dict, Any
from loguru import logger
from bisheng.api.services.invite_code.invite_code import InviteCodeService
from bisheng.common.services.config_service import settings
from bisheng.core.cache.redis_manager import get_redis_client_sync
from bisheng.core.storage.minio.minio_manager import get_minio_storage
from bisheng.linsight.domain.models.linsight_execute_task import LinsightExecuteTaskDao, ExecuteTaskStatusEnum, \
LinsightExecuteTask
from bisheng.linsight.domain.models.linsight_session_version import LinsightSessionVersionDao, SessionVersionStatusEnum, \
LinsightSessionVersion
from bisheng.utils import util
from bisheng_langchain.linsight.event import ExecStep
# The corresponding file parameter name of the Inscription file processing tool
local_file_tool_dict = {
"add_text_to_file": "file_path",
"replace_file_lines": "file_path"
}
# Step event extra processing tool corresponding to parameter name,
step_event_extra_tool_dict = {
"add_text_to_file": "file_path",
"replace_file_lines": "file_path",
"read_text_file": "file_path"
}
# Get all manipulated files in a task
async def get_all_files_from_session(execution_tasks: List[LinsightExecuteTask], file_details: List[Dict]) -> \
list[Any] | list[Exception | BaseException | None]:
"""
Get all manipulated files in a session
:param file_details:
:param execution_tasks: Execute Task List
:return: List with file details
"""
# Process File List
all_from_session_files = file_details
# Deduplication
seen = set()
all_from_session_files = [
file for file in all_from_session_files
if (file_tuple := (file["file_name"], file["file_path"], file["file_md5"])) not in seen and not seen.add(
file_tuple)
]
if not all_from_session_files:
logger.warning("No files found that were manipulated in the session")
return []
# Upload files toMinIO
async def upload_file_to_minio(file_info: Dict) -> dict | None:
"""Upload files toMinIOand returns file information"""
try:
minio_client = await get_minio_storage()
object_name = f"linsight/session_files/{execution_tasks[0].session_version_id}/{file_info['file_name']}"
# Use async upload if available, otherwise wrap sync call
await minio_client.put_object(
bucket_name=minio_client.bucket,
object_name=object_name,
file=file_info["file_path"]
)
file_info["file_url"] = object_name
return file_info
except Exception as e:
logger.error(f"Upload files toMinIOKalah {file_info['file_name']}: {e}")
return None
# Upload files in parallel toMinIO
upload_tasks = [
upload_file_to_minio(file_info)
for file_info in all_from_session_files
]
upload_results = await asyncio.gather(*upload_tasks, return_exceptions=True)
# Filter failed uploads
all_from_session_files = [
result for result in upload_results
if result is not None and not isinstance(result, Exception)
]
# Record failed uploads
failed_uploads = [
result for result in upload_results
if isinstance(result, Exception)
]
if failed_uploads:
logger.warning(f"Some files failed to upload: {len(failed_uploads)} files")
logger.debug(
f"Number of files manipulated in the session: {len(all_from_session_files)}Document Description: {all_from_session_files}")
return all_from_session_files
# Read File Directory File Details
async def read_file_directory(file_dir: str) -> List[Dict[str, str]]:
"""Read file details in file directory"""
if not file_dir or not os.path.exists(file_dir):
return []
files = util.read_files_in_directory(file_dir)
file_details = []
for file in files:
file_md5 = await util.async_calculate_md5(file)
file_details.append({
"file_name": os.path.basename(file),
"file_path": file,
"file_md5": file_md5,
"file_id": uuid.uuid4().hex[:8] # Generate unique filesID
})
return file_details
# Get the final result file
async def get_final_result_file(session_model: LinsightSessionVersion, file_details, answer) -> List[Dict]:
"""
Get the final result file
:param file_details:
:param session_model: LinsightSessionVersion Model Instance
:param answer: Answer content
:return: List containing final result file information
"""
# Final Result File
final_result_files = []
for file_info in file_details:
file_name: str = file_info["file_name"]
# Determine if the filename isanswerIn String
if file_name in answer:
# If the file name is in the answer, add it to the answer
final_result_files.append({
"file_name": file_name,
"file_path": file_info["file_path"],
"file_md5": file_info["file_md5"],
"file_id": file_info["file_id"]
})
async def upload_file_to_minio(final_file_info: Dict) -> dict | None:
"""Upload files toMinIOand returns file information"""
try:
object_name = f"linsight/final_result/{session_model.id}/{final_file_info['file_name']}"
# Use async upload if available, otherwise wrap sync call
minio_client = await get_minio_storage()
await minio_client.put_object(
bucket_name=minio_client.bucket,
object_name=object_name,
file=final_file_info["file_path"]
)
final_file_info["file_url"] = object_name
return final_file_info
except Exception as e:
logger.error(f"Upload files toMinIOKalah {final_file_info['file_name']}: {e}")
return None
# Upload files toMinIO (Parallel Processing)
if final_result_files:
upload_tasks = [
upload_file_to_minio(final_file_info)
for final_file_info in final_result_files
]
upload_results = await asyncio.gather(*upload_tasks, return_exceptions=True)
# Filter failed uploads
final_result_files = [
result for result in upload_results
if result is not None and not isinstance(result, Exception)
]
# Record failed uploads
failed_uploads = [
result for result in upload_results
if isinstance(result, Exception)
]
if failed_uploads:
logger.warning(f"Some files failed to upload: {len(failed_uploads)} files")
return final_result_files
# Additional Handling of Step Events
async def handle_step_event_extra(event: ExecStep, task_exec_obj) -> ExecStep:
"""
Additional logic for handling step events
:param task_exec_obj:
:param event: Event Object
"""
logger.debug(
f"extra processing of step events,call_id: {event.call_id}, name: {event.name}, status: {event.status}")
try:
if event.status == "end" and event.name in step_event_extra_tool_dict.keys():
file_path = event.params.get(step_event_extra_tool_dict[event.name], "")
if not file_path:
return event
file_name = os.path.basename(file_path)
logger.debug(f"Step event extra processing, filename: {file_name}")
# File path processing
if not os.path.isabs(file_path):
# relative paths, converting to absolute paths
file_path = os.path.join(task_exec_obj.file_dir, file_path)
file_path = os.path.normpath(file_path)
logger.debug(f"Step event extra processing, converted file path: {file_path}")
if not os.path.exists(file_path):
logger.error(f"Step event extra processing, file does not exist: {file_path}")
return event
file_md5 = await util.async_calculate_md5(file_path)
# Determine if the document has already been uploaded
step_event_extra_files = task_exec_obj.step_event_extra_files
if step_event_extra_files:
existing_file = next((f for f in step_event_extra_files if f["file_md5"] == file_md5), None)
if existing_file:
logger.debug(
f"Step event extra processing, file already exists: {existing_file['file_name']}, file_md5: {file_md5}")
event.extra_info["file_info"] = {
"file_name": file_name,
"file_md5": existing_file["file_md5"],
"file_url": existing_file["file_url"]
}
return event
object_name = f"linsight/step_event/{task_exec_obj.session_version_id}/{uuid.uuid4().hex[:8]}.{file_name.split('.')[-1]}"
logger.debug(f"Extra processing of step events, uploading files toMinIO: {object_name}")
minio_client = await get_minio_storage()
# Upload files toMinIO
await minio_client.put_object(
bucket_name=minio_client.bucket,
object_name=object_name,
file=file_path
)
event.extra_info["file_info"] = {
"file_name": file_name,
"file_md5": file_md5,
"file_url": object_name
}
# Add to Step Event Extra File List
task_exec_obj.step_event_extra_files.append(event.extra_info["file_info"])
except Exception as e:
logger.error(f"Step event extra handling exception: {e}")
# When an exception occurs, return to the original event without modification
return event
# Initiateworkerwhen checking for incomplete tasks and terminating
async def check_and_terminate_incomplete_tasks(node_id: str) -> None:
"""
Check for incomplete tasks and terminate
"""
from bisheng.linsight.worker import NodeManager
redis_client = get_redis_client_sync() # Get Redis Client
node_manager = NodeManager(redis_client, node_id) # Get Node Manager Instance
try:
# Get all incomplete tasks from the database
incomplete_tasks = await LinsightSessionVersionDao.get_session_versions_by_status(
status=SessionVersionStatusEnum.IN_PROGRESS
)
if not incomplete_tasks:
return
tasks_to_terminate = []
user_ids_to_rollback = set()
for session in incomplete_tasks:
session_id = session.id
# Check task ownership in Redis
owner_key = f"linsight:task:owner:{session_id}"
owner_node_id = await redis_client.get(owner_key)
should_terminate = False
if not owner_node_id:
# No owned node, task needs to be cleaned up.
should_terminate = True
logger.warning(f"Task {session_id} has no owner in Redis. Marking as failed.")
else:
# There is an owned node, check if the node is alive
is_alive = await node_manager.is_node_alive(owner_node_id)
if not is_alive:
# Node is dead, task needs to be cleaned up.
should_terminate = True
logger.warning(f"Task {session_id} owner {owner_node_id} is dead. Marking as failed.")
else:
# Node is alive, skip cleanup
logger.info(f"Task {session_id} is running on active node {owner_node_id}. Skipping.")
if should_terminate:
tasks_to_terminate.append(session_id)
user_ids_to_rollback.add(session.user_id)
# 3. 批量执行终止操作（只针对筛选出的任务）
if tasks_to_terminate:
await LinsightSessionVersionDao.batch_update_session_versions_status(
session_version_ids=tasks_to_terminate,
status=SessionVersionStatusEnum.FAILED,
output_result={"error_message": "Worker node crash detected"}
)
# 更新 execution task 状态
await LinsightExecuteTaskDao.batch_update_status_by_session_version_id(
session_version_ids=tasks_to_terminate,
status=ExecuteTaskStatusEnum.FAILED,
where=(
LinsightExecuteTask.status != ExecuteTaskStatusEnum.SUCCESS,
LinsightExecuteTask.status != ExecuteTaskStatusEnum.FAILED
)
)
logger.warning(f"Terminated {len(tasks_to_terminate)} incomplete tasks due to worker node crash.")
system_config = await settings.aget_all_config()
# DapatkanLinsight_invitation_code
linsight_invitation_code = system_config.get("linsight_invitation_code", False)
# Rollback invite code
if linsight_invitation_code:
for user_id in user_ids_to_rollback:
try:
await InviteCodeService.revoke_invite_code(user_id=user_id)
logger.info(f"User Rolled Back {user_id} Invitation code for")
except Exception as e:
logger.error(f"Rollback user {user_id} Invitation code failed for: {e}")
else:
logger.warning(
"Not enabled in system configuration Linsight Invitation code function, skip rollback operation")
logger.info("Check and terminate incomplete task action completed")
except Exception as e:
logger.error(f"Exception occurred while checking and terminating incomplete tasks: {e}")
return
================================================
FILE: src/backend/bisheng/linsight/worker.py
================================================
import argparse
import asyncio
import logging
import socket
import uuid
from multiprocessing import Process, Manager, set_start_method
from multiprocessing.managers import ValueProxy
from typing import Optional, Union
from bisheng.core.cache.redis_conn import RedisClient
from bisheng.core.cache.redis_manager import get_redis_client_sync
from bisheng.core.logger import set_logger_config
from bisheng.common.services.config_service import settings
from bisheng.linsight.domain.task_exec import LinsightWorkflowTask
logger = logging.getLogger(__name__)
class NodeManager:
_instance = None
def __init__(self, redis_client, node_id):
# generate unique node ID
self.node_id = node_id
self.redis = redis_client
self.heartbeat_key = f"linsight:node:heartbeat:{self.node_id}"
# Heartbeat interval (seconds)
self.interval = 5
# Redis key expiration time (seconds)
self.ttl = 15
@classmethod
def get_instance(cls, node_id):
if not cls._instance:
redis_client = get_redis_client_sync()
cls._instance = cls(redis_client, node_id)
return cls._instance
async def start_heartbeat(self):
"""Start the heartbeat task to indicate node liveness"""
logger.info(f"Starting heartbeat for node: {self.node_id}")
while True:
try:
# set heartbeat key with expiration
await self.redis.aset(self.heartbeat_key, "1", expiration=self.ttl)
except Exception as e:
logger.error(f"Heartbeat failed: {e}")
await asyncio.sleep(self.interval)
async def register_task_ownership(self, session_version_id):
"""Register task ownership to this node"""
key = f"linsight:task:owner:{session_version_id}"
# Set the node ID as the owner of the task with a TTL
await self.redis.aset(key, self.node_id, expiration=86400) # 1 day expiration
async def release_task_ownership(self, session_version_id):
"""Release task ownership"""
key = f"linsight:task:owner:{session_version_id}"
await self.redis.delete(key)
async def is_node_alive(self, target_node_id):
"""Check if a target node is alive based on its heartbeat"""
if not target_node_id:
return False
key = f"linsight:node:heartbeat:{target_node_id}"
exists = await self.redis.exists(key)
return exists > 0
# LinsightQueue queue
class LinsightQueue(object):
def __init__(self, name, namespace, redis):
self.__db: RedisClient = redis
self.key = '%s:%s' % (namespace, name)
async def qsize(self):
return await self.__db.allen(self.key) # Back to queuelistNumber of inner elements
async def put(self, data, timeout=None):
await self.__db.arpush(self.key, data, expiration=timeout) # Add a new element to the far right of the queue
async def get_wait(self, timeout=None):
# Returns the first element of the queue, if empty, wait until an element is queued (the timeout threshold istimeout, if isNonehas been waiting)
item = await self.__db.ablpop(self.key, timeout=timeout)
return item
async def get_nowait(self):
# Returns the first element of the queue directly, if the queue is emptyNone
item = await self.__db.alpop(self.key)
return item
# Get the position of a task's data in the queue
async def index(self, data):
"""
Get the position of a task's data in the queue
:param data: Task Data
:return: Position in queue, starting from 1; if not found, return 0
"""
items = await self.__db.alrange(self.key)
try:
index = items.index(data)
return index + 1 # Return index from1Getting Started
except ValueError:
return 0
# Delete a task data
async def remove(self, data):
"""
Delete a task data from the queue
:param data: Task Data
:return:
"""
await self.__db.alrem(self.key, data) # Remove the specified data from the queue
class ScheduleCenterProcess(Process):
def __init__(self, max_concurrency: ValueProxy = None, node_id: ValueProxy = None):
"""
Dispatch Center Process
:param max_concurrency: Maximum number of concurrent tasks allowed per process
"""
super().__init__()
self.daemon = True
self.queue: Optional[LinsightQueue] = None
# Semaphores
self.semaphore: Optional[asyncio.Semaphore] = None
self.node_manager: Optional[NodeManager] = None
self.max_concurrency: Optional[Union[int, ValueProxy]] = max_concurrency
self.node_id: Optional[ValueProxy] = node_id
def handle_task_result(self, task: asyncio.Task):
try:
result = task.result() # If there is an exception, it will be thrown here
except Exception as e:
logger.error(f"Task failed with exception: {e}")
finally:
# Release semaphore
if self.semaphore:
logger.info("Releasing semaphore after task completion.")
self.semaphore.release()
async def async_run(self):
"""
Asynchronous Run Method for Process
:return:
"""
logger.info("ScheduleCenterProcess started...")
# node manager heartbeat
node_manager = self.node_manager or NodeManager.get_instance(self.node_id.value)
while True:
await self.semaphore.acquire() # Acquire semaphore, limit concurrency
try:
session_version_id = await self.queue.get_wait()
if session_version_id is None:
logger.info("No session_version_id found in queue, waiting...")
self.semaphore.release()
continue
# Register task ownership
await node_manager.register_task_ownership(session_version_id)
exec_task = LinsightWorkflowTask()
logger.info(f"Processing session_version_id: {session_version_id} on node {node_manager.node_id}")
task = asyncio.create_task(
exec_task.async_run(session_version_id)
)
task.add_done_callback(self.handle_task_result) # Add callback to handle task completion
except Exception as e:
logger.error(f"Error in ScheduleCenterProcess: {e}")
if self.semaphore:
if self.semaphore._value < self.max_concurrency:
logger.info("Releasing semaphore due to error.")
self.semaphore.release()
continue
def run(self):
"""
Run Process
:return:
"""
set_logger_config(settings.logger_conf)
if self.max_concurrency is not None:
self.max_concurrency = self.max_concurrency.value # Dapatkan ValueProxy Actual value
else:
self.max_concurrency = 32
logger.warning("No max_concurrency provided, using default value of 32.")
self.semaphore = asyncio.Semaphore(self.max_concurrency)
logger.info(f"Semaphore initialized with max concurrency: {self.semaphore._value}")
redis_client = get_redis_client_sync()
self.queue = LinsightQueue('queue', namespace="linsight", redis=redis_client)
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
# 启动心跳
self.node_manager = NodeManager.get_instance(self.node_id.value)
loop.create_task(self.node_manager.start_heartbeat())
# 启动主逻辑
for _ in range(10000): # 你的原始逻辑
try:
loop.run_until_complete(self.async_run())
except Exception as e:
logger.error(f"Unhandled exception in main loop: {e}")
loop.close()
def start_schedule_center_process(worker_num: int = 4, max_concurrency: ValueProxy = None, node_id: ValueProxy = None):
"""
Start Schedule Center Process Workers
:param worker_num: Number of worker processes to start
:param max_concurrency: Maximum number of concurrent tasks allowed per process
:return:
Args:
node_id:
"""
logger.info(f"Starting {worker_num} ScheduleCenterProcess workers...")
if worker_num <= 0:
logger.error("worker_num must be greater than 0")
return
processes = []
for _ in range(worker_num):
process = ScheduleCenterProcess(max_concurrency, node_id)
process.start()
logger.info(f"Started ScheduleCenterProcess with PID: {process.pid}")
processes.append(process)
logger.info(f"Started {len(processes)} ScheduleCenterProcess workers successfully.")
return processes
if __name__ == '__main__':
set_start_method("spawn", force=True) # make sure that people are using the spawn Method Starts a New Process
parser = argparse.ArgumentParser()
parser.add_argument('--worker_num', type=int, default=4, help='Number of processes, defaults to4')
# Maximum number of concurrency for a single process
parser.add_argument('--max_concurrency', type=int, default=32,
help='Maximum number of concurrency for a single process, defaults to32')
args = parser.parse_args()
process_manager = Manager()
node_id = process_manager.Value('s', f"{socket.gethostname()}-{uuid.uuid4().hex[:8]}")
max_concurrency = process_manager.Value('i', args.max_concurrency)
# Check for incomplete tasks and terminate
from bisheng.linsight.domain.utils import check_and_terminate_incomplete_tasks
asyncio.run(check_and_terminate_incomplete_tasks(node_id.value))
try:
processes = start_schedule_center_process(worker_num=args.worker_num,
max_concurrency=max_concurrency,
node_id=node_id)
if processes:
for p in processes:
p.join() # Wait for all processes to end
except KeyboardInterrupt:
logger.info("ScheduleCenterProcess interrupted by user.")
logger.info("Stopping ScheduleCenterProcess workers...")
================================================
FILE: src/backend/bisheng/llm/README.md
================================================
# 模型管理模块
================================================
FILE: src/backend/bisheng/llm/__init__.py
================================================
================================================
FILE: src/backend/bisheng/llm/api/__init__.py
================================================
================================================
FILE: src/backend/bisheng/llm/api/router.py
================================================
from fastapi import APIRouter, Request, Depends, Body, Query, BackgroundTasks, UploadFile
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.schemas.api import resp_200, UnifiedResponseModel
from ..domain import LLMService
from ..domain.schemas import KnowledgeLLMConfig, AssistantLLMConfig, EvaluationLLMConfig, LLMServerCreateReq, \
WorkbenchModelConfig
router = APIRouter(prefix='/llm', tags=['LLM'])
@router.get('')
async def get_all_llm(request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user)):
ret = await LLMService.get_all_llm()
return resp_200(data=ret)
@router.post('')
async def add_llm_server(request: Request, login_user: UserPayload = Depends(UserPayload.get_admin_user),
server: LLMServerCreateReq = Body(..., description="Service Provider All Data")):
ret = await LLMService.add_llm_server(request, login_user, server)
return resp_200(data=ret)
@router.delete('')
async def delete_llm_server(request: Request, login_user: UserPayload = Depends(UserPayload.get_admin_user),
server_id: int = Body(..., embed=True, description="Service Provider UniqueID")):
await LLMService.delete_llm_server(request, login_user, server_id)
return resp_200()
@router.put('')
async def update_llm_server(request: Request, login_user: UserPayload = Depends(UserPayload.get_admin_user),
server: LLMServerCreateReq = Body(..., description="Service Provider All Data")):
ret = await LLMService.update_llm_server(request, login_user, server)
return resp_200(data=ret)
@router.get('/info')
async def get_one_llm(request: Request, login_user: UserPayload = Depends(UserPayload.get_admin_user),
server_id: int = Query(..., description="Service Provider UniqueID")):
ret = await LLMService.get_one_llm(server_id)
return resp_200(data=ret)
@router.post('/online')
async def update_model_online(request: Request, login_user: UserPayload = Depends(UserPayload.get_admin_user),
model_id: int = Body(..., embed=True, description="Model UniqueID"),
online: bool = Body(..., embed=True, description="Online or not")):
ret = await LLMService.update_model_online(model_id, online)
return resp_200(data=ret)
@router.get('/workbench', summary="Get workbench-related model configurations", response_model=UnifiedResponseModel)
async def get_workbench_llm():
""" Get Idea-Related Model Configurations """
ret = await LLMService.get_workbench_llm()
return resp_200(data=ret)
@router.post('/workbench', summary="Update workbench related model configurations", response_model=UnifiedResponseModel)
async def update_workbench_llm(
background_tasks: BackgroundTasks,
login_user: UserPayload = Depends(UserPayload.get_admin_user),
config_obj: WorkbenchModelConfig = Body(..., description="Model Configuration Object")):
""" Update Idea-Related Model Configurations """
ret = await LLMService.update_workbench_llm(login_user.user_id, config_obj, background_tasks)
return resp_200(data=ret)
@router.post('/workbench/asr')
async def invoke_workbench_asr(request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user),
file: UploadFile = None):
""" Call the workbench'sasrModels Convert Voice to Text """
text = await LLMService.invoke_workbench_asr(login_user, file)
return resp_200(data=text)
@router.post('/workbench/tts')
async def invoke_workbench_tts(request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user),
text: str = Body(..., embed=True, description="Text that needs to be synthesized")):
""" Call the workbench'sttsModels Convert text to speech """
audio_url = await LLMService.invoke_workbench_tts(login_user, text)
return resp_200(data=audio_url)
@router.get('/knowledge')
def get_knowledge_llm(request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user)):
ret = LLMService.get_knowledge_llm()
return resp_200(data=ret)
@router.post('/knowledge')
async def update_knowledge_llm(request: Request, login_user: UserPayload = Depends(UserPayload.get_admin_user),
data: KnowledgeLLMConfig = Body(..., description="Knowledge Base Default Model Configuration")):
""" Update default model configuration for knowledge base """
ret = await LLMService.update_knowledge_llm(data)
return resp_200(data=ret)
@router.get('/assistant')
async def get_assistant_llm(request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user)):
""" Get assistant related model configuration """
ret = await LLMService.get_assistant_llm()
return resp_200(data=ret)
@router.post('/assistant')
async def update_assistant_llm(request: Request, login_user: UserPayload = Depends(UserPayload.get_admin_user),
data: AssistantLLMConfig = Body(..., description="Assistant Default Model Configuration")):
""" Update assistant related model configurations """
ret = await LLMService.update_assistant_llm(data)
return resp_200(data=ret)
@router.get('/evaluation')
async def get_evaluation_llm(request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user)):
""" Get evaluation related model configurations """
ret = await LLMService.get_evaluation_llm()
return resp_200(data=ret)
@router.post('/evaluation')
async def update_evaluation_llm(request: Request, login_user: UserPayload = Depends(UserPayload.get_admin_user),
data: EvaluationLLMConfig = Body(..., description="Evaluate default model configuration")):
""" Update review related model configurations """
ret = await LLMService.update_evaluation_llm(data)
return resp_200(data=ret)
@router.get('/workflow')
async def get_workflow_llm(request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user)):
""" Get workflow-related model configurations """
ret = await LLMService.get_workflow_llm()
return resp_200(data=ret)
@router.post('/workflow')
async def update_workflow_llm(request: Request, login_user: UserPayload = Depends(UserPayload.get_admin_user),
data: EvaluationLLMConfig = Body(..., description="Workflow default model configuration")):
""" Update workflow-related model configurations """
ret = await LLMService.update_workflow_llm(data)
return resp_200(data=ret)
@router.get('/assistant/llm_list')
async def get_assistant_llm_list(request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user)):
""" Get a list of optional models for the assistant """
ret = await LLMService.get_assistant_llm_list(request, login_user)
return resp_200(data=ret)
================================================
FILE: src/backend/bisheng/llm/domain/__init__.py
================================================
from .services.llm import LLMService
__all__ = [
'LLMService'
]
================================================
FILE: src/backend/bisheng/llm/domain/const.py
================================================
# Some constants or enums for the module
from enum import Enum
# Service Provider Enumeration
class LLMServerType(Enum):
OPENAI = 'openai'
AZURE_OPENAI = 'azure_openai'
OLLAMA = 'ollama'
XINFERENCE = 'xinference'
LLAMACPP = 'llamacpp'
VLLM = 'vllm'
QWEN = 'qwen' # Ali Tongyi Qianqian
QIAN_FAN = 'qianfan' # Baidu Qianfan
ZHIPU = 'zhipu' # Zhi Spectrum Qing Yan
MINIMAX = 'minimax'
ANTHROPIC = 'anthropic'
DEEPSEEK = 'deepseek'
SPARK = 'spark' # Xunfei Starfire Large Model
BISHENG_RT = 'bisheng_rt'
TENCENT = 'tencent' # Tencent Cloud
MOONSHOT = 'moonshot' # Dark Side of the Moonkimi
VOLCENGINE = 'volcengine' # Large model of a volcanic engine
SILICON = 'silicon' # Silicon-based flow
MIND_IE = 'MindIE' # Ascendant Inference Engine MindIE
# Model Type Enumeration
class LLMModelType(Enum):
LLM = 'llm'
EMBEDDING = 'embedding'
RERANK = 'rerank'
ASR = 'asr'
TTS = 'tts'
class LLMModelStatus(Enum):
NORMAL = 0 # Normal
ERROR = 1 # Abnormal
UNKNOWN = 2 # Unknown
================================================
FILE: src/backend/bisheng/llm/domain/llm/__init__.py
================================================
from .asr import BishengASR
from .base import BishengBase
from .embedding import BishengEmbedding
from .llm import BishengLLM
from .tts import BishengTTS
__all__ = [
'BishengBase',
'BishengEmbedding',
'BishengLLM',
'BishengTTS',
'BishengASR',
]
================================================
FILE: src/backend/bisheng/llm/domain/llm/asr.py
================================================
from typing import Union, BinaryIO
import httpx
from pydantic import Field
from typing_extensions import Self
from bisheng.common.errcode.server import NoAsrModelConfigError, AsrModelConfigDeletedError, AsrModelTypeError, \
AsrProviderDeletedError, \
AsrModelOfflineError
from bisheng.core.ai import BaseASRClient, OpenAIASRClient, AliyunASRClient, AzureOpenAIASRClient
from bisheng.llm.domain.const import LLMModelType, LLMServerType
from bisheng.llm.domain.models import LLMModel, LLMServer
from .base import BishengBase
from ..utils import wrapper_bisheng_model_limit_check_async
async def _get_openai_params(params: dict, server_info: LLMServer, model_info: LLMModel) -> dict:
new_params = {
"api_key": params.get("openai_api_key"),
"model": model_info.model_name,
}
if params.get("openai_api_base"):
new_params["base_url"] = params["openai_api_base"]
if params.get("openai_proxy"):
new_params["http_client"] = httpx.AsyncClient(proxy=params["openai_proxy"])
return new_params
async def _get_azure_openai_params(params: dict, server_info: LLMServer, model_info: LLMModel) -> dict:
new_params = {
"api_key": params.get("openai_api_key"),
"api_version": params.get("openai_api_version"),
"azure_endpoint": params.get("azure_endpoint"),
"model": model_info.model_name,
}
return new_params
async def _get_qwen_params(params: dict, server_info: LLMServer, model_info: LLMModel) -> dict:
new_params = {
"api_key": params.get("openai_api_key") or params.get("api_key"),
"model": model_info.model_name,
}
return new_params
_asr_client_type = {
LLMServerType.OPENAI.value: {
"client": OpenAIASRClient,
"params_handler": _get_openai_params
},
LLMServerType.AZURE_OPENAI.value: {
"client": AzureOpenAIASRClient,
"params_handler": _get_azure_openai_params
},
LLMServerType.QWEN.value: {
"client": AliyunASRClient,
"params_handler": _get_qwen_params
}
}
class BishengASR(BishengBase):
asr: BaseASRClient = Field(..., description="asrInstances")
@classmethod
async def get_bisheng_asr(cls, **kwargs) -> Self:
model_id = kwargs.pop('model_id', 0)
if not model_id:
raise NoAsrModelConfigError()
model_info, server_info = await cls.get_model_server_info(model_id)
# ignore_onlineParameters are used to skip model presence checks
ignore_online = kwargs.pop('ignore_online', False)
if not model_info:
raise AsrModelConfigDeletedError()
if model_info.model_type != LLMModelType.ASR.value:
raise AsrModelTypeError(model_type=model_info.model_type)
if not server_info:
raise AsrProviderDeletedError()
if not ignore_online and not model_info.online:
raise AsrModelOfflineError(server_name=server_info.name, model_name=model_info.model_name)
# InisialisasiasrClient
asr_client = await cls.init_asr_client(model_info=model_info, server_info=server_info)
return cls(model_id=model_id, asr=asr_client, model_info=model_info, server_info=server_info, **kwargs)
@classmethod
async def init_asr_client(cls, model_info: LLMModel, server_info: LLMServer) -> BaseASRClient:
params = {}
if server_info.config:
if server_info.config:
params.update(server_info.config)
if model_info.config:
params.update(model_info.config)
if server_info.type not in _asr_client_type:
raise Exception(f'asrModel not supported{server_info.type}Type of service provider')
params_handler = _asr_client_type[server_info.type]['params_handler']
new_params = await params_handler(params, server_info, model_info)
client = _asr_client_type[server_info.type]['client'](**new_params)
return client
@wrapper_bisheng_model_limit_check_async
async def ainvoke(self, audio: Union[str, bytes, BinaryIO], **kwargs) -> str:
return await self.asr.transcribe(audio=audio, **kwargs)
================================================
FILE: src/backend/bisheng/llm/domain/llm/base.py
================================================
from typing import Optional, Dict
from pydantic import BaseModel, Field, ConfigDict
from typing_extensions import Self
from bisheng.common.constants.enums.telemetry import ApplicationTypeEnum
from ..models import LLMModel, LLMServer, LLMDao
class BishengBase(BaseModel):
model_config = ConfigDict(arbitrary_types_allowed=True, validate_by_name=True, validate_by_alias=True)
model_id: int = Field(description="Saved by backend servicemodelUniqueness quantificationID")
model_name: str = Field(default='', description='model name in mysql')
# field for telemetry logging
app_id: str = Field(..., description='application id')
app_type: ApplicationTypeEnum = Field(..., description='application type')
app_name: str = Field(..., description='application name')
user_id: int = Field(..., description='invoke user id')
# bishengStrongly related business parameters
model_info: Optional[LLMModel] = Field(default=None, description="Model Configuration Information")
server_info: Optional[LLMServer] = Field(default=None, description="Service Provider Information")
@classmethod
async def get_class_instance(cls, **kwargs: Dict) -> Self:
model_id: int | None = kwargs.pop('model_id', None)
model_info, server_info = await cls.get_model_server_info(model_id)
instance = cls(
model_id=model_id,
model_name=model_info.model_name,
model_info=model_info,
server_info=server_info,
**kwargs
)
return instance
@classmethod
async def get_model_server_info(cls, model_id: int | None) -> tuple[LLMModel | None, LLMServer | None]:
model_info = None
server_info = None
if not model_id:
return model_info, server_info
model_info = await LLMDao.aget_model_by_id(model_id)
if model_info:
server_info = await LLMDao.aget_server_by_id(model_info.server_id)
return model_info, server_info
@classmethod
def get_model_server_info_sync(cls, model_id: int | None) -> tuple[LLMModel | None, LLMServer | None]:
model_info = None
server_info = None
if not model_id:
return model_info, server_info
model_info = LLMDao.get_model_by_id(model_id)
if model_info:
server_info = LLMDao.get_server_by_id(model_info.server_id)
return model_info, server_info
async def update_model_status(self, status: int, remark: str = ''):
"""Update model status"""
if self.model_info.status != status:
self.model_info.status = status
await LLMDao.aupdate_model_status(self.model_id, status, remark[-500:]) # Limit note length to500characters.
def sync_update_model_status(self, status: int, remark: str = ''):
"""Update model status"""
if self.model_info.status != status:
self.model_info.status = status
LLMDao.update_model_status(self.model_id, status, remark[-500:])
def get_server_info_config(self):
if self.server_info and self.server_info.config:
return self.server_info.config
return {}
def get_model_info_config(self):
if self.model_info and self.model_info.config:
return self.model_info.config
return {}
================================================
FILE: src/backend/bisheng/llm/domain/llm/embedding.py
================================================
import json
from typing import Optional, Dict, List
import numpy as np
from langchain_core.embeddings import Embeddings
from loguru import logger
from pydantic import Field
from typing_extensions import Self
from bisheng.core.ai import OllamaEmbeddings, OpenAIEmbeddings, AzureOpenAIEmbeddings, DashScopeEmbeddings
from bisheng.llm.domain.const import LLMServerType, LLMModelType
from .base import BishengBase
from ..models import LLMModel, LLMServer
from ..utils import wrapper_bisheng_model_limit_check
def _get_user_kwargs(model_config: dict) -> dict:
user_kwargs = model_config.get('user_kwargs', {})
if isinstance(user_kwargs, str) and user_kwargs:
return json.loads(user_kwargs)
return user_kwargs if user_kwargs else {}
def _get_ollama_params(params: dict, server_config: dict, model_config: dict) -> dict:
params['base_url'] = server_config.get('base_url', '').rstrip('/')
user_kwargs = _get_user_kwargs(model_config)
user_kwargs.update(params)
return user_kwargs
def _get_openai_params(params: dict, server_config: dict, model_config: dict) -> dict:
if server_config:
params.update({
'api_key': server_config.get('openai_api_key') or server_config.get('api_key') or 'empty',
'base_url': server_config.get('openai_api_base', '') or server_config.get('base_url', ''),
})
params['base_url'] = params['base_url'].rstrip('/')
if server_config.get('openai_proxy'):
params['openai_proxy'] = server_config.get('openai_proxy')
params["check_embedding_ctx_length"] = False
user_kwargs = _get_user_kwargs(model_config)
user_kwargs.update(params)
return user_kwargs
def _get_azure_openai_params(params: dict, server_config: dict, model_config: dict) -> dict:
params.update({
'azure_endpoint': server_config.get('azure_endpoint').rstrip('/'),
'openai_api_key': server_config.get('openai_api_key'),
'openai_api_version': server_config.get('openai_api_version'),
'azure_deployment': params.pop('model'),
})
user_kwargs = _get_user_kwargs(model_config)
user_kwargs.update(params)
return user_kwargs
def _get_qwen_params(params: dict, server_config: dict, model_config: dict) -> dict:
params['dashscope_api_key'] = server_config.get('openai_api_key', '')
user_kwargs = _get_user_kwargs(model_config)
user_kwargs.update(params)
return user_kwargs
_node_type: Dict = {
# Open source inference framework
LLMServerType.OLLAMA.value: {"client": OllamaEmbeddings, "params_handler": _get_ollama_params},
LLMServerType.XINFERENCE.value: {"client": OpenAIEmbeddings, "params_handler": _get_openai_params},
LLMServerType.LLAMACPP.value: {"client": OpenAIEmbeddings, "params_handler": _get_openai_params},
LLMServerType.VLLM.value: {"client": OpenAIEmbeddings, "params_handler": _get_openai_params},
# OfficalAPISERVICES
LLMServerType.OPENAI.value: {"client": OpenAIEmbeddings, "params_handler": _get_openai_params},
LLMServerType.AZURE_OPENAI.value: {"client": AzureOpenAIEmbeddings, "params_handler": _get_azure_openai_params},
LLMServerType.QWEN.value: {"client": DashScopeEmbeddings, "params_handler": _get_qwen_params},
LLMServerType.QIAN_FAN.value: {"client": OpenAIEmbeddings, "params_handler": _get_openai_params},
LLMServerType.MINIMAX.value: {"client": OpenAIEmbeddings, "params_handler": _get_openai_params},
LLMServerType.ZHIPU.value: {"client": OpenAIEmbeddings, "params_handler": _get_openai_params},
LLMServerType.TENCENT.value: {"client": OpenAIEmbeddings, "params_handler": _get_openai_params},
LLMServerType.VOLCENGINE.value: {"client": OpenAIEmbeddings, "params_handler": _get_openai_params},
LLMServerType.SILICON.value: {"client": OpenAIEmbeddings, "params_handler": _get_openai_params},
}
class BishengEmbedding(BishengBase, Embeddings):
""" Use the embedding model that has been launched in model management """
embedding_ctx_length: int = Field(default=8192, description='embeddingModel Context Length')
max_retries: int = Field(default=6, description='embeddingNumber of failed model call retries')
request_timeout: int = Field(default=200, description='embeddingModel Call Timeout')
model_kwargs: dict = Field(default={}, description='embeddingModel Call Parameters')
embeddings: Optional[Embeddings] = Field(default=None)
@classmethod
async def get_bisheng_embedding(cls, **kwargs) -> Self:
return await cls.get_class_instance(**kwargs)
def __init__(self, **kwargs):
super().__init__(**kwargs)
self.model_id = kwargs.get('model_id')
if not self.model_id:
raise Exception('No bulkpost found in Trashembeddingmodel config')
if "model_info" in kwargs and "server_info" in kwargs:
self._init_client(model_info=kwargs.pop('model_info'), server_info=kwargs.pop('server_info'), **kwargs)
else:
model_info, server_info = self.get_model_server_info_sync(self.model_id)
self._init_client(model_info=model_info, server_info=server_info, **kwargs)
def _init_client(self, model_info, server_info, **kwargs):
ignore_online = kwargs.get('ignore_online', False)
if not model_info:
raise Exception('embeddingModel configuration has been deleted, please reconfigure the model')
if not server_info:
raise Exception('Service provider configuration has been deleted, please reconfigureembeddingModels')
if model_info.model_type != LLMModelType.EMBEDDING.value:
raise Exception(f'Support onlyEmbeddingModel of type, not supported{model_info.model_type}Type of model')
if not ignore_online and not model_info.online:
raise Exception(f'{server_info.name}under{model_info.model_name}The model is offline, please contact the administrator to launch the corresponding model')
logger.debug(
f'init_bisheng_embedding: server_id: {server_info.id}, model_id: {model_info.id}')
self.model_info: LLMModel = model_info
self.server_info: LLMServer = server_info
self.model_name = model_info.model_name
class_object = self._get_embedding_class(server_info.type)
params = self._get_embedding_params(server_info, **kwargs)
try:
self.embeddings = class_object(**params)
except Exception as e:
logger.exception('init_bisheng_embedding error')
raise Exception(f'Inisialisasibisheng embeddingComponent failed, please check the configuration or contact the administrator.Error message:{e}')
@staticmethod
def _get_embedding_class(server_type: str) -> type[Embeddings]:
node_type = _node_type.get(server_type)
if not node_type:
raise Exception(f'{server_type}Type of service provider is not supportedembedding')
class_object = node_type.get('client')
return class_object
def _get_default_params(self, server_config: dict, model_config: dict, **kwargs) -> dict:
params = {
"model": self.model_info.model_name,
}
return params
def _get_embedding_params(self, server_info: LLMServer, **kwargs) -> dict:
server_config = self.get_server_info_config()
model_config = self.get_model_info_config()
default_params = self._get_default_params(server_config, model_config, **kwargs)
params_handler = _node_type[server_info.type]['params_handler']
params = params_handler(default_params, server_config, model_config)
return params
@wrapper_bisheng_model_limit_check
def embed_documents(self, texts: List[str]) -> List[List[float]]:
"""embedding"""
ret = self.embeddings.embed_documents(texts)
# Whether the disc single vector is normalized
if ret:
vector = ret[0]
if np.linalg.norm(vector) != 1:
ret = [(np.array(doc) / np.linalg.norm(doc)).tolist() for doc in ret]
return ret
@wrapper_bisheng_model_limit_check
def embed_query(self, text: str) -> List[float]:
"""embedding"""
ret = self.embeddings.embed_query(text)
if np.linalg.norm(ret) != 1:
ret = (np.array(ret) / np.linalg.norm(ret)).tolist()
return ret
================================================
FILE: src/backend/bisheng/llm/domain/llm/llm.py
================================================
import json
from typing import List, Optional, Any, Sequence, Union, Dict, Type, Callable, Iterator, AsyncIterator
from langchain_core.callbacks import AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun
from langchain_core.language_models import BaseChatModel, LanguageModelInput
from langchain_core.messages import BaseMessage, ToolMessage, HumanMessage, BaseMessageChunk, AIMessage
from langchain_core.outputs import ChatResult, ChatGenerationChunk
from langchain_core.runnables import Runnable
from langchain_core.tools import BaseTool
from langchain_core.utils.function_calling import convert_to_openai_tool
from loguru import logger
from pydantic import Field
from typing_extensions import Self
from bisheng.common.errcode.server import NoLlmModelConfigError, LlmModelConfigDeletedError, LlmProviderDeletedError, \
LlmModelTypeError, LlmModelOfflineError, InitLlmError
from bisheng.core.ai import ChatOllama, ChatOpenAI, ChatOpenAICompatible, \
AzureChatOpenAI, ChatZhipuAI, MiniMaxChat, ChatAnthropic, MoonshotChat
from bisheng.core.ai.llm.custom_chat_deepseek import CustomChatDeepSeek
from bisheng.core.ai.llm.custom_chat_tongyi import CustomChatTongYi
from bisheng.llm.domain.const import LLMModelType, LLMServerType
from bisheng.llm.domain.models import LLMServer, LLMModel
from .base import BishengBase
from ..utils import wrapper_bisheng_model_limit_check, wrapper_bisheng_model_limit_check_async, \
wrapper_bisheng_model_generator, wrapper_bisheng_model_generator_async
def _get_user_kwargs(model_config: dict) -> dict:
user_kwargs = model_config.get('user_kwargs', {})
if isinstance(user_kwargs, str) and user_kwargs:
return json.loads(user_kwargs)
return user_kwargs if user_kwargs else {}
# Attention needs to be paid to the priority of the initialization parameters. Instantiation Incoming Highest -> The following configurations of the front-end interface -> Advanced parameters of the front-end interface have the lowest priority
def _get_ollama_params(params: dict, server_config: dict, model_config: dict) -> dict:
params['base_url'] = server_config.get('base_url', '').rstrip('/')
# some bugs
params['stream'] = params.pop('streaming', True)
if params.get('max_tokens'):
params['num_ctx'] = params.pop('max_tokens', None)
# User advanced custom configuration
user_kwargs = _get_user_kwargs(model_config)
user_kwargs.update(params)
return user_kwargs
def _get_xinference_params(params: dict, server_config: dict, model_config: dict) -> dict:
params = _get_openai_params(params, server_config, model_config)
if not params.get('api_key', None):
params['api_key'] = 'Empty'
user_kwargs = _get_user_kwargs(model_config)
user_kwargs.update(params)
return user_kwargs
def _get_openai_params(params: dict, server_config: dict, model_config: dict) -> dict:
if server_config:
params.update({
'api_key': server_config.get('openai_api_key') or server_config.get('api_key') or "empty",
'base_url': server_config.get('openai_api_base') or server_config.get('base_url'),
})
params['base_url'] = params['base_url'].rstrip('/')
if server_config.get('openai_proxy'):
params['openai_proxy'] = server_config.get('openai_proxy')
params['stream_usage'] = True
user_kwargs = _get_user_kwargs(model_config)
user_kwargs.update(params)
return user_kwargs
def _get_azure_openai_params(params: dict, server_config: dict, model_config: dict) -> dict:
params.update({
'azure_endpoint': server_config.get('azure_endpoint').rstrip('/'),
'openai_api_key': server_config.get('openai_api_key'),
'openai_api_version': server_config.get('openai_api_version'),
'azure_deployment': params.pop('model'),
'stream_usage': True,
})
user_kwargs = _get_user_kwargs(model_config)
user_kwargs.update(params)
return user_kwargs
def _get_qwen_params(params: dict, server_config: dict, model_config: dict) -> dict:
params['dashscope_api_key'] = server_config.get('openai_api_key', '')
params['model_kwargs'] = {
'enable_search': model_config.get('enable_web_search', False),
}
if params.get("streaming"):
params['model_kwargs']['incremental_output'] = True
if params.get('temperature'):
params['model_kwargs']['temperature'] = params.pop('temperature')
if params.get('max_tokens'):
params['model_kwargs']['max_tokens'] = params.pop('max_tokens')
user_kwargs = _get_user_kwargs(model_config)
if user_model_kwargs := user_kwargs.get('model_kwargs'):
user_model_kwargs.update(params['model_kwargs'])
params['model_kwargs'] = user_model_kwargs
user_kwargs.pop('model_kwargs', None)
user_kwargs.update(params)
return user_kwargs
def _get_minimax_params(params: dict, server_config: dict, model_config: dict) -> dict:
params['minimax_api_key'] = server_config.get('openai_api_key')
params['base_url'] = server_config.get('openai_api_base').rstrip('/')
if 'max_tokens' not in params:
params['max_tokens'] = 2048
if '/chat/completions' not in params['base_url']:
params['base_url'] = f"{params['base_url']}/chat/completions"
user_kwargs = _get_user_kwargs(model_config)
user_kwargs.update(params)
return user_kwargs
def _get_anthropic_params(params: dict, server_config: dict, model_config: dict) -> dict:
params.update(server_config)
user_kwargs = _get_user_kwargs(model_config)
user_kwargs.update(params)
return user_kwargs
def _get_zhipu_params(params: dict, server_config: dict, model_config: dict) -> dict:
params['zhipuai_api_key'] = server_config.get('openai_api_key')
params['zhipuai_api_base'] = server_config.get('openai_api_base').rstrip('/')
if 'chat/completions' not in params['zhipuai_api_base']:
params['zhipuai_api_base'] = f"{params['zhipuai_api_base'].rstrip('/')}/chat/completions"
user_kwargs = _get_user_kwargs(model_config)
user_kwargs.update(params)
return params
def _get_spark_params(params: dict, server_config: dict, model_config: dict) -> dict:
params.update({
'api_key': f'{server_config.get("api_key")}:{server_config.get("api_secret")}',
'base_url': server_config.get('openai_api_base').rstrip('/'),
})
user_kwargs = _get_user_kwargs(model_config)
user_kwargs.update(params)
return params
_llm_node_type: Dict = {
# Open source inference framework
LLMServerType.OLLAMA.value: {'client': ChatOllama, 'params_handler': _get_ollama_params},
LLMServerType.XINFERENCE.value: {'client': ChatOpenAI, 'params_handler': _get_xinference_params},
LLMServerType.LLAMACPP.value: {'client': ChatOpenAI, 'params_handler': _get_openai_params},
LLMServerType.VLLM.value: {'client': ChatOpenAICompatible, 'params_handler': _get_openai_params},
# OfficalapiSERVICES
LLMServerType.OPENAI.value: {'client': ChatOpenAICompatible, 'params_handler': _get_openai_params},
LLMServerType.AZURE_OPENAI.value: {'client': AzureChatOpenAI, 'params_handler': _get_azure_openai_params},
LLMServerType.QWEN.value: {'client': CustomChatTongYi, 'params_handler': _get_qwen_params},
LLMServerType.QIAN_FAN.value: {'client': ChatOpenAICompatible, 'params_handler': _get_openai_params},
LLMServerType.ZHIPU.value: {'client': ChatZhipuAI, 'params_handler': _get_zhipu_params},
LLMServerType.MINIMAX.value: {'client': MiniMaxChat, 'params_handler': _get_minimax_params},
LLMServerType.ANTHROPIC.value: {'client': ChatAnthropic, 'params_handler': _get_anthropic_params},
LLMServerType.DEEPSEEK.value: {'client': CustomChatDeepSeek, 'params_handler': _get_openai_params},
LLMServerType.SPARK.value: {'client': ChatOpenAICompatible, 'params_handler': _get_spark_params},
LLMServerType.TENCENT.value: {'client': ChatOpenAICompatible, 'params_handler': _get_openai_params},
LLMServerType.MOONSHOT.value: {'client': MoonshotChat, 'params_handler': _get_openai_params},
LLMServerType.VOLCENGINE.value: {'client': ChatOpenAICompatible, 'params_handler': _get_openai_params},
LLMServerType.SILICON.value: {'client': ChatOpenAICompatible, 'params_handler': _get_openai_params},
LLMServerType.MIND_IE.value: {'client': ChatOpenAICompatible, 'params_handler': _get_openai_params},
}
class BishengLLM(BishengBase, BaseChatModel):
"""
Use the llm model that has been launched in model management
"""
streaming: Optional[bool] = Field(default=None, description="Whether to use streaming output", alias="stream")
temperature: Optional[float] = Field(default=None, description="Model Generated Temperature")
llm: Optional[BaseChatModel] = Field(default=None)
@classmethod
async def get_bisheng_llm(cls, **kwargs) -> Self:
return await cls.get_class_instance(**kwargs)
def __init__(self, **kwargs):
super().__init__(**kwargs)
self.model_id = kwargs.get('model_id')
if not self.model_id:
raise NoLlmModelConfigError()
if "model_info" in kwargs and "server_info" in kwargs:
# Description is fromclass methodInitialized No need to query the database again
self._init_client(model_info=kwargs.pop("model_info"), server_info=kwargs.pop("server_info"), **kwargs)
else:
model_info, server_info = self.get_model_server_info_sync(self.model_id)
self._init_client(model_info, server_info, **kwargs)
def _init_client(self, model_info: LLMModel, server_info: LLMServer, **kwargs):
ignore_online = kwargs.get('ignore_online', False)
self.temperature = kwargs.get('temperature', None)
self.streaming = kwargs.get('streaming', None)
if not model_info:
raise LlmModelConfigDeletedError()
if not server_info:
raise LlmProviderDeletedError()
if model_info.model_type != LLMModelType.LLM.value:
raise LlmModelTypeError(model_type=model_info.model_type)
if not ignore_online and not model_info.online:
raise LlmModelOfflineError(server_name=server_info.name, model_name=model_info.model_name)
logger.debug(f'init_bisheng_llm: server_id: {server_info.id}, model_id: {model_info.id}')
self.model_info = model_info
self.model_name = model_info.model_name
self.server_info = server_info
class_object = self._get_llm_class(server_info.type)
params = self._get_llm_params(server_info, model_info)
try:
self.llm = class_object(**params)
except Exception as e:
logger.exception('init bisheng llm error')
raise InitLlmError(exception=e)
@staticmethod
def _get_llm_class(server_type: str) -> type[BaseChatModel]:
if server_type not in _llm_node_type:
raise Exception(f'not support llm type: {server_type}')
class_type = _llm_node_type[server_type]['client']
return class_type
def _get_llm_params(self, server_info: LLMServer, model_info: LLMModel) -> dict:
server_config = self.get_server_info_config()
model_config = self.get_model_info_config()
default_params = self._get_default_params(server_config, model_config)
params_handler = _llm_node_type[server_info.type]['params_handler']
params = params_handler(default_params, server_config, model_config)
return params
def _get_default_params(self, server_config: dict, model_config: dict) -> dict:
# Highest priority parameters because they are passed in when the object is instantiated
default_params: Dict[str, Any] = {
'model': self.model_info.model_name,
}
# The highest priority is passed in the instantiation, followed by the advanced configuration. Otherwise defaults totrue
if self.streaming is not None:
default_params['streaming'] = self.streaming
else:
user_kwargs = _get_user_kwargs(model_config)
if user_kwargs.get('streaming', None) is None:
default_params['streaming'] = True
else:
default_params['streaming'] = user_kwargs.get('streaming', None)
if self.temperature is not None:
default_params['temperature'] = self.temperature
if model_config.get('max_tokens'):
default_params['max_tokens'] = model_config.get('max_tokens')
return default_params
@property
def _llm_type(self):
return self.llm._llm_type
def parse_kwargs(self, messages: List[BaseMessage], kwargs: Dict[str, Any]) -> (List[BaseMessage], Dict[str, Any]):
if self.server_info.type == LLMServerType.MINIMAX.value:
if self.get_model_info_config().get('enable_web_search'):
if 'tools' not in kwargs:
kwargs.update({
'tools': [{'type': 'web_search'}],
})
else:
tool_exists = False
for tool in kwargs['tools']:
if tool.get('type') == 'web_search':
tool_exists = True
break
if not tool_exists:
kwargs['tools'].append({
'type': 'web_search',
})
elif self.server_info.type == LLMServerType.MOONSHOT.value:
if self.get_model_info_config().get('enable_web_search'):
if 'tools' not in kwargs:
kwargs.update({
'tools': [{
"type": "builtin_function",
"function": {
"name": "$web_search",
},
}],
})
else:
tool_exists = False
for tool in kwargs['tools']:
if tool.get('type') == 'builtin_function':
tool_exists = True
break
if not tool_exists:
kwargs['tools'].append({
"type": "builtin_function",
"function": {
"name": "$web_search",
},
})
# elif self.server_info.type == LLMServerType.QWEN.value:
# # ChatTongYi The input parameters for multimodality are special and need to be converted to support
# user_message = messages[-1]
# if isinstance(user_message, HumanMessage):
# if isinstance(user_message.content, list):
# for one in user_message.content:
# if one.get('type') == 'image' and one.get('data'):
# one['type'] = 'image'
# one['image'] = f"data:{one.get('mime_type')};{one.get('source_type')},{one.get('data')}"
# elif one.get('type') == 'image_url' and one.get('image_url'):
# one['type'] = 'image'
# one['image'] = one.pop('image_url', {}).get('url')
return messages, kwargs
@wrapper_bisheng_model_limit_check
def _generate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
stream: Optional[bool] = None,
**kwargs: Any,
) -> ChatResult:
messages, kwargs = self.parse_kwargs(messages, kwargs)
if self.server_info.type == LLMServerType.MOONSHOT.value:
ret = self.moonshot_generate(messages, stop, run_manager, **kwargs)
else:
ret = self.llm._generate(messages, stop, run_manager, **kwargs)
if self.server_info.type == LLMServerType.QWEN.value:
ret.generations[0].message = self.convert_qwen_result(ret.generations[0].message)
return ret
def moonshot_generate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
result = None
finish_reason = None
while finish_reason is None or finish_reason == 'tool_calls':
result = self.llm._generate(messages, stop, run_manager, **kwargs)
result_message = result.generations[0].message
finish_reason = result.generations[0].generation_info.get('finish_reason')
for tool_call in result_message.tool_calls:
tool_call_name = tool_call['name']
if tool_call_name == "$web_search":
messages.append(result_message)
messages.append(ToolMessage(
tool_call_id=tool_call['id'],
name=tool_call_name,
content=json.dumps(tool_call['args'], ensure_ascii=False),
))
else:
break
return result
@wrapper_bisheng_model_limit_check_async
async def _agenerate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
stream: Optional[bool] = None,
**kwargs: Any,
) -> ChatResult:
messages, kwargs = self.parse_kwargs(messages, kwargs)
if self.server_info.type == LLMServerType.MOONSHOT.value:
ret = await self.moonshot_agenerate(messages, stop, run_manager, **kwargs)
else:
ret = await self.llm._agenerate(messages, stop, run_manager, **kwargs)
if self.server_info.type == LLMServerType.QWEN.value:
ret.generations[0].message = self.convert_qwen_result(ret.generations[0].message)
return ret
async def moonshot_agenerate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
result = None
finish_reason = None
while finish_reason is None or finish_reason == 'tool_calls':
result = await self.llm._agenerate(messages, stop, run_manager, **kwargs)
result_message = result.generations[0].message
finish_reason = result.generations[0].generation_info.get('finish_reason')
for tool_call in result_message.tool_calls:
tool_call_name = tool_call['name']
if tool_call_name == "$web_search":
messages.append(result_message)
messages.append(ToolMessage(
tool_call_id=tool_call['id'],
name=tool_call_name,
content=json.dumps(tool_call['args'], ensure_ascii=False),
))
else:
break
return result
def bind_tools(
self,
tools: Sequence[Union[Dict[str, Any], Type, Callable, BaseTool]],
**kwargs: Any,
) -> Runnable[LanguageModelInput, BaseMessage]:
formatted_tools = [convert_to_openai_tool(tool) for tool in tools]
return super().bind(tools=formatted_tools, **kwargs)
def convert_qwen_result(self, message: BaseMessageChunk | BaseMessage) -> BaseMessageChunk | BaseMessage:
# ChatTongYi model vl model message.content is list
if isinstance(message.content, list):
message.content = ''.join([one.get('text', '') for one in message.content])
return message
@wrapper_bisheng_model_generator
def _stream(
self,
messages: list[BaseMessage],
stop: Optional[list[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> Iterator[ChatGenerationChunk]:
for one in self.llm._stream(messages, stop=stop, run_manager=run_manager, **kwargs):
if self.server_info.type == LLMServerType.QWEN.value:
one.message = self.convert_qwen_result(one.message)
yield one
@wrapper_bisheng_model_generator_async
async def _astream(
self,
messages: list[BaseMessage],
stop: Optional[list[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> AsyncIterator[ChatGenerationChunk]:
async for one in self.llm._astream(messages, stop=stop, run_manager=run_manager, **kwargs):
if self.server_info.type == LLMServerType.QWEN.value:
one.message = self.convert_qwen_result(one.message)
yield one
================================================
FILE: src/backend/bisheng/llm/domain/llm/rerank.py
================================================
from typing import Dict, Optional, Sequence
import dashscope
from langchain_core.callbacks import Callbacks
from langchain_core.documents import BaseDocumentCompressor, Document
from loguru import logger
from pydantic import Field
from typing_extensions import Self
from bisheng.core.ai import XinferenceRerank, CommonRerank, DashScopeRerank
from bisheng.llm.domain.const import LLMServerType, LLMModelType
from .llm import BishengBase
from ..utils import wrapper_bisheng_model_limit_check, wrapper_bisheng_model_limit_check_async
def _get_xinference_params(params: dict, server_config: dict, model_config: dict) -> dict:
params['base_url'] = server_config.get('openai_api_base') or server_config.get('base_url')
params['api_key'] = server_config.get('openai_api_key') or server_config.get('api_key') or 'empty'
params['model_uid'] = params.pop('model', '')
return params
def _get_common_params(params: dict, server_config: dict, model_config: dict) -> dict:
params['base_url'] = server_config.get('openai_api_base') or server_config.get('base_url')
params['api_key'] = server_config.get('openai_api_key') or server_config.get('api_key') or 'empty'
params['rerank_endpoint'] = '/rerank'
return params
def _get_qwen_params(params: dict, server_config: dict, model_config: dict) -> dict:
params['dashscope_api_key'] = server_config.get('openai_api_key', '')
params['top_n'] = None # return all documents
params["client"] = dashscope.TextReRank
return params
_node_type: Dict = {
# Open source inference framework
LLMServerType.XINFERENCE.value: {'client': XinferenceRerank, 'params_handler': _get_xinference_params},
LLMServerType.LLAMACPP.value: {'client': CommonRerank, 'params_handler': _get_common_params},
LLMServerType.VLLM.value: {'client': CommonRerank, 'params_handler': _get_common_params},
# OfficalapiSERVICES
LLMServerType.QWEN.value: {'client': DashScopeRerank, 'params_handler': _get_qwen_params},
LLMServerType.QIAN_FAN.value: {'client': CommonRerank, 'params_handler': _get_common_params},
LLMServerType.SILICON.value: {'client': CommonRerank, 'params_handler': _get_common_params},
}
class BishengRerank(BishengBase, BaseDocumentCompressor):
"""Rerank LLM wrapper class"""
rerank: Optional[BaseDocumentCompressor] = Field(None, description="Rerank client instance")
@classmethod
async def get_bisheng_rerank(cls, **kwargs) -> Self:
return await cls.get_class_instance(**kwargs)
def __init__(self, **kwargs):
super().__init__(**kwargs)
self.model_id = kwargs.get('model_id')
if not self.model_id:
raise Exception('No bulkpost found in Trashrerankmodel config')
if "model_info" in kwargs and "server_info" in kwargs:
self._init_client(model_info=kwargs.pop('model_info'), server_info=kwargs.pop('server_info'), **kwargs)
else:
model_info, server_info = self.get_model_server_info_sync(self.model_id)
self._init_client(model_info=model_info, server_info=server_info, **kwargs)
def _init_client(self, model_info, server_info, **kwargs):
ignore_online = kwargs.get('ignore_online', False)
if not model_info:
raise Exception('rerankModel configuration has been deleted, please reconfigure the model')
if not server_info:
raise Exception('Service provider configuration has been deleted, please reconfigurererankModels')
if model_info.model_type != LLMModelType.RERANK.value:
raise Exception(f'Support onlyRerankModel of type, not supported{model_info.model_type}Type of model')
if not ignore_online and not model_info.online:
raise Exception(f'{server_info.name}under{model_info.model_name}The model is offline, please contact the administrator to launch the corresponding model')
logger.debug(f'init_bisheng_rerank: server_id: {server_info.id}, model_id: {model_info.id}')
node_conf = _node_type.get(server_info.type)
if not node_conf:
raise Exception(f'Not supportedrerankService Provider:{server_info.type}')
self.model_info = model_info
self.model_name = model_info.model_name
self.server_info = server_info
params_handler = node_conf['params_handler']
client_class = node_conf['client']
params = {
"model": self.model_name,
}
params = params_handler(params, self.get_server_info_config(), self.get_model_info_config())
try:
self.rerank = client_class(**params)
except Exception as e:
logger.exception('init_bisheng_rerank error')
raise Exception(f'init bisheng rerank error，error msg：{e}')
@wrapper_bisheng_model_limit_check
def compress_documents(
self,
documents: Sequence[Document],
query: str,
callbacks: Optional[Callbacks] = None,
) -> Sequence[Document]:
return self.rerank.compress_documents(documents, query, callbacks=callbacks)
@wrapper_bisheng_model_limit_check_async
async def acompress_documents(
self,
documents: Sequence[Document],
query: str,
callbacks: Optional[Callbacks] = None,
) -> Sequence[Document]:
return await self.rerank.acompress_documents(documents, query, callbacks=callbacks)
================================================
FILE: src/backend/bisheng/llm/domain/llm/tts.py
================================================
import httpx
from pydantic import Field
from typing_extensions import Self
from bisheng.common.errcode.server import NoTtsModelConfigError, TtsModelConfigDeletedError, \
TtsModelTypeError, TtsProviderDeletedError, TtsModelOfflineError
from bisheng.core.ai import BaseTTSClient, OpenAITTSClient, \
AliyunTTSClient, AzureOpenAITTSClient
from bisheng.llm.domain.const import LLMModelType, LLMServerType
from bisheng.llm.domain.models import LLMServer, LLMModel
from .base import BishengBase
from ..utils import wrapper_bisheng_model_limit_check_async
async def _get_openai_params(params: dict, server_info: LLMServer, model_info: LLMModel) -> dict:
new_params = {
"api_key": params.get("openai_api_key"),
"model": model_info.model_name,
"base_url": params.get("openai_base_url"),
}
if params.get("openai_api_base"):
new_params["base_url"] = params["openai_api_base"]
if params.get("voice"):
new_params["voice"] = params["voice"]
if params.get("openai_proxy"):
new_params["http_client"] = httpx.AsyncClient(proxy=params["openai_proxy"])
return new_params
async def _get_azure_openai_params(params: dict, server_info: LLMServer, model_info: LLMModel) -> dict:
new_params = {
"api_key": params.get("openai_api_key"),
"api_version": params.get("openai_api_version"),
"azure_endpoint": params.get("azure_endpoint"),
"model": model_info.model_name,
}
if params.get("voice"):
new_params["voice"] = params["voice"]
return new_params
async def _get_qwen_params(params: dict, server_info: LLMServer, model_info: LLMModel) -> dict:
new_params = {
"api_key": params.get("openai_api_key") or params.get("api_key"),
"model": model_info.model_name,
}
if params.get("voice"):
new_params["voice"] = params["voice"]
return new_params
_tts_client_type = {
LLMServerType.OPENAI.value: {
"client": OpenAITTSClient,
"params_handler": _get_openai_params
},
LLMServerType.AZURE_OPENAI.value: {
"client": AzureOpenAITTSClient,
"params_handler": _get_azure_openai_params
},
LLMServerType.QWEN.value: {
"client": AliyunTTSClient,
"params_handler": _get_qwen_params
}
}
class BishengTTS(BishengBase):
tts: BaseTTSClient = Field(..., description="ttsInstances")
@classmethod
async def get_bisheng_tts(cls, **kwargs) -> Self:
model_id = kwargs.pop('model_id', 0)
if not model_id:
raise NoTtsModelConfigError()
model_info, server_info = await cls.get_model_server_info(model_id)
# ignore_onlineParameters are used to skip model presence checks
ignore_online = kwargs.get('ignore_online', False)
if not model_info:
raise TtsModelConfigDeletedError()
if model_info.model_type != LLMModelType.TTS.value:
raise TtsModelTypeError(model_type=model_info.model_type)
if not server_info:
raise TtsProviderDeletedError()
if not ignore_online and not model_info.online:
raise TtsModelOfflineError(server_name=server_info.name, model_name=model_info.model_name)
# InisialisasittsClient
tts_client = await cls.init_tts_client(model_info=model_info, server_info=server_info)
return cls(model_id=model_id, tts=tts_client, model_info=model_info, server_info=server_info, **kwargs)
@classmethod
async def init_tts_client(cls, model_info: LLMModel, server_info: LLMServer) -> BaseTTSClient:
params = {}
if server_info.config:
if server_info.config:
params.update(server_info.config)
if model_info.config:
params.update(model_info.config)
if server_info.type not in _tts_client_type:
raise Exception(f'TtsModel not supported{server_info.type}Type of service provider')
params_handler = _tts_client_type[server_info.type]['params_handler']
new_params = await params_handler(params, server_info, model_info)
client = _tts_client_type[server_info.type]['client'](**new_params)
return client
@wrapper_bisheng_model_limit_check_async
async def ainvoke(self, text: str, **kwargs) -> bytes:
return await self.tts.synthesize(text, **kwargs)
================================================
FILE: src/backend/bisheng/llm/domain/models/__init__.py
================================================
from .llm_server import LLMDao, LLMServer, LLMModel, LLMModelBase, LLMServerBase
__all__ = [
"LLMDao",
"LLMServer",
"LLMModel",
"LLMModelBase",
"LLMServerBase",
]
================================================
FILE: src/backend/bisheng/llm/domain/models/llm_server.py
================================================
from datetime import datetime
from typing import Dict, List, Optional
from sqlalchemy import CHAR, JSON, Column, DateTime, Text, UniqueConstraint, delete, text, update
from sqlmodel import Field, select, col
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session, get_async_db_session
from bisheng.llm.domain.const import LLMModelType
class LLMServerBase(SQLModelSerializable):
name: str = Field(default='', index=True, unique=True, description='Service name')
description: Optional[str] = Field(default='', sa_column=Column(Text), description='Service Description')
type: str = Field(sa_column=Column(CHAR(20)), description='Service Provider Type')
limit_flag: bool = Field(default=False, description='Whether to turn on the daily call limit')
limit: int = Field(default=0, description='Daily call limit')
config: Optional[Dict] = Field(default=None, sa_column=Column(JSON), description='Service Provider Public Configuration')
user_id: int = Field(default=0, description='creatorID')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class LLMModelBase(SQLModelSerializable):
server_id: Optional[int] = Field(default=None, nullable=False, index=True, description='SERVICESID')
name: str = Field(default='', description='Model Display Name')
description: Optional[str] = Field(default='', sa_column=Column(Text), description='Model Description')
model_name: str = Field(default='', description='Model name, parameters used when instantiating components')
model_type: str = Field(sa_column=Column(CHAR(20)), description='model type')
config: Optional[Dict] = Field(default=None, sa_column=Column(JSON), description='Service Provider Public Configuration')
status: int = Field(default=2, description='Model status.0Normal1abnormal:, 2: Unknown')
remark: Optional[str] = Field(default='', sa_column=Column(Text), description='Abnormal reason')
online: bool = Field(default=True, description='Online')
user_id: int = Field(default=0, description='creatorID')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class LLMServer(LLMServerBase, table=True):
__tablename__ = 'llm_server'
id: Optional[int] = Field(default=None, nullable=False, primary_key=True, description='Service UniqueID')
class LLMModel(LLMModelBase, table=True):
__tablename__ = 'llm_model'
__table_args__ = (UniqueConstraint('server_id', 'model_name', name='server_model_uniq'),)
id: Optional[int] = Field(default=None, nullable=False, primary_key=True, description='Model UniqueID')
class LLMDao:
@classmethod
def get_all_server(cls) -> List[LLMServer]:
""" Get all service providers """
statement = select(LLMServer).order_by(col(LLMServer.update_time).desc())
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
async def aget_all_server(cls) -> List[LLMServer]:
""" Get all providers asynchronously """
statement = select(LLMServer).order_by(col(LLMServer.update_time).desc())
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def insert_server_with_models(cls, server: LLMServer, models: List[LLMModel]):
""" Insert Service Provider and Model """
with get_sync_db_session() as session:
session.add(server)
session.flush()
for model in models:
model.server_id = server.id
session.add_all(models)
session.commit()
session.refresh(server)
return server
@classmethod
async def ainsert_server_with_models(cls, server: LLMServer, models: List[LLMModel]):
""" Insert service providers and models asynchronously """
async with get_async_db_session() as session:
session.add(server)
await session.flush()
for model in models:
model.server_id = server.id
session.add_all(models)
await session.commit()
await session.refresh(server)
return server
@classmethod
async def update_server_with_models(cls, server: LLMServer, models: List[LLMModel]):
""" Update Service Providers and Models """
async with get_async_db_session() as session:
session.add(server)
add_models = []
update_models = []
for model in models:
if model.id:
update_models.append(model)
else:
add_models.append(model)
# Delete model
await session.exec(
delete(LLMModel).where(col(LLMModel.server_id) == server.id,
col(LLMModel.id).not_in([model.id for model in update_models])))
# Add New Model
session.add_all(add_models)
# Update data for existing models
for one in update_models:
await session.exec(
update(LLMModel).where(LLMModel.id == one.id).values(
name=one.name,
description=one.description,
model_name=one.model_name,
model_type=one.model_type,
config=one.config))
await session.commit()
await session.refresh(server)
return server
@classmethod
def get_all_model(cls) -> List[LLMModel]:
""" Get all models """
statement = select(LLMModel)
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
def get_server_by_id(cls, server_id: int) -> Optional[LLMServer]:
""" According to serviceIDGet Service Providers """
statement = select(LLMServer).where(LLMServer.id == server_id)
with get_sync_db_session() as session:
return session.exec(statement).first()
@classmethod
async def aget_server_by_id(cls, server_id: int) -> Optional[LLMServer]:
""" According to serviceIDGet Service Providers """
statement = select(LLMServer).where(LLMServer.id == server_id)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.first()
@classmethod
def get_server_by_ids(cls, server_ids: List[int]) -> List[LLMServer]:
""" According to serviceIDGet Service Providers """
statement = select(LLMServer).where(col(LLMServer.id).in_(server_ids))
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
async def aget_server_by_ids(cls, server_ids: List[int]) -> List[LLMServer]:
""" According to serviceIDGet Service Providers """
statement = select(LLMServer).where(col(LLMServer.id).in_(server_ids))
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def get_server_by_name(cls, server_name: str) -> Optional[LLMServer]:
""" Get Service Provider by Service Name """
statement = select(LLMServer).where(LLMServer.name == server_name)
with get_sync_db_session() as session:
return session.exec(statement).first()
@classmethod
async def aget_server_by_name(cls, server_name: str) -> Optional[LLMServer]:
""" Get Service Provider by Service Name """
statement = select(LLMServer).where(LLMServer.name == server_name)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.first()
@classmethod
def get_model_by_id(cls, model_id: int) -> Optional[LLMModel]:
""" According to the modelIDGrabbed Objects """
statement = select(LLMModel).where(LLMModel.id == model_id)
with get_sync_db_session() as session:
return session.exec(statement).first()
@classmethod
async def aget_model_by_id(cls, model_id: int) -> Optional[LLMModel]:
""" According to the modelIDGrabbed Objects """
statement = select(LLMModel).where(LLMModel.id == model_id)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.first()
@classmethod
def get_model_by_ids(cls, model_ids: List[int]) -> List[LLMModel]:
""" According to the modelIDGrabbed Objects """
statement = select(LLMModel).where(col(LLMModel.id).in_(model_ids))
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
async def aget_model_by_ids(cls, model_ids: List[int]) -> List[LLMModel]:
""" According to the modelIDGrabbed Objects """
statement = select(LLMModel).where(col(LLMModel.id).in_(model_ids))
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def get_model_by_type(cls, model_type: LLMModelType) -> Optional[LLMModel]:
""" Get first created model based on model type """
statement = select(LLMModel).where(LLMModel.model_type == model_type.value).order_by(
col(LLMModel.id).asc())
with get_sync_db_session() as session:
return session.exec(statement).first()
@classmethod
async def aget_model_by_type(cls, model_type: LLMModelType) -> Optional[LLMModel]:
""" Get first created model based on model type """
statement = select(LLMModel).where(LLMModel.model_type == model_type.value).order_by(
col(LLMModel.id).asc())
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.first()
@classmethod
def get_model_by_server_ids(cls, server_ids: List[int]) -> List[LLMModel]:
""" According to serviceIDGrabbed Objects """
statement = select(LLMModel).where(col(LLMModel.server_id).in_(server_ids)).order_by(
col(LLMModel.update_time).desc())
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
async def aget_model_by_server_ids(cls, server_ids: List[int]) -> List[LLMModel]:
""" According to serviceIDGet the first model created """
statement = select(LLMModel).where(col(LLMModel.server_id).in_(server_ids)).order_by(
col(LLMModel.update_time).desc())
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def update_model_status(cls, model_id: int, status: int, remark: str = ''):
""" Update model status """
with get_sync_db_session() as session:
session.exec(
update(LLMModel).where(col(LLMModel.id) == model_id).values(status=status,
remark=remark))
session.commit()
@classmethod
async def aupdate_model_status(cls, model_id: int, status: int, remark: str = ''):
""" Asynchronously update model status """
async with get_async_db_session() as session:
await session.exec(
update(LLMModel).where(col(LLMModel.id) == model_id).values(status=status,
remark=remark))
await session.commit()
@classmethod
def update_model_online(cls, model_id: int, online: bool):
""" Update model online status """
with get_sync_db_session() as session:
session.exec(update(LLMModel).where(col(LLMModel.id) == model_id).values(online=online))
session.commit()
@classmethod
async def aupdate_model_online(cls, model_id: int, online: bool):
""" Asynchronous update model online status """
async with get_async_db_session() as session:
await session.exec(update(LLMModel).where(col(LLMModel.id) == model_id).values(online=online))
await session.commit()
@classmethod
def delete_server_by_id(cls, server_id: int):
""" According to serviceIDDelete Service Provider """
with get_sync_db_session() as session:
session.exec(delete(LLMServer).where(col(LLMServer.id) == server_id))
session.exec(delete(LLMModel).where(col(LLMModel.server_id) == server_id))
session.commit()
@classmethod
async def adelete_server_by_id(cls, server_id: int):
""" According to serviceIDDelete Service Provider """
async with get_async_db_session() as session:
await session.exec(delete(LLMServer).where(col(LLMServer.id) == server_id))
await session.exec(delete(LLMModel).where(col(LLMModel.server_id) == server_id))
await session.commit()
@classmethod
def delete_model_by_ids(cls, model_ids: List[int]):
""" According to the modelIDDelete model """
with get_sync_db_session() as session:
session.exec(delete(LLMModel).where(col(LLMModel.id).in_(model_ids)))
session.commit()
@classmethod
async def adelete_model_by_ids(cls, model_ids: List[int]):
""" According to the modelIDDelete model """
async with get_async_db_session() as session:
await session.exec(delete(LLMModel).where(col(LLMModel.id).in_(model_ids)))
await session.commit()
================================================
FILE: src/backend/bisheng/llm/domain/schemas.py
================================================
from typing import Optional, List
from pydantic import Field, BaseModel, model_validator
from bisheng.llm.domain.models import LLMModelBase, LLMServerBase
from bisheng.utils.mask_data import JsonFieldMasker
from bisheng_langchain.linsight.const import TaskMode
class LLMModelInfo(LLMModelBase):
id: Optional[int] = None
class LLMServerInfo(LLMServerBase):
id: Optional[int] = None
models: List[LLMModelInfo] = Field(default_factory=list, description='Model List')
# Sensitive Data Desensitization
@model_validator(mode='after')
def mask_sensitive_data(self):
if not self.config:
return self
mask_maker = JsonFieldMasker()
self.config = mask_maker.mask_json(self.config)
return self
class WSModel(BaseModel):
key: Optional[str] = None
id: str
name: Optional[str] = None
displayName: Optional[str] = None
class WorkbenchModelConfig(BaseModel):
"""
Inspiration Model Configuration
"""
# Task execution model
task_model: Optional[WSModel] = Field(default=None, description='Task execution model')
# RetrieveembeddingModels
embedding_model: Optional[WSModel] = Field(default=None, description='embeddingModels')
# Inspiration Execution Mode
linsight_executor_mode: Optional[TaskMode] = Field(default=None, description='Inspiration Execution Mode')
# Speech-to-text model
asr_model: Optional[WSModel] = Field(default=None, description='Speech-to-text model')
tts_model: Optional[WSModel] = Field(default=None, description='Text-to-speech model')
class LLMModelCreateReq(BaseModel):
id: Optional[int] = Field(default=None, description='Model UniqueID, Need to pass when updating')
name: str = Field(..., description='Model Display Name')
description: Optional[str] = Field(default='', description='Model Description')
model_name: str = Field(..., description='Model Name')
model_type: str = Field(..., description='model type')
online: bool = Field(default=True, description='Online')
config: Optional[dict] = Field(default=None, description='model config')
class LLMServerCreateReq(BaseModel):
id: Optional[int] = Field(default=None, description='service providerID, Need to pass when updating')
name: str = Field(..., description='Support service provider name')
description: Optional[str] = Field(default='', description='Service Provider Description')
type: str = Field(..., description='Service Provider Type')
limit_flag: Optional[bool] = Field(default=False, description='Whether to turn on the daily call limit')
limit: Optional[int] = Field(default=0, description='Daily call limit')
config: Optional[dict] = Field(default=None, description='Service Provider Configuration')
models: Optional[List[LLMModelCreateReq]] = Field(default_factory=list, description='List of models under Service Provider')
class KnowledgeLLMConfig(BaseModel):
embedding_model_id: Optional[int] = Field(None, description="Knowledge Base DefaultembeddingModel'sID")
source_model_id: Optional[int] = Field(None, description="the Knowledge Base Traceability Model'sID")
extract_title_model_id: Optional[int] = Field(None, description="Documentation Knowledge Base Extraction Header Model'sID")
qa_similar_model_id: Optional[int] = Field(None, description="QAThe Knowledge Base Similarity Question Model'sID")
abstract_prompt: Optional[str] = Field(None, description='Summary Prompt')
class AssistantLLMItem(BaseModel):
model_id: Optional[int] = Field(None, description="Model'sID")
agent_executor_type: Optional[str] = Field(default='ReAct',
description='Execution modefunction call or ReAct')
knowledge_max_content: Optional[int] = Field(default=15000, description='Maximum number of strings for knowledge base retrieval')
knowledge_sort_index: Optional[bool] = Field(default=False, description='Whether to reschedule after knowledge base retrieval')
streaming: Optional[bool] = Field(default=True, description='Whether to turn on streaming')
default: Optional[bool] = Field(default=False, description='Is default model')
class AssistantLLMConfig(BaseModel):
llm_list: Optional[List[AssistantLLMItem]] = Field(default_factory=list, description='Assistant OptionalLLMVertical')
auto_llm: Optional[AssistantLLMItem] = Field(None, description='Assistant Portrait Automatic Optimization Model Configuration')
class EvaluationLLMConfig(BaseModel):
model_id: Optional[int] = Field(None, description='The default model of the evaluation functionID')
================================================
FILE: src/backend/bisheng/llm/domain/services/__init__.py
================================================
from .llm import LLMService
__all__ = [
'LLMService',
]
================================================
FILE: src/backend/bisheng/llm/domain/services/llm.py
================================================
import json
import os
from typing import List, Optional, Dict
from fastapi import Request, BackgroundTasks, UploadFile
from langchain_core.documents import BaseDocumentCompressor, Document
from langchain_core.embeddings import Embeddings
from langchain_core.language_models import BaseChatModel
from loguru import logger
from bisheng.common.constants.enums.telemetry import ApplicationTypeEnum
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.http_error import NotFoundError, ServerError
from bisheng.common.errcode.llm import ServerExistError, ModelNameRepeatError, ServerAddError, ServerAddAllError
from bisheng.common.errcode.server import NoAsrModelConfigError, AsrModelConfigDeletedError, NoTtsModelConfigError, \
TtsModelConfigDeletedError
from bisheng.common.models.config import ConfigDao, ConfigKeyEnum, Config
from bisheng.core.cache.redis_manager import get_redis_client
from bisheng.core.storage.minio.minio_manager import get_minio_storage
from bisheng.knowledge.domain.models.knowledge import KnowledgeDao, KnowledgeTypeEnum
from bisheng.knowledge.domain.models.knowledge import KnowledgeState
from bisheng.llm.domain.const import LLMModelType
from bisheng.llm.domain.models import LLMDao, LLMServer, LLMModel
from bisheng.llm.domain.schemas import LLMServerInfo, LLMModelInfo, KnowledgeLLMConfig, AssistantLLMConfig, \
EvaluationLLMConfig, AssistantLLMItem, LLMServerCreateReq, WorkbenchModelConfig, WSModel
from bisheng.utils import generate_uuid, md5_hash
from bisheng.utils.mask_data import JsonFieldMasker
from ..llm import BishengASR, BishengLLM, BishengTTS, BishengEmbedding
from ..llm.rerank import BishengRerank
class LLMService:
@classmethod
async def get_all_llm(cls) -> List[LLMServerInfo]:
""" Get all the model data, Exclusion:keyand other sensitive information """
llm_servers = await LLMDao.aget_all_server()
ret = []
server_ids = []
for one in llm_servers:
server_ids.append(one.id)
ret.append(LLMServerInfo(**one.model_dump(exclude={'config'})))
llm_models = await LLMDao.aget_model_by_server_ids(server_ids)
server_dicts = {}
for one in llm_models:
if one.server_id not in server_dicts:
server_dicts[one.server_id] = []
server_dicts[one.server_id].append(LLMModelInfo(**one.model_dump(exclude={'config'})))
for one in ret:
one.models = server_dicts.get(one.id, [])
return ret
@classmethod
async def get_one_llm(cls, server_id: int) -> LLMServerInfo:
""" Get a service provider's details Containskeyand other sensitive configuration information """
llm = await LLMDao.aget_server_by_id(server_id)
if not llm:
raise NotFoundError.http_exception()
models = await LLMDao.aget_model_by_server_ids([server_id])
models = [LLMModelInfo(**one.model_dump()) for one in models]
return LLMServerInfo(**llm.model_dump(), models=models)
@classmethod
async def add_llm_server(cls, request: Request, login_user: UserPayload,
server: LLMServerCreateReq) -> LLMServerInfo:
""" Add a service provider """
exist_server = await LLMDao.aget_server_by_name(server.name)
if exist_server:
raise ServerExistError.http_exception()
model_dict = {}
for one in server.models:
if one.model_name not in model_dict:
model_dict[one.model_name] = LLMModel(**one.model_dump(), user_id=login_user.user_id)
else:
raise ModelNameRepeatError.http_exception()
db_server = LLMServer(**server.model_dump(exclude={'models'}))
db_server.user_id = login_user.user_id
db_server = await LLMDao.ainsert_server_with_models(db_server, list(model_dict.values()))
ret = await cls.get_one_llm(db_server.id)
success_models = []
success_msg = ''
failed_models = []
failed_msg = ''
# Try to instantiate the corresponding model, delete it if there is an error
common_params = {
'app_id': ApplicationTypeEnum.MODEL_TEST.value,
'app_name': ApplicationTypeEnum.MODEL_TEST.value,
'app_type': ApplicationTypeEnum.MODEL_TEST,
'user_id': login_user.user_id,
}
for one in ret.models:
try:
if one.model_type == LLMModelType.LLM.value:
await cls.get_bisheng_llm(model_id=one.id, ignore_online=True, **common_params)
elif one.model_type == LLMModelType.EMBEDDING.value:
await cls.get_bisheng_embedding(model_id=one.id, ignore_online=True, **common_params)
elif one.model_type == LLMModelType.ASR.value:
await cls.get_bisheng_asr(model_id=one.id, ignore_online=True, **common_params)
elif one.model_type == LLMModelType.TTS.value:
await cls.get_bisheng_tts(model_id=one.id, ignore_online=True, **common_params)
success_msg += f'{one.model_name},'
success_models.append(one)
except Exception as e:
logger.exception("init_model_error")
# If model initialization fails, do not add to the model list
failed_msg += f'<{one.model_name}>Add failed, Reason for failure:{str(e)}\n'
failed_models.append(one)
# Description Failed to add all models
if len(success_models) == 0 and failed_msg:
await LLMDao.adelete_server_by_id(ret.id)
raise ServerAddAllError.http_exception(failed_msg)
elif len(success_models) > 0 and failed_msg:
# Some models added successfully, Delete failed model information
ret.models = success_models
await LLMDao.adelete_model_by_ids(model_ids=[one.id for one in failed_models])
await cls.add_llm_server_hook(request, login_user, ret)
raise ServerAddError.http_exception(f"<{success_msg.rstrip(',')}>Added{failed_msg}")
await cls.add_llm_server_hook(request, login_user, ret)
return ret
@classmethod
async def delete_llm_server(cls, request: Request, login_user: UserPayload, server_id: int) -> bool:
""" Delete a service provider """
await LLMDao.adelete_server_by_id(server_id)
return True
@classmethod
async def add_llm_server_hook(cls, request: Request, login_user: UserPayload, server: LLMServerInfo) -> bool:
""" Add a service provider Next Actions """
handle_types = []
for one in server.models:
# test model status
await cls.test_model_status(one, login_user)
if one.model_type in handle_types:
continue
handle_types.append(one.model_type)
model_info = await LLMDao.aget_model_by_type(LLMModelType(one.model_type))
# Determine if this is the firstllmorembeddingModels
if model_info.id == one.id:
await cls.set_default_model(model_info)
return True
@classmethod
async def test_model_status(cls, model: LLMModel | LLMModelInfo, login_user: UserPayload):
common_params = {
'app_id': ApplicationTypeEnum.MODEL_TEST.value,
'app_name': ApplicationTypeEnum.MODEL_TEST.value,
'app_type': ApplicationTypeEnum.MODEL_TEST,
'user_id': login_user.user_id,
}
try:
if model.model_type == LLMModelType.LLM.value:
bisheng_model = await cls.get_bisheng_llm(model_id=model.id, ignore_online=True, **common_params)
await bisheng_model.ainvoke('hello')
elif model.model_type == LLMModelType.EMBEDDING.value:
bisheng_embed = await cls.get_bisheng_embedding(model_id=model.id, ignore_online=True, **common_params)
await bisheng_embed.aembed_query('hello')
elif model.model_type == LLMModelType.TTS.value:
bisheng_tts = await cls.get_bisheng_tts(model_id=model.id, ignore_online=True, **common_params)
await bisheng_tts.ainvoke('hello')
elif model.model_type == LLMModelType.ASR.value:
example_file_path = os.path.join(os.path.dirname(__file__), "./asr_example.wav")
with open(example_file_path, 'rb') as f:
bisheng_asr = await cls.get_bisheng_asr(model_id=model.id, ignore_online=True, **common_params)
await bisheng_asr.ainvoke(f)
elif model.model_type == LLMModelType.RERANK.value:
bisheng_rerank = await cls.get_bisheng_rerank(model_id=model.id, ignore_online=True, **common_params)
await bisheng_rerank.acompress_documents(documents=[Document(page_content="hello world")],
query="hello")
except Exception as e:
LLMDao.update_model_status(model.id, 1, str(e))
logger.exception(f'test model status: {model.id} {model.model_name}')
@classmethod
async def set_default_model(cls, model: LLMModel | LLMModelInfo):
""" Set default model configuration """
# Set defaultllmmodel config
if model.model_type == LLMModelType.LLM.value:
# Set default model configuration for knowledge base
knowledge_llm = await cls.aget_knowledge_llm()
knowledge_change = False
if not knowledge_llm.extract_title_model_id:
knowledge_llm.extract_title_model_id = model.id
knowledge_change = True
if not knowledge_llm.source_model_id:
knowledge_llm.source_model_id = model.id
knowledge_change = True
if not knowledge_llm.qa_similar_model_id:
knowledge_llm.qa_similar_model_id = model.id
knowledge_change = True
if knowledge_change:
await cls.update_knowledge_llm(knowledge_llm)
# Set default model configuration for reviews
evaluation_llm = await cls.get_evaluation_llm()
if not evaluation_llm.model_id:
evaluation_llm.model_id = model.id
await cls.update_evaluation_llm(evaluation_llm)
# Setting the default model configuration for the assistant
assistant_llm = await cls.get_assistant_llm()
assistant_change = False
if not assistant_llm.auto_llm:
assistant_llm.auto_llm = AssistantLLMItem(model_id=model.id)
assistant_change = True
if not assistant_llm.llm_list:
assistant_change = True
assistant_llm.llm_list = [
AssistantLLMItem(model_id=model.id, default=True)
]
if assistant_change:
await cls.update_assistant_llm(assistant_llm)
elif model.model_type == LLMModelType.EMBEDDING.value:
knowledge_llm = cls.get_knowledge_llm()
if not knowledge_llm.embedding_model_id:
knowledge_llm.embedding_model_id = model.id
await cls.update_knowledge_llm(knowledge_llm)
elif model.model_type == LLMModelType.TTS.value:
workbench_llm = await cls.get_workbench_llm()
if not workbench_llm.tts_model or not workbench_llm.tts_model.id:
workbench_llm.tts_model = WSModel(id=str(model.id), name=model.model_name)
await cls.update_workbench_llm(0, workbench_llm, BackgroundTasks())
elif model.model_type == LLMModelType.ASR.value:
workbench_llm = await cls.get_workbench_llm()
if not workbench_llm.asr_model or not workbench_llm.asr_model.id:
workbench_llm.asr_model = WSModel(id=str(model.id), name=model.model_name)
await cls.update_workbench_llm(0, workbench_llm, BackgroundTasks())
@classmethod
async def update_llm_server(cls, request: Request, login_user: UserPayload,
server: LLMServerCreateReq) -> LLMServerInfo:
""" Update Service Provider Information """
exist_server = await LLMDao.aget_server_by_id(server.id)
if not exist_server:
raise NotFoundError.http_exception()
old_models = await LLMDao.aget_model_by_server_ids([exist_server.id])
old_model_dict = {
one.id: one for one in old_models
}
if exist_server.name != server.name:
# If you change your name, determine if it already exists
name_server = await LLMDao.aget_server_by_name(server.name)
if name_server and name_server.id != server.id:
raise ServerExistError.http_exception(f'<{server.name}>already exists')
model_dict = {}
for one in server.models:
if one.model_name not in model_dict:
model_dict[one.model_name] = LLMModel(**one.model_dump())
# Explanation is to add a model
if not one.id:
model_dict[one.model_name].user_id = login_user.user_id
model_dict[one.model_name].server_id = exist_server.id
else:
raise ModelNameRepeatError.http_exception()
exist_server.name = server.name
exist_server.description = server.description
exist_server.type = server.type
exist_server.limit_flag = server.limit_flag
exist_server.limit = server.limit
mask_maker = JsonFieldMasker()
exist_server.config = mask_maker.update_json_with_masked(exist_server.config, server.config)
db_server = await LLMDao.update_server_with_models(exist_server, list(model_dict.values()))
new_server_info = await cls.get_one_llm(db_server.id)
# Determine if the model status needs to be re-determined
for one in new_server_info.models:
if one.id not in old_model_dict:
await cls.set_default_model(one)
# The new model, or the model name or type has changed
if (one.id not in old_model_dict or old_model_dict[one.id].model_name != one.model_name
or old_model_dict[one.id].model_type != one.model_type):
await cls.test_model_status(one, login_user)
return new_server_info
@classmethod
async def update_model_online(cls, model_id: int, online: bool) -> LLMModelInfo:
""" Update whether the model is online """
exist_model = await LLMDao.aget_model_by_id(model_id)
if not exist_model:
raise NotFoundError.http_exception()
exist_model.online = online
await LLMDao.aupdate_model_online(exist_model.id, online)
return LLMModelInfo(**exist_model.model_dump())
@classmethod
def get_knowledge_llm(cls) -> KnowledgeLLMConfig:
""" Get the default model configuration for the knowledge base """
ret = {}
config = ConfigDao.get_config(ConfigKeyEnum.KNOWLEDGE_LLM)
if config:
ret = json.loads(config.value)
return KnowledgeLLMConfig(**ret)
@classmethod
async def aget_knowledge_llm(cls) -> KnowledgeLLMConfig:
""" Get the default model configuration for the knowledge base """
ret = {}
config = await ConfigDao.aget_config(ConfigKeyEnum.KNOWLEDGE_LLM)
if config:
ret = json.loads(config.value)
return KnowledgeLLMConfig(**ret)
@classmethod
def get_knowledge_source_llm(cls, invoke_user_id: int) -> Optional[BaseChatModel]:
""" Get the default model configuration for Knowledge Base Traceability """
knowledge_llm = cls.get_knowledge_llm()
# If no model is configured, usejieba
if not knowledge_llm.source_model_id:
return None
return cls.get_bisheng_llm_sync(model_id=knowledge_llm.source_model_id,
app_id=ApplicationTypeEnum.RAG_TRACEABILITY.value,
app_name=ApplicationTypeEnum.RAG_TRACEABILITY.value,
app_type=ApplicationTypeEnum.RAG_TRACEABILITY,
user_id=invoke_user_id)
@classmethod
async def get_knowledge_source_llm_async(cls, invoke_user_id: int) -> Optional[BaseChatModel]:
""" Get the default model configuration for Knowledge Base Traceability """
knowledge_llm = await cls.aget_knowledge_llm()
# If no model is configured, usejieba
if not knowledge_llm.source_model_id:
return None
return await cls.get_bisheng_llm(model_id=knowledge_llm.source_model_id,
app_id=ApplicationTypeEnum.RAG_TRACEABILITY.value,
app_name=ApplicationTypeEnum.RAG_TRACEABILITY.value,
app_type=ApplicationTypeEnum.RAG_TRACEABILITY,
user_id=invoke_user_id)
@classmethod
def get_knowledge_similar_llm(cls, invoke_user_id: int) -> Optional[BaseChatModel]:
""" Get the default model configuration for knowledge base similar questions """
knowledge_llm = cls.get_knowledge_llm()
# If no model is configured, usejieba
if not knowledge_llm.qa_similar_model_id:
return None
return cls.get_bisheng_llm_sync(model_id=knowledge_llm.qa_similar_model_id,
app_id=ApplicationTypeEnum.KNOWLEDGE_BASE.value,
app_name=ApplicationTypeEnum.KNOWLEDGE_BASE.value,
app_type=ApplicationTypeEnum.KNOWLEDGE_BASE,
user_id=invoke_user_id)
@classmethod
def get_knowledge_default_embedding(cls, invoke_user_id: int) -> Optional[Embeddings]:
""" Get Knowledge Base DefaultsembeddingModels """
knowledge_llm = cls.get_knowledge_llm()
if not knowledge_llm.embedding_model_id:
return None
return cls.get_bisheng_knowledge_embedding_sync(model_id=knowledge_llm.embedding_model_id,
invoke_user_id=invoke_user_id)
@classmethod
async def _base_update_llm_config(cls, data: Dict, key: ConfigKeyEnum) -> Dict:
config = await ConfigDao.aget_config(key)
if config:
config.value = json.dumps(data)
else:
config = Config(key=key.value, value=json.dumps(data, ensure_ascii=False))
await ConfigDao.async_insert_config(config)
return data
@classmethod
async def update_knowledge_llm(cls, data: KnowledgeLLMConfig) \
-> KnowledgeLLMConfig:
""" Update default model configuration for knowledge base """
await cls._base_update_llm_config(data=data.model_dump(), key=ConfigKeyEnum.KNOWLEDGE_LLM)
return data
@classmethod
async def get_assistant_llm(cls) -> AssistantLLMConfig:
""" Get the default model configuration related to the assistant """
ret = {}
config = ConfigDao.get_config(ConfigKeyEnum.ASSISTANT_LLM)
if config:
ret = json.loads(config.value)
return AssistantLLMConfig(**ret)
@classmethod
def sync_get_assistant_llm(cls) -> AssistantLLMConfig:
""" Get the default model configuration related to the assistant """
ret = {}
config = ConfigDao.get_config(ConfigKeyEnum.ASSISTANT_LLM)
if config:
ret = json.loads(config.value)
return AssistantLLMConfig(**ret)
@classmethod
async def update_assistant_llm(cls, data: AssistantLLMConfig) \
-> AssistantLLMConfig:
""" Update default model configurations related to the assistant """
await cls._base_update_llm_config(data=data.model_dump(), key=ConfigKeyEnum.ASSISTANT_LLM)
return data
@classmethod
async def get_evaluation_llm(cls) -> EvaluationLLMConfig:
""" Get the default model configuration for the evaluation feature """
ret = {}
config = await ConfigDao.aget_config(ConfigKeyEnum.EVALUATION_LLM)
if config:
ret = json.loads(config.value)
return EvaluationLLMConfig(**ret)
@classmethod
def sync_get_evaluation_llm(cls) -> EvaluationLLMConfig:
""" Get the default model configuration for the evaluation feature """
ret = {}
config = ConfigDao.get_config(ConfigKeyEnum.EVALUATION_LLM)
if config:
ret = json.loads(config.value)
return EvaluationLLMConfig(**ret)
@classmethod
async def get_evaluation_llm_object(cls, invoke_user_id: int) -> BaseChatModel:
evaluation_llm = await cls.get_evaluation_llm()
if not evaluation_llm.model_id:
raise Exception('Evaluation model is not configured')
return await cls.get_bisheng_llm(model_id=evaluation_llm.model_id,
app_id=ApplicationTypeEnum.EVALUATION.value,
app_name=ApplicationTypeEnum.EVALUATION.value,
app_type=ApplicationTypeEnum.EVALUATION,
user_id=invoke_user_id)
@classmethod
async def get_bisheng_llm(cls, **kwargs) -> BaseChatModel:
""" Initialize LiftedllmConversation Model """
return await BishengLLM.get_bisheng_llm(**kwargs)
@classmethod
def get_bisheng_llm_sync(cls, **kwargs) -> BaseChatModel:
""" Initialize LiftedllmConversation Model """
return BishengLLM(**kwargs)
@classmethod
async def get_bisheng_linsight_llm(cls, invoke_user_id: int, **kwargs) -> BaseChatModel:
return await BishengLLM.get_bisheng_llm(app_id=ApplicationTypeEnum.LINSIGHT.value,
app_name=ApplicationTypeEnum.LINSIGHT.value,
app_type=ApplicationTypeEnum.LINSIGHT,
user_id=invoke_user_id,
**kwargs)
@classmethod
async def get_bisheng_rerank(cls, **kwargs) -> BaseDocumentCompressor:
return await BishengRerank.get_bisheng_rerank(**kwargs)
@classmethod
def get_bisheng_rerank_sync(cls, **kwargs) -> BaseDocumentCompressor:
return BishengRerank(**kwargs)
@classmethod
async def get_bisheng_embedding(cls, **kwargs) -> Embeddings:
""" Initialize LiftedembeddingModels """
return await BishengEmbedding.get_bisheng_embedding(**kwargs)
@classmethod
def get_bisheng_embedding_sync(cls, **kwargs) -> Embeddings:
""" Initialize LiftedembeddingModels """
return BishengEmbedding(**kwargs)
@classmethod
async def get_bisheng_daily_embedding(cls, invoke_user_id: int, model_id: int) -> Embeddings:
""" Get dailyembeddingModels """
return await cls.get_bisheng_embedding(model_id=model_id,
app_id=ApplicationTypeEnum.DAILY_CHAT.value,
app_name=ApplicationTypeEnum.DAILY_CHAT.value,
app_type=ApplicationTypeEnum.DAILY_CHAT,
user_id=invoke_user_id)
@classmethod
async def get_bisheng_linsight_embedding(cls, invoke_user_id: int, model_id: int) -> Embeddings:
""" Get Ideas DefaultembeddingModels """
return await cls.get_bisheng_embedding(model_id=model_id,
app_id=ApplicationTypeEnum.LINSIGHT.value,
app_name=ApplicationTypeEnum.LINSIGHT.value,
app_type=ApplicationTypeEnum.LINSIGHT,
user_id=invoke_user_id)
@classmethod
async def get_bisheng_knowledge_embedding(cls, invoke_user_id: int, model_id: int) -> Embeddings:
""" Get Knowledge Base DefaultsembeddingModels """
return await cls.get_bisheng_embedding(model_id=model_id,
app_id=ApplicationTypeEnum.KNOWLEDGE_BASE.value,
app_name=ApplicationTypeEnum.KNOWLEDGE_BASE.value,
app_type=ApplicationTypeEnum.KNOWLEDGE_BASE,
user_id=invoke_user_id)
@classmethod
def get_bisheng_knowledge_embedding_sync(cls, invoke_user_id: int, model_id: int) -> Embeddings:
""" Get Knowledge Base DefaultsembeddingModels """
return cls.get_bisheng_embedding_sync(model_id=model_id,
app_id=ApplicationTypeEnum.KNOWLEDGE_BASE.value,
app_name=ApplicationTypeEnum.KNOWLEDGE_BASE.value,
app_type=ApplicationTypeEnum.KNOWLEDGE_BASE,
user_id=invoke_user_id)
@classmethod
async def get_bisheng_asr(cls, **kwargs) -> BishengASR:
""" Initialize LiftedasrModels """
return await BishengASR.get_bisheng_asr(**kwargs)
@classmethod
async def get_bisheng_tts(cls, **kwargs) -> BishengTTS:
""" Initialize LiftedttsModels """
return await BishengTTS.get_bisheng_tts(**kwargs)
@classmethod
async def update_evaluation_llm(cls, data: EvaluationLLMConfig) \
-> EvaluationLLMConfig:
""" Update default model configuration for review feature """
await cls._base_update_llm_config(data=data.model_dump(), key=ConfigKeyEnum.EVALUATION_LLM)
return data
@classmethod
async def update_workflow_llm(cls, data: EvaluationLLMConfig) -> EvaluationLLMConfig:
""" Update workflow Default Model Configuration for """
await cls._base_update_llm_config(data=data.model_dump(), key=ConfigKeyEnum.WORKFLOW_LLM)
return data
@classmethod
async def get_workflow_llm(cls) -> EvaluationLLMConfig:
""" Get the default model configuration for the evaluation feature """
ret = {}
config = await ConfigDao.aget_config(ConfigKeyEnum.WORKFLOW_LLM)
if config:
ret = json.loads(config.value)
return EvaluationLLMConfig(**ret)
@classmethod
async def get_assistant_llm_list(cls, request: Request, login_user: UserPayload) -> List[LLMServerInfo]:
""" Get a list of optional models for the assistant """
assistant_llm = await cls.get_assistant_llm()
if not assistant_llm.llm_list:
return []
model_list = await LLMDao.aget_model_by_ids([one.model_id for one in assistant_llm.llm_list])
if not model_list:
return []
default_llm = next(filter(lambda x: x.default, assistant_llm.llm_list), None)
if not default_llm:
default_llm = assistant_llm.llm_list[0]
model_dict = {}
default_server = None
for one in model_list:
if one.server_id not in model_dict:
model_dict[one.server_id] = []
if one.id == default_llm.model_id:
default_server = one.server_id
model_dict[one.server_id].insert(0, LLMModelInfo(**one.model_dump(exclude={'config'})))
continue
model_dict[one.server_id].append(LLMModelInfo(**one.model_dump(exclude={'config'})))
server_list = await LLMDao.aget_server_by_ids(list(model_dict.keys()))
ret = []
for one in server_list:
if one.id == default_server:
ret.insert(0, LLMServerInfo(**one.model_dump(exclude={'config'}), models=model_dict[one.id]))
continue
ret.append(LLMServerInfo(**one.model_dump(exclude={'config'}), models=model_dict[one.id]))
return ret
@classmethod
async def update_workbench_llm(cls, invoke_user_id: int, config_obj: WorkbenchModelConfig,
background_tasks: BackgroundTasks):
"""
Update Invisible Model Configuration
:param invoke_user_id:
:param config_obj:
:param background_tasks:
:return:
"""
# Delay imports to avoid looping imports
from bisheng.worker.knowledge.rebuild_knowledge_worker import rebuild_knowledge_celery
config = await ConfigDao.aget_config(ConfigKeyEnum.LINSIGHT_LLM)
if not config:
config = Config(key=ConfigKeyEnum.LINSIGHT_LLM.value, value='{}')
if config_obj.embedding_model:
# Determine consistency
config_old_obj = WorkbenchModelConfig(**json.loads(config.value)) if config else WorkbenchModelConfig()
if (config_obj.embedding_model.id and config_old_obj.embedding_model is None or
config_obj.embedding_model.id != config_old_obj.embedding_model.id):
embeddings = await cls.get_bisheng_embedding(model_id=config_obj.embedding_model.id,
app_id=ApplicationTypeEnum.LINSIGHT.value,
app_name=ApplicationTypeEnum.LINSIGHT.value,
app_type=ApplicationTypeEnum.LINSIGHT,
user_id=invoke_user_id)
try:
await embeddings.aembed_query("test")
except Exception as e:
raise Exception(f"EmbeddingModel initialization failed: {str(e)}")
from bisheng.linsight.domain.services.sop_manage import SOPManageService
background_tasks.add_task(SOPManageService.rebuild_sop_vector_store_task, embeddings)
# Update Personal Knowledge Base
# 1.Upgrading alltypeare2(Private Repository)right of privacyknowledgeStatus and Model
private_knowledges = await KnowledgeDao.aget_all_knowledge(
knowledge_type=KnowledgeTypeEnum.PRIVATE
)
updated_count = 0
for knowledge in private_knowledges:
# Update status is rebuilding, model is newmodel_id
knowledge.state = KnowledgeState.REBUILDING.value
knowledge.model = config_obj.embedding_model.id
await KnowledgeDao.aupdate_one(knowledge)
updated_count += 1
# 3. For eachknowledgeStart asynchronous task
rebuild_knowledge_celery.delay(knowledge.id, int(knowledge.model), invoke_user_id)
logger.info(
f"Started rebuild task for knowledge_id={knowledge.id} with model_id={knowledge.model}")
logger.info(
f"Updated {updated_count} private knowledge bases to use new embedding model {config_obj.embedding_model.id}")
config.value = json.dumps(config_obj.model_dump(), ensure_ascii=False)
await ConfigDao.async_insert_config(config)
return config_obj
@classmethod
async def get_workbench_llm(cls) -> WorkbenchModelConfig:
"""
Get Workbench Model Configuration
:return:
"""
ret = {}
config = await ConfigDao.aget_config(ConfigKeyEnum.LINSIGHT_LLM)
if config:
ret = json.loads(config.value)
return WorkbenchModelConfig(**ret)
@classmethod
async def invoke_workbench_asr(cls, login_user: UserPayload, file: UploadFile) -> str:
""" Call the workbench'sasrModels Convert Voice to Text """
if not file:
raise ServerError.http_exception("no file upload")
workbench_llm = await cls.get_workbench_llm()
if not workbench_llm.asr_model or not workbench_llm.asr_model.id:
raise NoAsrModelConfigError.http_exception()
model_info = await LLMDao.aget_model_by_id(int(workbench_llm.asr_model.id))
if not model_info:
raise AsrModelConfigDeletedError.http_exception()
asr_client = await cls.get_bisheng_asr(model_id=int(workbench_llm.asr_model.id),
app_id=ApplicationTypeEnum.ASR.value,
app_name=ApplicationTypeEnum.ASR.value,
app_type=ApplicationTypeEnum.ASR,
user_id=login_user.user_id)
return await asr_client.ainvoke(file.file)
@classmethod
async def invoke_workbench_tts(cls, login_user: UserPayload, text: str) -> str:
"""
Call the workbench'sttsModels Convert text to speech
:return: minioPath to
"""
workbench_llm = await cls.get_workbench_llm()
redis_client = await get_redis_client()
if not workbench_llm.tts_model or not workbench_llm.tts_model.id:
raise NoTtsModelConfigError.http_exception()
model_info = await LLMDao.aget_model_by_id(model_id=int(workbench_llm.tts_model.id))
if not model_info:
raise TtsModelConfigDeletedError.http_exception()
# get from cache
voice = model_info.config.get("voice", "default") if model_info.config else "default"
cache_key = f"workbench_tts:{model_info.id}:{voice}:{md5_hash(text)}"
cache_value = await redis_client.aget(cache_key)
if cache_value:
return cache_value
tts_client = await cls.get_bisheng_tts(model_id=int(workbench_llm.tts_model.id),
app_id=ApplicationTypeEnum.TTS.value,
app_name=ApplicationTypeEnum.TTS.value,
app_type=ApplicationTypeEnum.TTS,
user_id=login_user.user_id)
audio_bytes = await tts_client.ainvoke(text)
# upload to minio
object_name = f"tts/{generate_uuid()}.mp3"
minio_client = await get_minio_storage()
await minio_client.put_object(object_name=object_name, file=audio_bytes, content_type="audio/mpeg",
bucket_name=minio_client.tmp_bucket)
cache_value = await minio_client.get_share_link(object_name, bucket=minio_client.tmp_bucket)
# The tmp bucket automatically clears files older than 7 days, so set the expiration time to 6 days
await redis_client.aset(cache_key, cache_value, expiration=6 * 24 * 3600)
return cache_value
================================================
FILE: src/backend/bisheng/llm/domain/utils.py
================================================
import functools
import time
from datetime import datetime
from typing import Any, Dict, Optional, Union
from uuid import UUID
from langchain_core.callbacks import BaseCallbackHandler
from langchain_core.outputs import ChatResult, GenerationChunk, ChatGenerationChunk
from loguru import logger
from bisheng.common.constants.enums.telemetry import StatusEnum, BaseTelemetryTypeEnum
from bisheng.common.schemas.telemetry.event_data_schema import ModelInvokeEventData
from bisheng.common.services import telemetry_service
from bisheng.core.cache.redis_manager import get_redis_client, get_redis_client_sync
from bisheng.core.logger import trace_id_var
from bisheng.llm.domain.const import LLMModelStatus
async def bisheng_model_limit_check(self: 'BishengBase'):
now = datetime.now().strftime("%Y-%m-%d")
if self.server_info.limit_flag:
# Number of calls checked
cache_key = f"model_limit:{now}:{self.server_info.id}"
redis_client = await get_redis_client()
use_num = await redis_client.aincr(cache_key)
if use_num > self.server_info.limit:
raise Exception(f'{self.server_info.name}/{self.model_info.model_name} Quota used up')
def sync_bisheng_model_limit_check(self: 'BishengBase'):
now = datetime.now().strftime("%Y-%m-%d")
if self.server_info.limit_flag:
# Number of calls checked
cache_key = f"model_limit:{now}:{self.server_info.id}"
use_num = get_redis_client_sync().incr(cache_key)
if use_num > self.server_info.limit:
raise Exception(f'{self.server_info.name}/{self.model_info.model_name} Quota used up')
def get_token_from_usage(token_usage: Dict[str, Any]) -> tuple[int, int, int, int]:
"""
FROMtoken_usageGet in DictionarytokenUsage
"""
input_token = token_usage.get('input_tokens', 0) or token_usage.get('prompt_tokens', 0)
output_token = token_usage.get('output_tokens', 0) or token_usage.get('completion_tokens', 0)
cache_token = token_usage.get('cached_token', 0) or token_usage.get("prompt_tokens_details", {}).get(
'cached_tokens', 0) or token_usage.get('input_tokens_details', {}).get('cache_read', 0)
total_token = token_usage.get('total_tokens', 0)
return input_token, output_token, cache_token, total_token
def parse_token_usage(result: Any) -> tuple[int, int, int, int]:
"""
analyzingtokenUsage
"""
input_token, output_token, cache_token, total_token = 0, 0, 0, 0
if isinstance(result, ChatResult):
for generation in result.generations:
token_usage = generation.generation_info.get('token_usage', {}) or generation.message.response_metadata.get(
'token_usage', {}) or generation.message.usage_metadata
tmp1, tmp2, tmp3, tmp4 = get_token_from_usage(token_usage)
input_token += tmp1
output_token += tmp2
cache_token += tmp3
total_token += tmp4
elif isinstance(result, ChatGenerationChunk):
token_usage = result.message.response_metadata.get('token_usage', {}) or result.generation_info.get(
'token_usage', {}) or result.message.usage_metadata
input_token, output_token, cache_token, total_token = get_token_from_usage(token_usage)
else:
logger.warning(f'unknown result type: {type(result)}')
return input_token, output_token, cache_token, total_token
class TelemetryCallback(BaseCallbackHandler):
"""
Telemetry Slider Callbacks
"""
def __init__(self, start_time: float):
self.start_time = start_time
self.first_token_time: Optional[int] = 0
def on_llm_new_token(
self,
token: str,
*,
chunk: Optional[Union[GenerationChunk, ChatGenerationChunk]] = None,
run_id: UUID,
parent_run_id: Optional[UUID] = None,
**kwargs: Any,
) -> Any:
if not self.first_token_time:
self.first_token_time = int((time.time() - self.start_time) * 1000)
def upload_telemetry_log(self: 'BishengBase', start_time: float, end_time: float, first_token_cost_time: int,
status: StatusEnum, is_stream: bool = False, result: Any = None):
"""
Upload Buried Point Log
"""
try:
logger.debug("start upload model invoke telemetry log")
input_token, output_token, cache_token, total_token = 0, 0, 0, 0
if self.model_info.model_type in ['llm']:
try:
input_token, output_token, cache_token, total_token = parse_token_usage(result)
except Exception as e:
logger.warning(f"parse token usage failed: {e}")
telemetry_service.log_event_sync(user_id=self.user_id, event_type=BaseTelemetryTypeEnum.MODEL_INVOKE,
trace_id=trace_id_var.get(),
event_data=ModelInvokeEventData(
model_id=self.model_id,
model_name=self.model_name,
model_type=self.model_info.model_type,
model_server_id=self.server_info.id,
model_server_name=self.server_info.name,
app_id=self.app_id,
app_name=self.app_name,
app_type=self.app_type,
start_time=int(start_time),
end_time=int(end_time),
first_token_cost_time=first_token_cost_time,
status=status,
is_stream=is_stream,
input_token=input_token,
output_token=output_token,
cache_token=cache_token,
total_token=total_token,
))
logger.debug("end upload model invoke telemetry log")
except Exception as e:
logger.exception(f"upload telemetry log failed")
def wrapper_bisheng_model_limit_check(func):
"""
Number of calls to check the decorator
"""
@functools.wraps(func)
def wrapper(*args, **kwargs):
self = args[0]
sync_bisheng_model_limit_check(self)
status = LLMModelStatus.NORMAL.value
remark = ""
start_time = time.time()
result = None
telemetry_status = StatusEnum.SUCCESS
telemetry_callback = None
try:
if self.model_info.model_type == 'llm':
telemetry_callback = TelemetryCallback(start_time=start_time)
if kwargs.get('run_manager') is not None:
kwargs['run_manager'].handlers.append(telemetry_callback)
result = func(*args, **kwargs)
return result
except Exception as e:
status = LLMModelStatus.ERROR.value
remark = str(e)
telemetry_status = StatusEnum.FAILED
raise e
finally:
end_time = time.time()
# Avoid blocking the main thread by uploading logs asynchronously using the thread pool
first_token_cost_time = telemetry_callback.first_token_time if telemetry_callback else 0
upload_telemetry_log(self, start_time, end_time, first_token_cost_time, telemetry_status, result=result)
self.sync_update_model_status(status, remark)
return wrapper
def wrapper_bisheng_model_limit_check_async(func):
"""
Number of calls to check the decorator
"""
@functools.wraps(func)
async def wrapper(*args, **kwargs):
self = args[0]
await bisheng_model_limit_check(self)
status = LLMModelStatus.NORMAL.value
remark = ""
start_time = time.time()
telemetry_status = StatusEnum.SUCCESS
result = None
telemetry_callback = None
try:
if self.model_info.model_type == 'llm':
telemetry_callback = TelemetryCallback(start_time=start_time)
if kwargs.get('run_manager') is not None:
kwargs['run_manager'].handlers.append(telemetry_callback)
result = await func(*args, **kwargs)
return result
except Exception as e:
status = LLMModelStatus.ERROR.value
remark = str(e)
telemetry_status = StatusEnum.FAILED
raise e
finally:
end_time = time.time()
first_token_cost_time = telemetry_callback.first_token_time if telemetry_callback else 0
upload_telemetry_log(self, start_time, end_time, first_token_cost_time, telemetry_status, result=result)
await args[0].update_model_status(status, remark)
return wrapper
def wrapper_bisheng_model_generator(func):
"""
Number of calls to check the decorator Decorative Synchronization Builder Functions
"""
@functools.wraps(func)
def wrapper(*args, **kwargs):
self = args[0]
sync_bisheng_model_limit_check(self)
status = LLMModelStatus.NORMAL.value
remark = ""
start_time = time.time()
first_token_cost_time = 0
telemetry_status = StatusEnum.SUCCESS
item = None
try:
for item in func(*args, **kwargs):
yield item
if first_token_cost_time == 0:
first_token_cost_time = int((time.time() - start_time) * 1000)
except Exception as e:
status = LLMModelStatus.ERROR.value
remark = str(e)
telemetry_status = StatusEnum.FAILED
raise e
finally:
end_time = time.time()
upload_telemetry_log(self, start_time, end_time, first_token_cost_time, telemetry_status, True, result=item)
self.sync_update_model_status(status, remark)
return wrapper
def wrapper_bisheng_model_generator_async(func):
"""
Number of calls to check the decorator Decorative Asynchronous Builder Functions
"""
@functools.wraps(func)
async def wrapper(*args, **kwargs):
self = args[0]
await bisheng_model_limit_check(self)
status = LLMModelStatus.NORMAL.value
remark = ""
start_time = time.time()
first_token_cost_time = 0
telemetry_status = StatusEnum.SUCCESS
item = None
try:
async for item in func(*args, **kwargs):
yield item
if first_token_cost_time == 0:
first_token_cost_time = int((time.time() - start_time) * 1000)
except Exception as e:
status = LLMModelStatus.ERROR.value
remark = str(e)
telemetry_status = StatusEnum.FAILED
raise e
finally:
end_time = time.time()
upload_telemetry_log(self, start_time, end_time, first_token_cost_time, telemetry_status, True, result=item)
await self.update_model_status(status, remark)
return wrapper
================================================
FILE: src/backend/bisheng/main.py
================================================
from contextlib import asynccontextmanager
from fastapi import FastAPI, HTTPException, Request, status
from fastapi.exceptions import RequestValidationError
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import JSONResponse, ORJSONResponse
from loguru import logger
from bisheng.api import router, router_rpc
from bisheng.common.errcode import BaseErrorCode
from bisheng.common.exceptions.auth import AuthJWTException
from bisheng.common.init_data import init_default_data
from bisheng.common.services.config_service import settings
from bisheng.core.context import initialize_app_context, close_app_context
from bisheng.core.logger import set_logger_config
from bisheng.services.utils import initialize_services, teardown_services
from bisheng.utils.http_middleware import CustomMiddleware, WebSocketLoggingMiddleware
from bisheng.utils.threadpool import thread_pool
def handle_http_exception(req: Request, exc: Exception) -> ORJSONResponse:
if isinstance(exc, HTTPException):
msg = {
'status_code': exc.status_code,
'status_message': exc.detail['error'] if isinstance(exc.detail, dict) else exc.detail
}
elif isinstance(exc, BaseErrorCode):
data = {'exception': str(exc), **exc.kwargs} if exc.kwargs else {'exception': str(exc)}
msg = {'status_code': exc.code, 'status_message': exc.message,
'data': data}
else:
logger.exception('Unhandled exception')
msg = {'status_code': 500, 'status_message': str(exc)}
logger.error(f'{req.method} {req.url} {str(exc)}')
return ORJSONResponse(content=msg)
def handle_request_validation_error(req: Request, exc: RequestValidationError) -> ORJSONResponse:
msg = {'status_code': status.HTTP_422_UNPROCESSABLE_ENTITY, 'status_message': exc.errors()}
logger.error(f'{req.method} {req.url} {str(exc.errors())[:100]}')
return ORJSONResponse(content=msg)
_EXCEPTION_HANDLERS = {
HTTPException: handle_http_exception,
RequestValidationError: handle_request_validation_error,
BaseErrorCode: handle_http_exception,
Exception: handle_http_exception
}
@asynccontextmanager
async def lifespan(app: FastAPI):
await initialize_app_context(config=settings)
initialize_services()
await init_default_data()
# LangfuseInstance.update()
yield
teardown_services()
thread_pool.tear_down()
await close_app_context()
def create_app():
"""Create the FastAPI app and include the router."""
app = FastAPI(
default_response_class=ORJSONResponse,
exception_handlers=_EXCEPTION_HANDLERS,
lifespan=lifespan,
)
origins = [
'*',
]
@app.get('/health')
def get_health():
return {'status': 'OK'}
app.add_middleware(
CORSMiddleware,
allow_origins=origins,
allow_credentials=False,
allow_methods=['*'],
allow_headers=['*'],
)
app.add_middleware(CustomMiddleware)
app.add_middleware(WebSocketLoggingMiddleware)
@app.exception_handler(AuthJWTException)
def authjwt_exception_handler(request: Request, exc: AuthJWTException):
return JSONResponse(status_code=401, content={'detail': str(exc)})
app.include_router(router)
app.include_router(router_rpc)
if settings.debug:
import tracemalloc
tracemalloc.start()
return app
app = create_app()
if __name__ == '__main__':
import uvicorn
set_logger_config(settings.logger_conf)
uvicorn.run(app, host='0.0.0.0', port=7860, workers=1, log_config=None)
================================================
FILE: src/backend/bisheng/mcp_manage/__init__.py
================================================
================================================
FILE: src/backend/bisheng/mcp_manage/clients/__init__.py
================================================
================================================
FILE: src/backend/bisheng/mcp_manage/clients/base.py
================================================
from abc import abstractmethod, ABC
from contextlib import asynccontextmanager
from typing import Any
from mcp import ClientSession
class BaseMcpClient(ABC):
"""
Base class for MCP clients.
"""
@abstractmethod
async def get_transport(self):
raise NotImplementedError("get_mcp_client_transport() must be implemented in subclasses.")
@asynccontextmanager
async def initialize(self):
"""
Initialize the client.
"""
async with self.get_transport() as (read, write):
async with ClientSession(read, write) as session:
await session.initialize()
yield session
async def list_tools(self):
async with self.initialize() as client_session:
tools = await client_session.list_tools()
return tools.tools
async def call_tool(self, name: str, arguments: dict[str, Any] | None = None) -> str:
"""
Call a tool.
"""
async with self.initialize() as client_session:
try:
resp = await client_session.call_tool(name, arguments)
except Exception as e:
return f"Tool call failed: {str(e)}"
return resp.model_dump_json()
================================================
FILE: src/backend/bisheng/mcp_manage/clients/sse.py
================================================
from contextlib import asynccontextmanager
from mcp.client.sse import sse_client
from bisheng.mcp_manage.clients.base import BaseMcpClient
class SseClient(BaseMcpClient):
"""
SSE client for connecting to the mcp server.
"""
def __init__(self, url: str, **kwargs):
"""
Initialize the SSE client.
:param url: The URL of the SSE server.
"""
super().__init__()
self.url = url
self.kwargs = kwargs
@asynccontextmanager
async def get_transport(self):
"""
Initialize the SSE client.
"""
async with sse_client(url=self.url, **self.kwargs) as (read, write):
yield read, write
================================================
FILE: src/backend/bisheng/mcp_manage/clients/stdio.py
================================================
from contextlib import asynccontextmanager
from mcp.client.stdio import stdio_client, StdioServerParameters
from bisheng.mcp_manage.clients.base import BaseMcpClient
class StdioClient(BaseMcpClient):
"""
SSE client for connecting to the mcp server.
"""
def __init__(self, **kwargs: dict):
"""
Initialize the SSE client.
:param url: The URL of the SSE server.
"""
super().__init__()
self.server_params = StdioServerParameters(**kwargs)
@asynccontextmanager
async def get_transport(self):
"""
Initialize the SSE client.
"""
async with stdio_client(server=self.server_params) as (read, write):
yield read, write
================================================
FILE: src/backend/bisheng/mcp_manage/clients/streamable.py
================================================
from contextlib import asynccontextmanager
from mcp.client.streamable_http import streamablehttp_client
from bisheng.mcp_manage.clients.base import BaseMcpClient
class StreamableClient(BaseMcpClient):
"""
SSE client for connecting to the mcp server.
"""
def __init__(self, url: str, **kwargs):
"""
Initialize the streamable http client.
:param url: The URL of the streamable server.
"""
super().__init__()
self.url = url
self.kwargs = kwargs
@asynccontextmanager
async def get_transport(self):
"""
Initialize the SSE client.
"""
async with streamablehttp_client(url=self.url, **self.kwargs) as (read, write, _):
yield read, write
================================================
FILE: src/backend/bisheng/mcp_manage/constant.py
================================================
from enum import Enum
class McpClientType(Enum):
SSE = 'sse'
STDIO = 'stdio'
STREAMABLE = 'streamable'
================================================
FILE: src/backend/bisheng/mcp_manage/langchain/__init__.py
================================================
================================================
FILE: src/backend/bisheng/mcp_manage/langchain/tool.py
================================================
import asyncio
import concurrent
import concurrent.futures
from typing import Any
from langchain_core.tools import StructuredTool
from pydantic import BaseModel, ConfigDict
from bisheng.mcp_manage.clients.base import BaseMcpClient
from bisheng_langchain.utils.openapi import convert_openapi_field_value
class McpTool(BaseModel):
model_config = ConfigDict(arbitrary_types_allowed=True)
name: str
description: str
mcp_client: BaseMcpClient
mcp_tool_name: str
arg_schema: dict = {}
def parse_kwargs_schema(self, kwargs: dict) -> None | dict:
"""convert the kwargs field value"""
if not kwargs:
return None
for k, v in kwargs.items():
k_type = self.arg_schema.get("properties", {}).get(k, {}).get("type")
kwargs[k] = convert_openapi_field_value(v, k_type)
return kwargs
def run(self, *args, **kwargs: Any) -> Any:
# todo call async method better when using in event pool
with concurrent.futures.ThreadPoolExecutor() as pool:
future = pool.submit(asyncio.run, self.arun(*args, **kwargs))
resp = future.result()
return resp
async def arun(self, *args, **kwargs: Any) -> Any:
"""Use the tool asynchronously."""
kwargs = self.parse_kwargs_schema(kwargs)
resp = await self.mcp_client.call_tool(self.mcp_tool_name, kwargs)
return resp
@classmethod
def get_mcp_tool(cls, name: str, description: str, mcp_client: BaseMcpClient,
mcp_tool_name: str, arg_schema: Any, **kwargs) -> StructuredTool:
"""Get a tool from the class."""
c = cls(name=name, description=description, mcp_client=mcp_client,
mcp_tool_name=mcp_tool_name, arg_schema=arg_schema)
return StructuredTool(name=c.name,
description=c.description,
func=c.run,
coroutine=c.arun,
args_schema=arg_schema,
**kwargs)
================================================
FILE: src/backend/bisheng/mcp_manage/manager.py
================================================
import json
from bisheng.mcp_manage.clients.base import BaseMcpClient
from bisheng.mcp_manage.clients.sse import SseClient
from bisheng.mcp_manage.clients.stdio import StdioClient
from bisheng.mcp_manage.clients.streamable import StreamableClient
from bisheng.mcp_manage.constant import McpClientType
class ClientManager:
@classmethod
async def connect_mcp_from_json(cls, client_json: dict | str):
""" Get the under the corresponding configurationmcpCONNECT """
return cls.sync_connect_mcp_from_json(client_json)
@classmethod
def sync_connect_mcp_from_json(cls, client_json: dict | str) -> BaseMcpClient:
""" Get the under the corresponding configurationmcpCONNECT """
if isinstance(client_json, str):
client_json = json.loads(client_json)
mcp_servers = client_json['mcpServers']
client_type = McpClientType.SSE.value
client_kwargs = {}
for _, kwargs in mcp_servers.items():
if 'type' in kwargs:
client_type = kwargs.pop('type', McpClientType.SSE.value)
elif 'command' in kwargs:
client_type = McpClientType.STDIO.value
kwargs.pop('name', '')
kwargs.pop('description', '')
client_kwargs = kwargs
break
return cls.sync_connect_mcp(client_type, **client_kwargs)
@classmethod
async def connect_mcp(cls, client_type: str, **kwargs) -> BaseMcpClient:
""" Get the mappingurlright of privacymcpCONNECT """
# Initialize the correspondingclient
return cls.sync_connect_mcp(client_type, **kwargs)
@classmethod
def sync_connect_mcp(cls, client_type: str, **kwargs) -> BaseMcpClient:
# Initialize the correspondingclient
if client_type == McpClientType.SSE.value:
client = SseClient(**kwargs)
elif client_type == McpClientType.STDIO.value:
client = StdioClient(**kwargs)
elif client_type == McpClientType.STREAMABLE.value:
client = StreamableClient(**kwargs)
else:
raise ValueError(f'client_type {client_type} not supported')
return client
================================================
FILE: src/backend/bisheng/open_endpoints/__init__.py
================================================
================================================
FILE: src/backend/bisheng/open_endpoints/api/__init__.py
================================================
================================================
FILE: src/backend/bisheng/open_endpoints/api/dependencies.py
================================================
from fastapi import Depends
from sqlmodel.ext.asyncio.session import AsyncSession
from bisheng.common.dependencies.core_deps import get_db_session
from bisheng.knowledge.domain.repositories.implementations.knowledge_file_repository_impl import \
KnowledgeFileRepositoryImpl
from bisheng.knowledge.domain.repositories.implementations.knowledge_repository_impl import KnowledgeRepositoryImpl
from bisheng.knowledge.domain.repositories.interfaces.knowledge_file_repository import KnowledgeFileRepository
from bisheng.knowledge.domain.repositories.interfaces.knowledge_repository import KnowledgeRepository
from bisheng.knowledge.domain.services.knowledge_file_service import KnowledgeFileService
from bisheng.knowledge.domain.services.knowledge_service import KnowledgeService
async def get_knowledge_repository(
session: AsyncSession = Depends(get_db_session),
) -> KnowledgeRepository:
"""Dapatkan KnowledgeRepositoryInstance Dependencies"""
return KnowledgeRepositoryImpl(session)
async def get_knowledge_file_repository(
session: AsyncSession = Depends(get_db_session),
) -> 'KnowledgeFileRepository':
"""Dapatkan KnowledgeFileRepositoryInstance Dependencies"""
return KnowledgeFileRepositoryImpl(session)
async def get_knowledge_service(
knowledge_repository: KnowledgeRepository = Depends(get_knowledge_repository),
knowledge_file_repository: KnowledgeFileRepository = Depends(get_knowledge_file_repository),
) -> 'KnowledgeService':
"""Dapatkan KnowledgeServiceInstance Dependencies"""
return KnowledgeService(knowledge_repository=knowledge_repository,
knowledge_file_repository=knowledge_file_repository)
async def get_knowledge_file_service(
knowledge_repository: KnowledgeRepository = Depends(get_knowledge_repository),
knowledge_file_repository: KnowledgeFileRepository = Depends(get_knowledge_file_repository),
) -> 'KnowledgeFileService':
"""Dapatkan KnowledgeFileServiceInstance Dependencies"""
return KnowledgeFileService(
knowledge_repository=knowledge_repository,
knowledge_file_repository=knowledge_file_repository,
)
================================================
FILE: src/backend/bisheng/open_endpoints/api/endpoints/__init__.py
================================================
================================================
FILE: src/backend/bisheng/open_endpoints/api/endpoints/assistant.py
================================================
# Login-free assistant related interface
import json
import time
from typing import Optional
from uuid import UUID
from fastapi import APIRouter, HTTPException, Query, Request, WebSocket, WebSocketException
from fastapi import status as http_status
from fastapi.responses import ORJSONResponse, StreamingResponse
from langchain_core.messages import AIMessage, HumanMessage, AIMessageChunk
from loguru import logger
from bisheng.api.services.assistant import AssistantService
from bisheng.api.services.assistant_agent import AssistantAgent
from bisheng.api.v1.chat import chat_manager
from bisheng.api.v1.schemas import (OpenAIChatCompletionReq,
OpenAIChatCompletionResp, OpenAIChoice)
from bisheng.chat.types import WorkType
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum, ApplicationTypeEnum
from bisheng.common.schemas.api import resp_200, PageData
from bisheng.common.schemas.telemetry.event_data_schema import ApplicationAliveEventData, ApplicationProcessEventData
from bisheng.common.services import telemetry_service
from bisheng.common.services.config_service import settings
from bisheng.core.logger import trace_id_var
from bisheng.open_endpoints.domain.utils import get_default_operator
from bisheng.utils import generate_uuid
from bisheng.utils import get_request_ip
router = APIRouter(prefix='/assistant', tags=['OpenAPI', 'Assistant'])
@router.post('/chat/completions')
async def assistant_chat_completions(request: Request, req_data: OpenAIChatCompletionReq):
"""
Compatible openaiInterface format, all errors must return non-http200Status code
Chat with your assistant
Fulfillment needs:
1. Determine if the model invoked by the assistant supports streaming calls
2. If not, follow the original logic
3. If supported andstream=True, with real streaming calls
4. stream=False Or return to the original logic?JSON
"""
assistant_id = UUID(req_data.model).hex
logger.info(
f'act=assistant_chat_completions assistant_id={req_data.model}, stream={req_data.stream}, ip={get_request_ip(request)}'
)
try:
# Get the default user information configured in the system configuration
login_user = get_default_operator()
except Exception as e:
return ORJSONResponse(status_code=500, content=str(e), media_type='application/json')
# Find Assistant Information
try:
assistant_info = await AssistantService.get_assistant_info(assistant_id, login_user)
except Exception as e:
return ORJSONResponse(status_code=500,
content=str(e),
media_type='application/json')
start_time = time.time()
try:
# Overlay Temperature Settings
if req_data.temperature != 0:
assistant_info.temperature = req_data.temperature
chat_history = []
question = ''
# Resolve the conversation history and the user's latest questions
for one in req_data.messages:
if one['role'] == 'user':
chat_history.append(HumanMessage(content=one['content']))
question = one['content']
elif one['role'] == 'assistant':
chat_history.append(AIMessage(content=one['content']))
# Remove user issue from history
if chat_history and chat_history[-1].content == question:
chat_history = chat_history[:-1]
# Initialization Assistantagent
agent = AssistantAgent(assistant_info, '', invoke_user_id=login_user.user_id)
await agent.init_assistant()
# Determine if the model supports streaming calls
model_supports_streaming = _check_model_supports_streaming(agent)
logger.debug(
f'act=assistant_chat_completions model_supports_streaming={model_supports_streaming}, stream={req_data.stream}, llm_type={type(agent.llm)}')
# Streaming is not supported for non-streaming calls or models
if not req_data.stream or not model_supports_streaming:
answer = await agent.run(question, chat_history)
answer = answer[-1].content
openai_resp_id = generate_uuid()
# Package the results asopenaiData Format
openai_resp = OpenAIChatCompletionResp(
id=openai_resp_id,
object='chat.completion',
created=int(time.time()),
model=req_data.model,
choices=[OpenAIChoice(index=0, message={
'role': 'assistant',
'content': answer
})],
)
# Non-streaming direct return results
if not req_data.stream:
return openai_resp
# The user requests streaming but the model does not support it, use pseudo-streaming to return
openai_resp.object = 'chat.completion.chunk'
openai_resp.choices = [OpenAIChoice(index=0, delta={'content': answer})]
async def _pseudo_event_stream():
yield f'data: {openai_resp.json()}\n\n'
yield 'data: [DONE]\n\n'
return StreamingResponse(_pseudo_event_stream(), media_type='text/event-stream')
# Model supports streaming and user-requested streaming, using real streaming calls
openai_resp_id = generate_uuid()
logger.info(f'act=assistant_chat_completions_streaming openai_resp_id={openai_resp_id}')
async def _streaming_event_generator():
"""Real Streaming Event Generator"""
logger.debug(f'[APIStreamed] _streaming_event_generatorto process')
try:
# Use True Streaming Calls
chunk_counter = 0
try:
async for message_chunk in agent.astream(question, chat_history):
chunk_counter += 1
if not message_chunk:
logger.debug(f'Empty message_chunk received')
continue
# Get the latest news
latest_message = message_chunk[-1] if isinstance(message_chunk, list) else message_chunk
if not isinstance(latest_message, AIMessageChunk):
continue
reasoning_content = latest_message.additional_kwargs.get("reasoning_content", "")
content = latest_message.content
chunk_data = {
"id": openai_resp_id,
"object": "chat.completion.chunk",
"created": int(time.time()),
"model": req_data.model,
"choices": [{
"index": 0,
"delta": {"content": content, "reasoning_content": reasoning_content},
"finish_reason": None
}]
}
# Use saferJSONSerialization to avoid transmission truncation
json_str = json.dumps(chunk_data, ensure_ascii=False, separators=(',', ':'))
yield f'data: {json_str}\n\n'
except Exception as astream_error:
logger.exception('[APIStreamed] agent.astream()Error calling')
raise astream_error
logger.info(f'[APIStreamed] astreamLoop ended, total processed{chunk_counter}Pcschunk')
# Send End Signal
end_chunk = {
"id": openai_resp_id,
"object": "chat.completion.chunk",
"created": int(time.time()),
"model": req_data.model,
"choices": [{
"index": 0,
"delta": {},
"finish_reason": "stop"
}]
}
yield f'data: {json.dumps(end_chunk, ensure_ascii=False)}\n\n'
yield 'data: [DONE]\n\n'
except Exception as exc:
logger.error(f'Streaming error: {exc}')
# Send error message
error_chunk = {
"id": openai_resp_id,
"object": "chat.completion.chunk",
"created": int(time.time()),
"model": req_data.model,
"choices": [{
"index": 0,
"delta": {"content": f"Error-free: {str(exc)}"},
"finish_reason": "stop"
}]
}
yield f'data: {json.dumps(error_chunk, ensure_ascii=False)}\n\n'
yield 'data: [DONE]\n\n'
try:
logger.info(f'[APIStreamed] BuatStreamingResponse, generator function: {_streaming_event_generator}')
return StreamingResponse(_streaming_event_generator(),
media_type='text/event-stream')
except Exception as exc:
logger.error(f'StreamingResponse creation error: {exc}')
return ORJSONResponse(status_code=500, content=str(exc))
finally:
end_time = time.time()
await telemetry_service.log_event(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.APPLICATION_ALIVE,
trace_id=trace_id_var.get(),
event_data=ApplicationAliveEventData(
app_id=assistant_id,
app_name=assistant_info.name,
app_type=ApplicationTypeEnum.ASSISTANT,
chat_id='',
start_time=int(start_time),
end_time=int(end_time)))
await telemetry_service.log_event(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.APPLICATION_PROCESS,
trace_id=trace_id_var.get(),
event_data=ApplicationProcessEventData(
app_id=assistant_id,
app_name=assistant_info.name,
app_type=ApplicationTypeEnum.ASSISTANT,
chat_id='',
start_time=int(start_time),
end_time=int(end_time),
process_time=int((end_time - start_time) * 1000)
))
def _check_model_supports_streaming(agent: AssistantAgent) -> bool:
"""
Check whether the model called by the helper supports streaming calls
Args:
agent: Assistant Proxy Instance
Returns:
bool: Does it support streaming calls?
"""
try:
# Othersagentright of privacyLLMDoes it support streaming?
if hasattr(agent, 'llm') and agent.llm:
# OthersBishengLLMright of privacystreamingProperty
if hasattr(agent.llm, 'streaming'):
return agent.llm.streaming
# Check Bottom Layerllmright of privacystreamProperty
elif hasattr(agent.llm, 'llm') and hasattr(agent.llm.llm, 'streaming'):
return agent.llm.llm.streaming
# If it cannot be determined, streaming is supported by default (most modernLLMare supported)
return True
except Exception as e:
logger.warning(f'Failed to check streaming support: {e}')
# Streaming is supported by default when an error occurs
return True
@router.get('/info/{assistant_id}')
async def get_assistant_info(request: Request, assistant_id: UUID):
"""
Getting Helper Information, Use the system configuration indefault_operator.userUser information to verify permissions
"""
assistant_id = assistant_id.hex
logger.info(f'act=get_default_operator assistant_id={assistant_id}, ip={get_request_ip(request)}')
# Determine if the configuration under is turned on
if not settings.get_from_db("default_operator").get("enable_guest_access"):
raise HTTPException(status_code=403, detail="No permission to access")
login_user = get_default_operator()
res = await AssistantService.get_assistant_info(assistant_id, login_user)
return resp_200(data=res)
@router.get('/list', status_code=200)
def get_assistant_list(request: Request,
name: str = Query(default=None, description='assistant name, fuzzy matching, Fuzzy matches with description'),
tag_id: int = Query(default=None, description='labelID'),
page: Optional[int] = Query(default=1, gt=0, description='Page'),
limit: Optional[int] = Query(default=10, gt=0, description='Listings Per Page'),
status: Optional[int] = Query(default=None, description='Is online status'),
user_id: int = None):
"""
Exposed interfaces for obtaining skill information
"""
logger.info(f'public_get_list ip: {request.client.host} user_id:{user_id}')
if not settings.get_from_db("default_operator").get("enable_guest_access"):
raise HTTPException(status_code=403, detail="No permission to access")
login_user = get_default_operator()
data, total = AssistantService.get_assistant(login_user, name, status, tag_id, page, limit)
return resp_200(PageData(data=data, total=total))
@router.websocket('/chat/{assistant_id}')
async def chat(*, websocket: WebSocket, assistant_id: str, chat_id: Optional[str] = None):
"""
Assistant'swsLogin-Free Interface
"""
logger.info(f'act=assistant_chat_ws assistant_id={assistant_id}, ip={get_request_ip(websocket)}')
login_user = get_default_operator()
try:
request = websocket
await chat_manager.dispatch_client(request, assistant_id, chat_id, login_user,
WorkType.GPTS, websocket)
except WebSocketException as exc:
logger.error(f'Websocket exception: {str(exc)}')
await websocket.close(code=http_status.WS_1011_INTERNAL_ERROR, reason=str(exc))
except Exception as exc:
logger.exception(f'Error in chat websocket: {str(exc)}')
message = exc.detail if isinstance(exc, HTTPException) else str(exc)
if 'Could not validate credentials' in str(exc):
await websocket.close(code=http_status.WS_1008_POLICY_VIOLATION, reason='Unauthorized')
else:
await websocket.close(code=http_status.WS_1011_INTERNAL_ERROR, reason=message)
================================================
FILE: src/backend/bisheng/open_endpoints/api/endpoints/chat.py
================================================
import json
from typing import List, Optional
from uuid import UUID
from fastapi import APIRouter, Body, WebSocket, status
from fastapi.middleware.wsgi import WSGIMiddleware
from loguru import logger
a = WSGIMiddleware
from bisheng.api.services.chat_imp import comment_answer
from bisheng.api.services.utils import set_flow_knowledge_id
from bisheng.api.v1.schemas import ChatInput, resp_200
from bisheng.chat.manager import ChatManager
from bisheng.common.services.config_service import settings
from bisheng.core.database import get_sync_db_session
from bisheng.database.models.flow import Flow
from bisheng.database.models.message import ChatMessage, ChatMessageDao
from bisheng.open_endpoints.domain.schemas.message import SyncMessage
from bisheng.processing.process import process_tweaks
router = APIRouter(prefix='/chat', tags=['OpenAPI', 'Chat'])
chat_manager = ChatManager()
expire = 600 # reids 60s Overdue
@router.websocket('/ws/{flow_id}')
async def union_websocket(flow_id: str,
websocket: WebSocket,
chat_id: Optional[str] = None,
tweak: Optional[str] = None,
knowledge_id: Optional[int] = None):
"""Websocket endpoint forF chat."""
if chat_id:
with get_sync_db_session() as session:
db_flow = session.get(Flow, flow_id)
if not db_flow:
await websocket.accept()
message = 'This skill has been deleted'
await websocket.close(code=status.WS_1008_POLICY_VIOLATION, reason=message)
if db_flow.status != 2:
await websocket.accept()
message = 'The current skill is not online and cannot be spoken to directly'
await websocket.close(code=status.WS_1008_POLICY_VIOLATION, reason=message)
graph_data = db_flow.data
try:
if tweak:
tweak = json.loads(tweak)
graph_data = process_tweaks(graph_data, tweak)
# vectordatabase update
if knowledge_id:
set_flow_knowledge_id(graph_data, knowledge_id)
await chat_manager.handle_websocket(
flow_id,
chat_id,
websocket,
settings.get_from_db('default_operator').get('user'),
gragh_data=graph_data,
source="api"
)
except Exception as exc:
logger.exception('union_websocket error: ')
await websocket.close(code=status.WS_1011_INTERNAL_ERROR, reason=str(exc))
# @router.get('/source')
# async def query_source(message_id: int, session: Session = Depends(get_session)):
# """source of message_id"""
# db_recall = session.query(RecallChunk).where(RecallChunk.message_id == message_id).all()
@router.post('/liked', status_code=200)
def like_response(*, data: dict):
message_id = data.get('message_id')
liked = data.get('liked')
with get_sync_db_session() as session:
message = session.get(ChatMessage, message_id)
message.liked = liked
session.add(message)
session.commit()
return {'status_code': 200, 'status_message': 'success'}
@router.post('/solved', status_code=200)
def solve_response(*, data: dict):
chat_id = data.get('chat_id')
solved = data.get('solved')
with get_sync_db_session() as session:
messages = session.query(ChatMessage).where(ChatMessage.chat_id == chat_id).all()
for message in messages:
message.solved = solved
with get_sync_db_session() as session:
session.add(message)
session.commit()
return {'status_code': 200, 'status_message': 'success'}
@router.post('/comment', status_code=200)
def comment(*, data: ChatInput):
comment_answer(data.message_id, data.comment)
return resp_200()
@router.post('/sync/messages', status_code=200)
def sync_message(*,
flow_id: UUID = Body(embed=True),
chat_id: str = Body(embed=True),
message_list: List[SyncMessage] = Body(embed=True),
user_id: int = Body(default=None, embed=True)):
flow_id = flow_id.hex
user_id = user_id if user_id else settings.get_from_db('default_operator').get('user')
batch_message = [
ChatMessage(is_bot=message.is_send,
source=0,
message=message.message,
extra=json.dumps(message.extra),
type='answer',
category='answer',
flow_id=flow_id,
user_id=user_id,
chat_id=chat_id,
create_time=message.create_time) for message in message_list
]
ChatMessageDao.insert_batch(batch_message)
return resp_200()
================================================
FILE: src/backend/bisheng/open_endpoints/api/endpoints/filelib.py
================================================
import asyncio
import json
import os
from typing import Dict, List, Optional
from fastapi import (APIRouter, BackgroundTasks, Body, File, Form, HTTPException, Query, Request,
UploadFile)
from loguru import logger
from starlette.responses import FileResponse
from bisheng.api.services import knowledge_imp
from bisheng.api.services.knowledge import KnowledgeService
from bisheng.api.services.knowledge_imp import (decide_vectorstores, delete_es, delete_vector,
text_knowledge)
from bisheng.api.v1.schemas import (ChunkInput, KnowledgeFileOne, KnowledgeFileProcess,
resp_200, resp_500, ExcelRule)
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum
from bisheng.common.errcode.http_error import ServerError
from bisheng.common.services import telemetry_service
from bisheng.common.services.config_service import settings
from bisheng.core.cache.utils import file_download, save_download_file, async_file_download
from bisheng.core.logger import trace_id_var
from bisheng.database.models.message import ChatMessageDao
from bisheng.interface.embeddings.custom import FakeEmbedding
from bisheng.knowledge.domain.models.knowledge import (KnowledgeCreate, KnowledgeDao, KnowledgeTypeEnum,
KnowledgeUpdate)
from bisheng.knowledge.domain.models.knowledge_file import (QAKnoweldgeDao, QAKnowledgeUpsert)
from bisheng.open_endpoints.domain.schemas.filelib import APIAddQAParam, APIAppendQAParam, QueryQAParam
from bisheng.open_endpoints.domain.utils import get_default_operator, get_default_operator_async
from bisheng.utils.util import sync_func_to_async
# build router
router = APIRouter(prefix='/filelib', tags=['OpenAPI', 'Knowledge'])
@router.post('/', status_code=201)
def create(request: Request, knowledge: KnowledgeCreate):
"""Create Knowledge Base."""
login_user = get_default_operator()
db_knowledge = KnowledgeService.create_knowledge(request, login_user, knowledge)
return resp_200(db_knowledge)
@router.put('/', status_code=201)
def update_knowledge(*, request: Request, knowledge: KnowledgeUpdate):
""" Update Knowledge Base."""
login_user = get_default_operator()
db_knowledge = KnowledgeService.update_knowledge(request, login_user, knowledge)
return resp_200(db_knowledge)
@router.get('/', status_code=200)
async def get_knowledge(*,
request: Request,
knowledge_type: int = Query(default=KnowledgeTypeEnum.NORMAL.value,
alias='type'),
name: str = None,
page_size: Optional[int] = 10,
page_num: Optional[int] = 1):
""" Read all knowledge base information. """
knowledge_type = KnowledgeTypeEnum(knowledge_type)
login_user = get_default_operator()
res, total = await KnowledgeService.get_knowledge(request, login_user, knowledge_type, name,
page_num, page_size)
return resp_200(data={'data': res, 'total': total})
@router.delete('/{knowledge_id}', status_code=200)
def delete_knowledge_api(*, request: Request, knowledge_id: int):
""" Delete Knowledge Base Information. """
login_user = get_default_operator()
KnowledgeService.delete_knowledge(request, login_user, knowledge_id)
return resp_200(message='knowledge deleted successfully')
# Empty all Knowledge Base file contents
@router.delete('/clear/{knowledge_id}', status_code=200)
def clear_knowledge_files(*, request: Request, knowledge_id: int):
""" Empty Knowledge Base Contents. """
login_user = get_default_operator()
KnowledgeService.delete_knowledge(request, login_user, knowledge_id, only_clear=True)
return resp_200(message='knowledge clear successfully')
@router.post('/file/{knowledge_id}')
async def upload_file(
request: Request,
knowledge_id: int,
separator: Optional[List[str]] = Form(default=None,
description='Split text rule, If not passed on, it is the default'),
separator_rule: Optional[List[str]] = Form(
default=None, description='Segmentation before or after the segmentation rule;before/after'),
chunk_size: Optional[int] = Form(default=None, description='Split text length, default if not passed'),
chunk_overlap: Optional[int] = Form(default=None,
description='Split text overlap length, default if not passed'),
callback_url: Optional[str] = Form(default=None, description='Return URL'),
file_url: Optional[str] = Form(default=None, description='File URL'),
file: Optional[UploadFile] = File(default=None, description='Upload file'),
background_tasks: BackgroundTasks = None,
retain_images: Optional[int] = Form(default=1, description='Keep document image'),
force_ocr: Optional[int] = Form(default=0, description='EnableOCR'),
enable_formula: Optional[int] = Form(default=1, description='latexFormula Recognition'),
filter_page_header_footer: Optional[int] = Form(default=0, description='Filter Header Footer'),
excel_rule: Optional[ExcelRule] = Form(default={}, description="excel rule"),
):
if file:
file_name = file.filename
if not file_name:
return resp_500(message='file name must be not empty')
# Cache Local
file_path = await sync_func_to_async(save_download_file)(save_download_file, file.file, 'bisheng', file_name)
else:
file_path, file_name = await async_file_download(file_url)
loging_user = await get_default_operator_async()
req_data = KnowledgeFileProcess(knowledge_id=knowledge_id,
separator=separator,
separator_rule=separator_rule,
chunk_size=chunk_size,
chunk_overlap=chunk_overlap,
retain_images=retain_images,
force_ocr=force_ocr,
enable_formula=enable_formula,
filter_page_header_footer=filter_page_header_footer,
callback_url=callback_url,
file_list=[KnowledgeFileOne(file_path=file_path, excel_rule=excel_rule)])
res = await sync_func_to_async(KnowledgeService.process_knowledge_file)(request=request,
login_user=loging_user,
background_tasks=background_tasks,
req_data=req_data)
return resp_200(data=res[0])
@router.delete('/file/{file_id}', status_code=200)
def delete_knowledge_file(request: Request, file_id: int):
""" Delete files in the Knowledge Base """
login_user = get_default_operator()
KnowledgeService.delete_knowledge_file(request, login_user, [file_id])
return resp_200()
@router.post('/delete_file', status_code=200)
def delete_file_batch_api(request: Request, file_ids: List[int]):
""" Bulk delete knowledge file information """
login_user = get_default_operator()
KnowledgeService.delete_knowledge_file(request, login_user, file_ids)
return resp_200()
@router.get('/file/list', status_code=200)
def get_filelist(request: Request,
knowledge_id: int,
keyword: str = None,
status: List[int] = Query(default=None),
page_size: int = 10,
page_num: int = 1):
""" Get knowledge base file information. """
login_user = get_default_operator()
data, total, flag = KnowledgeService.get_knowledge_files(request, login_user, knowledge_id,
keyword, status, page_num, page_size)
return resp_200(data={'data': data, 'total': total, 'writeable': flag})
@router.post('/chunks')
async def post_chunks(request: Request,
knowledge_id: int = Form(...),
metadata: str = Form(...),
separator: Optional[List[str]] = Form(default=None),
separator_rule: Optional[List[str]] = Form(default=None),
chunk_size: Optional[int] = Form(default=None),
chunk_overlap: Optional[int] = Form(default=None),
file: UploadFile = File(...)):
""" Upload files to the knowledge base and sync the interface """
file_name = file.filename
if not file_name:
return resp_500(message='file name must be not empty')
file_path = await sync_func_to_async(save_download_file)(file.file, 'bisheng', file_name)
login_user = await get_default_operator_async()
req_data = KnowledgeFileProcess(knowledge_id=knowledge_id,
separator=separator,
separator_rule=separator_rule,
chunk_size=chunk_size,
chunk_overlap=chunk_overlap,
file_list=[KnowledgeFileOne(file_path=file_path)],
extra=metadata)
res = await sync_func_to_async(KnowledgeService.sync_process_knowledge_file)(request, login_user, req_data)
return resp_200(data=res[0])
@router.post('/chunks_string')
async def post_string_chunks(request: Request, document: ChunkInput):
""" Get knowledge base file information. """
# String saved to file
content = '\n\n'.join([doc.page_content for doc in document.documents])
content_bytes = bytes(content, encoding='utf-8')
file_name = document.documents[0].metadata.get('source')
file_path = await sync_func_to_async(save_download_file)(content_bytes, 'bisheng', file_name)
login_user = await get_default_operator_async()
req_data = KnowledgeFileProcess(knowledge_id=document.knowledge_id,
separator=['\n\n'],
separator_rule=['after'],
file_list=[KnowledgeFileOne(file_path=file_path)],
extra=json.dumps(document.documents[0].metadata,
ensure_ascii=False))
knowledge, failed_files, process_files, _ = await sync_func_to_async(KnowledgeService.save_knowledge_file)(
login_user, req_data)
if failed_files:
return resp_200(data=failed_files[0])
res = await sync_func_to_async(text_knowledge)(knowledge, process_files[0], document.documents)
return resp_200(data=res)
@router.post('/chunk_clear', status_code=200)
async def clear_tmp_chunks_data(body: Dict):
# Delete via Interfacemilvus、es DATA
flow_id = body.get('flow_id')
chat_id = body.get('chat_id')
if flow_id and not chat_id:
# Clean temporary files under the skill
flow_id = flow_id.replace('-', '')
collection_name = f'tmp_{flow_id}_1'
delete_es(collection_name)
delete_vector(collection_name, None)
if chat_id:
# Query auto-generated
message = ChatMessageDao.get_latest_message_by_chatid(chat_id)
if message:
collection_name = f'tmp_{message.flow_id}_{chat_id}'
delete_es(collection_name)
delete_vector(collection_name, None)
return resp_200()
@router.get('/dump_vector', status_code=200)
def dump_vector_knowledge(collection_name: str, expr: str = None, store: str = 'Milvus'):
# dump vector db
embedding_tmp = FakeEmbedding()
vector_store = decide_vectorstores(collection_name, store, embedding_tmp)
if vector_store and vector_store.col:
fields = [
s.name for s in vector_store.col.schema.fields
if s.name not in ['pk', 'bbox', 'vector']
]
res_list = vector_store.col.query('file_id>1', output_fields=fields)
return resp_200(res_list)
else:
return resp_500('Parameter salah')
@router.get('/download_statistic')
def download_statistic_file(file_path: str):
suffix = file_path.split('.')[-1]
if suffix != 'log':
raise ServerError.http_exception(msg='only .log file supported download')
dir_path = file_path.replace('.log', '')
if dir_path.find(".") != -1 or not dir_path.startswith("/app/data"):
raise ServerError.http_exception(msg='invalid file path, file path must not contain .')
file_name = os.path.basename(file_path)
return FileResponse(file_path, filename=file_name)
@router.post('/add_qa')
def add_qa(*,
knowledge_id: int = Body(embed=True),
data: List[APIAddQAParam] = Body(embed=True),
user_id: Optional[int] = Body(default=None, embed=True)):
user_id = user_id if user_id else settings.get_from_db('default_operator').get('user')
knowledge = KnowledgeDao.query_by_id(knowledge_id)
logger.info('add_qa_data knowledge_id={} size={}', knowledge_id, len(data))
res = []
for item in data:
qa_insert = QAKnowledgeUpsert(knowledge_id=knowledge_id,
questions=[item.question],
answers=item.answer,
user_id=user_id,
extra_meta=json.dumps(item.extra),
source=3)
res.append(knowledge_imp.add_qa(knowledge, qa_insert))
return resp_200(res)
@router.post('/add_relative_qa')
def append_qa(*,
knowledge_id: int = Body(embed=True),
data: APIAppendQAParam = Body(embed=True),
user_id: Optional[int] = Body(default=None, embed=True)):
user_id = user_id if user_id else settings.get_from_db('default_operator').get('user')
knowledge = KnowledgeDao.query_by_id(knowledge_id)
qa_db = QAKnoweldgeDao.get_qa_knowledge_by_primary_id(data.id)
if not qa_db:
return HTTPException(404, detail='qa Right, nothing found.')
t = qa_db.dict()
t['answers'] = json.loads(t['answers'])
qa_insert = QAKnowledgeUpsert.validate(t)
qa_insert.questions.extend(data.relative_questions)
return resp_200(knowledge_imp.add_qa(knowledge, qa_insert))
@router.delete('/qa/{qa_id}', status_code=200)
def delete_qa_data(*, qa_id: int, question: Optional[str] = None):
""" Deleteqa Question to Information """
qa = QAKnoweldgeDao.get_qa_knowledge_by_primary_id(qa_id)
login_user = get_default_operator()
if not qa:
raise HTTPException(status_code=404, detail='qa Does not exist')
if question:
qa.questions = [q for q in qa.questions if q != question]
QAKnoweldgeDao.update(qa)
else:
QAKnoweldgeDao.delete_batch([qa_id])
telemetry_service.log_event_sync(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.DELETE_KNOWLEDGE_FILE,
trace_id=trace_id_var.get())
try:
knowledge = KnowledgeDao.query_by_id(qa.knowledge_id)
knowledge_imp.delete_vector_data(knowledge, file_ids=[qa_id])
if question:
knowledge_imp.QA_save_knowledge(knowledge, qa)
return resp_200()
except Exception as e:
return resp_500(message=f'error e={str(e)}')
@router.post('/update_qa', status_code=200)
def update_qa(
*,
id: int = Body(embed=True),
question: Optional[str] = Body(default=None, embed=True),
original_question: Optional[str] = Body(default=None, embed=True),
answer: Optional[List[str]] = Body(default=None, embed=True),
):
""" Deleteqa Question to Information """
qa = QAKnoweldgeDao.get_qa_knowledge_by_primary_id(id)
if not qa:
raise HTTPException(status_code=404, detail='qa Does not exist')
if original_question:
qa.questions = [q if q != question else question for q in qa.questions]
else:
qa.questions = [question]
if answer:
qa.answers = json.dumps(answer, ensure_ascii=False)
QAKnoweldgeDao.update(qa)
try:
knowledge = KnowledgeDao.query_by_id(qa.knowledge_id)
if question:
knowledge_imp.delete_vector_data(knowledge, file_ids=[id])
knowledge_imp.QA_save_knowledge(knowledge, qa)
return resp_200()
except Exception as e:
return resp_500(message=f'error e={str(e)}')
@router.get('/detail_qa', status_code=200)
def detail_qa(*, id: int):
""" Get questions on information """
qa = QAKnoweldgeDao.get_qa_knowledge_by_primary_id(id)
return resp_200(qa)
@router.post('/query_qa', status_code=200)
def query_qa(QueryQAParam: QueryQAParam):
""" Deleteqa Question to Information """
sources = [1, 2] # 3 Yes apiInverted
qa_list = QAKnoweldgeDao.query_by_condition_v1(source=sources,
create_start=QueryQAParam.timeRange[0],
create_end=QueryQAParam.timeRange[1])
if qa_list:
for q in qa_list:
q.answers = json.loads(q.answers)
return resp_200(qa_list)
================================================
FILE: src/backend/bisheng/open_endpoints/api/endpoints/flow.py
================================================
from uuid import UUID
from fastapi import APIRouter, HTTPException, Query, Request
from loguru import logger
from bisheng.api.services.flow import FlowService
from bisheng.common.services.config_service import settings
from bisheng.open_endpoints.api.endpoints.assistant import get_default_operator
router = APIRouter(prefix='/flows', tags=['OpenAPI', 'FlowV2'])
@router.get('/{flow_id}', status_code=200)
async def get_flow(request: Request, flow_id: UUID):
"""
Exposed interfaces for obtaining skill information
"""
flow_id = flow_id.hex
logger.info(f'public_get_flow ip: {request.client.host} flow_id:{flow_id}')
# Determine if the configuration under is turned on
if not settings.get_from_db("default_operator").get("enable_guest_access"):
raise HTTPException(status_code=403, detail="No permission to access")
default_user = get_default_operator()
return await FlowService.get_one_flow(default_user, flow_id)
@router.get('', status_code=200)
def get_flow_list(request: Request,
name: str = Query(default=None, description='accordingnameFind databases with fuzzy searches for descriptions'),
tag_id: int = Query(default=None, description='labelID'),
page_size: int = Query(default=10, description='Items per page'),
page_num: int = Query(default=1, description='Page'),
status: int = None,
user_id: int = None):
"""
Exposed interfaces for obtaining skill information
"""
logger.info(f'public_get_flow_list ip: {request.client.host} user_id={user_id}')
login_user = get_default_operator()
try:
return FlowService.get_all_flows(login_user, name, status, tag_id, page_num, page_size)
except Exception as e:
logger.error(e)
raise HTTPException(status_code=500, detail='Failed to get skills list')
================================================
FILE: src/backend/bisheng/open_endpoints/api/endpoints/knowledge.py
================================================
from typing import List
from fastapi import APIRouter, Depends, BackgroundTasks, Body
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.schemas.api import UnifiedResponseModel, resp_200
from bisheng.knowledge.domain.schemas.knowledge_schema import AddKnowledgeMetadataFieldsReq, \
UpdateKnowledgeMetadataFieldsReq, ModifyKnowledgeFileMetaDataReq
from bisheng.knowledge.domain.services.knowledge_file_service import KnowledgeFileService
from bisheng.knowledge.domain.services.knowledge_service import KnowledgeService
from bisheng.open_endpoints.api.dependencies import get_knowledge_service, get_knowledge_file_service
from bisheng.open_endpoints.domain.schemas.knowledge import DeleteUserMetadataReq
from bisheng.open_endpoints.domain.utils import get_default_operator_async
router = APIRouter(prefix='/knowledge', tags=['OpenAPI', 'Knowledge'])
# Add Knowledge Metadata Fields Endpoint
@router.post('/add_metadata_fields', response_model=UnifiedResponseModel)
async def add_metadata_fields(*,
default_user: UserPayload = Depends(get_default_operator_async),
req_data: AddKnowledgeMetadataFieldsReq,
knowledge_service: KnowledgeService = Depends(get_knowledge_service)):
"""
Add metadata fields to a knowledge base.
Args:
default_user:
req_data:
knowledge_service:
Returns:
"""
await knowledge_service.add_metadata_fields(default_user, req_data)
return resp_200(data=True)
@router.put("/modify_metadata_fields", response_model=UnifiedResponseModel)
async def update_metadata_fields(*,
default_user: UserPayload = Depends(get_default_operator_async),
req_data: UpdateKnowledgeMetadataFieldsReq,
knowledge_service: KnowledgeService = Depends(get_knowledge_service),
background_tasks: BackgroundTasks):
"""
Update metadata fields in a knowledge base.
Args:
default_user:
req_data:
knowledge_service:
background_tasks:
Returns:
"""
await knowledge_service.update_metadata_fields(
default_user, req_data, background_tasks)
return resp_200(data=True)
@router.delete('/delete_metadata_fields', response_model=UnifiedResponseModel)
async def delete_metadata_fields(*,
default_user: UserPayload = Depends(get_default_operator_async),
knowledge_id: int = Body(..., embed=True, description="Knowledge ID"),
field_names: List[str] = Body(..., embed=True,
description="List of field names to delete"),
knowledge_service: KnowledgeService = Depends(get_knowledge_service),
background_tasks: BackgroundTasks):
"""
Delete metadata fields from a knowledge base.
Args:
default_user:
knowledge_id:
field_names:
knowledge_service:
background_tasks:
Returns:
"""
await knowledge_service.delete_metadata_fields(
default_user, knowledge_id, field_names, background_tasks)
return resp_200(data=True)
@router.get('/get_metadata_fields/{knowledge_id}', response_model=UnifiedResponseModel)
async def list_metadata_fields(*,
default_user: UserPayload = Depends(get_default_operator_async),
knowledge_id: int,
knowledge_service: KnowledgeService = Depends(get_knowledge_service)):
"""
List metadata fields of a knowledge base.
Args:
default_user:
knowledge_id:
knowledge_service:
Returns:
"""
metadata_fields = await knowledge_service.list_metadata_fields(default_user, knowledge_id)
return resp_200(data=metadata_fields)
@router.post('/file/add_user_metadata', response_model=UnifiedResponseModel)
async def add_file_user_metadata(*,
default_user: UserPayload = Depends(get_default_operator_async),
knowledge_id: int = Body(..., embed=True, description="Knowledge ID"),
add_metadata_list: List[ModifyKnowledgeFileMetaDataReq] = Body(...,
description="File User Metadata List"),
knowledge_file_service: KnowledgeFileService = Depends(get_knowledge_file_service)):
"""
Add user metadata to a knowledge file.
Args:
knowledge_id:
add_metadata_list:
default_user:
knowledge_file_service:
Returns:
"""
await knowledge_file_service.add_file_user_metadata(
default_user, knowledge_id, add_metadata_list)
return resp_200(data=True)
@router.put('/file/modify_user_metadata', response_model=UnifiedResponseModel)
async def modify_file_user_metadata(*,
default_user: UserPayload = Depends(get_default_operator_async),
knowledge_id: int = Body(..., embed=True, description="Knowledge ID"),
modify_metadata_list: List[ModifyKnowledgeFileMetaDataReq] = Body(...,
description="File User Metadata List"),
knowledge_file_service: KnowledgeFileService = Depends(
get_knowledge_file_service)):
"""
Modify user metadata of a knowledge file.
Args:
modify_metadata_list:
knowledge_id:
default_user:
knowledge_file_service:
Returns:
"""
await knowledge_file_service.batch_modify_file_user_metadata(default_user, knowledge_id,
modify_metadata_list)
return resp_200(data=True)
@router.delete('/file/delete_user_metadata', response_model=UnifiedResponseModel)
async def delete_file_user_metadata(*,
default_user: UserPayload = Depends(get_default_operator_async),
knowledge_id: int = Body(..., embed=True, description="Knowledge ID"),
delete_user_metadatas: List[DeleteUserMetadataReq] = Body(...,
description="Delete User Metadata List"),
knowledge_file_service: KnowledgeFileService = Depends(
get_knowledge_file_service)):
"""
Delete user metadata from a knowledge file.
Args:
default_user:
knowledge_id:
delete_user_metadatas:
knowledge_file_service:
Returns:
"""
await knowledge_file_service.batch_delete_file_user_metadata(
default_user, knowledge_id, delete_user_metadatas)
return resp_200(data=True)
@router.post('/file/list_user_metadata', response_model=UnifiedResponseModel)
async def list_file_user_metadata(*,
default_user: UserPayload = Depends(get_default_operator_async),
knowledge_id: int = Body(..., embed=True, description="Knowledge ID"),
knowledge_file_ids: List[int] = Body(..., description="Knowledge File IDs"),
knowledge_file_service: KnowledgeFileService = Depends(
get_knowledge_file_service)):
"""
List user metadata of knowledge files.
Args:
default_user:
knowledge_id:
knowledge_file_ids:
knowledge_file_service:
Returns:
"""
metadata_list = await knowledge_file_service.list_knowledge_file_user_metadata(
default_user, knowledge_id, knowledge_file_ids)
return resp_200(data=metadata_list)
================================================
FILE: src/backend/bisheng/open_endpoints/api/endpoints/llm.py
================================================
from fastapi import APIRouter, Request, UploadFile, Body
from bisheng.common.schemas.api import resp_200
from bisheng.llm.domain import LLMService
from bisheng.open_endpoints.domain.utils import get_default_operator
router = APIRouter(prefix='/llm', tags=['OpenAPI', 'llm'])
@router.post('/workbench/asr')
async def invoke_workbench_asr(request: Request, file: UploadFile = None):
""" Call the workbench'sasrModels Convert Voice to Text """
login_user = get_default_operator()
text = await LLMService.invoke_workbench_asr(login_user, file)
return resp_200(data=text)
@router.post('/workbench/tts')
async def invoke_workbench_tts(request: Request, text: str = Body(..., embed=True, description="Text that needs to be synthesized")):
""" Call the workbench'sttsModels Convert text to speech """
login_user = get_default_operator()
audio_url = await LLMService.invoke_workbench_tts(login_user, text)
return resp_200(data=audio_url)
================================================
FILE: src/backend/bisheng/open_endpoints/api/endpoints/workflow.py
================================================
import time
import uuid
from typing import Optional, List
from uuid import UUID
from fastapi import APIRouter, Request, Body, Path, WebSocket, WebSocketException
from fastapi import status as http_status
from fastapi.responses import ORJSONResponse
from loguru import logger
from starlette.responses import StreamingResponse
from bisheng.api.services.workflow import WorkFlowService
from bisheng.api.v1.chat import chat_manager
from bisheng.api.v1.schema.workflow import WorkflowStream, WorkflowEvent, WorkflowEventType
from bisheng.api.v1.schemas import resp_200
from bisheng.chat.types import WorkType
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum, ApplicationTypeEnum
from bisheng.common.errcode.http_error import NotFoundError, ServerError
from bisheng.common.schemas.telemetry.event_data_schema import ApplicationAliveEventData
from bisheng.common.services import telemetry_service
from bisheng.core.logger import trace_id_var
from bisheng.database.models.flow import FlowDao, FlowType
from bisheng.open_endpoints.domain.utils import get_default_operator, get_default_operator_async
from bisheng.worker.workflow.redis_callback import RedisCallback
from bisheng.worker.workflow.tasks import execute_workflow, continue_workflow, workflow_stateful_worker
from bisheng.workflow.common.workflow import WorkflowStatus
router = APIRouter(prefix='/workflow', tags=['OpenAPI', 'Workflow'])
@router.post('/invoke')
async def invoke_workflow(request: Request,
workflow_id: UUID = Body(..., description='Workflow UniqueID'),
override: Optional[dict] = Body(default=None, description='override node params'),
stream: Optional[bool] = Body(default=True, description='Whether to stream calls'),
user_input: Optional[dict] = Body(default=None, description='User input', alias='input'),
message_id: Optional[int] = Body(default=None,
description='MessageID,Once,Unique identifier of user input message'),
session_id: Optional[str] = Body(default=None,
description='会话ID,一次workflow调用的唯一标识')):
login_user = get_default_operator()
workflow_id = workflow_id.hex
# Query workflow information
workflow_info = await FlowDao.aget_flow_by_id(workflow_id)
if not workflow_info:
raise NotFoundError.http_exception()
if workflow_info.flow_type != FlowType.WORKFLOW.value:
raise NotFoundError.http_exception()
# Resolve Outchat_idAndunique_id
if not session_id:
chat_id = uuid.uuid4().hex
unique_id = f'{chat_id}_async_task_id'
session_id = unique_id
else:
chat_id = session_id.split('_', 1)[0]
unique_id = session_id
start_time = time.time()
logger.debug(f'invoke_workflow: {workflow_id}, {session_id}')
workflow = RedisCallback(unique_id, workflow_id, chat_id, login_user.user_id, source="api")
execute_worker = await workflow_stateful_worker.find_task_node(chat_id)
# Query workflow status
status_info = workflow.get_workflow_status()
if not status_info:
# Initialize workflow
workflow.set_workflow_data(workflow_info.data, override=override)
workflow.set_workflow_status(WorkflowStatus.WAITING.value)
# Start asynchronous task
execute_workflow.apply_async([unique_id, workflow_id, chat_id, login_user.user_id, "api"],
queue=execute_worker)
elif status_info['status'] == WorkflowStatus.INPUT.value:
if not user_input:
raise ServerError(msg="workflow waiting for user input, but user input not provided")
# Set user input
if not message_id:
raise ServerError(msg="message_id is required when providing user input")
await workflow.async_set_user_input(user_input, message_id, verify_input=True)
await workflow.async_set_workflow_status(WorkflowStatus.INPUT_OVER.value)
continue_workflow.apply_async([unique_id, workflow_id, chat_id, login_user.user_id, "api"],
queue=execute_worker)
logger.debug(f'waiting workflow over or input: {workflow_id}, {session_id}')
async def handle_workflow_event(event_list: List):
async for event in workflow.get_response_until_break():
if event.category == WorkflowEventType.NodeRun.value:
continue
# Non-streaming requests, filtering out streaming output events generated by nodes
if not stream and event.category == WorkflowEventType.StreamMsg.value and event.type == 'stream':
continue
workflow_stream = WorkflowStream(session_id=session_id,
data=WorkFlowService.convert_chat_response_to_workflow_event(event))
event_list.append(workflow_stream.data)
yield f'data: {workflow_stream.model_dump_json()}\n\n'
tmp_status_info = await workflow.async_get_workflow_status()
if tmp_status_info['status'] in [WorkflowStatus.SUCCESS.value, WorkflowStatus.FAILED.value]:
await workflow.async_clear_workflow_status()
if tmp_status_info['status'] == WorkflowStatus.SUCCESS.value:
workflow_stream = WorkflowStream(session_id=session_id,
data=WorkflowEvent(event=WorkflowEventType.Close.value))
event_list.append(workflow_stream.data)
yield f'data: {workflow_stream.model_dump_json()}\n\n'
res = []
# Non-streaming returns a cumulative list of events
if not stream:
async for _ in handle_workflow_event(res):
pass
end_time = time.time()
await telemetry_service.log_event(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.APPLICATION_ALIVE,
trace_id=trace_id_var.get(),
event_data=ApplicationAliveEventData(
app_id=workflow_id,
app_name=workflow_info.name,
app_type=ApplicationTypeEnum.WORKFLOW,
chat_id=session_id,
start_time=int(start_time),
end_time=int(end_time)))
return resp_200(data={
'session_id': session_id,
'events': res
})
try:
return StreamingResponse(handle_workflow_event(res), media_type='text/event-stream')
except Exception as exc:
logger.exception(f'invoke_workflow error: {str(exc)}')
return ORJSONResponse(status_code=500, content=str(exc))
finally:
end_time = time.time()
await telemetry_service.log_event(user_id=login_user.user_id,
event_type=BaseTelemetryTypeEnum.APPLICATION_ALIVE,
trace_id=trace_id_var.get(),
event_data=ApplicationAliveEventData(
app_id=workflow_id,
app_name=workflow_info.name,
app_type=ApplicationTypeEnum.WORKFLOW,
chat_id=session_id,
start_time=int(start_time),
end_time=int(end_time)))
@router.post('/stop')
async def stop_workflow(request: Request,
workflow_id: UUID = Body(..., description='Workflow UniqueID'),
session_id: str = Body(description='SessionsID,Once,workflowUnique identifier of the call')):
workflow_id = workflow_id.hex
login_user = await get_default_operator_async()
chat_id = session_id.split('_', 1)[0]
unique_id = session_id
workflow = RedisCallback(unique_id, workflow_id, chat_id, login_user.user_id, source="api")
await workflow.async_set_workflow_stop()
return resp_200()
@router.websocket('/chat/{workflow_id}')
async def workflow_ws(*,
workflow_id: UUID = Path(..., description='Workflow UniqueID'),
websocket: WebSocket,
chat_id: Optional[str] = None):
""" Use Exempt Login Link """
try:
workflow_id = workflow_id.hex
# Authorize.jwt_required(auth_from='websocket', websocket=websocket)
# payload = Authorize.get_jwt_subject()
login_user = await get_default_operator_async()
await chat_manager.dispatch_client(websocket, workflow_id, chat_id, login_user, WorkType.WORKFLOW, websocket)
except WebSocketException as exc:
logger.error(f'Websocket exception: {str(exc)}')
await websocket.close(code=http_status.WS_1011_INTERNAL_ERROR, reason=str(exc))
except Exception as e:
logger.error(f'Websocket handle error: {str(e)}')
await websocket.close(code=http_status.WS_1011_INTERNAL_ERROR, reason=str(e))
================================================
FILE: src/backend/bisheng/open_endpoints/api/router.py
================================================
from bisheng.open_endpoints.api.endpoints.assistant import router as assistant_router_rpc
from bisheng.open_endpoints.api.endpoints.chat import router as chat_router_rpc
from bisheng.open_endpoints.api.endpoints.filelib import router as filelib_router_rpc
from bisheng.open_endpoints.api.endpoints.flow import router as flow_router
from bisheng.open_endpoints.api.endpoints.workflow import router as workflow_router_rpc
from .endpoints.knowledge import router as knowledge_router_rpc
__all__ = [
'knowledge_router_rpc', 'chat_router_rpc', 'flow_router',
'assistant_router_rpc', 'workflow_router_rpc', 'filelib_router_rpc'
]
================================================
FILE: src/backend/bisheng/open_endpoints/domain/__init__.py
================================================
================================================
FILE: src/backend/bisheng/open_endpoints/domain/schemas/__init__.py
================================================
================================================
FILE: src/backend/bisheng/open_endpoints/domain/schemas/filelib.py
================================================
from typing import Dict, List, Optional
from pydantic import BaseModel
class APIAddQAParam(BaseModel):
question: str
answer: List[str]
extra: Optional[Dict] = {}
class APIAppendQAParam(BaseModel):
relative_questions: List[str] = []
id: str = None
class QueryQAParam(BaseModel):
timeRange: List[str]
================================================
FILE: src/backend/bisheng/open_endpoints/domain/schemas/knowledge.py
================================================
from pydantic import BaseModel
class DeleteUserMetadataReq(BaseModel):
knowledge_file_id: int
field_names: list[str]
================================================
FILE: src/backend/bisheng/open_endpoints/domain/schemas/message.py
================================================
from pydantic import BaseModel
class SyncMessage(BaseModel):
is_send: bool
message: str
create_time: str
extra: dict
================================================
FILE: src/backend/bisheng/open_endpoints/domain/utils.py
================================================
from fastapi.exceptions import HTTPException
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.services.config_service import settings
from bisheng.user.domain.models.user import UserDao
def get_default_operator() -> UserPayload:
user_id = settings.get_from_db('default_operator').get('user')
if not user_id:
raise HTTPException(status_code=500, detail='Not configureddefault_operatorIIuserConfigure')
# Find default user information
login_user = UserDao.get_user(user_id)
if not login_user:
raise HTTPException(status_code=500, detail='not founddefault_operatorIIuserUser Information')
login_user = UserPayload.init_login_user_sync(user_id=login_user.user_id, user_name=login_user.user_name)
return login_user
async def get_default_operator_async() -> UserPayload:
user_id = (await settings.aget_from_db('default_operator')).get('user')
if not user_id:
raise HTTPException(status_code=500, detail='Not configureddefault_operatorIIuserConfigure')
# Find default user information
login_user = await UserDao.aget_user(user_id)
if not login_user:
raise HTTPException(status_code=500, detail='not founddefault_operatorIIuserUser Information')
login_user = await UserPayload.init_login_user(user_id=login_user.user_id, user_name=login_user.user_name)
return login_user
================================================
FILE: src/backend/bisheng/patches/__init__.py
================================================
================================================
FILE: src/backend/bisheng/patches/langchain_openai.patch
================================================
--- base.py 2024-12-18 19:01:22.000000000 +0800
+++ base_modify.py 2025-02-10 11:11:47.899000000 +0800
@@ -259,6 +259,8 @@
]
except KeyError:
pass
+ if _dict.get('reasoning_content'):
+ additional_kwargs['reasoning_content'] = _dict['reasoning_content']
if role == "user" or default_class == HumanMessageChunk:
return HumanMessageChunk(content=content, id=id_)
================================================
FILE: src/backend/bisheng/pptx2md/__init__.py
================================================
# Copyright 2024 Liu Siyao
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from .entry import convert
from .types import ConversionConfig
__all__ = ['convert', 'ConversionConfig']
================================================
FILE: src/backend/bisheng/pptx2md/__main__.py
================================================
# Copyright 2024 Liu Siyao
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import argparse
import logging
from pathlib import Path
from bisheng.pptx2md.entry import convert
from bisheng.pptx2md.log import setup_logging
from bisheng.pptx2md.types import ConversionConfig
setup_logging(compat_tqdm=True)
logger = logging.getLogger(__name__)
def parse_args() -> ConversionConfig:
arg_parser = argparse.ArgumentParser(description='Convert pptx to markdown')
arg_parser.add_argument('pptx_path', type=Path, help='path to the pptx file to be converted')
arg_parser.add_argument('-t', '--title', type=Path, help='path to the custom title list file')
arg_parser.add_argument('-o', '--output', type=Path, help='path of the output file')
arg_parser.add_argument('-i', '--image-dir', type=Path, help='where to put images extracted')
arg_parser.add_argument('--image-width', type=int, help='maximum image with in px')
arg_parser.add_argument('--disable-image', action="store_true", help='disable image extraction')
arg_parser.add_argument('--disable-wmf',
action="store_true",
help='keep wmf formatted image untouched(avoid exceptions under linux)')
arg_parser.add_argument('--disable-color', action="store_true", help='do not add color HTML tags')
arg_parser.add_argument('--disable-escaping',
action="store_true",
help='do not attempt to escape special characters')
arg_parser.add_argument('--disable-notes', action="store_true", help='do not add presenter notes')
arg_parser.add_argument('--enable-slides', action="store_true", help='deliniate slides `\n---\n`')
arg_parser.add_argument('--try-multi-column', action="store_true", help='try to detect multi-column slides')
arg_parser.add_argument('--wiki', action="store_true", help='generate output as wikitext(TiddlyWiki)')
arg_parser.add_argument('--mdk', action="store_true", help='generate output as madoko markdown')
arg_parser.add_argument('--qmd', action="store_true", help='generate output as quarto markdown presentation')
arg_parser.add_argument('--min-block-size',
type=int,
default=15,
help='the minimum character number of a text block to be converted')
arg_parser.add_argument("--page", type=int, default=None, help="only convert the specified page")
arg_parser.add_argument(
"--keep-similar-titles",
action="store_true",
help="keep similar titles (allow for repeated slide titles - One or more - Add (cont.) to the title)")
args = arg_parser.parse_args()
# Determine output path if not specified
if args.output is None:
extension = '.tid' if args.wiki else '.qmd' if args.qmd else '.md'
args.output = Path(f'out{extension}')
return ConversionConfig(
pptx_path=args.pptx_path,
output_path=args.output,
image_dir=args.image_dir or args.output.parent / 'img',
title_path=args.title,
image_width=args.image_width,
disable_image=args.disable_image,
disable_wmf=args.disable_wmf,
disable_color=args.disable_color,
disable_escaping=args.disable_escaping,
disable_notes=args.disable_notes,
enable_slides=args.enable_slides,
try_multi_column=args.try_multi_column,
is_wiki=args.wiki,
is_mdk=args.mdk,
is_qmd=args.qmd,
min_block_size=args.min_block_size,
page=args.page,
keep_similar_titles=args.keep_similar_titles,
)
def main():
config = parse_args()
convert(config)
if __name__ == '__main__':
main()
================================================
FILE: src/backend/bisheng/pptx2md/entry.py
================================================
# Copyright 2024 Liu Siyao
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import logging
import bisheng.pptx2md.outputter as outputter
from bisheng.pptx2md.parser import parse
from bisheng.pptx2md.types import ConversionConfig
from bisheng.pptx2md.utils import load_pptx, prepare_titles
logger = logging.getLogger(__name__)
def convert(config: ConversionConfig):
if config.title_path:
config.custom_titles = prepare_titles(config.title_path)
prs = load_pptx(config.pptx_path)
logger.info("conversion started")
ast = parse(config, prs)
if str(config.output_path).endswith('.json'):
with open(config.output_path, 'w') as f:
f.write(ast.model_dump_json(indent=2))
logger.info(f'presentation data saved to {config.output_path}')
return
if config.is_wiki:
out = outputter.WikiFormatter(config)
elif config.is_mdk:
out = outputter.MadokoFormatter(config)
elif config.is_qmd:
out = outputter.QuartoFormatter(config)
else:
out = outputter.MarkdownFormatter(config)
out.output(ast)
logger.info(f'converted document saved to {config.output_path}')
================================================
FILE: src/backend/bisheng/pptx2md/image.py
================================================
# Copyright 2024 Liu Siyao
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# --coding:utf-8--
# author = ''
def convert_wmf_to_png(input_file, output_png_path):
"""
Convert WMF data to a PNG file.
"""
# from PIL import ImageGrab
# shape.Copy()
# image = ImageGrab.grabclipboard()
# #image.save('{}.jpg'.format(filename), 'jpeg')
# image.save(output_png_path)
# from PIL import Image
# Image.open(input_file).save(output_png_path)
from wand.image import Image
with Image(filename=input_file) as img:
img.format = 'png'
img.save(filename=output_png_path)
================================================
FILE: src/backend/bisheng/pptx2md/log.py
================================================
# Copyright 2024 Liu Siyao
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import logging
import sys
from tqdm import tqdm
class TqdmStreamHandler(logging.StreamHandler):
def emit(self, record):
try:
msg = self.format(record)
tqdm.write(msg)
self.flush()
except Exception:
self.handleError(record)
def setup_logging(compat_tqdm=True):
root_logger = logging.getLogger()
for handler in root_logger.handlers[:]:
root_logger.removeHandler(handler)
formatter = logging.Formatter(
'%(asctime)s | %(levelname)s | %(name)s | %(message)s',
datefmt='%Y-%m-%d %H:%M:%S',
)
if compat_tqdm:
stream_handler = TqdmStreamHandler(sys.stdout)
else:
stream_handler = logging.StreamHandler(sys.stdout)
stream_handler.setLevel(logging.INFO)
stream_handler.setFormatter(formatter)
root_logger.setLevel(logging.INFO)
root_logger.addHandler(stream_handler)
================================================
FILE: src/backend/bisheng/pptx2md/multi_column.py
================================================
# Copyright 2024 Liu Siyao
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import logging
from operator import attrgetter
from typing import Optional
import numpy as np
import pptx
from pptx import Presentation
from pptx.enum.shapes import MSO_SHAPE_TYPE, PP_PLACEHOLDER
from pptx.util import Length
from scipy.optimize import curve_fit
from bisheng.pptx2md.types import MultiColumnSlide
logger = logging.getLogger(__name__)
def normal_pdf(x_vector, mu=0, sigma=1):
return (1 / (sigma * np.sqrt(2 * np.pi))) * np.exp(-((x_vector - mu) / sigma)**2 / 2)
def f(x_vector, theta0, theta1, sigma0, sigma1):
# sigma = 100
return 0.5 * ((1 / (sigma0 * np.sqrt(2 * np.pi))) * np.exp(-((x_vector - theta0) / sigma0)**2 / 2) +
(1 / (sigma1 * np.sqrt(2 * np.pi))) * np.exp(-((x_vector - theta1) / sigma1)**2 / 2))
def f_gauss1(x_vector, theta0, sigma0):
salida = normal_pdf(x_vector, theta0, sigma0)
return (salida)
def f_gauss2(x_vector, theta0, theta1, sigma0, sigma1):
salida = (normal_pdf(x_vector, theta0, sigma0) + normal_pdf(x_vector, theta1, sigma1)) / 2
return (salida)
def f_gauss3(x_vector, theta0, theta1, theta2, sigma0, sigma1, sigma2):
salida = (normal_pdf(x_vector, theta0, sigma0) + normal_pdf(x_vector, theta1, sigma1) +
normal_pdf(x_vector, theta2, sigma2)) / 3
return (salida)
def compute_pdf_overlap(pdf_fun1, pdf_fun2):
fun_array = np.vstack([pdf_fun1, pdf_fun2])
intersection = np.min(fun_array, axis=0)
pdf_overlap = np.sum(intersection)
return (pdf_overlap)
def fit_column_model(x_val, g_val):
q1 = np.quantile(x_val, 0.25)
q2 = np.median(x_val)
q3 = np.quantile(x_val, 0.75)
# print("Using q1: %d, q2: %d, q3: %d"%(q1, q2, q3))
try:
params1, cov1 = curve_fit(f_gauss1, x_val, g_val, [q2, q2 - q1])
except:
params1 = [q2, q2 - 1]
try:
params2, cov2 = curve_fit(f_gauss2, x_val, g_val, [q1, q3, q1, q1])
except:
params2 = [q1, q3, q1, q1]
try:
params3, cov3 = curve_fit(f_gauss3, x_val, g_val, [q1, q2, q3, q1, q2 - q1, q1])
except:
params3 = [q1, q2, q3, q1, q2 - q1, q1]
# Extract area under the curve of the intersection
auc1 = compute_pdf_overlap(f_gauss1(x_val, *params1), g_val)
auc2 = compute_pdf_overlap(f_gauss2(x_val, *params2), g_val)
auc3 = compute_pdf_overlap(f_gauss3(x_val, *params3), g_val)
print("Using auc1: %.2f, auc2: %.2f, auc3: %.2f" % (auc1, auc2, auc3))
if auc1 > 0.86:
print("Selected 1")
return (params1)
elif auc2 > 0.86:
print("Selected 2")
return (params2)
elif auc3 > 0.86:
print("Selected 3")
return (params3)
else:
idx = np.argmax([auc1, auc2, auc3])
all_params = [params1, params2, params3]
print("Selected %d" % (idx + 1))
return (all_params[idx])
def ungroup_shapes(shapes):
res = []
for shape in shapes:
try:
if shape.shape_type == MSO_SHAPE_TYPE.GROUP:
res.extend(ungroup_shapes(shape.shapes))
else:
res.append(shape)
except Exception as e:
print(f'failed to load shape {shape}, skipped. error: {e}')
return res
def is_two_column_text(slide):
if slide.slide_layout.name != "TITLE":
all_mu = list()
all_sigma = list()
for shape in sorted(ungroup_shapes(slide.shapes), key=attrgetter('top', 'left')):
if shape.shape_type == MSO_SHAPE_TYPE.PLACEHOLDER:
if shape.placeholder_format.type == PP_PLACEHOLDER.TITLE:
if shape.has_text_frame:
print('SLIDE TITLE: %s' % shape.text_frame.text)
continue
if shape.shape_type == MSO_SHAPE_TYPE.PICTURE or shape.has_text_frame:
centroid_x = shape.left + shape.width / 2
all_mu.append(Length(centroid_x).mm)
all_sigma.append(Length(shape.width / 4).mm) # Gaussiana - 2sigma
return (all_mu, all_sigma)
else:
return False
def assign_shapes(slide, params, ncols=2, slide_width_mm=1000):
shapes_dict = {"shapes_pre": list(), "shapes_l": list(), "shapes_c": list(), "shapes_r": list()}
shapes = sorted(ungroup_shapes(slide.shapes), key=attrgetter('top', 'left'))
print("Ncols is %d" % ncols)
if ncols == 1:
shapes_dict["shapes_pre"] = sorted(shapes,
key=lambda x: x.shape_type == MSO_SHAPE_TYPE.PLACEHOLDER,
reverse=True)
return (shapes_dict)
elif ncols == 2:
param_means = params[0:2]
param_sds = params[2:]
elif ncols == 3:
param_means = params[0:3]
param_sds = params[3:]
else:
raise (ValueError, "Error in the number of columns")
x_vector = np.arange(1, slide_width_mm)
for shape in slide.shapes:
if shape.shape_type == MSO_SHAPE_TYPE.PLACEHOLDER:
if shape.placeholder_format.type == PP_PLACEHOLDER.TITLE:
if shape.has_text_frame:
print('SLIDE TITLE: %s' % shape.text_frame.text)
shapes_dict["shapes_pre"].insert(0, shape)
else:
shapes_dict["shapes_pre"].append(shape)
continue
if shape.shape_type == MSO_SHAPE_TYPE.PICTURE or shape.has_text_frame:
centroid_x = shape.left + shape.width / 2
curr_mu = Length(centroid_x).mm
curr_sigma = Length(shape.width / 4).mm # Gaussian - 2sigma
area_u_c = np.zeros(ncols)
for idx, param_mu in enumerate(param_means):
area_u_c[idx] = compute_pdf_overlap(normal_pdf(x_vector, mu=param_mu, sigma=param_sds[idx]),
normal_pdf(x_vector, curr_mu, curr_sigma))
max_score_column = np.argmax(area_u_c)
if max_score_column == 0:
shapes_dict["shapes_l"].append(shape)
elif max_score_column == 1:
if ncols == 2:
shapes_dict["shapes_r"].append(shape)
elif ncols == 3:
shapes_dict["shapes_c"].append(shape)
else:
raise (ValueError, "Not allowed number of columns")
elif max_score_column == 2:
shapes_dict["shapes_r"].append(shape)
else:
raise (ValueError, "Max number of columns does not correspond to the number of columns")
return (shapes_dict)
def get_multi_column_slide_if_present(prs: Presentation, raw_slide, process_shapes) -> Optional[MultiColumnSlide]:
pdf_modelo = is_two_column_text(raw_slide)
if not pdf_modelo:
return None
slide_width_mm = pptx.util.Length(prs.slide_width)
t_vector = np.arange(1, slide_width_mm)
# Model to infer number of columns
salida = map(lambda mu, sigma: normal_pdf(t_vector, mu, sigma), pdf_modelo[0], pdf_modelo[1])
sum_of_gaussian = np.mean(list(salida), axis=0)
parameters = fit_column_model(t_vector, sum_of_gaussian)
num_cols = int(len(parameters) / 2)
if num_cols == 1:
return None
slide = MultiColumnSlide(preface=[], columns=[], notes=[])
dict_shapes = assign_shapes(raw_slide, parameters, num_cols, slide_width_mm=slide_width_mm)
slide.preface = process_shapes(dict_shapes["shapes_pre"])
if num_cols == 2:
slide.columns = [process_shapes(dict_shapes["shapes_l"]), process_shapes(dict_shapes["shapes_r"])]
elif num_cols == 3:
slide.columns = [
process_shapes(dict_shapes["shapes_l"]),
process_shapes(dict_shapes["shapes_c"]),
process_shapes(dict_shapes["shapes_r"])
]
return slide
================================================
FILE: src/backend/bisheng/pptx2md/outputter.py
================================================
# Copyright 2024 Liu Siyao
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import re
import urllib.parse
from typing import List
from rapidfuzz import fuzz
from bisheng.pptx2md.types import ConversionConfig, ElementType, ParsedPresentation, SlideElement, SlideType, TextRun
from bisheng.pptx2md.utils import rgb_to_hex
class Formatter:
def __init__(self, config: ConversionConfig):
os.makedirs(config.output_path.parent, exist_ok=True)
self.ofile = open(config.output_path, 'w', encoding='utf8')
self.config = config
def output(self, presentation_data: ParsedPresentation):
self.put_header()
last_element = None
last_title = None
for slide_idx, slide in enumerate(presentation_data.slides):
all_elements = []
if slide.type == SlideType.General:
all_elements = slide.elements
elif slide.type == SlideType.MultiColumn:
all_elements = slide.preface + slide.columns
for element in all_elements:
if last_element and last_element.type == ElementType.ListItem and element.type != ElementType.ListItem:
self.put_list_footer()
match element.type:
case ElementType.Title:
element.content = element.content.strip()
if element.content:
if last_title and last_title.level == element.level and fuzz.ratio(
last_title.content, element.content, score_cutoff=92):
# skip if the title is the same as the last one
# Allow for repeated slide titles - One or more - Add (cont.) to the title
if self.config.keep_similar_titles:
self.put_title(f'{element.content} (cont.)', element.level)
else:
self.put_title(element.content, element.level)
last_title = element
case ElementType.ListItem:
if not (last_element and last_element.type == ElementType.ListItem):
self.put_list_header()
self.put_list(self.get_formatted_runs(element.content), element.level)
case ElementType.Paragraph:
self.put_para(self.get_formatted_runs(element.content))
case ElementType.Image:
self.put_image(element.path, element.width)
case ElementType.Table:
self.put_table([[self.get_formatted_runs(cell) for cell in row] for row in element.content])
last_element = element
if not self.config.disable_notes and slide.notes:
self.put_para('---')
for note in slide.notes:
self.put_para(note)
if slide_idx < len(presentation_data.slides) - 1 and self.config.enable_slides:
self.put_para("\n---\n")
self.close()
def put_header(self):
pass
def put_title(self, text, level):
pass
def put_list(self, text, level):
pass
def put_list_header(self):
self.put_para('')
def put_list_footer(self):
self.put_para('')
def get_formatted_runs(self, runs: List[TextRun]):
res = ''
for run in runs:
text = run.text
if text == '':
continue
if not self.config.disable_escaping:
text = self.get_escaped(text)
if run.style.hyperlink:
text = self.get_hyperlink(text, run.style.hyperlink)
if run.style.is_accent:
text = self.get_accent(text)
elif run.style.is_strong:
text = self.get_strong(text)
if run.style.color_rgb and not self.config.disable_color:
text = self.get_colored(text, run.style.color_rgb)
res += text
return res.strip()
def put_para(self, text):
pass
def put_image(self, path, max_width):
pass
def put_table(self, table):
pass
def get_accent(self, text):
pass
def get_strong(self, text):
pass
def get_colored(self, text, rgb):
pass
def get_hyperlink(self, text, url):
pass
def get_escaped(self, text):
pass
def write(self, text):
self.ofile.write(text)
def flush(self):
self.ofile.flush()
def close(self):
self.ofile.close()
class MarkdownFormatter(Formatter):
# write outputs to markdown
def __init__(self, config: ConversionConfig):
super().__init__(config)
self.esc_re1 = re.compile(r'([\\\*`!_\{\}\[\]\(\)#\+-\.])')
self.esc_re2 = re.compile(r'(<[^>]+>)')
def put_title(self, text, level):
self.ofile.write('#' * level + ' ' + text + '\n\n')
def put_list(self, text, level):
self.ofile.write(' ' * level + '* ' + text.strip() + '\n')
def put_para(self, text):
self.ofile.write(text + '\n\n')
def put_image(self, path, max_width=None):
if max_width is None:
self.ofile.write(f'})\n\n')
else:
self.ofile.write(f'\n\n')
def put_table(self, table):
gen_table_row = lambda row: '| ' + ' | '.join([c.replace('\n', ' ') for c in row]) + ' |'
self.ofile.write(gen_table_row(table[0]) + '\n')
self.ofile.write(gen_table_row([':-:' for _ in table[0]]) + '\n')
self.ofile.write('\n'.join([gen_table_row(row) for row in table[1:]]) + '\n\n')
def get_accent(self, text):
return ' _' + text + '_ '
def get_strong(self, text):
return ' __' + text + '__ '
def get_colored(self, text, rgb):
return ' %s ' % (rgb_to_hex(rgb), text)
def get_hyperlink(self, text, url):
return '[' + text + '](' + url + ')'
def esc_repl(self, match):
return '\\' + match.group(0)
def get_escaped(self, text):
text = re.sub(self.esc_re1, self.esc_repl, text)
text = re.sub(self.esc_re2, self.esc_repl, text)
return text
class WikiFormatter(Formatter):
# write outputs to wikitext
def __init__(self, config: ConversionConfig):
super().__init__(config)
self.esc_re = re.compile(r'<([^>]+)>')
def put_title(self, text, level):
self.ofile.write('!' * level + ' ' + text + '\n\n')
def put_list(self, text, level):
self.ofile.write('*' * (level + 1) + ' ' + text.strip() + '\n')
def put_para(self, text):
self.ofile.write(text + '\n\n')
def put_image(self, path, max_width):
if max_width is None:
self.ofile.write(f'\n\n')
else:
self.ofile.write(f'\n\n')
def get_accent(self, text):
return ' __' + text + '__ '
def get_strong(self, text):
return ' \'\'' + text + '\'\' '
def get_colored(self, text, rgb):
return ' @@color:%s; %s @@ ' % (rgb_to_hex(rgb), text)
def get_hyperlink(self, text, url):
return '[[' + text + '|' + url + ']]'
def esc_repl(self, match):
return "''''" + match.group(0)
def get_escaped(self, text):
text = re.sub(self.esc_re, self.esc_repl, text)
return text
class MadokoFormatter(Formatter):
# write outputs to madoko markdown
def __init__(self, config: ConversionConfig):
super().__init__(config)
self.ofile.write('[TOC]\n\n')
self.esc_re1 = re.compile(r'([\\\*`!_\{\}\[\]\(\)#\+-\.])')
self.esc_re2 = re.compile(r'(<[^>]+>)')
def put_title(self, text, level):
self.ofile.write('#' * level + ' ' + text + '\n\n')
def put_list(self, text, level):
self.ofile.write(' ' * level + '* ' + text.strip() + '\n')
def put_para(self, text):
self.ofile.write(text + '\n\n')
def put_image(self, path, max_width):
if max_width is None:
self.ofile.write(f'\n\n')
elif max_width < 500:
self.ofile.write(f'\n\n')
else:
self.ofile.write('~ Figure {caption: image caption}\n')
self.ofile.write('{width:%spx;}\n' % (path, max_width))
self.ofile.write('~\n\n')
def get_accent(self, text):
return ' _' + text + '_ '
def get_strong(self, text):
return ' __' + text + '__ '
def get_colored(self, text, rgb):
return ' %s ' % (rgb_to_hex(rgb), text)
def get_hyperlink(self, text, url):
return '[' + text + '](' + url + ')'
def esc_repl(self, match):
return '\\' + match.group(0)
def get_escaped(self, text):
text = re.sub(self.esc_re1, self.esc_repl, text)
text = re.sub(self.esc_re2, self.esc_repl, text)
return text
class QuartoFormatter(Formatter):
# write outputs to quarto markdown - reveal js
def __init__(self, config: ConversionConfig):
super().__init__(config)
self.esc_re1 = re.compile(r'([\\\*`!_\{\}\[\]\(\)#\+-\.])')
self.esc_re2 = re.compile(r'(<[^>]+>)')
def output(self, presentation_data: ParsedPresentation):
self.put_header()
last_title = None
def put_elements(elements: List[SlideElement]):
nonlocal last_title
last_element = None
for element in elements:
if last_element and last_element.type == ElementType.ListItem and element.type != ElementType.ListItem:
self.put_list_footer()
match element.type:
case ElementType.Title:
element.content = element.content.strip()
if element.content:
if last_title and last_title.level == element.level and fuzz.ratio(
last_title.content, element.content, score_cutoff=92):
# skip if the title is the same as the last one
# Allow for repeated slide titles - One or more - Add (cont.) to the title
if self.config.keep_similar_titles:
self.put_title(f'{element.content} (cont.)', element.level)
else:
self.put_title(element.content, element.level)
last_title = element
case ElementType.ListItem:
if not (last_element and last_element.type == ElementType.ListItem):
self.put_list_header()
self.put_list(self.get_formatted_runs(element.content), element.level)
case ElementType.Paragraph:
self.put_para(self.get_formatted_runs(element.content))
case ElementType.Image:
self.put_image(element.path, element.width)
case ElementType.Table:
self.put_table([[self.get_formatted_runs(cell) for cell in row] for row in element.content])
last_element = element
for slide_idx, slide in enumerate(presentation_data.slides):
if slide.type == SlideType.General:
put_elements(slide.elements)
elif slide.type == SlideType.MultiColumn:
put_elements(slide.preface)
if len(slide.columns) == 2:
width = '50%'
elif len(slide.columns) == 3:
width = '33%'
else:
raise ValueError(f'Unsupported number of columns: {len(slide.columns)}')
self.put_para(':::: {.columns}')
for column in slide.columns:
self.put_para(f'::: {{.column width="{width}"}}')
put_elements(column)
self.put_para(':::')
self.put_para('::::')
if not self.config.disable_notes and slide.notes:
self.put_para("::: {.notes}")
for note in slide.notes:
self.put_para(note)
self.put_para(":::")
if slide_idx < len(presentation_data.slides) - 1 and self.config.enable_slides:
self.put_para("\n---\n")
self.close()
def put_header(self):
self.ofile.write('''---
title: "Presentation Title"
author: "Author"
format:
revealjs:
slide-number: c/t
width: 1600
height: 900
logo: img/logo.png
footer: "Organization"
incremental: true
theme: [simple]
---
''')
def put_title(self, text, level):
self.ofile.write('#' * level + ' ' + text + '\n\n')
def put_list(self, text, level):
self.ofile.write(' ' * level + '* ' + text.strip() + '\n')
def put_para(self, text):
self.ofile.write(text + '\n\n')
def put_image(self, path, max_width=None):
if max_width is None:
self.ofile.write(f'})\n\n')
else:
self.ofile.write(f'\n\n')
def put_table(self, table):
gen_table_row = lambda row: '| ' + ' | '.join([c.replace('\n', ' ') for c in row]) + ' |'
self.ofile.write(gen_table_row(table[0]) + '\n')
self.ofile.write(gen_table_row([':-:' for _ in table[0]]) + '\n')
self.ofile.write('\n'.join([gen_table_row(row) for row in table[1:]]) + '\n\n')
def get_accent(self, text):
return ' _' + text + '_ '
def get_strong(self, text):
return ' __' + text + '__ '
def get_colored(self, text, rgb):
return ' %s ' % (rgb_to_hex(rgb), text)
def get_hyperlink(self, text, url):
return '[' + text + '](' + url + ')'
def esc_repl(self, match):
return '\\' + match.group(0)
def get_escaped(self, text):
text = re.sub(self.esc_re1, self.esc_repl, text)
text = re.sub(self.esc_re2, self.esc_repl, text)
return text
================================================
FILE: src/backend/bisheng/pptx2md/parser.py
================================================
# Copyright 2024 Liu Siyao
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import print_function
import logging
import os
from functools import partial
from operator import attrgetter
from typing import List, Union
import pptx
from PIL import Image
from pptx import Presentation
from pptx.enum.dml import MSO_COLOR_TYPE, MSO_THEME_COLOR
from pptx.enum.shapes import MSO_SHAPE_TYPE, PP_PLACEHOLDER
from pptx.parts.image import Image as PPTXImage, Image
from pptx.util import lazyproperty
from rapidfuzz import process as fuze_process
from tqdm import tqdm
from bisheng.pptx2md.multi_column import get_multi_column_slide_if_present
from bisheng.pptx2md.types import (
ConversionConfig,
GeneralSlide,
ImageElement,
ListItemElement,
ParagraphElement,
ParsedPresentation,
SlideElement,
TableElement,
TextRun,
TextStyle,
TitleElement,
)
logger = logging.getLogger(__name__)
picture_count = 0
# CustomizableImageResolvepptxIssue with some image formats not supported in
class CustomImage(PPTXImage):
def __init__(self, blob: bytes, filename: str = None):
super(CustomImage, self).__init__(blob=blob, filename=filename)
@lazyproperty
def ext(self) -> str:
"""Canonical file extension for this image e.g. `'png'`.
The returned extension is all lowercase and is the canonical extension for the content type
of this image, regardless of what extension may have been used in its filename, if any.
"""
ext_map = {
"BMP": "bmp",
"GIF": "gif",
"JPEG": "jpg",
"PNG": "png",
"TIFF": "tiff",
"WMF": "wmf",
"WEBP": "webp",
}
format = self._format
if format not in ext_map:
tmpl = "unsupported image format, expected one of: %s, got '%s'"
raise ValueError(tmpl % (ext_map.keys(), format))
return ext_map[format]
def is_title(shape):
if shape.is_placeholder and (shape.placeholder_format.type == PP_PLACEHOLDER.TITLE or
shape.placeholder_format.type == PP_PLACEHOLDER.SUBTITLE or
shape.placeholder_format.type == PP_PLACEHOLDER.VERTICAL_TITLE or
shape.placeholder_format.type == PP_PLACEHOLDER.CENTER_TITLE):
return True
return False
def is_text_block(config: ConversionConfig, shape):
if shape.has_text_frame:
if shape.is_placeholder and shape.placeholder_format.type == PP_PLACEHOLDER.BODY:
return True
if len(shape.text) > config.min_block_size:
return True
return False
def is_list_block(shape) -> bool:
levels = []
for para in shape.text_frame.paragraphs:
if para.level not in levels:
levels.append(para.level)
if para.level != 0 or len(levels) > 1:
return True
return False
def is_accent(font):
if font.underline or font.italic or (
font.color.type == MSO_COLOR_TYPE.SCHEME and
(font.color.theme_color == MSO_THEME_COLOR.ACCENT_1 or font.color.theme_color == MSO_THEME_COLOR.ACCENT_2 or
font.color.theme_color == MSO_THEME_COLOR.ACCENT_3 or font.color.theme_color == MSO_THEME_COLOR.ACCENT_4 or
font.color.theme_color == MSO_THEME_COLOR.ACCENT_5 or font.color.theme_color == MSO_THEME_COLOR.ACCENT_6)):
return True
return False
def is_strong(font):
if font.bold or (font.color.type == MSO_COLOR_TYPE.SCHEME and (font.color.theme_color == MSO_THEME_COLOR.DARK_1 or
font.color.theme_color == MSO_THEME_COLOR.DARK_2)):
return True
return False
def get_text_runs(para) -> List[TextRun]:
runs = []
for run in para.runs:
result = TextRun(text=run.text, style=TextStyle())
if result.text == '':
continue
try:
if run.hyperlink.address:
result.style.hyperlink = run.hyperlink.address
except:
result.style.hyperlink = 'error:ppt-link-parsing-issue'
if is_accent(run.font):
result.style.is_accent = True
if is_strong(run.font):
result.style.is_strong = True
if run.font.color.type == MSO_COLOR_TYPE.RGB:
result.style.color_rgb = run.font.color.rgb
runs.append(result)
return runs
def process_title(config: ConversionConfig, shape, slide_idx) -> TitleElement:
text = shape.text_frame.text.strip()
if config.custom_titles:
res = fuze_process.extractOne(text, config.custom_titles.keys(), score_cutoff=92)
if not res:
return TitleElement(content=text.strip(), level=max(config.custom_titles.values()) + 1)
else:
logger.info(f'Title in slide {slide_idx} "{text}" is converted to "{res[0]}" as specified in title file.')
return TitleElement(content=res[0].strip(), level=config.custom_titles[res[0]])
else:
return TitleElement(content=text.strip(), level=1)
def process_text_blocks(config: ConversionConfig, shape, slide_idx) -> List[Union[ListItemElement, ParagraphElement]]:
results = []
if is_list_block(shape):
for para in shape.text_frame.paragraphs:
if para.text.strip() == '':
continue
text = get_text_runs(para)
results.append(ListItemElement(content=text, level=para.level))
else:
# paragraph block
for para in shape.text_frame.paragraphs:
if para.text.strip() == '':
continue
text = get_text_runs(para)
results.append(ParagraphElement(content=text))
return results
def process_picture(config: ConversionConfig, shape, slide_idx) -> Union[ImageElement, None]:
if config.disable_image:
return None
global picture_count
image = CustomImage(shape.image.blob, shape.image.filename)
file_prefix = ''.join(os.path.basename(config.pptx_path).split('.')[:-1])
pic_name = file_prefix + f'_{picture_count}'
pic_ext = image.ext
if not os.path.exists(config.image_dir):
os.makedirs(config.image_dir)
output_path = config.image_dir / f'{pic_name}.{pic_ext}'
common_path = os.path.commonpath([config.output_path, config.image_dir])
img_outputter_path = os.path.relpath(output_path, common_path).replace('\\', '/')
with open(output_path, 'wb') as f:
f.write(image.blob)
picture_count += 1
# normal images
if pic_ext != 'wmf':
return ImageElement(path=img_outputter_path, width=config.image_width)
# wmf images, try to convert, if failed, output as original
try:
try:
from PIL import Image
Image.open(output_path).save(os.path.splitext(output_path)[0] + '.png')
return ImageElement(path=os.path.splitext(img_outputter_path)[0] + '.png', width=config.image_width)
except Exception: # Image failed, try another
from wand.image import Image
with Image(filename=output_path) as img:
img.format = 'png'
img.save(filename=os.path.splitext(output_path)[0] + '.png')
logger.info(f'Image {output_path} in slide {slide_idx} converted to png.')
return ImageElement(path=os.path.splitext(img_outputter_path)[0] + '.png', width=config.image_width)
except Exception:
logger.warning(f'Cannot convert wmf image {output_path} in slide {slide_idx} to png, skipped.')
return None
def process_table(config: ConversionConfig, shape, slide_idx) -> Union[TableElement, None]:
table = [[sum([get_text_runs(p)
for p in cell.text_frame.paragraphs], [])
for cell in row.cells]
for row in shape.table.rows]
if len(table) > 0:
return TableElement(content=table)
return None
def ungroup_shapes(shapes) -> List[SlideElement]:
res = []
for shape in shapes:
try:
if shape.shape_type == MSO_SHAPE_TYPE.GROUP:
res.extend(ungroup_shapes(shape.shapes))
else:
res.append(shape)
except Exception as e:
logger.warning(f'failed to load shape {shape}, skipped. error: {e}')
return res
def process_shapes(config: ConversionConfig, current_shapes, slide_id: int) -> List[SlideElement]:
results = []
for shape in current_shapes:
if is_title(shape):
results.append(process_title(config, shape, slide_id))
elif is_text_block(config, shape):
results.extend(process_text_blocks(config, shape, slide_id))
elif shape.shape_type == MSO_SHAPE_TYPE.PICTURE:
try:
pic = process_picture(config, shape, slide_id)
if pic:
results.append(pic)
except AttributeError as e:
logger.warning(f'Failed to process picture, skipped: {e}')
elif shape.shape_type == MSO_SHAPE_TYPE.TABLE:
table = process_table(config, shape, slide_id)
if table:
results.append(table)
else:
try:
ph = shape.placeholder_format
if ph.type == PP_PLACEHOLDER.OBJECT and hasattr(shape, "image") and getattr(shape, "image"):
pic = process_picture(config, shape, slide_id)
if pic:
results.append(pic)
except:
pass
return results
def parse(config: ConversionConfig, prs: Presentation) -> ParsedPresentation:
result = ParsedPresentation(slides=[])
for idx, slide in enumerate(tqdm(prs.slides, desc='Converting slides')):
if config.page is not None and idx + 1 != config.page:
continue
shapes = []
try:
shapes = sorted(ungroup_shapes(slide.shapes), key=attrgetter('top', 'left'))
except:
logger.warning('Bad shapes encountered in this slide. Please check or remove them and try again.')
logger.warning('shapes:')
try:
for sp in slide.shapes:
logger.warning(sp.shape_type)
logger.warning(sp.top, sp.left, sp.width, sp.height)
except:
logger.warning('failed to print all bad shapes.')
if not config.try_multi_column:
result_slide = GeneralSlide(elements=process_shapes(config, shapes, idx + 1))
else:
multi_column_slide = get_multi_column_slide_if_present(
prs, slide, partial(process_shapes, config=config, slide_id=idx + 1))
if multi_column_slide:
result_slide = multi_column_slide
else:
result_slide = GeneralSlide(elements=process_shapes(config, shapes, idx + 1))
if not config.disable_notes and slide.has_notes_slide:
text = slide.notes_slide.notes_text_frame.text
if text:
result_slide.notes.append(text)
result.slides.append(result_slide)
return result
================================================
FILE: src/backend/bisheng/pptx2md/types.py
================================================
# Copyright 2024 Liu Siyao
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import annotations
from enum import Enum
from pathlib import Path
from typing import List, Optional, Union
from pydantic import BaseModel
class ConversionConfig(BaseModel):
"""Configuration for PowerPoint to Markdown conversion."""
pptx_path: Path
"""Path to the pptx file to be converted"""
output_path: Path
"""Path of the output file"""
image_dir: Optional[Path]
"""Where to put images extracted"""
title_path: Optional[Path] = None
"""Path to the custom title list file"""
image_width: Optional[int] = None
"""Maximum image width in px"""
disable_image: bool = False
"""Disable image extraction"""
disable_wmf: bool = False
"""Keep wmf formatted image untouched (avoid exceptions under linux)"""
disable_color: bool = False
"""Do not add color HTML tags"""
disable_escaping: bool = False
"""Do not attempt to escape special characters"""
disable_notes: bool = False
"""Do not add presenter notes"""
enable_slides: bool = False
"""Deliniate slides with `\n---\n`"""
is_wiki: bool = False
"""Generate output as wikitext (TiddlyWiki)"""
is_mdk: bool = False
"""Generate output as madoko markdown"""
is_qmd: bool = False
"""Generate output as quarto markdown presentation"""
min_block_size: int = 15
"""The minimum character number of a text block to be converted"""
page: Optional[int] = None
"""Only convert the specified page"""
custom_titles: dict[str, int] = {}
"""Mapping of custom titles to their heading levels"""
try_multi_column: bool = False
"""Try to detect multi-column slides"""
keep_similar_titles: bool = False
"""Keep similar titles (allow for repeated slide titles - One or more - Add (cont.) to the title)"""
class ElementType(str, Enum):
Title = "Title"
ListItem = "ListItem"
Paragraph = "Paragraph"
Image = "Image"
Table = "Table"
class TextStyle(BaseModel):
is_accent: bool = False
is_strong: bool = False
color_rgb: Optional[tuple[int, int, int]] = None
hyperlink: Optional[str] = None
class TextRun(BaseModel):
text: str
style: TextStyle
class Position(BaseModel):
left: float
top: float
width: float
height: float
class BaseElement(BaseModel):
type: ElementType
position: Optional[Position] = None
style: Optional[TextStyle] = None
class TitleElement(BaseElement):
type: ElementType = ElementType.Title
content: str
level: int
class ListItemElement(BaseElement):
type: ElementType = ElementType.ListItem
content: List[TextRun]
level: int = 1
class ParagraphElement(BaseElement):
type: ElementType = ElementType.Paragraph
content: List[TextRun]
class ImageElement(BaseElement):
type: ElementType = ElementType.Image
path: str
width: Optional[int] = None
original_ext: str = "" # For tracking original file extension (e.g. wmf)
alt_text: str = "" # For accessibility
class TableElement(BaseElement):
type: ElementType = ElementType.Table
content: List[List[List[TextRun]]] # rows -> cols -> rich text
SlideElement = Union[TitleElement, ListItemElement, ParagraphElement, ImageElement, TableElement]
class SlideType(str, Enum):
MultiColumn = "MultiColumn"
General = "General"
class MultiColumnSlide(BaseModel):
type: SlideType = SlideType.MultiColumn
preface: List[SlideElement]
columns: List[SlideElement]
notes: List[str] = []
class GeneralSlide(BaseModel):
type: SlideType = SlideType.General
elements: List[SlideElement]
notes: List[str] = []
Slide = Union[GeneralSlide, MultiColumnSlide]
class ParsedPresentation(BaseModel):
slides: List[Slide]
================================================
FILE: src/backend/bisheng/pptx2md/utils.py
================================================
# Copyright 2024 Liu Siyao
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import logging
import os
import re
import shutil
import tempfile
import uuid
from pathlib import Path
from pptx import Presentation
logger = logging.getLogger(__name__)
def fix_null_rels(file_path):
temp_dir_name = tempfile.mkdtemp()
shutil.unpack_archive(file_path, temp_dir_name, 'zip')
rels = [
os.path.join(dp, f)
for dp, dn, filenames in os.walk(temp_dir_name)
for f in filenames
if os.path.splitext(f)[1] == '.rels'
]
pat = re.compile(r'<\S*Relationship[^>]+Target\S*=\S*"NULL"[^>]*/>', re.I)
for fn in rels:
f = open(fn, 'r+')
content = f.read()
res = pat.search(content)
if res is not None:
content = pat.sub('', content)
f.seek(0)
f.truncate()
f.write(content)
f.close()
tfn = uuid.uuid4().hex
shutil.make_archive(tfn, 'zip', temp_dir_name)
shutil.rmtree(temp_dir_name)
tgt = f'{file_path[:-5]}_purged.pptx'
shutil.move(f'{tfn}.zip', tgt)
return tgt
def load_pptx(file_path: str) -> Presentation:
if not os.path.exists(file_path):
logger.error(f'source file {file_path} not exist!')
logger.error(f'absolute path: {os.path.abspath(file_path)}')
raise FileNotFoundError(file_path)
try:
prs = Presentation(file_path)
except KeyError as err:
if len(err.args) > 0 and re.match(r'There is no item named .*NULL.* in the archive', str(err.args[0])):
logger.info('corrupted links found, trying to purge...')
try:
res_path = fix_null_rels(file_path)
logger.info(f'purged file saved to {res_path}.')
prs = Presentation(res_path)
except:
logger.error(
'failed to purge corrupted links, you can report this at https://github.com/ssine/pptx2md/issues')
raise err
else:
logger.error('unknown error, you can report this at https://github.com/ssine/pptx2md/issues')
raise err
return prs
def prepare_titles(title_path: Path) -> dict[str, int]:
titles: dict[str, int] = {}
with open(title_path, 'r', encoding='utf8') as f:
indent = -1
for line in f.readlines():
cnt = 0
while line[cnt] == ' ':
cnt += 1
if cnt == 0:
titles[line.strip()] = 1
else:
if indent == -1:
indent = cnt
titles[line.strip()] = 2
else:
titles[line.strip()] = cnt // indent + 1
return titles
def rgb_to_hex(rgb):
r, g, b = rgb
return f'#{r:02x}{g:02x}{b:02x}'
================================================
FILE: src/backend/bisheng/processing/__init__.py
================================================
================================================
FILE: src/backend/bisheng/processing/base.py
================================================
from typing import Tuple, Union
from langchain_core.runnables import RunnableConfig
from loguru import logger
from bisheng.api.v1.callback import AsyncStreamingLLMCallbackHandler
from bisheng.processing.process import fix_memory_inputs, format_actions
async def get_result_and_steps(langchain_object, inputs: Union[dict, str], **kwargs):
"""Get result and thought from extracted json"""
try:
if hasattr(langchain_object, 'verbose'):
langchain_object.verbose = True
if hasattr(langchain_object, 'return_intermediate_steps'):
# https://github.com/hwchase17/langchain/issues/2068
# Deactivating until we have a frontend solution
# to display intermediate steps
langchain_object.return_intermediate_steps = True
try:
fix_memory_inputs(langchain_object)
except Exception as exc:
logger.error(exc)
async_callbacks = [AsyncStreamingLLMCallbackHandler(**kwargs)]
output = await langchain_object.ainvoke(inputs, config=RunnableConfig(callbacks=async_callbacks))
intermediate_steps = (output.get('intermediate_steps', [])
if isinstance(output, dict) else [])
source_document = (output.get('source_documents', '') if isinstance(output, dict) else '')
# Default copywriting for cases where the return is empty
if isinstance(output, dict):
result = output.get(langchain_object.output_keys[0])
else:
result = output
try:
if intermediate_steps and isinstance(intermediate_steps[0], Tuple):
thought = format_actions(intermediate_steps)
else:
thought = intermediate_steps
except Exception as exc:
logger.exception(exc)
thought = ''
except Exception as exc:
logger.exception(exc)
raise ValueError(f'Error: {str(exc)}') from exc
return result, thought, source_document
================================================
FILE: src/backend/bisheng/processing/process.py
================================================
import asyncio
import json
from pathlib import Path
from typing import Any, Coroutine, Dict, List, Optional, Tuple, Union
from langchain.chains.base import Chain
from langchain.schema import AgentAction, Document
from langchain.vectorstores.base import VectorStore
from loguru import logger
from pydantic import BaseModel
from sqlmodel import select
from bisheng.core.database import get_sync_db_session
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync
from bisheng.database.models.message import ChatMessage
from bisheng.database.models.report import Report as ReportModel
from bisheng.interface.run import build_sorted_vertices, get_memory_key, update_memory_keys
from bisheng.services.deps import get_session_service
from bisheng.template.field.base import TemplateField
from bisheng.utils.docx_temp import test_replace_string
from bisheng_langchain.input_output import Report
def fix_memory_inputs(langchain_object):
"""
Given a LangChain object, this function checks if it has a memory attribute and if that memory key exists in the
object's input variables. If so, it does nothing. Otherwise, it gets a possible new memory key using the
get_memory_key function and updates the memory keys using the update_memory_keys function.
"""
if not hasattr(langchain_object, 'memory') or langchain_object.memory is None:
return
try:
if (hasattr(langchain_object.memory, 'memory_key')
and langchain_object.memory.memory_key in langchain_object.input_variables):
return
except AttributeError:
input_variables = (langchain_object.prompt.input_variables if hasattr(
langchain_object, 'prompt') else langchain_object.input_keys)
if langchain_object.memory.memory_key in input_variables:
return
possible_new_mem_key = get_memory_key(langchain_object)
if possible_new_mem_key is not None:
update_memory_keys(langchain_object, possible_new_mem_key)
def format_actions(actions: List[Tuple[AgentAction, str]]) -> str:
"""Format a list of (AgentAction, answer) tuples into a string."""
output = []
for action, answer in actions:
log = action.log
tool = action.tool
tool_input = action.tool_input
output.append(f'Log: {log}')
if 'Action' not in log and 'Action Input' not in log:
output.append(f'Tool: {tool}')
output.append(f'Tool Input: {tool_input}')
output.append(f'Answer: {answer}')
output.append('') # Add a blank line
return '\n'.join(output)
def get_result_and_thought(langchain_object: Any, inputs: dict):
"""Get result and thought from extracted json"""
try:
if hasattr(langchain_object, 'verbose'):
langchain_object.verbose = True
if hasattr(langchain_object, 'return_intermediate_steps'):
langchain_object.return_intermediate_steps = True
fix_memory_inputs(langchain_object)
try:
# all use chat handlers
# action = 'default'
from bisheng.api.v1 import callback
callbacks = [callback.StreamingLLMCallbackHandler(None, flow_id=None, chat_id=None)]
output = langchain_object(inputs, return_only_outputs=True, callbacks=callbacks)
except ValueError as exc:
# make the error message more informative
logger.debug(f'Error: {str(exc)}')
raise exc
except Exception as exc:
raise ValueError(f'Error: {str(exc)}') from exc
return output
def get_input_str_if_only_one_input(inputs: dict) -> Optional[str]:
"""Get input string if only one input is provided"""
return list(inputs.values())[0] if len(inputs) == 1 else None
def get_build_result(data_graph, session_id):
# If session_id is provided, load the langchain_object from the session
# using build_sorted_vertices_with_caching.get_result_by_session_id
# if it returns something different than None, return it
# otherwise, build the graph and return the result
if session_id:
logger.debug(f'Loading LangChain object from session {session_id}')
result = build_sorted_vertices(data_graph=data_graph)
if result is not None:
logger.debug('Loaded LangChain object')
return result
logger.debug('Building langchain object')
return build_sorted_vertices(data_graph)
def process_inputs(inputs: Optional[dict], artifacts: Dict[str, Any], input_key: str) -> dict:
if inputs is None:
inputs = {}
for key, value in artifacts.items():
if key == 'repr':
continue
elif key not in inputs or not inputs[key]:
inputs[key] = value
# For 7mmx7mmx10mm of blood clot,apiPengaturandefault inputto prevent the entry of changes after skill changes
if input_key not in inputs and 'default_input' in inputs:
inputs[input_key] = inputs.pop('default_input')
elif 'default_input' in inputs:
inputs.pop('default_input')
return inputs
def generate_result(langchain_object: Union[Chain, VectorStore], inputs: dict):
if isinstance(langchain_object, Chain):
if inputs is None:
raise ValueError('Inputs must be provided for a Chain')
logger.debug('Generating result and thought')
result = get_result_and_thought(langchain_object, inputs)
logger.debug('Generated result and thought')
elif isinstance(langchain_object, VectorStore):
result = langchain_object.search(**inputs)
elif isinstance(langchain_object, Document):
result = langchain_object.dict()
else:
logger.warning(f'Unknown langchain_object type: {type(langchain_object)}')
if isinstance(langchain_object, Coroutine):
result = asyncio.run(langchain_object)
result = langchain_object
return result
class Result(BaseModel):
result: Any = None
session_id: str
async def process_graph_cached(
data_graph: Dict[str, Any],
inputs: Optional[dict] = None,
clear_cache=False,
session_id=None,
flow_id=None,
history_count=10,
) -> Result:
session_service = get_session_service()
if clear_cache:
session_service.clear_session(session_id)
if session_id is None:
session_id = session_service.generate_key(session_id=session_id, data_graph=data_graph)
# Load the graph using SessionService
session = await session_service.load_session(session_id,
data_graph,
artifacts={},
process_file=True,
flow_id=flow_id,
chat_id=session_id)
graph, artifacts = session if session else (None, None)
if not graph:
raise ValueError('Graph not found in the session')
built_object = await graph.abuild()
input_key_object = built_object.input_keys[0]
# memery input
if hasattr(built_object, 'memory') and built_object.memory is not None:
fix_memory_inputs(built_object)
with get_sync_db_session() as session:
history = session.exec(
select(ChatMessage).where(
ChatMessage.chat_id == session_id,
ChatMessage.category.in_(['question', 'answer'])).order_by(
ChatMessage.id.desc()).limit(history_count)).all()
history = list(reversed(history))
next_loop = -1
for index, chat_message in enumerate(history):
if index + 1 >= len(history):
continue
if index <= next_loop:
continue
if not chat_message.is_bot and history[index + 1].is_bot:
next_loop = index + 1
if not chat_message.message or not chat_message.message.startswith('{'):
continue
inputs_hsitory = json.loads(chat_message.message)
outputs_history = {built_object.output_keys[0]: history[next_loop].message}
built_object.memory.save_context(inputs_hsitory, outputs_history)
if isinstance(built_object, Report):
processed_inputs = process_inputs(inputs, artifacts or {}, input_key_object)
result = generate_result(built_object, processed_inputs)
# build report
with get_sync_db_session() as db_session:
template = db_session.exec(
select(ReportModel).where(ReportModel.flow_id == flow_id).order_by(
ReportModel.id.desc())).first()
if not template:
logger.error('template not found flow_id={}', flow_id)
raise ValueError(f'template not found flow_id={flow_id}')
minio_client = get_minio_storage_sync()
template_muban = await minio_client.get_share_link(template.object_name, clear_host=False)
report_name = built_object.report_name
report_name = report_name if report_name.endswith('.docx') else f'{report_name}.docx'
result = (result.get(built_object.output_keys[0]) if isinstance(result, dict) else result)
test_replace_string(template_muban, result, report_name)
result = {built_object.output_keys[0]: await minio_client.get_share_link(report_name, clear_host=False)}
elif any(
(vertex.id.startswith('InputNode')
for vertex in graph.vertices)) and (not inputs
or all(len(ins) == 0 for ins in inputs.values())):
input_batch = []
for vertex in graph.vertices:
if vertex.id.startswith('InputNode'):
questions = await vertex.get_result()
for question in questions:
input_batch.append({built_object.input_keys[0]: question})
report = ''
for question in input_batch:
logger.info('produce auto question question={}', question)
processed_inputs = process_inputs(question, artifacts or {}, input_key_object)
result = generate_result(built_object, processed_inputs)
report = f"""{report}### {question} \n {result} \n """
result = report
else:
processed_inputs = process_inputs(inputs, artifacts or {}, input_key_object)
result = generate_result(built_object, processed_inputs)
# langchain_object is now updated with the new memory
# we need to update the cache with the updated langchain_object
session_service.update_session(session_id, (graph, artifacts))
return Result(result=result, session_id=session_id)
async def load_flow_from_json(flow: Union[Path, str, dict],
tweaks: Optional[dict] = None,
build=True):
"""
Load flow from a JSON file or a JSON object.
:param flow: JSON file path or JSON object
:param tweaks: Optional tweaks to be processed
:param build: If True, build the graph, otherwise return the graph object
:return: Langchain object or Graph object depending on the build parameter
"""
# If input is a file path, load JSON from the file
if isinstance(flow, (str, Path)):
with open(flow, 'r', encoding='utf-8') as f:
flow_graph = json.load(f)
# If input is a dictionary, assume it's a JSON object
elif isinstance(flow, dict):
flow_graph = flow
else:
raise TypeError('Input must be either a file path (str) or a JSON object (dict)')
graph_data = flow_graph['data']
if tweaks is not None:
graph_data = process_tweaks(graph_data, tweaks)
from bisheng.api.utils import build_flow_no_yield
graph = await build_flow_no_yield(graph_data=graph_data,
artifacts={},
process_file=True,
flow_id='tmp',
chat_id=None)
if build:
langchain_object = graph.build()
for object in langchain_object:
if hasattr(object._built_object, 'input_keys'):
langchain_object = object._built_object
break
if hasattr(langchain_object, 'verbose'):
langchain_object.verbose = True
if hasattr(langchain_object, 'return_intermediate_steps'):
# Deactivating until we have a frontend solution
# to display intermediate steps
langchain_object.return_intermediate_steps = False
fix_memory_inputs(langchain_object)
return langchain_object
return graph
def validate_input(graph_data: Dict[str, Any],
tweaks: Dict[str, Dict[str, Any]]) -> List[Dict[str, Any]]:
if not isinstance(graph_data, dict) or not isinstance(tweaks, dict):
raise ValueError('graph_data and tweaks should be dictionaries')
nodes = graph_data.get('data', {}).get('nodes') or graph_data.get('nodes')
if not isinstance(nodes, list):
raise ValueError(
"graph_data should contain a list of nodes under 'data' key or directly under 'nodes' key"
)
return nodes
def apply_tweaks(node: Dict[str, Any], node_tweaks: Dict[str, Any]) -> None:
template_data = node.get('data', {}).get('node', {}).get('template')
if not isinstance(template_data, dict):
logger.warning(f"Template data for node {node.get('id')} should be a dictionary")
return
for tweak_name, tweak_value in node_tweaks.items():
if tweak_name and tweak_value and tweak_name in template_data:
key = tweak_name if tweak_name == 'file_path' else 'value'
template_data[tweak_name][key] = tweak_value
elif tweak_name and tweak_value:
template_data[tweak_name] = TemplateField(field_type=type(tweak_value).__name__,
name=tweak_name,
value=tweak_value).to_dict()
def process_tweaks(graph_data: Dict[str, Any], tweaks: Dict[str, Dict[str,
Any]]) -> Dict[str, Any]:
"""
This function is used to tweak the graph data using the node id and the tweaks dict.
:param graph_data: The dictionary containing the graph data. It must contain a 'data' key with
'nodes' as its child or directly contain 'nodes' key. Each node should have an 'id' and 'data'.
:param tweaks: A dictionary where the key is the node id and the value is a dictionary of the tweaks.
The inner dictionary contains the name of a certain parameter as the key and the value to be tweaked.
:return: The modified graph_data dictionary.
:raises ValueError: If the input is not in the expected format.
"""
nodes = validate_input(graph_data, tweaks)
for node in nodes:
if isinstance(node, dict) and isinstance(node.get('id'), str):
node_id = node['id']
if node_tweaks := tweaks.get(node_id):
apply_tweaks(node, node_tweaks)
else:
logger.warning("Each node should be a dictionary with an 'id' key of type str")
return graph_data
================================================
FILE: src/backend/bisheng/run_celery.py
================================================
from bisheng.worker.main import bisheng_celery
if __name__ == '__main__':
bisheng_celery.start(
argv=['worker', '-l', 'info', '-c', '20', '-P', 'threads', '-Q', 'knowledge_celery,workflow_celery,celery'])
# bisheng_celery.worker_main(
# argv=["worker", "--loglevel=info", "--logfile=./logs/celery.log", '--pool=threads', '--concurrency=4',"-Q=workflow_celery"])
# worker.main(celery_app)
# celery -A run_celery.celery_app worker -l info -c 16
# celery -A run_celery.celery_app beat # Schedule cron job Rilis
# celery -A run_celery.celery_app worker -l info -P gevent # Scheduling Execution Tasks
================================================
FILE: src/backend/bisheng/run_celery_beat.py
================================================
from bisheng.worker.main import bisheng_celery
if __name__ == '__main__':
bisheng_celery.start(argv=['beat', '--loglevel=debug'])
# bisheng_celery.worker_main(
# argv=["worker", "--loglevel=info", "--logfile=./logs/celery.log", '--pool=threads', '--concurrency=4',"-Q=workflow_celery"])
# worker.main(celery_app)
# celery -A run_celery.celery_app worker -l info -c 16
# celery -A run_celery.celery_app beat # Schedule cron job Rilis
# celery -A run_celery.celery_app worker -l info -P gevent # Scheduling Execution Tasks
================================================
FILE: src/backend/bisheng/script/__init__.py
================================================
================================================
FILE: src/backend/bisheng/script/base_telemetry_events_reindex.py
================================================
from bisheng.core.search.elasticsearch.manager import get_statistics_es_connection_sync
INDEX_MAPPING = {
"mappings": { # Defining the indexed Mapping
"properties": {
"event_id": {"type": "keyword"},
"event_type": {"type": "keyword"},
"trace_id": {"type": "keyword"},
"timestamp": {"type": "date", "format": "strict_date_optional_time||epoch_second"},
"user_context": {
"type": "object",
"properties": {
"user_id": {"type": "integer"},
"user_name": {"type": "keyword"},
"user_group_infos": {
"type": "object",
"properties": {
"user_group_id": {"type": "integer"},
"user_group_name": {"type": "keyword"}
}
},
"user_role_infos": {
"type": "object",
"properties": {
"role_id": {"type": "integer"},
"role_name": {"type": "keyword"},
"group_id": {"type": "integer"},
}
}
}
},
"event_data": {
"type": "object",
"dynamic": True
}
}
}
}
import time
def wait_for_task(
es,
task_id: str,
poll_interval: int = 10,
timeout: int = 3600,
):
"""
轮询 ES task 状态，直到完成或超时
"""
start_time = time.time()
while True:
task_info = es.tasks.get(task_id=task_id)
completed = task_info.get("completed", False)
if completed:
response = task_info.get("response", {})
failures = response.get("failures", [])
total = response.get("total", 0)
created = response.get("created", 0)
updated = response.get("updated", 0)
print(
f"[REINDEX DONE] total={total}, created={created}, updated={updated}"
)
if failures:
raise RuntimeError(f"Reindex failures: {failures}")
return response
if time.time() - start_time > timeout:
raise TimeoutError(f"Reindex task timeout: {task_id}")
status = task_info.get("task", {}).get("status", {})
print(
f"[REINDEX RUNNING] "
f"total={status.get('total', 0)} "
f"created={status.get('created', 0)} "
f"updated={status.get('updated', 0)}"
)
time.sleep(poll_interval)
def count_docs(es, index):
return es.count(index=index)["count"]
if __name__ == '__main__':
es_conn = get_statistics_es_connection_sync()
temp_index_name = "base_telemetry_events_v1"
original_index_name = "base_telemetry_events"
# 1. 记录原始数据量
try:
source_count = count_docs(es_conn, original_index_name)
print(f"Original doc count: {source_count}")
except:
source_count = 0
print("Original index might not exist.")
# 创建临时索引
if not es_conn.indices.exists(index=temp_index_name):
es_conn.indices.create(index=temp_index_name, body=INDEX_MAPPING)
print(f"Created temporary index: {temp_index_name}")
# 使用Elasticsearch的_reindex API进行数据迁移
reindex_body = {
"source": {
"index": original_index_name
},
"dest": {
"index": temp_index_name
}
}
resp = es_conn.options(request_timeout=3600).reindex(
body=reindex_body,
wait_for_completion=False
)
task_id = resp["task"]
print(f"Reindex started, task_id={task_id}")
wait_for_task(
es_conn,
task_id=task_id,
poll_interval=5,
timeout=3600
)
# 删除原始索引
es_conn.indices.delete(index=original_index_name)
# 将临时索引重命名为原始索引名
es_conn.indices.put_alias(index=temp_index_name, name=original_index_name)
print(f"Reindexed data to {original_index_name} successfully.")
================================================
FILE: src/backend/bisheng/script/convert_sys_embeddings.py
================================================
from bisheng.common.services.config_service import settings
from bisheng.knowledge.domain.models.knowledge import KnowledgeDao
from bisheng.llm.domain.const import LLMServerType, LLMModelType
from bisheng.llm.domain.models import LLMServer, LLMDao, LLMModel
def parse_openai_embedding_conf(name, model_name, embedding_conf):
# Explanation isazureright of privacyapiSERVICES
if embedding_conf.get('openai_api_type') in ('azure', 'azure_ad', 'azuread'):
server = LLMServer(
name=f"AzureOpenAI",
description='System upgrade automatically added',
type=LLMServerType.AZURE_OPENAI.value,
config={
"azure_endpoint": embedding_conf.pop("azure_endpoint", ''),
"openai_api_key": embedding_conf.pop("openai_api_key", ''),
"openai_api_version": embedding_conf.pop("openai_api_version", '')
},
user_id=1,
)
model = LLMModel(
name=name,
model_name=model_name,
model_type=LLMModelType.EMBEDDING.value,
online=True,
user_id=1,
)
else:
server = LLMServer(
name=f"OpenAI",
description='System upgrade automatically added',
type=LLMServerType.OPENAI.value,
config={
"openai_api_key": embedding_conf.pop("openai_api_key", ''),
"openai_api_base": embedding_conf.pop("openai_api_base", ''),
"openai_proxy": embedding_conf.pop("openai_proxy", ''),
},
user_id=1,
)
model = LLMModel(
name=name,
model_name=model_name,
model_type=LLMModelType.EMBEDDING.value,
online=True,
user_id=1,
)
return server, model
def parse_rt_embedding_conf(name, model_name, embedding_conf):
server = LLMServer(
name=f"RT",
description='System upgrade automatically added',
type=LLMServerType.BISHENG_RT.value,
config={
"host_base_url": embedding_conf.get("host_base_url", ''),
},
user_id=1,
)
model = LLMModel(
name=name,
model_name=model_name,
model_type=LLMModelType.EMBEDDING.value,
online=True,
user_id=1,
)
return server, model
# In the system configuration,embeddingconfiguration item, to the service provider in the model management, level-up034Execute this script
def convert_sys_embeddings_to_mysql():
knowledge_conf = settings.get_knowledge()
embeddings = knowledge_conf.get('embeddings', {})
if not embeddings:
print('no found embeddings')
return
# Query if there is an existing knowledge base
all_knowledge = KnowledgeDao.get_all_knowledge()
if not all_knowledge:
return
# Warehousing all system configurations first
need_add_server = {}
need_add_server_index = {}
for name, embedding_conf in embeddings.items():
model_name = embedding_conf.get('model')
if not model_name and name == 'text-embedding-ada-002':
model_name = 'text-embedding-ada-002'
if not model_name:
print("not foundmodelFirst name, not inserted into model management")
continue
# Explanation isopenaiOfficial Services of
if name == 'text-embedding-ada-002' or embedding_conf.get('component') == 'openai':
server, model = parse_openai_embedding_conf(name, model_name, embedding_conf)
else:
# Explanation isrtDeployedembeddingModels
server, model = parse_rt_embedding_conf(name, model_name, embedding_conf)
if server.type not in need_add_server_index:
need_add_server_index[server.type] = 0
need_add_server_index[server.type] += 1
server.name = f"{server.name}_{need_add_server_index[server.type]}"
llm_server = LLMDao.insert_server_with_models(server, [model])
llm_model_list = LLMDao.get_model_by_server_ids([llm_server.id])
for one in llm_model_list:
if one.name == name:
need_add_server[name] = one
# Reset Model Configuration for Knowledge Base
update_knowledge = []
for one in all_knowledge:
if one.model in need_add_server:
print(f"Modify Knowledge Base【{one.name}】 Model Configuration for")
one.model = need_add_server[one.model].id
update_knowledge.append(one)
if update_knowledge:
KnowledgeDao.update_knowledge_list(update_knowledge)
if not need_add_server_index.get(LLMServerType.BISHENG_RT.value):
# Add a defaultRTservice provider
server = LLMServer(
name=f"RT_OLD",
description='The system upgrade is automatically added, and it is not recommended to use it in the future',
type=LLMServerType.BISHENG_RT.value,
config={
"host_base_url": 'http://xxxx:8000',
},
user_id=1,
)
llm_server = LLMDao.insert_server_with_models(server, [])
if __name__ == '__main__':
convert_sys_embeddings_to_mysql()
================================================
FILE: src/backend/bisheng/script/knowledge.sh
================================================
#!/bin/bash
export PYTHONPATH="./"
run_mode=${1:-convert_all}
knowledge_id=${2:-0}
if [ "$run_mode" = "convert_all" ]; then
echo "Converting all knowledge"
python bisheng/script/knowledge_data_convert.py --mode convert_all
elif [ "$run_mode" = "convert_one" ]; then
echo "Converting one knowledge : $knowledge_id"
python bisheng/script/knowledge_data_convert.py --mode convert_one --id ${knowledge_id}
elif [ "$run_mode" = "scan_all" ]; then
echo "Scanning all knowledge files..."
python bisheng/script/knowledge_data_fix.py --mode scan_all
elif [ "$run_mode" = "scan_one" ]; then
echo "Scanning single knowledge : $knowledge_id"
python bisheng/script/knowledge_data_fix.py --mode scan_one --id ${knowledge_id}
elif [ "$run_mode" = "fix_all" ]; then
echo "Fixing all knowledge files..."
python bisheng/script/knowledge_data_fix.py --mode fix_all
elif [ "$run_mode" = "fix_one" ]; then
echo "Fixing single knowledge : $knowledge_id"
python bisheng/script/knowledge_data_fix.py --mode fix_one --id ${knowledge_id}
else
echo "Invalid run mode. Use 'convert_all', 'convert_one', 'scan_all', 'scan_one', 'fix_all', or 'fix_one'."
exit 1
fi
================================================
FILE: src/backend/bisheng/script/knowledge_data_convert.py
================================================
import argparse
import json
import traceback
from typing import List, Dict
from bisheng.common.constants.vectorstore_metadata import KNOWLEDGE_RAG_METADATA_SCHEMA
from bisheng.knowledge.domain.knowledge_rag import KnowledgeRag
from bisheng.knowledge.domain.models.knowledge import Knowledge, KnowledgeDao, KnowledgeTypeEnum
from bisheng.knowledge.domain.models.knowledge_file import KnowledgeFile
from bisheng.llm.domain import LLMService
from bisheng.script.knowledge_data_fix import get_all_knowledge_files, _get_milvus_chunks_data, _get_es_chunks_data
user_map = {}
def get_user_name_by_id(user_id: int):
from bisheng.user.domain.models.user import UserDao
if user_id in user_map:
return user_map[user_id]
tmp_user = UserDao.get_user(user_id)
if tmp_user:
user_map[user_id] = tmp_user.user_name
return tmp_user.user_name
return f"unknown user: {user_id}"
def convert_new_metadata(old_metadata: Dict, file: KnowledgeFile, knowledge: Knowledge):
user_metadata = {}
if old_metadata.get("extra"):
try:
user_metadata = json.loads(old_metadata["extra"])
except Exception:
pass
return {
"document_id": file.id,
"document_name": file.file_name,
"abstract": old_metadata.get("title", ""),
"chunk_index": old_metadata.get("chunk_index", 0),
"bbox": old_metadata.get("bbox", ""),
"page": old_metadata.get("page", 0),
"knowledge_id": knowledge.id,
"upload_time": int(file.create_time.timestamp()),
"update_time": int(file.update_time.timestamp()),
"uploader": get_user_name_by_id(file.user_id),
"updater": get_user_name_by_id(file.user_id),
"user_metadata": user_metadata,
}
def convert_milvus_data(knowledge: Knowledge, all_file: List[KnowledgeFile], new_collection_name: str):
print(f"converting ID:{knowledge.id} right of privacyMilvusDATA...")
# convert_milvus_chunk
# New.. MilvusObjects
embedding = LLMService.get_bisheng_knowledge_embedding_sync(0, model_id=int(knowledge.model))
new_milvus = KnowledgeRag.init_milvus_vectorstore(new_collection_name, embedding,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
for file in all_file:
all_file_chunk = _get_milvus_chunks_data(knowledge, all_fields_expect_pk=True, file_id=file.id)
if not all_file_chunk:
print(f"The knowledge base upon ID:{knowledge.id} from here{file.id}NoMilvusdata, skipping the conversion.")
continue
if all_file_chunk[0].get("document_id"):
print(f"The knowledge base upon ID:{knowledge.id} right of privacyMilvusData is already in the new format, skipping conversion.")
return None
new_texts = []
new_metadata = []
new_embedding = []
for one in all_file_chunk:
new_texts.append(one["text"])
new_metadata.append(convert_new_metadata(one, file, knowledge))
new_embedding.append(one["vector"])
new_milvus.add_embeddings(texts=new_texts, embeddings=new_embedding, metadatas=new_metadata)
# Add conversion logic here
print(f"The knowledge base upon ID:{knowledge.id} right of privacyMilvusData conversion complete.")
return new_milvus
def convert_es_data(knowledge: Knowledge, all_file: List[KnowledgeFile], new_index_name: str):
print(f"converting ID:{knowledge.id} right of privacyElasticsearchDATA...")
# convert_es_chunk
es_vectorstore = KnowledgeRag.init_es_vectorstore_sync(new_index_name,
metadata_schemas=KNOWLEDGE_RAG_METADATA_SCHEMA)
for file in all_file:
all_file_chunk = _get_es_chunks_data(knowledge, source=True, file_id=file.id)
if not all_file_chunk:
print(f"The knowledge base upon ID:{knowledge.id} from here{file.id}Noesdata, skipping the conversion.")
continue
if all_file_chunk[0].get("document_id"):
print(f"The knowledge base upon ID:{knowledge.id} right of privacyESData is already in the new format, skipping conversion.")
return None
new_texts = []
new_metadata = []
for one in all_file_chunk:
new_texts.append(one["_source"]["text"])
new_metadata.append(convert_new_metadata(one["_source"]["metadata"], file, knowledge))
es_vectorstore.add_texts(texts=new_texts, metadatas=new_metadata)
print(f"The knowledge base upon ID:{knowledge.id} right of privacyElasticsearchData conversion complete.")
return es_vectorstore
def convert_one_knowledge_data(knowledge: Knowledge):
print(f"Start Conversion ID:{knowledge.id} {knowledge.name} Data...")
if knowledge.collection_name.startswith("partition_"):
print(f"!!! Skip partition knowledge base ID:{knowledge.id} Data conversion. Please repair the data first")
return
if knowledge.type == KnowledgeTypeEnum.QA.value:
print(f"SkipQAThe knowledge base upon ID:{knowledge.id} Data conversion.!!!")
return
all_file = get_all_knowledge_files(knowledge.id)
old_collection_name = knowledge.collection_name
old_index_name = knowledge.index_name
new_collection_name = f"{old_collection_name}_new"
new_index_name = f"{old_index_name}_new"
try:
old_milvus_vector = KnowledgeRag.init_knowledge_milvus_vectorstore_sync(0, knowledge)
old_fields = old_milvus_vector.col.schema.fields
old_fields = {field.name: field for field in old_fields}
if "document_id" in old_fields:
print(f"The knowledge base upon ID:{knowledge.id} The data is already in the new format, skipping the conversion.")
return
milvus_vector = convert_milvus_data(knowledge, all_file, new_collection_name)
es_vector = convert_es_data(knowledge, all_file, new_index_name)
if milvus_vector and es_vector:
knowledge.collection_name = new_collection_name
knowledge.index_name = new_index_name
KnowledgeDao.update_one(knowledge)
# clear old data
milvus_vector.client.drop_collection(old_collection_name)
if es_vector.client.indices.exists(index=old_index_name):
es_vector.client.indices.delete(index=old_index_name)
else:
print(f"The knowledge base upon ID:{knowledge.id} Data does not need to be updated.")
except Exception as e:
print(f"The knowledge base upon ID:{knowledge.id} Data conversion failed, error reason:{e}")
traceback.print_exc()
# Add conversion logic here
print(f"The knowledge base upon ID:{knowledge.id} Data conversion complete.")
def convert_all_knowledge_data():
all_knowledge = KnowledgeDao.get_all_knowledge()
total = len(all_knowledge)
for index, knowledge in enumerate(all_knowledge):
print(
f"convert progress: {round((index + 1) / total * 100, 2)}% knowledge id: {knowledge.id} name: {knowledge.name}")
convert_one_knowledge_data(knowledge)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--mode', type=str, default="convert_all",
help='modalities.convert_all: Convert data from all knowledge bases;convert_one: Convert data from a knowledge base')
# Maximum number of concurrency for a single process
parser.add_argument('--id', type=int, default=0, help='The knowledge base uponID, parameter is required if operating a single knowledge base')
args = parser.parse_args()
if args.mode == "convert_all":
convert_all_knowledge_data()
elif args.mode == "convert_one":
tmp_knowledge = KnowledgeDao.query_by_id(args.id)
if not tmp_knowledge:
print(f"The knowledge base upon ID:{args.id} It does not exist and cannot be converted.")
exit(0)
convert_one_knowledge_data(tmp_knowledge)
else:
print("modeParameter salah,can only be convert_all OR convert_one")
================================================
FILE: src/backend/bisheng/script/knowledge_data_fix.py
================================================
import argparse
import os
import traceback
from typing import List
import openpyxl
from pymilvus import Collection
from bisheng.api.services.knowledge_imp import decide_vectorstores, QA_save_knowledge, delete_vector_data
from bisheng.knowledge.domain.models.knowledge import KnowledgeDao, Knowledge, KnowledgeTypeEnum
from bisheng.knowledge.domain.models.knowledge_file import QAKnoweldgeDao, KnowledgeFileDao, QAKnowledge, KnowledgeFile, \
QAStatus, KnowledgeFileStatus
from bisheng.llm.domain import LLMService
from bisheng.worker.knowledge.file_worker import insert_milvus
_output_path = os.path.join(os.path.dirname(__file__), 'output')
os.makedirs(_output_path, exist_ok=True)
def get_all_knowledge() -> List[Knowledge]:
all_knowledge = []
page = 1
limit = 1000
while True:
knowledge = KnowledgeDao.get_all_knowledge(page=page, limit=limit)
if not knowledge:
break
page += 1
all_knowledge.extend(knowledge)
return all_knowledge
def get_all_qa_knowledge_files(knowledge_id: int) -> List[QAKnowledge]:
all_files = []
page = 1
limit = 1000
while True:
files = QAKnoweldgeDao.get_qa_knowledge_by_knowledge_id(knowledge_id, page=page, page_size=limit)
if not files:
break
page += 1
for file in files:
if file.status != QAStatus.ENABLED.value:
continue
all_files.append(file)
return all_files
def get_all_knowledge_files(knowledge_id: int) -> List[KnowledgeFile]:
all_files = []
page = 1
limit = 1000
while True:
files = KnowledgeFileDao.get_file_by_filters(knowledge_id, page=page, page_size=limit)
if not files:
break
page += 1
for file in files:
if file.status != KnowledgeFileStatus.SUCCESS.value:
continue
all_files.append(file)
return all_files
def _get_es_chunks_data(knowledge: Knowledge, es_obj=None, source: bool = False, file_id: int = None):
if not es_obj:
embedding = LLMService.get_bisheng_knowledge_embedding_sync(0, model_id=int(knowledge.model))
es_obj = decide_vectorstores(knowledge.index_name or knowledge.collection_name, "ElasticKeywordsSearch",
embedding)
es_client = es_obj.client
all_chunks = []
def handle_hits(hits):
for hit in hits:
all_chunks.append({
"file_id": hit['fields'].get('metadata.file_id', [''])[0],
"source": hit['fields'].get('metadata.source', [''])[0],
"extra": hit['fields'].get('metadata.extra', [''])[0],
"_id": hit['_id'],
"_source": hit['_source'] if source else None,
})
query = {"match_all": {}}
if file_id:
query = {"term": {"metadata.file_id": file_id}}
result = es_client.search(index=knowledge.index_name or knowledge.collection_name,
query=query,
size=5000,
scroll="5m",
source=source,
fields=["_id", "metadata.source", "metadata.file_id", "metadata.extra"])
handle_hits(result["hits"]["hits"])
scroll_id = result['_scroll_id']
while True:
result = es_client.scroll(scroll_id=scroll_id, scroll='1m')
tmp_hits = result['hits']['hits']
if not tmp_hits:
break
handle_hits(tmp_hits)
scroll_id = result['_scroll_id']
es_client.clear_scroll(scroll_id=scroll_id)
return all_chunks
def _get_milvus_chunks_data(knowledge: Knowledge, milvus_obj=None, all_fields_expect_pk: bool = False,
file_id: int = None):
if not milvus_obj:
embedding = LLMService.get_bisheng_knowledge_embedding_sync(0, model_id=int(knowledge.model))
milvus_obj = decide_vectorstores(knowledge.collection_name, "Milvus", embedding)
all_milvus_chunks = []
output_fields = ["file_id", "extra", "source", "pk"]
if all_fields_expect_pk:
output_fields = [s.name for s in milvus_obj.col.schema.fields if s.name != "pk"]
expr = f"knowledge_id=='{knowledge.id}'" if knowledge.collection_name.startswith("partition_") else None
if file_id:
expr = f"{expr} and file_id=={file_id}" if expr else f"file_id == {file_id}"
iterator = milvus_obj.col.query_iterator(
batch_size=1000,
expr=expr,
output_fields=output_fields,
timeout=30,
)
while True:
result = iterator.next()
if not result:
iterator.close()
break
all_milvus_chunks.extend(result)
return all_milvus_chunks
def _scan_knowledge_error_data(knowledge: Knowledge, all_file_data: List[KnowledgeFile | QAKnowledge], milvus_obj,
es_obj):
all_milvus_chunks = _get_milvus_chunks_data(knowledge, milvus_obj)
judge_qa_chunk = lambda x: not x.get("source")
if knowledge.type != KnowledgeTypeEnum.QA.value:
judge_qa_chunk = lambda x: x.get("source")
all_milvus_chunks_map = {
item["file_id"]: item for item in all_milvus_chunks if judge_qa_chunk(item) # source Absent Explanation YesQACorrect data, otherwise it is the data of the document knowledge base
}
all_es_chunks = _get_es_chunks_data(knowledge, es_obj)
all_es_chunks_map = {
item["file_id"]: item for item in all_es_chunks if judge_qa_chunk(item) # source Absent Explanation YesQACorrect data, otherwise it is the data of the document knowledge base
}
no_data = []
no_milvus_data = []
no_es_data = []
for file in all_file_data:
milvus_flag = file.id in all_milvus_chunks_map
if milvus_flag:
all_milvus_chunks_map.pop(file.id)
es_flag = file.id in all_es_chunks_map
if es_flag:
all_es_chunks_map.pop(file.id)
if not milvus_flag and not es_flag:
no_data.append(file)
elif not milvus_flag:
no_milvus_data.append(file)
elif not es_flag:
no_es_data.append(file)
return no_data, no_milvus_data, no_es_data, all_milvus_chunks_map, all_es_chunks_map
def scan_qa_knowledge_error_data(knowledge: Knowledge, milvus_obj, es_obj):
""" scan all qa knowledge data and find those that not exist in milvus """
all_qa = get_all_qa_knowledge_files(knowledge.id)
return _scan_knowledge_error_data(knowledge, all_qa, milvus_obj, es_obj)
def scan_normal_knowledge_error_data(knowledge: Knowledge, milvus_obj, es_obj):
""" scan all normal knowledge data and find those that not exist in milvus """
all_files = get_all_knowledge_files(knowledge.id)
return _scan_knowledge_error_data(knowledge, all_files, milvus_obj, es_obj)
def _file_status(file: KnowledgeFile):
if file.status == KnowledgeFileStatus.PROCESSING.value:
return "Parsing"
elif file.status == KnowledgeFileStatus.SUCCESS.value:
return "Parsing Successful"
elif file.status == KnowledgeFileStatus.FAILED.value:
return "Parse Failure"
elif file.status == KnowledgeFileStatus.REBUILDING.value:
return "Rebuilding"
elif file.status == KnowledgeFileStatus.WAITING.value:
return "In queue:"
elif file.status == KnowledgeFileStatus.TIMEOUT.value:
return "Timed out"
else:
return "Unknown Status"
def _qa_status(file: QAKnowledge):
if file.status == QAStatus.ENABLED.value:
return "Enable"
elif file.status == QAStatus.DISABLED.value:
return "Disable"
elif file.status == QAStatus.PROCESSING.value:
return "Sedang diproses"
elif file.status == QAStatus.FAILED.value:
return "Failed to Process"
else:
return "Unknown Status"
def _knowledge_common_row(knowledge: Knowledge, note: str = ""):
one_common_row = [knowledge.id, knowledge.name, knowledge.collection_name, knowledge.index_name]
if knowledge.type == KnowledgeTypeEnum.QA.value:
one_common_row.append("QAThe knowledge base upon")
elif knowledge.type == KnowledgeTypeEnum.NORMAL.value:
one_common_row.append("Docly Knowledge Base")
elif knowledge.type == KnowledgeTypeEnum.PRIVATE.value:
one_common_row.append("Personal Knowledge Base")
else:
one_common_row.append("Unknown type knowledge base")
one_common_row.append(knowledge.create_time.strftime("%Y-%m-%d %H:%M:%S"))
one_common_row.append(knowledge.update_time.strftime("%Y-%m-%d %H:%M:%S"))
one_common_row.append(note)
return one_common_row
def _file_row(knowledge: Knowledge, file: KnowledgeFile | QAKnowledge, milvus_flag: str, es_flag: str):
return [
file.id,
file.file_name if knowledge.type != KnowledgeTypeEnum.QA.value else file.questions[0],
_file_status(file) if knowledge.type != KnowledgeTypeEnum.QA.value else _qa_status(file),
file.create_time.strftime("%Y-%m-%d %H:%M:%S"),
file.update_time.strftime("%Y-%m-%d %H:%M:%S"),
milvus_flag,
es_flag,
]
def _init_knowledge_obj(knowledge: Knowledge):
try:
embedding = LLMService.get_bisheng_knowledge_embedding_sync(0, model_id=int(knowledge.model))
except Exception as e:
print(
f"!!!! skip knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name} because embedding model error: {e}")
raise Exception(f"Skip this Knowledge Base, Reason:embeddingModel error: {e}")
try:
milvus_obj = decide_vectorstores(knowledge.collection_name, "Milvus", embedding)
if milvus_obj.col is None:
raise Exception("Skip this Knowledge Base, Reason:Milvus collection name not exist")
collection_info = milvus_obj.col.schema
fields = collection_info.fields
fields = {field.name: 1 for field in fields}
if "extra" not in fields or "source" not in fields or "file_id" not in fields:
raise Exception("Skip this Knowledge Base, Reason:Milvus fields not found file_id or source or extra")
except Exception as e:
print(
f"!!!! skip knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name} because milvus connection error: {e}")
raise Exception(f"Skip this Knowledge Base, Reason:MilvusConnection Error: {e}")
try:
es_obj = decide_vectorstores(knowledge.index_name or knowledge.collection_name, "ElasticKeywordsSearch",
embedding)
except Exception as e:
print(
f"!!!! skip knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name} because es connection error: {e}")
raise Exception(f"Skip this Knowledge Base, Reason:ESConnection Error: {e}")
return milvus_obj, es_obj
def _save_knowledge_error_data(rows: List[List[str]], file_name: str):
header_rows = [
['The knowledge base uponID', 'Library Name', 'collection_name', 'index_name', 'Knowledge Base Post Type', 'Knowledge Base Created Time', 'Knowledge Base Updated Time',
'Knowledge Base Notes', 'Doc.ID', 'File Name',
'Document Status',
'File Created Time', 'File Updated Time', 'Milvuspresence or does it', 'ESpresence or does it']
]
if not rows:
print("no error data found")
return
rows = header_rows + rows
wb = openpyxl.workbook.Workbook()
sh = wb.active
for row in rows:
sh.append(row)
file_path = os.path.join(_output_path, file_name)
wb.save(file_path)
print(f"=========== knowledge error data file saved to: {file_path}")
def scan_one_knowledge(knowledge: Knowledge = None, knowledge_id: int = None) -> List[List[str]]:
""" return error data rows for one knowledge """
if not knowledge:
knowledge = KnowledgeDao.query_by_id(knowledge_id)
if not knowledge:
print(f"knowledge_id: {knowledge_id} not exist")
return []
try:
milvus_obj, es_obj = _init_knowledge_obj(knowledge)
except Exception as e:
print(
f"!!!! skip knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name} because error: {e}")
return [_knowledge_common_row(knowledge, str(e))]
if knowledge.type == KnowledgeTypeEnum.QA.value:
no_data, no_milvus_data, no_es_data, milvus_extra, es_extra = scan_qa_knowledge_error_data(knowledge,
milvus_obj,
es_obj)
else:
no_data, no_milvus_data, no_es_data, milvus_extra, es_extra = scan_normal_knowledge_error_data(knowledge,
milvus_obj,
es_obj)
if not no_data and not no_milvus_data and not no_es_data:
return []
print(f"!!!! find error data knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name}")
note = ""
if milvus_extra and es_extra:
note = "Milvus And ES Redundant data present"
elif milvus_extra:
note = "Milvus Redundant data present"
elif es_extra:
note = "ES Redundant data present"
one_common_row = _knowledge_common_row(knowledge, note)
def _parse_one_row(file, milvus_flag: str, es_flag: str):
one_row = one_common_row.copy()
one_row.extend(_file_row(knowledge, file, milvus_flag, es_flag))
return one_row
rows = []
for one in no_data:
rows.append(_parse_one_row(one, "No ", "No "))
for one in no_milvus_data:
rows.append(_parse_one_row(one, "No ", "Yes "))
for one in no_es_data:
rows.append(_parse_one_row(one, "Yes ", "No "))
if not rows and note:
return [one_common_row]
return rows
def scan_knowledge_error_data():
""" scan all knowledge data and find those that not exist in milvus """
all_knowledge = get_all_knowledge()
total = len(all_knowledge)
rows = []
for index, knowledge in enumerate(all_knowledge):
print(
f"{round(index / total * 100, 2)}% ---- start scan knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name}")
error_rows = scan_one_knowledge(knowledge)
if error_rows:
rows.extend(error_rows)
else:
print(f"==== no error data knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name}")
_save_knowledge_error_data(rows, "all_knowledge_error_data.xlsx")
return None
def _sync_milvus_new_collection_name(knowledge: Knowledge, milvus_obj) -> bool:
""" copy knowledge data to new milvus collection name """
embedding = LLMService.get_bisheng_knowledge_embedding_sync(0, model_id=int(knowledge.model))
new_collection_name = knowledge.index_name or knowledge.collection_name
if new_collection_name == knowledge.collection_name:
return milvus_obj
# create new collection name
new_col = Collection(name=new_collection_name, schema=milvus_obj.col.schema, using=milvus_obj.alias,
consistency_level=milvus_obj.consistency_level)
new_milvus_obj = decide_vectorstores(new_collection_name, "Milvus", embedding)
output_fields = [s.name for s in milvus_obj.col.schema.fields if s.name != "pk"]
try:
# get old chunks
all_chunks = _get_milvus_chunks_data(knowledge, milvus_obj, all_fields_expect_pk=True)
insert_milvus(all_chunks, output_fields, new_milvus_obj)
knowledge.collection_name = new_collection_name
knowledge = KnowledgeDao.update_one(knowledge)
return new_milvus_obj
except Exception as e:
print(
f"!!!! copy vector data knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name} error: {e}")
new_milvus_obj.col.delete()
return False
def _fix_knowledge_data(knowledge: Knowledge, milvus_obj, es_obj, all_files: List[QAKnowledge | KnowledgeFile]):
""" fix all qa knowledge data """
if knowledge.collection_name.startswith("partition_"):
print("start copy milvus data to new collection_name")
milvus_obj = _sync_milvus_new_collection_name(knowledge, milvus_obj)
if not milvus_obj:
raise Exception(f"copy milvus data to new collection_name failed")
print("finish copy milvus data to new collection_name")
all_milvus_chunk = _get_milvus_chunks_data(knowledge, milvus_obj, all_fields_expect_pk=True)
all_milvus_chunk_map = {}
all_es_chunk = _get_es_chunks_data(knowledge, es_obj, source=True)
all_es_chunk_map = {}
judge_qa_chunk = lambda x: not x.get("source")
if knowledge.type != KnowledgeTypeEnum.QA.value:
judge_qa_chunk = lambda x: x.get("source")
remove_pk = []
remove_id = []
for item in all_milvus_chunk:
# have source, is doc chunk
if not judge_qa_chunk(item):
remove_pk.append(item.get("pk"))
continue
if item["file_id"] not in all_milvus_chunk_map:
all_milvus_chunk_map[item["file_id"]] = []
all_milvus_chunk_map[item["file_id"]].append(item)
for item in all_es_chunk:
# no source, is qa chunk
if not judge_qa_chunk(item):
remove_id.append(item.get("_id"))
continue
if item["file_id"] not in all_es_chunk_map:
all_es_chunk_map[item["file_id"]] = []
all_es_chunk_map[item["file_id"]].append(item)
old_all_milvus_chunk_map = all_milvus_chunk_map.copy()
old_all_es_chunk_map = all_es_chunk_map.copy()
no_data = []
no_milvus_data = []
no_es_data = []
for file in all_files:
milvus_flag = file.id in all_milvus_chunk_map
if milvus_flag:
all_milvus_chunk_map.pop(file.id)
es_flag = file.id in all_es_chunk_map
if es_flag:
all_es_chunk_map.pop(file.id)
if not milvus_flag and not es_flag:
no_data.append(file)
elif not milvus_flag:
no_milvus_data.append(file)
elif not es_flag:
no_es_data.append(file)
if remove_pk:
print(
f"----remove extra milvus data knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name} pk: {remove_pk}")
milvus_obj.col.delete(expr=f"pk in {remove_pk}")
if remove_id:
print(
f"----remove extra es data knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name} _id: {remove_id}")
es_obj.client.delete_by_query(index=knowledge.index_name or knowledge.collection_name,
query={"terms": {"_id": remove_id}})
if all_milvus_chunk_map:
print(f"---- fix knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name} milvus extra data found")
for file_id, chunks in all_milvus_chunk_map.items():
milvus_obj.col.delete(expr=f"file_id=={file_id}")
if all_es_chunk_map:
print(f"---- fix knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name} es extra data found")
es_obj.client.delete_by_query(index=knowledge.index_name or knowledge.collection_name,
query={"terms": {"metadata.file_id": list(all_es_chunk_map.keys())}})
return no_data, no_milvus_data, no_es_data, old_all_milvus_chunk_map, old_all_es_chunk_map, milvus_obj
def fix_qa_knowledge_data(knowledge: Knowledge, milvus_obj, es_obj):
all_qa = get_all_qa_knowledge_files(knowledge.id)
no_data, no_milvus_data, no_es_data, _, _, _ = _fix_knowledge_data(knowledge, milvus_obj, es_obj, all_qa)
for qa in no_data:
print(f"try insert qa {qa.id} into vector store")
QA_save_knowledge(knowledge, qa)
for qa in no_milvus_data:
print(f"try insert qa {qa.id} into vector store")
delete_vector_data(knowledge, [qa.id])
QA_save_knowledge(knowledge, qa)
for qa in no_es_data:
print(f"try insert qa {qa.id} into vector store")
delete_vector_data(knowledge, [qa.id])
QA_save_knowledge(knowledge, qa)
def fix_normal_knowledge_data(knowledge: Knowledge, milvus_obj, es_obj):
all_file = get_all_knowledge_files(knowledge.id)
no_data, no_milvus_data, no_es_data, all_milvus, all_es, milvus_obj = _fix_knowledge_data(knowledge, milvus_obj,
es_obj,
all_file)
for file in no_data:
print(f"update file to failed status file_id: {file.id}; file_name: {file.file_name}")
KnowledgeFileDao.update_file_status([file.id], KnowledgeFileStatus.FAILED,
"The file data was found to be missing when the knowledge base data was repaired. The file status has been updated to parse failed. Please parse again.")
for file in no_milvus_data:
es_chunks = all_es.get(file.id, [])
if not es_chunks:
file.status = KnowledgeFileStatus.FAILED.value
file.remark = "The file data was found to be missing when the knowledge base data was repaired. The file status has been updated to parse failed. Please parse again."
KnowledgeFileDao.update(file)
print(f"!!!!skip file_id: {file.id}; file_name: {file.file_name} because es data not found")
continue
texts = []
metadatas = []
for chunk in es_chunks:
texts.append(chunk["_source"]["text"])
metadatas.append(chunk["_source"]["metadata"])
try:
milvus_obj.add_texts(texts=texts, metadatas=metadatas)
except Exception as e:
file.status = KnowledgeFileStatus.FAILED.value
file.remark = str(e)
KnowledgeFileDao.update(file)
for file in no_es_data:
milvus_chunks = all_milvus.get(file.id, [])
if not milvus_chunks:
print(f"!!!!skip file_id: {file.id}; file_name: {file.file_name} because milvus data not found")
file.status = KnowledgeFileStatus.FAILED.value
KnowledgeFileDao.update(file)
continue
texts = []
metadatas = []
for chunk in milvus_chunks:
texts.append(chunk.pop("text"))
chunk.pop("pk")
chunk.pop("vector")
metadatas.append(chunk)
es_obj.add_texts(texts=texts, metadatas=metadatas)
def fix_one_knowledge(knowledge: Knowledge = None, knowledge_id: int = None):
if not knowledge:
knowledge = KnowledgeDao.query_by_id(knowledge_id)
if not knowledge:
print(f"knowledge_id: {knowledge_id} not exist")
return
try:
milvus_obj, es_obj = _init_knowledge_obj(knowledge)
except Exception as e:
print(
f"!!!! skip knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name} because error: {e}")
return
if not knowledge.index_name:
knowledge.index_name = knowledge.collection_name
KnowledgeDao.update_one(knowledge)
try:
print(f"---- start fix knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name}")
if knowledge.type == KnowledgeTypeEnum.QA.value:
fix_qa_knowledge_data(knowledge, milvus_obj, es_obj)
else:
fix_normal_knowledge_data(knowledge, milvus_obj, es_obj)
print(f"---- finish fix knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name}")
except Exception as e:
print(f"---- error fix knowledge_id: {knowledge.id}; knowledge_name: {knowledge.name}")
traceback.print_exc()
def fix_knowledge_error_data():
all_knowledge = get_all_knowledge()
total = len(all_knowledge)
for index, knowledge in enumerate(all_knowledge):
print(f"---- start fix knowledge process: {round(index / total * 100, 2)}%")
fix_one_knowledge(knowledge)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--mode', type=str, default="scan_all",
help='modalities.scan_all: Scan all Knowledge Base error data;fix_all: Fix all Knowledge Base error data;fix_one: Fixing single knowledge base error data;scan_one: Scan single KB error data')
# Maximum number of concurrency for a single process
parser.add_argument('--id', type=int, default=0, help='The knowledge base uponID, parameter is required if operating a single knowledge base')
args = parser.parse_args()
if args.mode == "scan_all":
scan_knowledge_error_data()
elif args.mode == "fix_all":
fix_knowledge_error_data()
elif args.mode == "fix_one":
fix_one_knowledge(None, args.id)
elif args.mode == "scan_one":
tmp_rows = scan_one_knowledge(None, args.id)
_save_knowledge_error_data(tmp_rows, f"{args.id}_knowledge_error_data.xlsx")
else:
print("modeParameter error, can only bescan_all、fix_all、fix_one、scan_oneOne of them,")
================================================
FILE: src/backend/bisheng/script/mid_table.sh
================================================
export PYTHONPATH="./"
python bisheng/script/sync_increment_table.py
================================================
FILE: src/backend/bisheng/script/sync_increment_table.py
================================================
from datetime import datetime
from bisheng.api.services.workflow import WorkFlowService
from bisheng.knowledge.domain.services.knowledge_service import KnowledgeService
from bisheng.user.domain.services.user import UserService
from bisheng.worker import sync_mid_user_interact_dtl
from bisheng.worker.telemetry.mid_table import sync_mid_user_increment, sync_mid_knowledge_increment, \
sync_mid_app_increment
def sync_user_increment_table_all():
"""
Sync user increment table
"""
first_user = UserService.get_first_user()
if not first_user:
print("No users found, skipping user increment table sync.")
return
start_date = first_user.create_time.isoformat()
end_date = datetime.now().isoformat()
sync_mid_user_increment(start_date, end_date)
def sync_knowledge_increment_table_all():
"""
Sync knowledge increment table
"""
first_knowledge = KnowledgeService.get_first_knowledge()
if not first_knowledge:
print("No knowledge entries found, skipping knowledge increment table sync.")
return
start_date = first_knowledge.create_time.isoformat()
end_date = datetime.now().isoformat()
sync_mid_knowledge_increment(start_date, end_date)
def sync_app_increment_table_all():
"""
Sync all increment tables
"""
first_app = WorkFlowService.get_first_app()
if not first_app:
print("No apps found, skipping app increment table sync.")
return
start_date = first_app['create_time'].isoformat()
end_date = datetime.now().isoformat()
sync_mid_app_increment(start_date, end_date)
def sync_user_interact_dtl_all():
first_date = datetime(2025, 12, 1).isoformat()
end_date = datetime.now().isoformat()
sync_mid_user_interact_dtl(first_date, end_date)
if __name__ == '__main__':
sync_user_increment_table_all()
sync_knowledge_increment_table_all()
sync_app_increment_table_all()
sync_user_interact_dtl_all()
================================================
FILE: src/backend/bisheng/script/telemetry_events_reindex.sh
================================================
#!/bin/bash
export PYTHONPATH="./"
echo "Reindexing telemetry events..."
python bisheng/script/base_telemetry_events_reindex.py
echo "Reindexing completed."
================================================
FILE: src/backend/bisheng/server.py
================================================
from gunicorn.app.base import BaseApplication # type: ignore
class BishengApplication(BaseApplication):
def __init__(self, app, options=None):
self.options = options or {}
self.application = app
super().__init__()
def load_config(self):
config = {
key: value
for key, value in self.options.items()
if key in self.cfg.settings and value is not None
}
for key, value in config.items():
self.cfg.set(key.lower(), value)
def load(self):
return self.application
================================================
FILE: src/backend/bisheng/services/__init__.py
================================================
from .manager import service_manager
from .schema import ServiceType
__all__ = ['service_manager', 'ServiceType']
================================================
FILE: src/backend/bisheng/services/auth/__init__.py
================================================
================================================
FILE: src/backend/bisheng/services/auth/factory.py
================================================
from bisheng.services.auth.service import AuthService
from bisheng.services.factory import ServiceFactory
class AuthServiceFactory(ServiceFactory):
name = 'auth_service'
def __init__(self):
super().__init__(AuthService)
def create(self, settings_service):
return AuthService(settings_service)
================================================
FILE: src/backend/bisheng/services/auth/service.py
================================================
from typing import TYPE_CHECKING
from bisheng.services.base import Service
if TYPE_CHECKING:
from bisheng.services.settings.service import SettingsService
class AuthService(Service):
name = 'auth_service'
def __init__(self, settings_service: 'SettingsService'):
self.settings_service = settings_service
================================================
FILE: src/backend/bisheng/services/auth/utils.py
================================================
from datetime import datetime, timedelta, timezone
from typing import Annotated, Coroutine, Optional, Union
from uuid import UUID
from bisheng.services.database.models.api_key.crud import check_key
from bisheng.services.database.models.api_key.model import ApiKey
from bisheng.services.database.models.user.crud import (get_user_by_id, get_user_by_username,
update_user_last_login_at)
from bisheng.services.database.models.user.model import User
from bisheng.services.deps import get_session, get_settings_service
from cryptography.fernet import Fernet
from fastapi import Depends, HTTPException, Security, status
from fastapi.security import APIKeyHeader, APIKeyQuery, OAuth2PasswordBearer
from jose import JWTError, jwt
from sqlmodel import Session
oauth2_login = OAuth2PasswordBearer(tokenUrl='api/v1/login', auto_error=False)
API_KEY_NAME = 'x-api-key'
api_key_query = APIKeyQuery(name=API_KEY_NAME, scheme_name='API key query', auto_error=False)
api_key_header = APIKeyHeader(name=API_KEY_NAME, scheme_name='API key header', auto_error=False)
# Source: https://github.com/mrtolkien/fastapi_simple_security/blob/master/fastapi_simple_security/security_api_key.py
async def api_key_security(
query_param: str = Security(api_key_query),
header_param: str = Security(api_key_header),
db: Session = Depends(get_session),
) -> Optional[User]:
settings_service = get_settings_service()
result: Optional[Union[ApiKey, User]] = None
if settings_service.auth_settings.AUTO_LOGIN:
# Get the first user
if not settings_service.auth_settings.SUPERUSER:
raise HTTPException(
status_code=status.HTTP_400_BAD_REQUEST,
detail='Missing first superuser credentials',
)
result = get_user_by_username(db, settings_service.auth_settings.SUPERUSER)
elif not query_param and not header_param:
raise HTTPException(
status_code=status.HTTP_403_FORBIDDEN,
detail='An API key must be passed as query or header',
)
elif query_param:
result = check_key(db, query_param)
else:
result = check_key(db, header_param)
if not result:
raise HTTPException(
status_code=status.HTTP_403_FORBIDDEN,
detail='Invalid or missing API key',
)
if isinstance(result, ApiKey):
return result.user
elif isinstance(result, User):
return result
async def get_current_user(
token: str = Security(oauth2_login),
query_param: str = Security(api_key_query),
header_param: str = Security(api_key_header),
db: Session = Depends(get_session),
) -> User:
if token:
return await get_current_user_by_jwt(token, db)
else:
if not query_param and not header_param:
raise HTTPException(
status_code=status.HTTP_403_FORBIDDEN,
detail='An API key as query or header, or a JWT token must be passed',
)
user = await api_key_security(query_param, header_param, db)
if user:
return user
raise HTTPException(
status_code=status.HTTP_403_FORBIDDEN,
detail='Invalid or missing API key',
)
async def get_current_user_by_jwt(
token: Annotated[str, Depends(oauth2_login)],
db: Session = Depends(get_session),
) -> User:
settings_service = get_settings_service()
credentials_exception = HTTPException(
status_code=status.HTTP_401_UNAUTHORIZED,
detail='Could not validate credentials',
headers={'WWW-Authenticate': 'Bearer'},
)
if isinstance(token, Coroutine):
token = await token
if settings_service.auth_settings.SECRET_KEY is None:
raise credentials_exception
try:
payload = jwt.decode(
token,
settings_service.auth_settings.SECRET_KEY,
algorithms=[settings_service.auth_settings.ALGORITHM],
)
user_id: UUID = payload.get('sub') # type: ignore
token_type: str = payload.get('type') # type: ignore
if expires := payload.get('exp', None):
expires_datetime = datetime.fromtimestamp(expires, timezone.utc)
# TypeError: can't compare offset-naive and offset-aware datetimes
if datetime.now(timezone.utc) > expires_datetime:
raise credentials_exception
if user_id is None or token_type:
raise credentials_exception
except JWTError as e:
raise credentials_exception from e
user = get_user_by_id(db, user_id) # type: ignore
if user is None or not user.is_active:
raise credentials_exception
return user
def get_current_active_user(current_user: Annotated[User, Depends(get_current_user)]):
if not current_user.is_active:
raise HTTPException(status_code=400, detail='Inactive user')
return current_user
def get_current_active_superuser(current_user: Annotated[User, Depends(get_current_user)]) -> User:
if not current_user.is_active:
raise HTTPException(status_code=401, detail='Inactive user')
if not current_user.is_superuser:
raise HTTPException(status_code=400, detail="The user doesn't have enough privileges")
return current_user
def verify_password(plain_password, hashed_password):
settings_service = get_settings_service()
return settings_service.auth_settings.pwd_context.verify(plain_password, hashed_password)
def get_password_hash(password):
settings_service = get_settings_service()
return settings_service.auth_settings.pwd_context.hash(password)
def create_token(data: dict, expires_delta: timedelta):
settings_service = get_settings_service()
to_encode = data.copy()
expire = datetime.now(timezone.utc) + expires_delta
to_encode['exp'] = expire
return jwt.encode(
to_encode,
settings_service.auth_settings.SECRET_KEY,
algorithm=settings_service.auth_settings.ALGORITHM,
)
def create_super_user(
username: str,
password: str,
db: Session = Depends(get_session),
) -> User:
super_user = get_user_by_username(db, username)
if not super_user:
super_user = User(
username=username,
password=get_password_hash(password),
is_superuser=True,
is_active=True,
last_login_at=None,
)
db.add(super_user)
db.commit()
db.refresh(super_user)
return super_user
def create_user_longterm_token(db: Session = Depends(get_session)) -> dict:
settings_service = get_settings_service()
username = settings_service.auth_settings.SUPERUSER
password = settings_service.auth_settings.SUPERUSER_PASSWORD
if not username or not password:
raise HTTPException(
status_code=status.HTTP_400_BAD_REQUEST,
detail='Missing first superuser credentials',
)
super_user = create_super_user(db=db, username=username, password=password)
access_token_expires_longterm = timedelta(days=365)
access_token = create_token(
data={'sub': str(super_user.id)},
expires_delta=access_token_expires_longterm,
)
# Update: last_login_at
update_user_last_login_at(super_user.id, db)
return {
'access_token': access_token,
'refresh_token': None,
'token_type': 'bearer',
}
def create_user_api_key(user_id: UUID) -> dict:
access_token = create_token(
data={'sub': str(user_id), 'role': 'api_key'},
expires_delta=timedelta(days=365 * 2),
)
return {'api_key': access_token}
def get_user_id_from_token(token: str) -> UUID:
try:
user_id = jwt.get_unverified_claims(token)['sub']
return UUID(user_id)
except (KeyError, JWTError, ValueError):
return UUID(int=0)
def create_user_tokens(user_id: UUID, db: Session = Depends(get_session), update_last_login: bool = False) -> dict:
settings_service = get_settings_service()
access_token_expires = timedelta(minutes=settings_service.auth_settings.ACCESS_TOKEN_EXPIRE_MINUTES)
access_token = create_token(
data={'sub': str(user_id)},
expires_delta=access_token_expires,
)
refresh_token_expires = timedelta(minutes=settings_service.auth_settings.REFRESH_TOKEN_EXPIRE_MINUTES)
refresh_token = create_token(
data={'sub': str(user_id), 'type': 'rf'},
expires_delta=refresh_token_expires,
)
# Update: last_login_at
if update_last_login:
update_user_last_login_at(user_id, db)
return {
'access_token': access_token,
'refresh_token': refresh_token,
'token_type': 'bearer',
}
def create_refresh_token(refresh_token: str, db: Session = Depends(get_session)):
settings_service = get_settings_service()
try:
payload = jwt.decode(
refresh_token,
settings_service.auth_settings.SECRET_KEY,
algorithms=[settings_service.auth_settings.ALGORITHM],
)
user_id: UUID = payload.get('sub') # type: ignore
token_type: str = payload.get('type') # type: ignore
if user_id is None or token_type is None:
raise HTTPException(status_code=status.HTTP_401_UNAUTHORIZED, detail='Invalid refresh token')
return create_user_tokens(user_id, db)
except JWTError as e:
raise HTTPException(
status_code=status.HTTP_401_UNAUTHORIZED,
detail='Invalid refresh token',
) from e
def authenticate_user(username: str, password: str, db: Session = Depends(get_session)) -> Optional[User]:
user = get_user_by_username(db, username)
if not user:
return None
if not user.is_active:
if not user.last_login_at:
raise HTTPException(status_code=400, detail='Waiting for approval')
raise HTTPException(status_code=400, detail='Inactive user')
return user if verify_password(password, user.password) else None
def add_padding(s):
# Calculate the number of padding characters needed
padding_needed = 4 - len(s) % 4
return s + '=' * padding_needed
def get_fernet(settings_service=Depends(get_settings_service)):
SECRET_KEY = settings_service.auth_settings.SECRET_KEY
# It's important that your secret key is 32 url-safe base64-encoded bytes
padded_secret_key = add_padding(SECRET_KEY)
fernet = Fernet(padded_secret_key)
return fernet
def encrypt_api_key(api_key: str, settings_service=Depends(get_settings_service)):
fernet = get_fernet(settings_service)
# Two-way encryption
encrypted_key = fernet.encrypt(api_key.encode())
return encrypted_key
def decrypt_api_key(encrypted_api_key: str, settings_service=Depends(get_settings_service)):
fernet = get_fernet(settings_service)
# Two-way decryption
if isinstance(encrypted_api_key, str):
encoded_bytes = encrypted_api_key.encode()
else:
encoded_bytes = encrypted_api_key
decrypted_key = fernet.decrypt(encoded_bytes).decode()
return decrypted_key
================================================
FILE: src/backend/bisheng/services/base.py
================================================
from abc import ABC
class Service(ABC):
name: str
ready: bool = False
def teardown(self):
pass
def set_ready(self):
self.ready = True
================================================
FILE: src/backend/bisheng/services/cache/__init__.py
================================================
from langflow.services.cache.service import InMemoryCache
from . import factory, service
__all__ = [
'factory',
'service',
'InMemoryCache',
]
================================================
FILE: src/backend/bisheng/services/cache/base.py
================================================
import abc
from bisheng.services.base import Service
class BaseCacheService(Service):
"""
Abstract base class for a cache.
"""
name = 'cache_service'
@abc.abstractmethod
def get(self, key):
"""
Retrieve an item from the cache.
Args:
key: The key of the item to retrieve.
Returns:
The value associated with the key, or None if the key is not found.
"""
@abc.abstractmethod
def set(self, key, value):
"""
Add an item to the cache.
Args:
key: The key of the item.
value: The value to cache.
"""
@abc.abstractmethod
def upsert(self, key, value):
"""
Add an item to the cache if it doesn't exist, or update it if it does.
Args:
key: The key of the item.
value: The value to cache.
"""
@abc.abstractmethod
def delete(self, key):
"""
Remove an item from the cache.
Args:
key: The key of the item to remove.
"""
@abc.abstractmethod
def clear(self):
"""
Clear all items from the cache.
"""
@abc.abstractmethod
def __contains__(self, key):
"""
Check if the key is in the cache.
Args:
key: The key of the item to check.
Returns:
True if the key is in the cache, False otherwise.
"""
@abc.abstractmethod
def __getitem__(self, key):
"""
Retrieve an item from the cache using the square bracket notation.
Args:
key: The key of the item to retrieve.
"""
@abc.abstractmethod
def __setitem__(self, key, value):
"""
Add an item to the cache using the square bracket notation.
Args:
key: The key of the item.
value: The value to cache.
"""
@abc.abstractmethod
def __delitem__(self, key):
"""
Remove an item from the cache using the square bracket notation.
Args:
key: The key of the item to remove.
"""
================================================
FILE: src/backend/bisheng/services/cache/factory.py
================================================
from typing import TYPE_CHECKING
from bisheng.services.cache.service import BaseCacheService, InMemoryCache, RedisCache
from bisheng.services.factory import ServiceFactory
from loguru import logger
if TYPE_CHECKING:
from bisheng.services.settings.service import SettingsService
class CacheServiceFactory(ServiceFactory):
def __init__(self):
super().__init__(BaseCacheService)
def create(self, settings_service: 'SettingsService'):
# Here you would have logic to create and configure a CacheService
# based on the settings_service
if settings_service.settings.CACHE_TYPE == 'redis':
logger.debug('Creating Redis cache')
redis_cache = RedisCache(
host=settings_service.settings.REDIS_HOST,
port=settings_service.settings.REDIS_PORT,
db=settings_service.settings.REDIS_DB,
expiration_time=settings_service.settings.REDIS_CACHE_EXPIRE,
)
if redis_cache.is_connected():
logger.debug('Redis cache is connected')
return redis_cache
logger.warning('Redis cache is not connected, falling back to in-memory cache')
return InMemoryCache()
elif settings_service.settings.CACHE_TYPE == 'memory':
return InMemoryCache()
================================================
FILE: src/backend/bisheng/services/cache/service.py
================================================
import pickle
import threading
import time
from collections import OrderedDict
from bisheng.services.base import Service
from bisheng.services.cache.base import BaseCacheService
from loguru import logger
class InMemoryCache(BaseCacheService, Service):
"""
A simple in-memory cache using an OrderedDict.
This cache supports setting a maximum size and expiration time for cached items.
When the cache is full, it uses a Least Recently Used (LRU) eviction policy.
Thread-safe using a threading Lock.
Attributes:
max_size (int, optional): Maximum number of items to store in the cache.
expiration_time (int, optional): Time in seconds after which a cached item expires. Default is 1 hour.
Example:
cache = InMemoryCache(max_size=3, expiration_time=5)
# setting cache values
cache.set("a", 1)
cache.set("b", 2)
cache["c"] = 3
# getting cache values
a = cache.get("a")
b = cache["b"]
"""
def __init__(self, max_size=None, expiration_time=60 * 60):
"""
Initialize a new InMemoryCache instance.
Args:
max_size (int, optional): Maximum number of items to store in the cache.
expiration_time (int, optional): Time in seconds after which a cached item expires. Default is 1 hour.
"""
self._cache = OrderedDict()
self._lock = threading.RLock()
self.max_size = max_size
self.expiration_time = expiration_time
def get(self, key):
"""
Retrieve an item from the cache.
Args:
key: The key of the item to retrieve.
Returns:
The value associated with the key, or None if the key is not found or the item has expired.
"""
with self._lock:
return self._get_without_lock(key)
def _get_without_lock(self, key):
"""
Retrieve an item from the cache without acquiring the lock.
"""
if item := self._cache.get(key):
if self.expiration_time is None or time.time() - item['time'] < self.expiration_time:
# Move the key to the end to make it recently used
self._cache.move_to_end(key)
# Check if the value is pickled
if isinstance(item['value'], bytes):
value = pickle.loads(item['value'])
else:
value = item['value']
return value
else:
self.delete(key)
return None
def set(self, key, value, pickle=False):
"""
Add an item to the cache.
If the cache is full, the least recently used item is evicted.
Args:
key: The key of the item.
value: The value to cache.
"""
with self._lock:
if key in self._cache:
# Remove existing key before re-inserting to update order
self.delete(key)
elif self.max_size and len(self._cache) >= self.max_size:
# Remove least recently used item
self._cache.popitem(last=False)
# pickle locally to mimic Redis
if pickle:
value = pickle.dumps(value)
self._cache[key] = {'value': value, 'time': time.time()}
def upsert(self, key, value):
"""
Inserts or updates a value in the cache.
If the existing value and the new value are both dictionaries, they are merged.
Args:
key: The key of the item.
value: The value to insert or update.
"""
with self._lock:
existing_value = self._get_without_lock(key)
if existing_value is not None and isinstance(existing_value, dict) and isinstance(value, dict):
existing_value.update(value)
value = existing_value
self.set(key, value)
def get_or_set(self, key, value):
"""
Retrieve an item from the cache. If the item does not exist,
set it with the provided value.
Args:
key: The key of the item.
value: The value to cache if the item doesn't exist.
Returns:
The cached value associated with the key.
"""
with self._lock:
if key in self._cache:
return self.get(key)
self.set(key, value)
return value
def delete(self, key):
"""
Remove an item from the cache.
Args:
key: The key of the item to remove.
"""
with self._lock:
self._cache.pop(key, None)
def clear(self):
"""
Clear all items from the cache.
"""
with self._lock:
self._cache.clear()
def __contains__(self, key):
"""Check if the key is in the cache."""
return key in self._cache
def __getitem__(self, key):
"""Retrieve an item from the cache using the square bracket notation."""
return self.get(key)
def __setitem__(self, key, value):
"""Add an item to the cache using the square bracket notation."""
self.set(key, value)
def __delitem__(self, key):
"""Remove an item from the cache using the square bracket notation."""
self.delete(key)
def __len__(self):
"""Return the number of items in the cache."""
return len(self._cache)
def __repr__(self):
"""Return a string representation of the InMemoryCache instance."""
return f'InMemoryCache(max_size={self.max_size}, expiration_time={self.expiration_time})'
class RedisCache(BaseCacheService, Service):
"""
A Redis-based cache implementation.
This cache supports setting an expiration time for cached items.
Attributes:
expiration_time (int, optional): Time in seconds after which a cached item expires. Default is 1 hour.
Example:
cache = RedisCache(expiration_time=5)
# setting cache values
cache.set("a", 1)
cache.set("b", 2)
cache["c"] = 3
# getting cache values
a = cache.get("a")
b = cache["b"]
"""
def __init__(self, host='localhost', port=6379, db=0, expiration_time=60 * 60):
"""
Initialize a new RedisCache instance.
Args:
host (str, optional): Redis host.
port (int, optional): Redis port.
db (int, optional): Redis DB.
expiration_time (int, optional): Time in seconds after which a
ached item expires. Default is 1 hour.
"""
try:
import redis
except ImportError as exc:
raise ImportError(
'RedisCache requires the redis-py package.'
' Please install bisheng with the deploy extra: pip install bisheng[deploy]'
) from exc
logger.warning(
'RedisCache is an experimental feature and may not work as expected.'
' Please report any issues to our GitHub repository.'
)
self._client = redis.StrictRedis(host=host, port=port, db=db)
self.expiration_time = expiration_time
# check connection
def is_connected(self):
"""
Check if the Redis client is connected.
"""
import redis
try:
self._client.ping()
return True
except redis.exceptions.ConnectionError:
return False
def get(self, key):
"""
Retrieve an item from the cache.
Args:
key: The key of the item to retrieve.
Returns:
The value associated with the key, or None if the key is not found.
"""
value = self._client.get(key)
return pickle.loads(value) if value else None
def set(self, key, value):
"""
Add an item to the cache.
Args:
key: The key of the item.
value: The value to cache.
"""
try:
if pickled := pickle.dumps(value):
result = self._client.setex(key, self.expiration_time, pickled)
if not result:
raise ValueError('RedisCache could not set the value.')
except TypeError as exc:
raise TypeError('RedisCache only accepts values that can be pickled. ') from exc
def upsert(self, key, value):
"""
Inserts or updates a value in the cache.
If the existing value and the new value are both dictionaries, they are merged.
Args:
key: The key of the item.
value: The value to insert or update.
"""
existing_value = self.get(key)
if existing_value is not None and isinstance(existing_value, dict) and isinstance(value, dict):
existing_value.update(value)
value = existing_value
self.set(key, value)
def delete(self, key):
"""
Remove an item from the cache.
Args:
key: The key of the item to remove.
"""
self._client.delete(key)
def clear(self):
"""
Clear all items from the cache.
"""
self._client.flushdb()
def __contains__(self, key):
"""Check if the key is in the cache."""
return False if key is None else self._client.exists(key)
def __getitem__(self, key):
"""Retrieve an item from the cache using the square bracket notation."""
return self.get(key)
def __setitem__(self, key, value):
"""Add an item to the cache using the square bracket notation."""
self.set(key, value)
def __delitem__(self, key):
"""Remove an item from the cache using the square bracket notation."""
self.delete(key)
def __repr__(self):
"""Return a string representation of the RedisCache instance."""
return f'RedisCache(expiration_time={self.expiration_time})'
================================================
FILE: src/backend/bisheng/services/chat/__init__.py
================================================
================================================
FILE: src/backend/bisheng/services/chat/cache.py
================================================
from contextlib import contextmanager
from typing import Any, Awaitable, Callable, List, Optional
import pandas as pd
from bisheng.services.base import Service
from PIL import Image
class Subject:
"""Base class for implementing the observer pattern."""
def __init__(self):
self.observers: List[Callable[[], None]] = []
def attach(self, observer: Callable[[], None]):
"""Attach an observer to the subject."""
self.observers.append(observer)
def detach(self, observer: Callable[[], None]):
"""Detach an observer from the subject."""
self.observers.remove(observer)
def notify(self):
"""Notify all observers about an event."""
for observer in self.observers:
if observer is None:
continue
observer()
class AsyncSubject:
"""Base class for implementing the async observer pattern."""
def __init__(self):
self.observers: List[Callable[[], Awaitable]] = []
def attach(self, observer: Callable[[], Awaitable]):
"""Attach an observer to the subject."""
self.observers.append(observer)
def detach(self, observer: Callable[[], Awaitable]):
"""Detach an observer from the subject."""
self.observers.remove(observer)
async def notify(self):
"""Notify all observers about an event."""
for observer in self.observers:
if observer is None:
continue
await observer()
class CacheService(Subject, Service):
"""Manages cache for different clients and notifies observers on changes."""
name = 'cache_service'
def __init__(self):
super().__init__()
self._cache = {}
self.current_client_id = None
self.current_cache = {}
@contextmanager
def set_client_id(self, client_id: str):
"""
Context manager to set the current client_id and associated cache.
Args:
client_id (str): The client identifier.
"""
previous_client_id = self.current_client_id
self.current_client_id = client_id
self.current_cache = self._cache.setdefault(client_id, {})
try:
yield
finally:
self.current_client_id = previous_client_id
self.current_cache = self._cache.get(self.current_client_id, {})
def add(self, name: str, obj: Any, obj_type: str, extension: Optional[str] = None):
"""
Add an object to the current client's cache.
Args:
name (str): The cache key.
obj (Any): The object to cache.
obj_type (str): The type of the object.
"""
object_extensions = {
'image': 'png',
'pandas': 'csv',
}
if obj_type in object_extensions:
_extension = object_extensions[obj_type]
else:
_extension = type(obj).__name__.lower()
self.current_cache[name] = {
'obj': obj,
'type': obj_type,
'extension': extension or _extension,
}
self.notify()
def add_pandas(self, name: str, obj: Any):
"""
Add a pandas DataFrame or Series to the current client's cache.
Args:
name (str): The cache key.
obj (Any): The pandas DataFrame or Series object.
"""
if isinstance(obj, (pd.DataFrame, pd.Series)):
self.add(name, obj.to_csv(), 'pandas', extension='csv')
else:
raise ValueError('Object is not a pandas DataFrame or Series')
def add_image(self, name: str, obj: Any, extension: str = 'png'):
"""
Add a PIL Image to the current client's cache.
Args:
name (str): The cache key.
obj (Any): The PIL Image object.
"""
if isinstance(obj, Image.Image):
self.add(name, obj, 'image', extension=extension)
else:
raise ValueError('Object is not a PIL Image')
def get(self, name: str):
"""
Get an object from the current client's cache.
Args:
name (str): The cache key.
Returns:
The cached object associated with the given cache key.
"""
return self.current_cache[name]
def get_last(self):
"""
Get the last added item in the current client's cache.
Returns:
The last added item in the cache.
"""
return list(self.current_cache.values())[-1]
cache_service = CacheService()
================================================
FILE: src/backend/bisheng/services/chat/config.py
================================================
class ChatConfig:
streaming: bool = True
================================================
FILE: src/backend/bisheng/services/chat/factory.py
================================================
from bisheng.services.chat.service import ChatService
from bisheng.services.factory import ServiceFactory
class ChatServiceFactory(ServiceFactory):
def __init__(self):
super().__init__(ChatService)
def create(self):
# Here you would have logic to create and configure a ChatService
return ChatService()
================================================
FILE: src/backend/bisheng/services/chat/service.py
================================================
import asyncio
import uuid
from collections import defaultdict
from typing import Any, Dict, List
import orjson
from bisheng.api.v1.schemas import ChatMessage, ChatResponse, FileResponse
from bisheng.interface.utils import pil_to_base64
from bisheng.services import ServiceType, service_manager
from bisheng.services.base import Service
from bisheng.services.chat.cache import Subject
from bisheng.services.chat.utils import process_graph
from fastapi import WebSocket, status
from loguru import logger
from starlette.websockets import WebSocketState
from .cache import cache_service
class ChatHistory(Subject):
def __init__(self):
super().__init__()
self.history: Dict[str, List[ChatMessage]] = defaultdict(list)
def add_message(self, client_id: str, message: ChatMessage):
"""Add a message to the chat history."""
self.history[client_id].append(message)
if not isinstance(message, FileResponse):
self.notify()
def get_history(self, client_id: str, filter_messages=True) -> List[ChatMessage]:
"""Get the chat history for a client."""
if history := self.history.get(client_id, []):
if filter_messages:
return [msg for msg in history if msg.type not in ['start', 'stream']]
return history
else:
return []
def empty_history(self, client_id: str):
"""Empty the chat history for a client."""
self.history[client_id] = []
class ChatService(Service):
name = 'chat_service'
def __init__(self):
self.active_connections: Dict[str, WebSocket] = {}
self.connection_ids: Dict[str, str] = {}
self.chat_history = ChatHistory()
self.chat_cache = cache_service
self.chat_cache.attach(self.update)
self.cache_service = service_manager.get(ServiceType.CACHE_SERVICE)
def on_chat_history_update(self):
"""Send the last chat message to the client."""
client_id = self.chat_cache.current_client_id
if client_id in self.active_connections:
chat_response = self.chat_history.get_history(client_id, filter_messages=False)[-1]
if chat_response.is_bot:
# Process FileResponse
if isinstance(chat_response, FileResponse):
# If data_type is pandas, convert to csv
if chat_response.data_type == 'pandas':
chat_response.data = chat_response.data.to_csv()
elif chat_response.data_type == 'image':
# Base64 encode the image
chat_response.data = pil_to_base64(chat_response.data)
# get event loop
loop = asyncio.get_event_loop()
coroutine = self.send_json(client_id, chat_response)
asyncio.run_coroutine_threadsafe(coroutine, loop)
def update(self):
if self.chat_cache.current_client_id in self.active_connections:
self.last_cached_object_dict = self.chat_cache.get_last()
# Add a new ChatResponse with the data
chat_response = FileResponse(
message=None,
type='file',
data=self.last_cached_object_dict['obj'],
data_type=self.last_cached_object_dict['type'],
)
self.chat_history.add_message(self.chat_cache.current_client_id, chat_response)
async def connect(self, client_id: str, websocket: WebSocket):
self.active_connections[client_id] = websocket
# This is to avoid having multiple clients with the same id
self.connection_ids[client_id] = f'{client_id}-{uuid.uuid4()}'
def disconnect(self, client_id: str):
self.active_connections.pop(client_id, None)
self.connection_ids.pop(client_id, None)
async def send_message(self, client_id: str, message: str):
websocket = self.active_connections[client_id]
await websocket.send_text(message)
async def send_json(self, client_id: str, message: ChatMessage):
websocket = self.active_connections[client_id]
await websocket.send_json(message.model_dump())
async def close_connection(self, client_id: str, code: int, reason: str):
if websocket := self.active_connections[client_id]:
try:
await websocket.close(code=code, reason=reason)
self.disconnect(client_id)
except RuntimeError as exc:
# This is to catch the following error:
# Unexpected ASGI message 'websocket.close', after sending 'websocket.close'
if 'after sending' in str(exc):
logger.error(f'Error closing connection: {exc}')
async def process_message(self, client_id: str, payload: Dict, langchain_object: Any):
# Process the graph data and chat message
chat_inputs = payload.pop('inputs', {})
chatkey = payload.pop('chatKey', None)
chat_inputs = ChatMessage(message=chat_inputs, chatKey=chatkey)
self.chat_history.add_message(client_id, chat_inputs)
# graph_data = payload
start_resp = ChatResponse(message=None, type='start', intermediate_steps='')
await self.send_json(client_id, start_resp)
# is_first_message = len(self.chat_history.get_history(client_id=client_id)) <= 1
# Generate result and thought
try:
logger.debug('Generating result and thought')
result, intermediate_steps, raw_output = await process_graph(
langchain_object=langchain_object,
chat_inputs=chat_inputs,
client_id=client_id,
session_id=self.connection_ids[client_id],
)
self.set_cache(client_id, langchain_object)
except Exception as e:
# Log stack trace
logger.exception(e)
self.chat_history.empty_history(client_id)
raise e
# Send a response back to the frontend, if needed
intermediate_steps = intermediate_steps or ''
history = self.chat_history.get_history(client_id, filter_messages=False)
file_responses = []
if history:
# Iterate backwards through the history
for msg in reversed(history):
if isinstance(msg, FileResponse):
if msg.data_type == 'image':
# Base64 encode the image
if isinstance(msg.data, str):
continue
msg.data = pil_to_base64(msg.data)
file_responses.append(msg)
if msg.type == 'start':
break
response = ChatResponse(
message=result,
intermediate_steps=intermediate_steps.strip(),
type='end',
files=file_responses,
)
await self.send_json(client_id, response)
self.chat_history.add_message(client_id, response)
def set_cache(self, client_id: str, langchain_object: Any) -> bool:
"""
Set the cache for a client.
"""
# client_id is the flow id but that already exists in the cache
# so we need to change it to something else
result_dict = {
'result': langchain_object,
'type': type(langchain_object),
}
self.cache_service.upsert(client_id, result_dict)
return client_id in self.cache_service
async def handle_websocket(self, client_id: str, websocket: WebSocket):
await self.connect(client_id, websocket)
try:
chat_history = self.chat_history.get_history(client_id)
# iterate and make BaseModel into dict
chat_history = [chat.model_dump() for chat in chat_history]
await websocket.send_json(chat_history)
while True:
json_payload = await websocket.receive_json()
if isinstance(json_payload, str):
payload = orjson.loads(json_payload)
elif isinstance(json_payload, dict):
payload = json_payload
if 'clear_history' in payload and payload['clear_history']:
self.chat_history.history[client_id] = []
continue
with self.chat_cache.set_client_id(client_id):
if langchain_object := self.cache_service.get(client_id).get('result'):
await self.process_message(client_id, payload, langchain_object)
else:
raise RuntimeError(f'Could not find a build result for client_id {client_id}')
except Exception as exc:
# Handle any exceptions that might occur
logger.exception(f'Error handling websocket: {exc}')
if websocket.client_state == WebSocketState.CONNECTED:
await self.close_connection(
client_id=client_id,
code=status.WS_1011_INTERNAL_ERROR,
reason=str(exc),
)
elif websocket.client_state == WebSocketState.DISCONNECTED:
self.disconnect(client_id)
finally:
try:
# first check if the connection is still open
if websocket.client_state == WebSocketState.CONNECTED:
await self.close_connection(
client_id=client_id,
code=status.WS_1000_NORMAL_CLOSURE,
reason='Client disconnected',
)
except Exception as exc:
logger.error(f'Error closing connection: {exc}')
self.disconnect(client_id)
def dict_to_markdown_table(my_dict):
markdown_table = '| Key | Value |\n|---|---|\n'
for key, value in my_dict.items():
markdown_table += f'| {key} | {value} |\n'
return markdown_table
def list_of_dicts_to_markdown_table(dict_list):
if not dict_list:
return 'No data provided.'
# Extract headers from the keys of the first dictionary
headers = dict_list[0].keys()
markdown_table = '| ' + ' | '.join(headers) + ' |\n'
markdown_table += '| ' + ' | '.join('---' for _ in headers) + ' |\n'
for row_dict in dict_list:
row = [str(row_dict.get(header, '')) for header in headers]
markdown_table += '| ' + ' | '.join(row) + ' |\n'
return markdown_table
================================================
FILE: src/backend/bisheng/services/chat/utils.py
================================================
from bisheng.api.v1.schemas import ChatMessage
from bisheng.interface.utils import try_setting_streaming_options
from bisheng.processing.base import get_result_and_steps
from loguru import logger
async def process_graph(
langchain_object,
chat_inputs: ChatMessage,
client_id: str,
session_id: str,
):
langchain_object = try_setting_streaming_options(langchain_object)
logger.debug('Loaded langchain object')
if langchain_object is None:
# Raise user facing error
raise ValueError('There was an error loading the langchain_object. Please, check all the nodes and try again.')
# Generate result and thought
try:
if chat_inputs.message is None:
logger.debug('No message provided')
chat_inputs.message = {}
logger.debug('Generating result and thought')
result, intermediate_steps, raw_output = await get_result_and_steps(
langchain_object,
chat_inputs.message,
client_id=client_id,
session_id=session_id,
)
logger.debug('Generated result and intermediate_steps')
return result, intermediate_steps, raw_output
except Exception as e:
# Log stack trace
logger.exception(e)
raise e
================================================
FILE: src/backend/bisheng/services/credentials/__init__.py
================================================
================================================
FILE: src/backend/bisheng/services/credentials/factory.py
================================================
from typing import TYPE_CHECKING
from bisheng.services.credentials.service import CredentialService
from bisheng.services.factory import ServiceFactory
if TYPE_CHECKING:
from bisheng.services.settings.service import SettingsService
class CredentialServiceFactory(ServiceFactory):
def __init__(self):
super().__init__(CredentialService)
def create(self, settings_service: 'SettingsService'):
return CredentialService(settings_service)
================================================
FILE: src/backend/bisheng/services/credentials/service.py
================================================
from typing import TYPE_CHECKING, Optional, Union
from uuid import UUID
from bisheng.services.auth import utils as auth_utils
from bisheng.services.base import Service
from bisheng.services.database.models.credential.model import Credential
from bisheng.services.deps import get_session
from fastapi import Depends
from sqlmodel import Session, select
if TYPE_CHECKING:
from bisheng.services.settings.service import SettingsService
class CredentialService(Service):
name = 'credential_service'
def __init__(self, settings_service: 'SettingsService'):
self.settings_service = settings_service
def get_credential(self, user_id: Union[UUID, str], name: str, session: Session = Depends(get_session)) -> str:
# we get the credential from the database
# credential = session.query(Credential).filter(Credential.user_id == user_id, Credential.name == name).first()
credential = session.exec(
select(Credential).where(Credential.user_id == user_id, Credential.name == name)
).first()
# we decrypt the value
if not credential or not credential.value:
raise ValueError(f'{name} credential not found.')
decrypted = auth_utils.decrypt_api_key(credential.value, settings_service=self.settings_service)
return decrypted
def list_credentials(
self, user_id: Union[UUID, str], session: Session = Depends(get_session)
) -> list[Optional[str]]:
credentials = session.exec(select(Credential).where(Credential.user_id == user_id)).all()
return [credential.name for credential in credentials]
================================================
FILE: src/backend/bisheng/services/deps.py
================================================
from bisheng.services import ServiceType, service_manager
from bisheng.services.session.service import SessionService
from bisheng.services.task.service import TaskService
# from sqlmodel import Session
# def get_credential_service() -> "CredentialService":
# return service_manager.get(ServiceType.CREDENTIAL_SERVICE) # type: ignore
# def get_plugins_service() -> "PluginService":
# return service_manager.get(ServiceType.PLUGIN_SERVICE) # type: ignore
# def get_settings_service() -> "SettingsService":
# try:
# # type: ignore
# return service_manager.get(ServiceType.SETTINGS_SERVICE)
# except ValueError:
# # initialize settings service
# from langflow.services.manager import initialize_settings_service
# initialize_settings_service()
# # type: ignore
# return service_manager.get(ServiceType.SETTINGS_SERVICE)
# def get_db_service() -> "DatabaseService":
# return service_manager.get(ServiceType.DATABASE_SERVICE) # type: ignore
# def get_session() -> Generator["Session", None, None]:
# db_service = get_db_service()
# yield from db_service.get_session()
# def get_cache_service() -> "BaseCacheService":
# return service_manager.get(ServiceType.CACHE_SERVICE) # type: ignore
def get_session_service() -> SessionService:
return service_manager.get(ServiceType.SESSION_SERVICE) # type: ignore
def get_task_service() -> TaskService:
return service_manager.get(ServiceType.TASK_SERVICE) # type: ignore
# def get_chat_service() -> "ChatService":
# return service_manager.get(ServiceType.CHAT_SERVICE) # type: ignore
# def get_store_service() -> "StoreService":
# return service_manager.get(ServiceType.STORE_SERVICE) # type: ignore
================================================
FILE: src/backend/bisheng/services/factory.py
================================================
from typing import TYPE_CHECKING
if TYPE_CHECKING:
from bisheng.services.base import Service
class ServiceFactory:
def __init__(self, service_class):
self.service_class = service_class
def create(self, *args, **kwargs) -> 'Service':
raise NotImplementedError
================================================
FILE: src/backend/bisheng/services/manager.py
================================================
from typing import TYPE_CHECKING, Dict, List, Optional
from bisheng.services.schema import ServiceType
from loguru import logger
if TYPE_CHECKING:
from bisheng.services.base import Service
from bisheng.services.factory import ServiceFactory
class ServiceManager:
"""
Manages the creation of different services.
"""
def __init__(self):
self.services: Dict[str, 'Service'] = {}
self.factories = {}
self.dependencies = {}
def register_factory(
self,
service_factory: 'ServiceFactory',
dependencies: Optional[List[ServiceType]] = None,
):
"""
Registers a new factory with dependencies.
"""
if dependencies is None:
dependencies = []
service_name = service_factory.service_class.name
self.factories[service_name] = service_factory
self.dependencies[service_name] = dependencies
def get(self, service_name: ServiceType) -> 'Service':
"""
Get (or create) a service by its name.
"""
if service_name not in self.services:
self._create_service(service_name)
return self.services[service_name]
def _create_service(self, service_name: ServiceType):
"""
Create a new service given its name, handling dependencies.
"""
logger.debug(f'Create service {service_name}')
self._validate_service_creation(service_name)
# Create dependencies first
for dependency in self.dependencies.get(service_name, []):
if dependency not in self.services:
self._create_service(dependency)
# Collect the dependent services
dependent_services = {
dep.value: self.services[dep]
for dep in self.dependencies.get(service_name, [])
}
# Create the actual service
self.services[service_name] = self.factories[service_name].create(**dependent_services)
self.services[service_name].set_ready()
def _validate_service_creation(self, service_name: ServiceType):
"""
Validate whether the service can be created.
"""
if service_name not in self.factories:
raise ValueError(f"No factory registered for the service class '{service_name.name}'")
def update(self, service_name: ServiceType):
"""
Update a service by its name.
"""
if service_name in self.services:
logger.debug(f'Update service {service_name}')
self.services.pop(service_name, None)
self.get(service_name)
def teardown(self):
"""
Teardown all the services.
"""
for service in self.services.values():
if service is None:
continue
logger.debug(f'Teardown service {service.name}')
try:
service.teardown()
except Exception as exc:
logger.exception(exc)
self.services = {}
self.factories = {}
self.dependencies = {}
service_manager = ServiceManager()
def reinitialize_services():
"""
Reinitialize all the services needed.
"""
# service_manager.update(ServiceType.SETTINGS_SERVICE)
# service_manager.update(ServiceType.DATABASE_SERVICE)
# service_manager.update(ServiceType.CACHE_SERVICE)
# service_manager.update(ServiceType.CHAT_SERVICE)
service_manager.update(ServiceType.SESSION_SERVICE)
# service_manager.update(ServiceType.AUTH_SERVICE)
service_manager.update(ServiceType.TASK_SERVICE)
# # Test cache connection
# service_manager.get(ServiceType.CACHE_SERVICE)
# # Test database connection
# service_manager.get(ServiceType.DATABASE_SERVICE)
# # Test cache connection
# service_manager.get(ServiceType.CACHE_SERVICE)
# # Test database connection
# service_manager.get(ServiceType.DATABASE_SERVICE)
# def initialize_settings_service():
# """
# Initialize the settings manager.
# """
# from bisheng.services.settings import factory as settings_factory
# service_manager.register_factory(settings_factory.SettingsServiceFactory())
def initialize_session_service():
"""
Initialize the session manager.
"""
# from bisheng.services.cache import factory as cache_factory
from bisheng.services.session import factory as session_service_factory # type: ignore
# initialize_settings_service()
# service_manager.register_factory(cache_factory.CacheServiceFactory(),
# dependencies=[ServiceType.SETTINGS_SERVICE])
service_manager.register_factory(session_service_factory.SessionServiceFactory(), []
# dependencies=[ServiceType.CACHE_SERVICE],
)
================================================
FILE: src/backend/bisheng/services/plugins/__init__.py
================================================
================================================
FILE: src/backend/bisheng/services/plugins/base.py
================================================
from typing import Any
class BasePlugin:
def initialize(self):
pass
def teardown(self):
pass
def get(self) -> Any:
pass
class CallbackPlugin(BasePlugin):
def get_callback(self, _id=None):
pass
================================================
FILE: src/backend/bisheng/services/plugins/factory.py
================================================
from typing import TYPE_CHECKING
from bisheng.services.factory import ServiceFactory
from bisheng.services.plugins.service import PluginService
if TYPE_CHECKING:
from bisheng.services.settings.service import SettingsService
class PluginServiceFactory(ServiceFactory):
def __init__(self):
super().__init__(PluginService)
def create(self, settings_service: 'SettingsService'):
service = PluginService(settings_service)
return service
================================================
FILE: src/backend/bisheng/services/plugins/langfuse_plugin.py
================================================
from typing import Optional
from bisheng.services.deps import get_settings_service
from bisheng.services.plugins.base import CallbackPlugin
from loguru import logger
class LangfuseInstance:
_instance = None
@classmethod
def get(cls):
logger.debug('Getting Langfuse instance')
if cls._instance is None:
cls.create()
return cls._instance
@classmethod
def create(cls):
try:
logger.debug('Creating Langfuse instance')
from langfuse import Langfuse # type: ignore
settings_manager = get_settings_service()
if settings_manager.settings.LANGFUSE_PUBLIC_KEY and settings_manager.settings.LANGFUSE_SECRET_KEY:
logger.debug('Langfuse credentials found')
cls._instance = Langfuse(
public_key=settings_manager.settings.LANGFUSE_PUBLIC_KEY,
secret_key=settings_manager.settings.LANGFUSE_SECRET_KEY,
host=settings_manager.settings.LANGFUSE_HOST,
)
else:
logger.debug('No Langfuse credentials found')
cls._instance = None
except ImportError:
logger.debug('Langfuse not installed')
cls._instance = None
@classmethod
def update(cls):
logger.debug('Updating Langfuse instance')
cls._instance = None
cls.create()
@classmethod
def teardown(cls):
logger.debug('Tearing down Langfuse instance')
if cls._instance is not None:
cls._instance.flush()
cls._instance = None
class LangfusePlugin(CallbackPlugin):
def initialize(self):
LangfuseInstance.create()
def teardown(self):
LangfuseInstance.teardown()
def get(self):
return LangfuseInstance.get()
def get_callback(self, _id: Optional[str] = None):
if _id is None:
_id = 'default'
from langfuse.callback import CreateTrace # type: ignore
logger.debug('Initializing langfuse callback')
try:
langfuse_instance = self.get()
if langfuse_instance is not None and hasattr(langfuse_instance, 'trace'):
trace = langfuse_instance.trace(CreateTrace(name='bisheng-' + _id, id=_id))
if trace:
return trace.getNewHandler()
except Exception as exc:
logger.error(f'Error initializing langfuse callback: {exc}')
return None
================================================
FILE: src/backend/bisheng/services/plugins/service.py
================================================
import importlib
import inspect
import os
from typing import TYPE_CHECKING, Union
from bisheng.services.base import Service
from bisheng.services.plugins.base import BasePlugin, CallbackPlugin
from loguru import logger
if TYPE_CHECKING:
from bisheng.services.settings.service import SettingsService
class PluginService(Service):
name = 'plugin_service'
def __init__(self, settings_service: 'SettingsService'):
self.plugins: dict[str, BasePlugin] = {}
# plugin_dir = settings_service.settings.PLUGIN_DIR
self.plugin_dir = os.path.dirname(__file__)
self.plugins_base_module = 'bisheng.services.plugins'
self.load_plugins()
def load_plugins(self):
base_files = ['base.py', 'service.py', 'factory.py', '__init__.py']
for module in os.listdir(self.plugin_dir):
if module.endswith('.py') and module not in base_files:
plugin_name = module[:-3]
module_path = f'{self.plugins_base_module}.{plugin_name}'
try:
mod = importlib.import_module(module_path)
for attr_name in dir(mod):
attr = getattr(mod, attr_name)
if (
inspect.isclass(attr)
and issubclass(attr, BasePlugin)
and attr not in [CallbackPlugin, BasePlugin]
):
self.register_plugin(plugin_name, attr())
except Exception as exc:
logger.error(f'Error loading plugin {plugin_name}: {exc}')
def register_plugin(self, plugin_name, plugin_instance):
self.plugins[plugin_name] = plugin_instance
plugin_instance.initialize()
def get_plugin(self, plugin_name) -> Union[BasePlugin, None]:
return self.plugins.get(plugin_name)
def get(self, plugin_name):
if plugin := self.get_plugin(plugin_name):
return plugin.get()
return None
def teardown(self):
for plugin in self.plugins.values():
plugin.teardown()
def get_callbacks(self, _id=None):
callbacks = []
for plugin in self.plugins.values():
if isinstance(plugin, CallbackPlugin):
callback = plugin.get_callback(_id=_id)
if callback:
callbacks.append(callback)
return callbacks
================================================
FILE: src/backend/bisheng/services/schema.py
================================================
from enum import Enum
class ServiceType(str, Enum):
"""
Enum for the different types of services that can be
registered with the service manager.
"""
# AUTH_SERVICE = "auth_service"
# CACHE_SERVICE = "cache_service"
# SETTINGS_SERVICE = "settings_service"
# DATABASE_SERVICE = "database_service"
# CHAT_SERVICE = "chat_service"
SESSION_SERVICE = 'session_service'
TASK_SERVICE = 'task_service'
# PLUGIN_SERVICE = "plugin_service"
# STORE_SERVICE = "store_service"
# CREDENTIAL_SERVICE = "credential_service"
================================================
FILE: src/backend/bisheng/services/session/__init__.py
================================================
================================================
FILE: src/backend/bisheng/services/session/factory.py
================================================
from bisheng.services.factory import ServiceFactory
from bisheng.services.session.service import SessionService
# if TYPE_CHECKING:
# from bisheng.services.cache.service import BaseCacheService
class SessionServiceFactory(ServiceFactory):
def __init__(self):
super().__init__(SessionService)
def create(self):
return SessionService()
================================================
FILE: src/backend/bisheng/services/session/service.py
================================================
from bisheng.api.utils import build_flow_no_yield
from bisheng.core.cache.redis_manager import get_redis_client_sync
from bisheng.services.base import Service
from bisheng.services.session.utils import compute_dict_hash, session_id_generator
# if TYPE_CHECKING:
# from bisheng.services.cache.base import BaseCacheService
class SessionService(Service):
name = 'session_service'
def __init__(self):
self.cache_service = get_redis_client_sync()
async def load_session(self, key, data_graph, **kwargs):
# Check if the data is cached
if key in self.cache_service:
return await self.cache_service.aget(key)
if key is None:
key = self.generate_key(session_id=None, data_graph=data_graph)
# If not cached, build the graph and cache it
# graph, artifacts = await build_sorted_vertices(data_graph)
# Complete with custom initialization methodsapiAlignment with Chat
artifacts = {}
graph = await build_flow_no_yield(graph_data=data_graph, **kwargs)
await self.cache_service.aset(key, (graph, artifacts))
return graph, artifacts
def build_key(self, session_id, data_graph):
json_hash = compute_dict_hash(data_graph)
return f"{session_id}{'_' if session_id else ''}{json_hash}"
def generate_key(self, session_id, data_graph):
# Hash the JSON and combine it with the session_id to create a unique key
if session_id is None:
# generate a 5 char session_id to concatenate with the json_hash
session_id = session_id_generator()
return self.build_key(session_id, data_graph=data_graph).lower()
def update_session(self, session_id, value):
self.cache_service.set(session_id, value)
def clear_session(self, session_id):
self.cache_service.delete(session_id)
================================================
FILE: src/backend/bisheng/services/session/utils.py
================================================
import hashlib
import random
import string
from bisheng.common.utils.util import orjson_dumps
from bisheng.core.cache.utils import filter_json
def session_id_generator(size=6):
return ''.join(random.SystemRandom().choices(string.ascii_uppercase + string.digits, k=size))
def compute_dict_hash(graph_data):
graph_data = filter_json(graph_data)
cleaned_graph_json = orjson_dumps(graph_data, sort_keys=True)
return hashlib.sha256(cleaned_graph_json.encode('utf-8')).hexdigest()
================================================
FILE: src/backend/bisheng/services/settings/__init__.py
================================================
from . import factory, service
__all__ = ['factory', 'service']
================================================
FILE: src/backend/bisheng/services/settings/auth.py
================================================
import secrets
from pathlib import Path
from typing import Optional
from bisheng.services.settings.constants import DEFAULT_SUPERUSER, DEFAULT_SUPERUSER_PASSWORD
from bisheng.services.settings.utils import read_secret_from_file, write_secret_to_file
from loguru import logger
from passlib.context import CryptContext
from pydantic import Field, field_validator
from pydantic_settings import SettingsConfigDict, BaseSettings
class AuthSettings(BaseSettings):
# Login settings
CONFIG_DIR: str
SECRET_KEY: str = Field(
default='',
description='Secret key for JWT. If not provided, a random one will be generated.',
frozen=False,
)
ALGORITHM: str = 'HS256'
ACCESS_TOKEN_EXPIRE_MINUTES: int = 60
REFRESH_TOKEN_EXPIRE_MINUTES: int = 60 * 12 * 7
# API Key to execute /process endpoint
API_KEY_SECRET_KEY: Optional[str] = 'b82818e0ad4ff76615c5721ee21004b07d84cd9b87ba4d9cb42374da134b841a'
API_KEY_ALGORITHM: str = 'HS256'
API_V1_STR: str = '/api/v1'
# If AUTO_LOGIN = True
# > The application does not request login and logs in automatically as a super user.
AUTO_LOGIN: bool = True
NEW_USER_IS_ACTIVE: bool = False
SUPERUSER: str = DEFAULT_SUPERUSER
SUPERUSER_PASSWORD: str = DEFAULT_SUPERUSER_PASSWORD
pwd_context: CryptContext = CryptContext(schemes=['bcrypt'], deprecated='auto')
model_config = SettingsConfigDict(validate_assignment=True, extra='ignore', env_prefix='bisheng_')
def reset_credentials(self):
self.SUPERUSER = DEFAULT_SUPERUSER
self.SUPERUSER_PASSWORD = DEFAULT_SUPERUSER_PASSWORD
@field_validator('SECRET_KEY', mode='before')
@classmethod
def get_secret_key(cls, value, values):
config_dir = values.get('CONFIG_DIR')
if not config_dir:
logger.debug('No CONFIG_DIR provided, not saving secret key')
return value or secrets.token_urlsafe(32)
secret_key_path = Path(config_dir) / 'secret_key'
if value:
logger.debug('Secret key provided')
write_secret_to_file(secret_key_path, value)
else:
logger.debug('No secret key provided, generating a random one')
if secret_key_path.exists():
value = read_secret_from_file(secret_key_path)
logger.debug('Loaded secret key')
if not value:
value = secrets.token_urlsafe(32)
write_secret_to_file(secret_key_path, value)
logger.debug('Saved secret key')
else:
value = secrets.token_urlsafe(32)
write_secret_to_file(secret_key_path, value)
logger.debug('Saved secret key')
return value
================================================
FILE: src/backend/bisheng/services/settings/base.py
================================================
import contextlib
import json
import os
from pathlib import Path
from shutil import copy2
from typing import List, Optional
import orjson
import yaml
from loguru import logger
from pydantic import field_validator, validator
from pydantic_settings import BaseSettings, SettingsConfigDict
# BASE_COMPONENTS_PATH = str(Path(__file__).parent / "components")
BASE_COMPONENTS_PATH = str(Path(__file__).parent.parent.parent / 'components')
class Settings(BaseSettings):
CHAINS: dict = {}
AGENTS: dict = {}
PROMPTS: dict = {}
LLMS: dict = {}
TOOLS: dict = {}
MEMORIES: dict = {}
EMBEDDINGS: dict = {}
VECTORSTORES: dict = {}
DOCUMENTLOADERS: dict = {}
WRAPPERS: dict = {}
RETRIEVERS: dict = {}
TOOLKITS: dict = {}
TEXTSPLITTERS: dict = {}
UTILITIES: dict = {}
OUTPUT_PARSERS: dict = {}
CUSTOM_COMPONENTS: dict = {}
# Define the default LANGFLOW_DIR
CONFIG_DIR: Optional[str] = None
DEV: bool = False
DATABASE_URL: Optional[str] = None
CACHE_TYPE: str = 'memory'
REMOVE_API_KEYS: bool = False
COMPONENTS_PATH: List[str] = []
LANGCHAIN_CACHE: str = 'InMemoryCache'
# Redis
REDIS_HOST: str = 'localhost'
REDIS_PORT: int = 6379
REDIS_DB: int = 0
REDIS_CACHE_EXPIRE: int = 3600
# PLUGIN_DIR: Optional[str] = None
LANGFUSE_SECRET_KEY: Optional[str] = None
LANGFUSE_PUBLIC_KEY: Optional[str] = None
LANGFUSE_HOST: Optional[str] = None
STORE: Optional[bool] = True
STORE_URL: Optional[str] = 'https://api.langflow.store'
DOWNLOAD_WEBHOOK_URL: Optional[
str
] = 'https://api.langflow.store/flows/trigger/ec611a61-8460-4438-b187-a4f65e5559d4'
LIKE_WEBHOOK_URL: Optional[str] = 'https://api.langflow.store/flows/trigger/64275852-ec00-45c1-984e-3bff814732da'
@field_validator('CONFIG_DIR', mode="before")
@classmethod
def set_langflow_dir(cls, value):
if not value:
from platformdirs import user_cache_dir
# Define the app name and author
app_name = 'langflow'
app_author = 'logspace'
# Get the cache directory for the application
cache_dir = user_cache_dir(app_name, app_author)
# Create a .langflow directory inside the cache directory
value = Path(cache_dir)
value.mkdir(parents=True, exist_ok=True)
if isinstance(value, str):
value = Path(value)
if not value.exists():
value.mkdir(parents=True, exist_ok=True)
return str(value)
@field_validator('DATABASE_URL', mode='before')
@classmethod
def set_database_url(cls, value, values):
if not value:
logger.debug('No database_url provided, trying LANGFLOW_DATABASE_URL env variable')
if langflow_database_url := os.getenv('LANGFLOW_DATABASE_URL'):
value = langflow_database_url
logger.debug('Using LANGFLOW_DATABASE_URL env variable.')
else:
logger.debug('No DATABASE_URL env variable, using sqlite database')
# Originally, we used sqlite:///./langflow.db
# so we need to migrate to the new format
# if there is a database in that location
if not values['CONFIG_DIR']:
raise ValueError('CONFIG_DIR not set, please set it or provide a DATABASE_URL')
new_path = f"{values['CONFIG_DIR']}/langflow.db"
if Path('./langflow.db').exists():
if Path(new_path).exists():
logger.debug(f'Database already exists at {new_path}, using it')
else:
try:
logger.debug('Copying existing database to new location')
copy2('./langflow.db', new_path)
logger.debug(f'Copied existing database to {new_path}')
except Exception:
logger.error('Failed to copy database, using default path')
new_path = './langflow.db'
value = f'sqlite:///{new_path}'
return value
@field_validator('COMPONENTS_PATH', mode='before')
@classmethod
def set_components_path(cls, value):
if os.getenv('LANGFLOW_COMPONENTS_PATH'):
logger.debug('Adding LANGFLOW_COMPONENTS_PATH to components_path')
langflow_component_path = os.getenv('LANGFLOW_COMPONENTS_PATH')
if Path(langflow_component_path).exists() and langflow_component_path not in value:
if isinstance(langflow_component_path, list):
for path in langflow_component_path:
if path not in value:
value.append(path)
logger.debug(f'Extending {langflow_component_path} to components_path')
elif langflow_component_path not in value:
value.append(langflow_component_path)
logger.debug(f'Appending {langflow_component_path} to components_path')
if not value:
value = [BASE_COMPONENTS_PATH]
logger.debug('Setting default components path to components_path')
elif BASE_COMPONENTS_PATH not in value:
value.append(BASE_COMPONENTS_PATH)
logger.debug('Adding default components path to components_path')
logger.debug(f'Components path: {value}')
return value
model_config = SettingsConfigDict(validate_assignment=True, extra='ignore', env_prefix='LANGFLOW_')
# @model_validator()
# @classmethod
# def validate_lists(cls, values):
# for key, value in values.items():
# if key != "dev" and not value:
# values[key] = []
# return values
def update_from_yaml(self, file_path: str, dev: bool = False):
new_settings = load_settings_from_yaml(file_path)
self.CHAINS = new_settings.CHAINS or {}
self.AGENTS = new_settings.AGENTS or {}
self.PROMPTS = new_settings.PROMPTS or {}
self.LLMS = new_settings.LLMS or {}
self.TOOLS = new_settings.TOOLS or {}
self.MEMORIES = new_settings.MEMORIES or {}
self.WRAPPERS = new_settings.WRAPPERS or {}
self.TOOLKITS = new_settings.TOOLKITS or {}
self.TEXTSPLITTERS = new_settings.TEXTSPLITTERS or {}
self.UTILITIES = new_settings.UTILITIES or {}
self.EMBEDDINGS = new_settings.EMBEDDINGS or {}
self.VECTORSTORES = new_settings.VECTORSTORES or {}
self.DOCUMENTLOADERS = new_settings.DOCUMENTLOADERS or {}
self.RETRIEVERS = new_settings.RETRIEVERS or {}
self.OUTPUT_PARSERS = new_settings.OUTPUT_PARSERS or {}
self.CUSTOM_COMPONENTS = new_settings.CUSTOM_COMPONENTS or {}
self.COMPONENTS_PATH = new_settings.COMPONENTS_PATH or []
self.DEV = dev
def update_settings(self, **kwargs):
logger.debug('Updating settings')
for key, value in kwargs.items():
# value may contain sensitive information, so we don't want to log it
if not hasattr(self, key):
logger.debug(f'Key {key} not found in settings')
continue
logger.debug(f'Updating {key}')
if isinstance(getattr(self, key), list):
# value might be a '[something]' string
with contextlib.suppress(json.decoder.JSONDecodeError):
value = orjson.loads(str(value))
if isinstance(value, list):
for item in value:
if isinstance(item, Path):
item = str(item)
if item not in getattr(self, key):
getattr(self, key).append(item)
logger.debug(f'Extended {key}')
else:
if isinstance(value, Path):
value = str(value)
if value not in getattr(self, key):
getattr(self, key).append(value)
logger.debug(f'Appended {key}')
else:
setattr(self, key, value)
logger.debug(f'Updated {key}')
logger.debug(f'{key}: {getattr(self, key)}')
def save_settings_to_yaml(settings: Settings, file_path: str):
with open(file_path, 'w') as f:
settings_dict = settings.model_dump()
yaml.dump(settings_dict, f)
def load_settings_from_yaml(file_path: str) -> Settings:
# Check if a string is a valid path or a file name
if '/' not in file_path:
# Get current path
current_path = os.path.dirname(os.path.abspath(__file__))
file_path = os.path.join(current_path, file_path)
with open(file_path, 'r') as f:
settings_dict = yaml.safe_load(f)
settings_dict = {k.upper(): v for k, v in settings_dict.items()}
for key in settings_dict:
if key not in Settings.model_fields.keys():
raise KeyError(f'Key {key} not found in settings')
logger.debug(f'Loading {len(settings_dict[key])} {key} from {file_path}')
return Settings(**settings_dict)
================================================
FILE: src/backend/bisheng/services/settings/constants.py
================================================
DEFAULT_SUPERUSER = 'admin'
DEFAULT_SUPERUSER_PASSWORD = '1234'
================================================
FILE: src/backend/bisheng/services/settings/factory.py
================================================
from pathlib import Path
from bisheng.services.factory import ServiceFactory
from bisheng.services.settings.service import SettingsService
class SettingsServiceFactory(ServiceFactory):
def __init__(self):
super().__init__(SettingsService)
def create(self):
# Here you would have logic to create and configure a SettingsService
bisheng_dir = Path(__file__).parent.parent.parent
return SettingsService.load_settings_from_yaml(str(bisheng_dir / 'config.yaml'))
================================================
FILE: src/backend/bisheng/services/settings/service.py
================================================
import os
import yaml
from bisheng.services.base import Service
from bisheng.services.settings.auth import AuthSettings
from bisheng.services.settings.base import Settings
from loguru import logger
class SettingsService(Service):
name = 'settings_service'
def __init__(self, settings: Settings, auth_settings: AuthSettings):
super().__init__()
self.settings = settings
self.auth_settings = auth_settings
@classmethod
def load_settings_from_yaml(cls, file_path: str) -> 'SettingsService':
# Check if a string is a valid path or a file name
if '/' not in file_path:
# Get current path
current_path = os.path.dirname(os.path.abspath(__file__))
file_path = os.path.join(current_path, file_path)
with open(file_path, 'r') as f:
settings_dict = yaml.safe_load(f)
settings_dict = {k.upper(): v for k, v in settings_dict.items()}
for key in settings_dict:
if key not in Settings.model_fields.keys():
raise KeyError(f'Key {key} not found in settings')
logger.debug(f'Loading {len(settings_dict[key])} {key} from {file_path}')
settings = Settings(**settings_dict)
if not settings.CONFIG_DIR:
raise ValueError('CONFIG_DIR must be set in settings')
auth_settings = AuthSettings(
CONFIG_DIR=settings.CONFIG_DIR,
)
return cls(settings, auth_settings)
================================================
FILE: src/backend/bisheng/services/settings/utils.py
================================================
import os
import platform
from pathlib import Path
from loguru import logger
def set_secure_permissions(file_path):
if platform.system() in ['Linux', 'Darwin']: # Unix/Linux/Mac
os.chmod(file_path, 0o600)
elif platform.system() == 'Windows':
import win32api
import win32con
import win32security
user, domain, _ = win32security.LookupAccountName('', win32api.GetUserName())
sd = win32security.GetFileSecurity(file_path, win32security.DACL_SECURITY_INFORMATION)
dacl = win32security.ACL()
# Set the new DACL for the file: read and write access for the owner, no access for everyone else
dacl.AddAccessAllowedAce(
win32security.ACL_REVISION,
win32con.GENERIC_READ | win32con.GENERIC_WRITE,
user,
)
sd.SetSecurityDescriptorDacl(1, dacl, 0)
win32security.SetFileSecurity(file_path, win32security.DACL_SECURITY_INFORMATION, sd)
else:
print('Unsupported OS')
def write_secret_to_file(path: Path, value: str) -> None:
with path.open('wb') as f:
f.write(value.encode('utf-8'))
try:
set_secure_permissions(path)
except Exception:
logger.error('Failed to set secure permissions on secret key')
def read_secret_from_file(path: Path) -> str:
with path.open('r') as f:
return f.read()
================================================
FILE: src/backend/bisheng/services/store/__init__.py
================================================
================================================
FILE: src/backend/bisheng/services/store/exceptions.py
================================================
class CustomException(Exception):
def __init__(self, detail, status_code):
super().__init__(detail)
self.status_code = status_code
# Define custom exceptions with status codes
class UnauthorizedError(CustomException):
def __init__(self, detail='Unauthorized access'):
super().__init__(detail, 401)
class ForbiddenError(CustomException):
def __init__(self, detail='Forbidden'):
super().__init__(detail, 403)
class APIKeyError(CustomException):
def __init__(self, detail='API key error'):
super().__init__(detail, 400) # ! Should be 401
class FilterError(CustomException):
def __init__(self, detail='Filter error'):
super().__init__(detail, 400)
================================================
FILE: src/backend/bisheng/services/store/factory.py
================================================
from typing import TYPE_CHECKING
from bisheng.services.factory import ServiceFactory
from bisheng.services.store.service import StoreService
if TYPE_CHECKING:
from bisheng.services.settings.service import SettingsService
class StoreServiceFactory(ServiceFactory):
def __init__(self):
super().__init__(StoreService)
def create(self, settings_service: 'SettingsService'):
return StoreService(settings_service)
================================================
FILE: src/backend/bisheng/services/store/schema.py
================================================
from typing import List, Optional
from uuid import UUID
from pydantic import field_validator, BaseModel
class TagResponse(BaseModel):
id: UUID
name: Optional[str] = None
class UsersLikesResponse(BaseModel):
likes_count: Optional[int] = None
liked_by_user: Optional[bool] = None
class CreateComponentResponse(BaseModel):
id: UUID
class TagsIdResponse(BaseModel):
tags_id: Optional[TagResponse] = None
class ListComponentResponse(BaseModel):
id: Optional[UUID] = None
name: Optional[str] = None
description: Optional[str] = None
liked_by_count: Optional[int] = None
liked_by_user: Optional[bool] = None
is_component: Optional[bool] = None
metadata: Optional[dict] = {}
user_created: Optional[dict] = {}
tags: Optional[List[TagResponse]] = None
downloads_count: Optional[int] = None
last_tested_version: Optional[str] = None
private: Optional[bool] = None
# tags comes as a TagsIdResponse but we want to return a list of TagResponse
@field_validator('tags', mode="before")
@classmethod
def tags_to_list(cls, v):
# Check if all values are have id and name
# if so, return v else transform to TagResponse
if not v:
return v
if all(['id' in tag and 'name' in tag for tag in v]):
return v
else:
return [TagResponse(**tag.get('tags_id')) for tag in v if tag.get('tags_id')]
class ListComponentResponseModel(BaseModel):
count: Optional[int] = 0
authorized: bool
results: Optional[List[ListComponentResponse]] = None
class DownloadComponentResponse(BaseModel):
id: UUID
name: Optional[str] = None
description: Optional[str] = None
data: Optional[dict] = None
is_component: Optional[bool] = None
metadata: Optional[dict] = {}
class StoreComponentCreate(BaseModel):
name: str
description: Optional[str] = None
data: dict
tags: Optional[List[str]] = None
parent: Optional[UUID] = None
is_component: Optional[bool] = None
last_tested_version: Optional[str] = None
private: Optional[bool] = True
================================================
FILE: src/backend/bisheng/services/store/service.py
================================================
import json
from contextlib import asynccontextmanager
from contextvars import ContextVar
from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple
from uuid import UUID
import httpx
from bisheng.services.base import Service
from bisheng.services.store.exceptions import APIKeyError, FilterError, ForbiddenError
from bisheng.services.store.schema import (CreateComponentResponse, DownloadComponentResponse,
ListComponentResponse, ListComponentResponseModel,
StoreComponentCreate)
from bisheng.services.store.utils import (process_component_data, process_tags_for_post,
update_components_with_user_data)
from httpx import HTTPError, HTTPStatusError
from loguru import logger
if TYPE_CHECKING:
from bisheng.services.settings.service import SettingsService
user_data_var: ContextVar[Optional[Dict[str, Any]]] = ContextVar('user_data', default=None)
@asynccontextmanager
async def user_data_context(store_service: 'StoreService', api_key: Optional[str] = None):
# Fetch and set user data to the context variable
if api_key:
try:
user_data, _ = await store_service._get(
f'{store_service.base_url}/users/me', api_key, params={'fields': 'id'}
)
user_data_var.set(user_data[0])
except HTTPStatusError as exc:
if exc.response.status_code == 403:
raise ValueError('Invalid API key')
try:
yield
finally:
# Clear the user data from the context variable
user_data_var.set(None)
class StoreService(Service):
"""This is a service that integrates bisheng with the store which
is a Directus instance. It allows to search, get and post components to
the store."""
name = 'store_service'
def __init__(self, settings_service: 'SettingsService'):
self.settings_service = settings_service
self.base_url = self.settings_service.settings.STORE_URL
self.download_webhook_url = self.settings_service.settings.DOWNLOAD_WEBHOOK_URL
self.like_webhook_url = self.settings_service.settings.LIKE_WEBHOOK_URL
self.components_url = f'{self.base_url}/items/components'
self.default_fields = [
'id',
'name',
'description',
'user_created.username',
'is_component',
'tags.tags_id.name',
'tags.tags_id.id',
'count(liked_by)',
'count(downloads)',
'metadata',
'last_tested_version',
'private',
]
# Create a context manager that will use the api key to
# get the user data and all requests inside the context manager
# will make a property return that data
# Without making the request multiple times
async def check_api_key(self, api_key: str):
# Check if the api key is valid
# If it is, return True
# If it is not, return False
try:
user_data, _ = await self._get(f'{self.base_url}/users/me', api_key, params={'fields': 'id'})
return 'id' in user_data[0]
except HTTPStatusError as exc:
if exc.response.status_code in [403, 401]:
return False
else:
raise ValueError(f'Unexpected status code: {exc.response.status_code}')
except Exception as exc:
raise ValueError(f'Unexpected error: {exc}')
async def _get(
self, url: str, api_key: Optional[str] = None, params: Optional[Dict[str, Any]] = None
) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:
"""Utility method to perform GET requests."""
if api_key:
headers = {'Authorization': f'Bearer {api_key}'}
else:
headers = {}
async with httpx.AsyncClient() as client:
try:
response = await client.get(url, headers=headers, params=params)
response.raise_for_status()
except HTTPError as exc:
raise exc
except Exception as exc:
raise ValueError(f'GET failed: {exc}')
json_response = response.json()
result = json_response['data']
metadata = {}
if 'meta' in json_response:
metadata = json_response['meta']
if isinstance(result, dict):
return [result], metadata
return result, metadata
async def call_webhook(self, api_key: str, webhook_url: str, component_id: UUID) -> None:
# The webhook is a POST request with the data in the body
# For now we are calling it just for testing
try:
headers = {'Authorization': f'Bearer {api_key}'}
async with httpx.AsyncClient() as client:
response = await client.post(webhook_url, headers=headers, json={'component_id': str(component_id)})
response.raise_for_status()
return response.json()
except HTTPError as exc:
raise exc
except Exception as exc:
logger.debug(f'Webhook failed: {exc}')
def build_tags_filter(self, tags: List[str]):
tags_filter: Dict[str, Any] = {'tags': {'_and': []}}
for tag in tags:
tags_filter['tags']['_and'].append({'_some': {'tags_id': {'name': {'_eq': tag}}}})
return tags_filter
async def count_components(
self,
filter_conditions: List[Dict[str, Any]],
api_key: Optional[str] = None,
use_api_key: Optional[bool] = False,
) -> int:
params = {'aggregate': json.dumps({'count': '*'})}
if filter_conditions:
params['filter'] = json.dumps({'_and': filter_conditions})
api_key = api_key if use_api_key else None
results, _ = await self._get(self.components_url, api_key, params)
return int(results[0].get('count', 0))
@staticmethod
def build_search_filter_conditions(query: str):
# instead of build the param ?search=query, we will build the filter
# that will use _icontains (case insensitive)
conditions: Dict[str, Any] = {'_or': []}
conditions['_or'].append({'name': {'_icontains': query}})
conditions['_or'].append({'description': {'_icontains': query}})
conditions['_or'].append({'tags': {'tags_id': {'name': {'_icontains': query}}}})
conditions['_or'].append({'user_created': {'username': {'_icontains': query}}})
return conditions
def build_filter_conditions(
self,
component_id: Optional[str] = None,
search: Optional[str] = None,
private: Optional[bool] = None,
tags: Optional[List[str]] = None,
is_component: Optional[bool] = None,
filter_by_user: Optional[bool] = False,
liked: Optional[bool] = False,
store_api_Key: Optional[str] = None,
):
filter_conditions = []
if search is not None:
search_conditions = self.build_search_filter_conditions(search)
filter_conditions.append(search_conditions)
if private is not None:
filter_conditions.append({'private': {'_eq': private}})
if tags:
tags_filter = self.build_tags_filter(tags)
filter_conditions.append(tags_filter)
if component_id is not None:
filter_conditions.append({'id': {'_eq': component_id}})
if is_component is not None:
filter_conditions.append({'is_component': {'_eq': is_component}})
if liked and store_api_Key:
liked_filter = self.build_liked_filter()
filter_conditions.append(liked_filter)
elif liked and not store_api_Key:
raise APIKeyError('You must provide an API key to filter by likes')
if filter_by_user and store_api_Key:
user_data = user_data_var.get()
if not user_data:
raise ValueError('No user data')
filter_conditions.append({'user_created': {'_eq': user_data['id']}})
elif filter_by_user and not store_api_Key:
raise APIKeyError('You must provide an API key to filter your components')
else:
filter_conditions.append({'private': {'_eq': False}})
return filter_conditions
def build_liked_filter(self):
user_data = user_data_var.get()
# params["filter"] = json.dumps({"user_created": {"_eq": user_data["id"]}})
if not user_data:
raise ValueError('No user data')
return {'liked_by': {'directus_users_id': {'_eq': user_data['id']}}}
async def query_components(
self,
api_key: Optional[str] = None,
sort: Optional[List[str]] = None,
page: int = 1,
limit: int = 15,
fields: Optional[List[str]] = None,
filter_conditions: Optional[List[Dict[str, Any]]] = None,
use_api_key: Optional[bool] = False,
) -> Tuple[List[ListComponentResponse], Dict[str, Any]]:
params: Dict[str, Any] = {
'page': page,
'limit': limit,
'fields': ','.join(fields) if fields is not None else ','.join(self.default_fields),
'meta': 'filter_count', # !This is DEPRECATED so we should remove it ASAP
}
# ?aggregate[count]=likes
if sort:
params['sort'] = ','.join(sort)
# Only public components or the ones created by the user
# check for "public" or "Public"
if filter_conditions:
params['filter'] = json.dumps({'_and': filter_conditions})
# If not liked, this means we are getting public components
# so we don't need to risk passing an invalid api_key
# and getting 401
api_key = api_key if use_api_key else None
results, metadata = await self._get(self.components_url, api_key, params)
if isinstance(results, dict):
results = [results]
results_objects = [ListComponentResponse(**result) for result in results]
return results_objects, metadata
async def get_liked_by_user_components(self, component_ids: List[str], api_key: str) -> List[str]:
# Get fields id
# filter should be "id is in component_ids AND liked_by directus_users_id token is api_key"
# return the ids
user_data = user_data_var.get()
if not user_data:
raise ValueError('No user data')
params = {
'fields': 'id',
'filter': json.dumps(
{
'_and': [
{'id': {'_in': component_ids}},
{'liked_by': {'directus_users_id': {'_eq': user_data['id']}}},
]
}
),
}
results, _ = await self._get(self.components_url, api_key, params)
return [result['id'] for result in results]
# Which of the components is parent of the user's components
async def get_components_in_users_collection(self, component_ids: List[str], api_key: str):
user_data = user_data_var.get()
if not user_data:
raise ValueError('No user data')
params = {
'fields': 'id',
'filter': json.dumps(
{
'_and': [
{'user_created': {'_eq': user_data['id']}},
{'parent': {'_in': component_ids}},
]
}
),
}
results, _ = await self._get(self.components_url, api_key, params)
return [result['id'] for result in results]
async def download(self, api_key: str, component_id: UUID) -> DownloadComponentResponse:
url = f'{self.components_url}/{component_id}'
params = {'fields': ','.join(['id', 'name', 'description', 'data', 'is_component', 'metadata'])}
if not self.download_webhook_url:
raise ValueError('DOWNLOAD_WEBHOOK_URL is not set')
component, _ = await self._get(url, api_key, params)
await self.call_webhook(api_key, self.download_webhook_url, component_id)
if len(component) > 1:
raise ValueError('Something went wrong while downloading the component')
component_dict = component[0]
download_component = DownloadComponentResponse(**component_dict)
# Check if metadata is an empty dict
if download_component.metadata in [None, {}] and download_component.data is not None:
# If it is, we need to build the metadata
try:
download_component.metadata = process_component_data(download_component.data.get('nodes', []))
except KeyError:
raise ValueError('Invalid component data. No nodes found')
return download_component
async def upload(self, api_key: str, component_data: StoreComponentCreate) -> CreateComponentResponse:
headers = {'Authorization': f'Bearer {api_key}'}
component_dict = component_data.model_dump(exclude_unset=True)
# Parent is a UUID, but the store expects a string
response = None
if component_dict.get('parent'):
component_dict['parent'] = str(component_dict['parent'])
component_dict = process_tags_for_post(component_dict)
try:
# response = httpx.post(self.components_url, headers=headers, json=component_dict)
# response.raise_for_status()
async with httpx.AsyncClient() as client:
response = await client.post(self.components_url, headers=headers, json=component_dict)
response.raise_for_status()
component = response.json()['data']
return CreateComponentResponse(**component)
except HTTPError as exc:
if response:
try:
errors = response.json()
message = errors['errors'][0]['message']
if message == 'An unexpected error occurred.':
# This is a bug in Directus that returns this error
# when an error was thrown in the flow
message = 'You already have a component with this name. Please choose a different name.'
raise FilterError(message)
except UnboundLocalError:
pass
raise ValueError(f'Upload failed: {exc}')
async def update(
self, api_key: str, component_id: UUID, component_data: StoreComponentCreate
) -> CreateComponentResponse:
# Patch is the same as post, but we need to add the id to the url
headers = {'Authorization': f'Bearer {api_key}'}
component_dict = component_data.model_dump(exclude_unset=True)
# Parent is a UUID, but the store expects a string
response = None
if component_dict.get('parent'):
component_dict['parent'] = str(component_dict['parent'])
component_dict = process_tags_for_post(component_dict)
try:
# response = httpx.post(self.components_url, headers=headers, json=component_dict)
# response.raise_for_status()
async with httpx.AsyncClient() as client:
response = await client.patch(
self.components_url + f'/{component_id}', headers=headers, json=component_dict
)
response.raise_for_status()
component = response.json()['data']
return CreateComponentResponse(**component)
except HTTPError as exc:
if response:
try:
errors = response.json()
message = errors['errors'][0]['message']
if message == 'An unexpected error occurred.':
# This is a bug in Directus that returns this error
# when an error was thrown in the flow
message = 'You already have a component with this name. Please choose a different name.'
raise FilterError(message)
except UnboundLocalError:
pass
raise ValueError(f'Upload failed: {exc}')
async def get_tags(self) -> List[Dict[str, Any]]:
url = f'{self.base_url}/items/tags'
params = {'fields': ','.join(['id', 'name'])}
tags, _ = await self._get(url, api_key=None, params=params)
return tags
async def get_user_likes(self, api_key: str) -> List[Dict[str, Any]]:
url = f'{self.base_url}/users/me'
params = {
'fields': ','.join(['id', 'likes']),
}
likes, _ = await self._get(url, api_key, params)
return likes
async def get_component_likes_count(self, component_id: str, api_key: Optional[str] = None) -> int:
url = f'{self.components_url}/{component_id}'
params = {
'fields': ','.join(['id', 'count(liked_by)']),
}
result, _ = await self._get(url, api_key=api_key, params=params)
if len(result) == 0:
raise ValueError('Component not found')
likes = result[0]['liked_by_count']
# likes_by_count is a string
# try to convert it to int
try:
likes = int(likes)
except ValueError:
raise ValueError(f'Unexpected value for likes count: {likes}')
return likes
async def like_component(self, api_key: str, component_id: str) -> bool:
# if it returns a list with one id, it means the like was successful
# if it returns an int, it means the like was removed
if not self.like_webhook_url:
raise ValueError('LIKE_WEBHOOK_URL is not set')
headers = {'Authorization': f'Bearer {api_key}'}
# response = httpx.post(
# self.like_webhook_url,
# json={"component_id": str(component_id)},
# headers=headers,
# )
# response.raise_for_status()
async with httpx.AsyncClient() as client:
response = await client.post(
self.like_webhook_url,
json={'component_id': str(component_id)},
headers=headers,
)
response.raise_for_status()
if response.status_code == 200:
result = response.json()
if isinstance(result, list):
return True
elif isinstance(result, int):
return False
else:
raise ValueError(f'Unexpected result: {result}')
else:
raise ValueError(f'Unexpected status code: {response.status_code}')
async def get_list_component_response_model(
self,
component_id: Optional[str] = None,
search: Optional[str] = None,
private: Optional[bool] = None,
tags: Optional[List[str]] = None,
is_component: Optional[bool] = None,
fields: Optional[List[str]] = None,
filter_by_user: bool = False,
liked: bool = False,
store_api_key: Optional[str] = None,
sort: Optional[List[str]] = None,
page: int = 1,
limit: int = 15,
):
async with user_data_context(api_key=store_api_key, store_service=self):
filter_conditions: List[Dict[str, Any]] = self.build_filter_conditions(
component_id=component_id,
search=search,
private=private,
tags=tags,
is_component=is_component,
filter_by_user=filter_by_user,
liked=liked,
store_api_Key=store_api_key,
)
result: List[ListComponentResponse] = []
authorized = False
metadata: Dict = {}
comp_count = 0
try:
result, metadata = await self.query_components(
api_key=store_api_key,
page=page,
limit=limit,
sort=sort,
fields=fields,
filter_conditions=filter_conditions,
use_api_key=liked or filter_by_user,
)
if metadata:
comp_count = metadata.get('filter_count', 0)
except HTTPStatusError as exc:
if exc.response.status_code == 403:
raise ForbiddenError('You are not authorized to access this public resource') from exc
elif exc.response.status_code == 401:
raise APIKeyError(
'You are not authorized to access this resource. Please check your API key.'
) from exc
except Exception as exc:
raise ValueError(f'Unexpected error: {exc}') from exc
try:
if result and not metadata:
if len(result) >= limit:
comp_count = await self.count_components(
api_key=store_api_key,
filter_conditions=filter_conditions,
use_api_key=liked or filter_by_user,
)
else:
comp_count = len(result)
elif not metadata:
comp_count = 0
except HTTPStatusError as exc:
if exc.response.status_code == 403:
raise ForbiddenError('You are not authorized to access this public resource')
elif exc.response.status_code == 401:
raise APIKeyError('You are not authorized to access this resource. Please check your API key.')
if store_api_key:
# Now, from the result, we need to get the components
# the user likes and set the liked_by_user to True
# if any of the components does not have an id, it means
# we should not update the components
if not result or any(component.id is None for component in result):
authorized = await self.check_api_key(store_api_key)
else:
try:
updated_result = await update_components_with_user_data(
result, self, store_api_key, liked=liked
)
authorized = True
result = updated_result
except Exception:
# If we get an error here, it means the user is not authorized
authorized = False
return ListComponentResponseModel(results=result, authorized=authorized, count=comp_count)
================================================
FILE: src/backend/bisheng/services/store/utils.py
================================================
from typing import TYPE_CHECKING, List
import httpx
if TYPE_CHECKING:
from bisheng.services.store.schema import ListComponentResponse
from bisheng.services.store.service import StoreService
def process_tags_for_post(component_dict):
tags = component_dict.pop('tags', None)
if tags and all(isinstance(tag, str) for tag in tags):
component_dict['tags'] = [{'tags_id': tag} for tag in tags]
return component_dict
async def update_components_with_user_data(
components: List['ListComponentResponse'],
store_service: 'StoreService',
store_api_Key: str,
liked: bool,
):
"""
Updates the components with the user data (liked_by_user and in_users_collection)
"""
component_ids = [str(component.id) for component in components]
if liked:
# If liked is True, this means all we got were liked_by_user components
# So we can set liked_by_user to True for all components
liked_by_user_ids = component_ids
else:
liked_by_user_ids = await store_service.get_liked_by_user_components(
component_ids=component_ids,
api_key=store_api_Key,
)
# Now we need to set the liked_by_user attribute
for component in components:
component.liked_by_user = str(component.id) in liked_by_user_ids
return components
# Get the latest released version of bisheng (https://pypi.org/project/bisheng/)
def get_lf_version_from_pypi():
try:
response = httpx.get('https://pypi.org/pypi/bisheng/json')
if response.status_code != 200:
return None
return response.json()['info']['version']
except Exception:
return None
def process_component_data(nodes_list):
names = [node['id'].split('-')[0] for node in nodes_list]
metadata = {}
for name in names:
if name in metadata:
metadata[name]['count'] += 1
else:
metadata[name] = {'count': 1}
metadata['total'] = len(names)
return metadata
================================================
FILE: src/backend/bisheng/services/task/__init__.py
================================================
================================================
FILE: src/backend/bisheng/services/task/backends/__init__.py
================================================
================================================
FILE: src/backend/bisheng/services/task/backends/anyio.py
================================================
from typing import Any, Callable, Optional, Tuple
import anyio
from bisheng.services.task.backends.base import TaskBackend
from loguru import logger
class AnyIOTaskResult:
def __init__(self, scope):
self._scope = scope
self._status = 'PENDING'
self._result = None
self._exception = None
@property
def status(self) -> str:
if self._status == 'DONE':
return 'FAILURE' if self._exception is not None else 'SUCCESS'
return self._status
@property
def result(self) -> Any:
return self._result
def ready(self) -> bool:
return self._status == 'DONE'
async def run(self, func, *args, **kwargs):
try:
self._result = await func(*args, **kwargs)
except Exception as e:
self._exception = e
finally:
self._status = 'DONE'
class AnyIOBackend(TaskBackend):
name = 'anyio'
def __init__(self):
self.tasks = {}
async def launch_task(
self, task_func: Callable[..., Any], *args: Any, **kwargs: Any
) -> Tuple[Optional[str], Optional[AnyIOTaskResult]]:
"""
Launch a new task in an asynchronous manner.
Parameters:
task_func: The asynchronous function to run.
*args: Positional arguments to pass to task_func.
**kwargs: Keyword arguments to pass to task_func.
Returns:
A tuple containing a unique task ID and the task result object.
"""
async with anyio.create_task_group() as tg:
try:
task_result = AnyIOTaskResult(tg)
tg.start_soon(task_result.run, task_func, *args, **kwargs)
task_id = str(id(task_result))
self.tasks[task_id] = task_result
logger.info(f'Task {task_id} started.')
return task_id, task_result
except Exception as e:
logger.error(f'An error occurred while launching the task: {e}')
return None, None
def get_task(self, task_id: str) -> Any:
return self.tasks.get(task_id)
================================================
FILE: src/backend/bisheng/services/task/backends/base.py
================================================
from abc import ABC, abstractmethod
from typing import Any, Callable
class TaskBackend(ABC):
@abstractmethod
def launch_task(self, task_func: Callable[..., Any], *args: Any, **kwargs: Any):
pass
@abstractmethod
def get_task(self, task_id: str) -> Any:
pass
================================================
FILE: src/backend/bisheng/services/task/backends/celery.py
================================================
from typing import Any, Callable
from bisheng.services.task.backends.base import TaskBackend
from bisheng.worker import celery_app
from celery.result import AsyncResult # type: ignore
class CeleryBackend(TaskBackend):
name = 'celery'
def __init__(self):
self.celery_app = celery_app
def launch_task(self, task_func: Callable[..., Any], *args: Any, **kwargs: Any) -> tuple[str, AsyncResult]:
# I need to type the delay method to make it easier
from celery import Task # type: ignore
if not hasattr(task_func, 'delay'):
raise ValueError(f'Task function {task_func} does not have a delay method')
task: Task = task_func.delay(*args, **kwargs)
return task.id, AsyncResult(task.id, app=self.celery_app)
def get_task(self, task_id: str) -> Any:
return AsyncResult(task_id, app=self.celery_app)
================================================
FILE: src/backend/bisheng/services/task/factory.py
================================================
from bisheng.services.factory import ServiceFactory
from bisheng.services.task.service import TaskService
class TaskServiceFactory(ServiceFactory):
def __init__(self):
super().__init__(TaskService)
def create(self):
# Here you would have logic to create and configure a TaskService
return TaskService()
================================================
FILE: src/backend/bisheng/services/task/service.py
================================================
from typing import Any, Callable, Coroutine, Union
from loguru import logger
from bisheng.services.base import Service
from bisheng.services.task.backends.anyio import AnyIOBackend
from bisheng.services.task.backends.base import TaskBackend
def check_celery_availability():
try:
raise Exception("celery disabled")
except Exception as exc:
logger.debug(f'Celery not available: {exc}')
status = {'availability': None}
return status
try:
status = check_celery_availability()
USE_CELERY = status.get('availability') is not None
except ImportError:
USE_CELERY = False
class TaskService(Service):
name = 'task_service'
def __init__(self):
self.backend = self.get_backend()
self.use_celery = USE_CELERY
@property
def backend_name(self) -> str:
return self.backend.name
def get_backend(self) -> TaskBackend:
if USE_CELERY:
from bisheng.services.task.backends.celery import CeleryBackend
logger.debug('Using Celery backend')
return CeleryBackend()
logger.debug('Using AnyIO backend')
return AnyIOBackend()
# In your TaskService class
async def launch_and_await_task(
self,
task_func: Callable[..., Any],
*args: Any,
**kwargs: Any,
) -> Any:
if not self.use_celery:
return None, await task_func(*args, **kwargs)
if not hasattr(task_func, 'apply'):
raise ValueError(f'Task function {task_func} does not have an apply method')
task = task_func.apply(args=args, kwargs=kwargs)
result = task.get()
# if result is coroutine
if isinstance(result, Coroutine):
result = await result
return task.id, result
async def launch_task(self, task_func: Callable[..., Any], *args: Any, **kwargs: Any) -> Any:
logger.debug(f'Launching task {task_func} with args {args} and kwargs {kwargs}')
logger.debug(f'Using backend {self.backend}')
task = self.backend.launch_task(task_func, *args, **kwargs)
return await task if isinstance(task, Coroutine) else task
def get_task(self, task_id: Union[int, str]) -> Any:
return self.backend.get_task(task_id)
================================================
FILE: src/backend/bisheng/services/task/utils.py
================================================
import contextlib
from typing import TYPE_CHECKING
if TYPE_CHECKING:
with contextlib.suppress(ImportError):
from celery import Celery # type: ignore
def get_celery_worker_status(app: 'Celery'):
i = app.control.inspect()
availability = app.control.ping()
stats = i.stats()
registered_tasks = i.registered()
active_tasks = i.active()
scheduled_tasks = i.scheduled()
return {
'availability': availability,
'stats': stats,
'registered_tasks': registered_tasks,
'active_tasks': active_tasks,
'scheduled_tasks': scheduled_tasks,
}
================================================
FILE: src/backend/bisheng/services/utils.py
================================================
# from bisheng.services.auth.utils import create_super_user, verify_password
# from bisheng.services.database.utils import initialize_database
from bisheng.services.manager import service_manager
# from bisheng.services.settings.constants import DEFAULT_SUPERUSER, DEFAULT_SUPERUSER_PASSWORD
from loguru import logger
def get_factories_and_deps():
# from bisheng.services.auth import factory as auth_factory
# from bisheng.services.cache import factory as cache_factory
# from bisheng.services.chat import factory as chat_factory
# from bisheng.services.credentials import factory as credentials_factory
# from bisheng.services.database import factory as database_factory
# from bisheng.services.plugins import factory as plugins_factory
from bisheng.services.session import factory as session_service_factory # type: ignore
# from bisheng.services.settings import factory as settings_factory
# from bisheng.services.store import factory as store_factory
from bisheng.services.task import factory as task_factory
return [
# (settings_factory.SettingsServiceFactory(), []),
# (
# auth_factory.AuthServiceFactory(),
# [ServiceType.SETTINGS_SERVICE],
# ),
# (
# database_factory.DatabaseServiceFactory(),
# [ServiceType.SETTINGS_SERVICE],
# ),
# (
# cache_factory.CacheServiceFactory(),
# [ServiceType.SETTINGS_SERVICE],
# ),
# (chat_factory.ChatServiceFactory(), []),
(task_factory.TaskServiceFactory(), []),
(
session_service_factory.SessionServiceFactory(), []
# [ServiceType.CACHE_SERVICE],
),
# (plugins_factory.PluginServiceFactory(), [ServiceType.SETTINGS_SERVICE]),
# (store_factory.StoreServiceFactory(), [ServiceType.SETTINGS_SERVICE]),
# (credentials_factory.CredentialServiceFactory(), [ServiceType.SETTINGS_SERVICE]),
]
# def get_or_create_super_user(session: Session, username, password, is_default):
# from bisheng.services.database.models.user.model import User
# user = session.exec(select(User).where(User.username == username)).first()
# if user and user.is_superuser:
# return None # Superuser already exists
# if user and is_default:
# if user.is_superuser:
# if verify_password(password, user.password):
# return None
# else:
# # Superuser exists but password is incorrect
# # which means that the user has changed the
# # base superuser credentials.
# # This means that the user has already created
# # a superuser and changed the password in the UI
# # so we don't need to do anything.
# logger.debug(
# "Superuser exists but password is incorrect. "
# "This means that the user has changed the "
# "base superuser credentials."
# )
# return None
# else:
# logger.debug("User with superuser credentials exists but is not a superuser.")
# return None
# if user:
# if verify_password(password, user.password):
# raise ValueError("User with superuser credentials exists but is not a superuser.")
# else:
# raise ValueError("Incorrect superuser credentials")
# if is_default:
# logger.debug("Creating default superuser.")
# else:
# logger.debug("Creating superuser.")
# try:
# return create_super_user(username, password, db=session)
# except Exception as exc:
# if "UNIQUE constraint failed: user.username" in str(exc):
# # This is to deal with workers running this
# # at startup and trying to create the superuser
# # at the same time.
# logger.debug("Superuser already exists.")
# return None
# def setup_superuser(settings_service, session: Session):
# if settings_service.auth_settings.AUTO_LOGIN:
# logger.debug("AUTO_LOGIN is set to True. Creating default superuser.")
# else:
# # Remove the default superuser if it exists
# teardown_superuser(settings_service, session)
# username = settings_service.auth_settings.SUPERUSER
# password = settings_service.auth_settings.SUPERUSER_PASSWORD
# is_default = (username == DEFAULT_SUPERUSER) and (password == DEFAULT_SUPERUSER_PASSWORD)
# try:
# user = get_or_create_super_user(session=session, username=username, password=password, is_default=is_default)
# if user is not None:
# logger.debug("Superuser created successfully.")
# except Exception as exc:
# logger.exception(exc)
# raise RuntimeError("Could not create superuser. Please create a superuser manually.") from exc
# finally:
# settings_service.auth_settings.reset_credentials()
# def teardown_superuser(settings_service, session):
# """
# Teardown the superuser.
# """
# # If AUTO_LOGIN is True, we will remove the default superuser
# # from the database.
# if not settings_service.auth_settings.AUTO_LOGIN:
# try:
# logger.debug("AUTO_LOGIN is set to False. Removing default superuser if exists.")
# username = DEFAULT_SUPERUSER
# from bisheng.services.database.models.user.model import User
# user = session.exec(select(User).where(User.username == username)).first()
# if user and user.is_superuser is True:
# session.delete(user)
# session.commit()
# logger.debug("Default superuser removed successfully.")
# else:
# logger.debug("Default superuser not found.")
# except Exception as exc:
# logger.exception(exc)
# raise RuntimeError("Could not remove default superuser.") from exc
def teardown_services():
"""
Teardown all the services.
"""
# try:
# teardown_superuser(get_settings_service(), next(get_session()))
# except Exception as exc:
# logger.exception(exc)
try:
service_manager.teardown()
except Exception as exc:
logger.exception(exc)
# def initialize_settings_service():
# """
# Initialize the settings manager.
# """
# from bisheng.services.settings import factory as settings_factory
# service_manager.register_factory(settings_factory.SettingsServiceFactory())
def initialize_session_service():
"""
Initialize the session manager.
"""
# from bisheng.services.cache import factory as cache_factory
from bisheng.services.session import factory as session_service_factory # type: ignore
# initialize_settings_service()
# service_manager.register_factory(cache_factory.CacheServiceFactory(),
# dependencies=[ServiceType.SETTINGS_SERVICE])
service_manager.register_factory(session_service_factory.SessionServiceFactory(), []
# dependencies=[ServiceType.CACHE_SERVICE],
)
def initialize_services(fix_migration: bool = False):
"""
Initialize all the services needed.
"""
for factory, dependencies in get_factories_and_deps():
try:
service_manager.register_factory(factory, dependencies=dependencies)
except Exception as exc:
logger.exception(exc)
raise RuntimeError(
'Could not initialize services. Please check your settings.') from exc
# Test cache connection
# service_manager.get(ServiceType.CACHE_SERVICE)
# Setup the superuser
# try:
# initialize_database(fix_migration=fix_migration)
# except Exception as exc:
# logger.exception(exc)
# raise exc
# setup_superuser(service_manager.get(ServiceType.SETTINGS_SERVICE), next(get_session()))
# try:
# get_db_service().migrate_flows_if_auto_login()
# except Exception as exc:
# logger.error(f"Error migrating flows: {exc}")
# raise RuntimeError("Error migrating flows") from exc
================================================
FILE: src/backend/bisheng/share_link/__init__.py
================================================
================================================
FILE: src/backend/bisheng/share_link/api/__init__.py
================================================
================================================
FILE: src/backend/bisheng/share_link/api/dependencies.py
================================================
from typing import Annotated, Union
from fastapi import Depends, Header
from sqlmodel.ext.asyncio.session import AsyncSession
from bisheng.common.dependencies.core_deps import get_db_session
from bisheng.share_link.domain.models.share_link import ShareLink
from bisheng.share_link.domain.repositories.implementations.share_link_repository_impl import ShareLinkRepositoryImpl
from bisheng.share_link.domain.repositories.interfaces.share_link_repository import ShareLinkRepository
from bisheng.share_link.domain.services.share_link_service import ShareLinkService
async def get_share_link_repository(
session: AsyncSession = Depends(get_db_session),
) -> ShareLinkRepository:
"""DapatkanShareLinkRepositoryInstance Dependencies"""
return ShareLinkRepositoryImpl(session)
async def get_share_link_service(
share_link_repository: ShareLinkRepository = Depends(get_share_link_repository),
) -> 'ShareLinkService':
"""DapatkanShareLinkServiceInstance Dependencies"""
return ShareLinkService(share_link_repository=share_link_repository)
# Resolve share links in request headerstokenand return the corresponding sharing link information
async def header_share_token_parser(
share_token: Annotated[str | None, Header(alias="share-token")] = None,
share_link_service: ShareLinkService = Depends(get_share_link_service),
) -> Union['ShareLink', None]:
"""According to the request headershare-tokenGet shared link info"""
if not share_token:
return None
share_link = await share_link_service.get_share_link_by_token(share_token)
return share_link
================================================
FILE: src/backend/bisheng/share_link/api/endpoints/__init__.py
================================================
================================================
FILE: src/backend/bisheng/share_link/api/endpoints/share_link.py
================================================
from fastapi import APIRouter, Depends
from fastapi.params import Body
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.schemas.api import resp_200, UnifiedResponseModel
from bisheng.share_link.api.dependencies import get_share_link_service
from bisheng.share_link.api.schemas.share_link_schema import GenerateShareLinkRequest
router = APIRouter()
@router.post('/generate_share_link', summary='generate share link',
response_model=UnifiedResponseModel)
async def generate_share_link(
req_param: GenerateShareLinkRequest = Body(..., description="generate share link request"),
login_user: UserPayload = Depends(UserPayload.get_login_user),
share_link_service=Depends(get_share_link_service)
):
"""
Generate share link
:param login_user:
:param share_link_service:
:param req_param:
:return:
"""
share_link = await share_link_service.generate_share_link(req_param, login_user)
return resp_200(data=share_link)
@router.get('/{share_token}', summary='get share link info',
response_model=UnifiedResponseModel)
async def get_share_link_info(
share_token: str,
share_link_service=Depends(get_share_link_service)
):
"""
Get share link info by share token
:param share_token:
:param share_link_service:
:return:
"""
share_link = await share_link_service.get_share_link_by_token(share_token)
return resp_200(data=share_link)
================================================
FILE: src/backend/bisheng/share_link/api/router.py
================================================
from fastapi import APIRouter
from .endpoints.share_link import router as share_link_router
router = APIRouter(prefix='/share-link', tags=['ShareLink'])
router.include_router(share_link_router)
================================================
FILE: src/backend/bisheng/share_link/api/schemas/__init__.py
================================================
================================================
FILE: src/backend/bisheng/share_link/api/schemas/share_link_schema.py
================================================
from typing import Optional, Dict
from pydantic import BaseModel, Field
from bisheng.share_link.domain.models.share_link import ResourceTypeEnum, ShareMode
class GenerateShareLinkRequest(BaseModel):
"""Generate shared link request model"""
resource_type: ResourceTypeEnum = Field(..., description="Resource Type")
resource_id: str = Field(..., description="reasourseID")
share_mode: ShareMode = Field(default=ShareMode.READ_ONLY, description="sharing mode")
expire_time: int = Field(default=0, description="Expiration time, in seconds,0Indicates never expires")
meta_data: Optional[Dict] = Field(None, description="Metadata to store additional information")
================================================
FILE: src/backend/bisheng/share_link/domain/__init__.py
================================================
================================================
FILE: src/backend/bisheng/share_link/domain/models/__init__.py
================================================
================================================
FILE: src/backend/bisheng/share_link/domain/models/share_link.py
================================================
import uuid
from datetime import datetime
from enum import Enum
from typing import Optional, Dict
from sqlalchemy import Enum as SQLEnum, DateTime, text, JSON
from sqlalchemy import CHAR, Column
from sqlmodel import Field
from bisheng.common.models.base import SQLModelSerializable
class ResourceTypeEnum(str, Enum):
"""Resource Type Enumeration"""
# Inspirational Sessions
LINSIGHT_SESSION = "linsight_session"
# Workbench Conversation
WORKBENCH_CHAT = "workbench_chat"
# The Workflow
WORKFLOW = "workflow"
# Skill
SKILL = "skill"
# assistant
ASSISTANT = "assistant"
class ShareMode(str, Enum):
"""Shared Mode Enumeration"""
# Read-Only
READ_ONLY = "read_only"
# Can Edit
EDITABLE = "editable"
class ShareLinkStatusEnum(str, Enum):
"""Shared link status enumeration"""
# Effective
ACTIVE = "active"
# Tidak berlaku
INACTIVE = "inactive"
# Kedaluwarsa
EXPIRED = "expired"
class ShareLink(SQLModelSerializable, table=True):
"""
Shared Link Model
"""
__tablename__ = 'share_link'
id: str = Field(default_factory=lambda: uuid.uuid4().hex, description='Share linkID',
sa_column=Column(CHAR(36), unique=True, nullable=False, primary_key=True))
share_token: str = Field(sa_column=Column(CHAR(36), index=True, unique=True), description='Share linkToken')
resource_id: str = Field(sa_column=Column(CHAR(36), index=True), description='reasourseID')
resource_type: ResourceTypeEnum = Field(..., sa_column=Column(SQLEnum(ResourceTypeEnum)), description='Resource Type')
share_mode: ShareMode = Field(..., sa_column=Column(SQLEnum(ShareMode)), description='sharing mode')
status: ShareLinkStatusEnum = Field(default=ShareLinkStatusEnum.ACTIVE,
sa_column=Column(SQLEnum(ShareLinkStatusEnum)), description='Share link status')
access_count: int = Field(default=0, description='Number of visits')
meta_data: Optional[Dict] = Field(default=None, description='Shared link metadata', sa_type=JSON, nullable=True)
expire_time: int = Field(default=0, description='Expiration time, in seconds,0Indicates never expires')
create_user_id: str = Field(..., sa_column=Column(CHAR(36)), description='Create UserID')
create_time: datetime = Field(default_factory=datetime.now, description='Creation Time',
sa_column=Column(DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=True, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
================================================
FILE: src/backend/bisheng/share_link/domain/repositories/__init__.py
================================================
================================================
FILE: src/backend/bisheng/share_link/domain/repositories/implementations/__init__.py
================================================
================================================
FILE: src/backend/bisheng/share_link/domain/repositories/implementations/share_link_repository_impl.py
================================================
from sqlmodel.ext.asyncio.session import AsyncSession
from bisheng.common.repositories.implementations.base_repository_impl import BaseRepositoryImpl
from bisheng.share_link.domain.models.share_link import ShareLink
from bisheng.share_link.domain.repositories.interfaces.share_link_repository import ShareLinkRepository
class ShareLinkRepositoryImpl(BaseRepositoryImpl[ShareLink, str], ShareLinkRepository):
"""Shared link repository implementation"""
def __init__(self, session: AsyncSession):
super().__init__(session, ShareLink)
================================================
FILE: src/backend/bisheng/share_link/domain/repositories/interfaces/__init__.py
================================================
================================================
FILE: src/backend/bisheng/share_link/domain/repositories/interfaces/share_link_repository.py
================================================
from abc import ABC
from bisheng.common.repositories.interfaces.base_repository import BaseRepository
from bisheng.share_link.domain.models.share_link import ShareLink
class ShareLinkRepository(BaseRepository[ShareLink, str], ABC):
"""Shared Link Warehouse Interface"""
pass
================================================
FILE: src/backend/bisheng/share_link/domain/services/__init__.py
================================================
================================================
FILE: src/backend/bisheng/share_link/domain/services/share_link_service.py
================================================
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode.http_error import NotFoundError
from bisheng.common.utils import util as common_util
from bisheng.share_link.api.schemas.share_link_schema import GenerateShareLinkRequest
from bisheng.share_link.domain.models.share_link import ShareLink
from bisheng.share_link.domain.repositories.interfaces.share_link_repository import ShareLinkRepository
class ShareLinkService:
def __init__(self,
share_link_repository: 'ShareLinkRepository'):
self.share_link_repository = share_link_repository
async def generate_share_link(self, generate_share_link: GenerateShareLinkRequest,
login_user: UserPayload) -> ShareLink:
"""Generate sharable link"""
share_token = common_util.generate_short_high_entropy_string()
share_link = ShareLink(
share_token=share_token,
resource_id=generate_share_link.resource_id,
resource_type=generate_share_link.resource_type,
share_mode=generate_share_link.share_mode,
expire_time=generate_share_link.expire_time,
meta_data=generate_share_link.meta_data,
create_user_id=login_user.user_id
)
return await self.share_link_repository.save(share_link)
async def get_share_link_by_token(self, share_token: str) -> ShareLink:
"""
accordingshare_tokenGet shared link info
:param share_token:
:return:
"""
share_link = await self.share_link_repository.find_one(share_token=share_token)
if not share_link:
raise NotFoundError()
return share_link
================================================
FILE: src/backend/bisheng/sql.json
================================================
[
{
"知识库问答_严谨版": {
"id": "RetrievalQA-bb180",
"type": "RetrievalQA",
"node": {
"output_types": [],
"display_name": "知识库问答_严谨版",
"documentation": "",
"base_classes": [
"RetrievalQA",
"BaseRetrievalQA",
"Chain",
"function"
],
"description": "Chain for question-answering against an index.",
"template": {
"tags_RetrievalQA-ypS4m": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags",
"proxy": {
"id": "RetrievalQA-ypS4m",
"field": "tags"
}
},
"memory_RetrievalQA-ypS4m": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory",
"proxy": {
"id": "RetrievalQA-ypS4m",
"field": "memory"
}
},
"verbose_RetrievalQA-ypS4m": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose",
"proxy": {
"id": "RetrievalQA-ypS4m",
"field": "verbose"
}
},
"metadata_RetrievalQA-ypS4m": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata",
"proxy": {
"id": "RetrievalQA-ypS4m",
"field": "metadata"
}
},
"callbacks_RetrievalQA-ypS4m": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks",
"proxy": {
"id": "RetrievalQA-ypS4m",
"field": "callbacks"
}
},
"input_key_RetrievalQA-ypS4m": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_key",
"proxy": {
"id": "RetrievalQA-ypS4m",
"field": "input_key"
}
},
"input_node_RetrievalQA-ypS4m": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question",
"proxy": {
"id": "RetrievalQA-ypS4m",
"field": "input_node"
}
},
"output_key_RetrievalQA-ypS4m": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "output_key",
"proxy": {
"id": "RetrievalQA-ypS4m",
"field": "output_key"
}
},
"return_source_documents_RetrievalQA-ypS4m": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_source_documents",
"proxy": {
"id": "RetrievalQA-ypS4m",
"field": "return_source_documents"
}
},
"token_max_CombineDocsChain-bMaXk": {
"l2": false,
"info": "只对Stuff类型生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": 7000,
"l2_name": "token_max",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max",
"proxy": {
"id": "CombineDocsChain-bMaXk",
"field": "token_max"
}
},
"chain_type_CombineDocsChain-bMaXk": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "chain_type",
"proxy": {
"id": "CombineDocsChain-bMaXk",
"field": "chain_type"
}
},
"document_prompt_CombineDocsChain-bMaXk": {
"info": "",
"list": false,
"name": "document_prompt",
"type": "BasePromptTemplate",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "document_prompt",
"proxy": {
"id": "CombineDocsChain-bMaXk",
"field": "document_prompt"
}
},
"context_PromptTemplate-gM5zg": {
"info": "",
"list": false,
"name": "context",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "context",
"proxy": {
"id": "PromptTemplate-gM5zg",
"field": "context"
}
},
"question_PromptTemplate-gM5zg": {
"info": "",
"list": false,
"name": "question",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "question",
"proxy": {
"id": "PromptTemplate-gM5zg",
"field": "question"
}
},
"template_PromptTemplate-gM5zg": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "参考文本：\n{context}\n\n----------------------\n# Role：客服\n\n# Background :\n- 你是一名客服人员。你的日常工作是根据【参考文本】的内容回复用户的各类咨询问题。\n\n# Goals:\n- 以上【参考文本】中是从企业知识库中查找到的相关信息，只能结合以上信息进行回答，若以上内容为空或其中没有找到能回答【用户问题】的内容时，则回复“没有找到相关内容”，不能根据你自己的知识自己发挥。\n- 特别要注意区分【用户问题】与【参考文本】中不同的日期、人名、公司名这些关键信息。\n- 用中文回答问题，并且答案要严谨、专业、清晰、可读性好。\n- 拥有排版审美, 会利用序号, 缩进, 分隔线和换行符等等来美化信息排版。\n\n----------------------\n\n用户问题: {question}\n你的回答:",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "template",
"proxy": {
"id": "PromptTemplate-gM5zg",
"field": "template"
}
},
"output_parser_PromptTemplate-gM5zg": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser",
"proxy": {
"id": "PromptTemplate-gM5zg",
"field": "output_parser"
}
},
"input_variables_PromptTemplate-gM5zg": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"context",
"question"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_variables",
"proxy": {
"id": "PromptTemplate-gM5zg",
"field": "input_variables"
}
},
"template_format_PromptTemplate-gM5zg": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "template_format",
"proxy": {
"id": "PromptTemplate-gM5zg",
"field": "template_format"
}
},
"partial_variables_PromptTemplate-gM5zg": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "partial_variables",
"proxy": {
"id": "PromptTemplate-gM5zg",
"field": "partial_variables"
}
},
"validate_template_PromptTemplate-gM5zg": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "validate_template",
"proxy": {
"id": "PromptTemplate-gM5zg",
"field": "validate_template"
}
},
"ids_ElasticKeywordsSearch-dVhq9": {
"l2": false,
"info": "",
"list": true,
"name": "ids",
"show": false,
"type": "str",
"l2_name": "ids",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ids",
"proxy": {
"id": "ElasticKeywordsSearch-dVhq9",
"field": "ids"
}
},
"documents_ElasticKeywordsSearch-dVhq9": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents",
"proxy": {
"id": "ElasticKeywordsSearch-dVhq9",
"field": "documents"
}
},
"embedding_ElasticKeywordsSearch-dVhq9": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding",
"proxy": {
"id": "ElasticKeywordsSearch-dVhq9",
"field": "embedding"
}
},
"metadatas_ElasticKeywordsSearch-dVhq9": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas",
"proxy": {
"id": "ElasticKeywordsSearch-dVhq9",
"field": "metadatas"
}
},
"index_name_ElasticKeywordsSearch-dVhq9": {
"l2": true,
"info": "",
"list": false,
"name": "index_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "知识库",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_name",
"proxy": {
"id": "ElasticKeywordsSearch-dVhq9",
"field": "index_name"
}
},
"ssl_verify_ElasticKeywordsSearch-dVhq9": {
"l2": false,
"info": "",
"list": false,
"name": "ssl_verify",
"show": true,
"type": "str",
"value": "",
"l2_name": "ssl_verify",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ssl_verify",
"proxy": {
"id": "ElasticKeywordsSearch-dVhq9",
"field": "ssl_verify"
}
},
"search_kwargs_ElasticKeywordsSearch-dVhq9": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": false,
"type": "code",
"value": "{\"k\":6,\"query_strategy\":\"match_phrase\",\"must_or_should\":\"must\"}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs",
"proxy": {
"id": "ElasticKeywordsSearch-dVhq9",
"field": "search_kwargs"
}
},
"refresh_indices_ElasticKeywordsSearch-dVhq9": {
"l2": false,
"info": "",
"list": false,
"name": "refresh_indices",
"show": false,
"type": "bool",
"value": true,
"l2_name": "refresh_indices",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "refresh_indices",
"proxy": {
"id": "ElasticKeywordsSearch-dVhq9",
"field": "refresh_indices"
}
},
"elasticsearch_url_ElasticKeywordsSearch-dVhq9": {
"l2": false,
"info": "",
"list": false,
"name": "elasticsearch_url",
"show": true,
"type": "str",
"value": "",
"l2_name": "elasticsearch_url",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ES_connection_url",
"proxy": {
"id": "ElasticKeywordsSearch-dVhq9",
"field": "elasticsearch_url"
}
},
"question_PromptTemplate-4iBFj": {
"info": "",
"list": false,
"name": "question",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "question",
"proxy": {
"id": "PromptTemplate-4iBFj",
"field": "question"
}
},
"template_PromptTemplate-4iBFj": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "分析给定Question，提取Question中包含的KeyWords，输出列表形式\n\nExamples:\nQuestion: 数据项素2022年营收是多少？\nKeyWords: ['数据项素', '2022', '营收, '营业收入']\nQuestion: 深圳出差住宿标准是什么？\nKeyWords: ['深圳', '出差', '住宿标准', '报销标准']\n\n\n----------------\nQuestion: {question}\nKeyWords: ",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "template",
"proxy": {
"id": "PromptTemplate-4iBFj",
"field": "template"
}
},
"output_parser_PromptTemplate-4iBFj": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser",
"proxy": {
"id": "PromptTemplate-4iBFj",
"field": "output_parser"
}
},
"input_variables_PromptTemplate-4iBFj": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"question"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_variables",
"proxy": {
"id": "PromptTemplate-4iBFj",
"field": "input_variables"
}
},
"template_format_PromptTemplate-4iBFj": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "template_format",
"proxy": {
"id": "PromptTemplate-4iBFj",
"field": "template_format"
}
},
"partial_variables_PromptTemplate-4iBFj": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "partial_variables",
"proxy": {
"id": "PromptTemplate-4iBFj",
"field": "partial_variables"
}
},
"validate_template_PromptTemplate-4iBFj": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "validate_template",
"proxy": {
"id": "PromptTemplate-4iBFj",
"field": "validate_template"
}
},
"callbacks_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "callbacks",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"list": true,
"file_path": "",
"display_name": "Callbacks",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "callbacks"
}
},
"async_client_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "async_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Async Client",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "async_client"
}
},
"cache_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": true,
"file_path": "",
"password": false,
"name": "cache",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"display_name": "Cache",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "cache"
}
},
"client_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Client",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "client"
}
},
"default_headers_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_headers",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Headers",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "default_headers"
}
},
"default_query_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_query",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Query",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "default_query"
}
},
"http_client_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "http_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Http Client",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "http_client"
}
},
"max_retries_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 2,
"password": false,
"name": "max_retries",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Retries",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "max_retries"
}
},
"max_tokens_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": true,
"name": "max_tokens",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Tokens",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "max_tokens"
},
"value": ""
},
"metadata_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "metadata",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Metadata",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "metadata"
}
},
"model_kwargs_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "model_kwargs",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Model Kwargs",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "model_kwargs"
}
},
"model_name_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "gpt-3.5-turbo-0613",
"password": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"name": "model_name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Model Name",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "model_name"
}
},
"n_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 1,
"password": false,
"name": "n",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "N",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "n"
}
},
"name_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Name",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "name"
}
},
"openai_api_base_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_api_base",
"display_name": "OpenAI API Base",
"advanced": true,
"dynamic": false,
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "openai_api_base"
}
},
"openai_api_key_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "",
"password": true,
"name": "openai_api_key",
"display_name": "OpenAI API Key",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "openai_api_key"
}
},
"openai_organization_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "openai_organization",
"display_name": "OpenAI Organization",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "openai_organization"
}
},
"openai_proxy_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_proxy",
"display_name": "OpenAI Proxy",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "openai_proxy"
}
},
"request_timeout_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "request_timeout",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Request Timeout",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "request_timeout"
}
},
"streaming_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": false,
"password": false,
"name": "streaming",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Streaming",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "streaming"
}
},
"tags_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tags",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Tags",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "tags"
}
},
"temperature_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": 0.7,
"password": false,
"name": "temperature",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Temperature",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "temperature"
}
},
"tiktoken_model_name_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tiktoken_model_name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Tiktoken Model Name",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "tiktoken_model_name"
}
},
"verbose_ChatOpenAI-d9de6": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "verbose",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Verbose",
"proxy": {
"id": "ChatOpenAI-d9de6",
"field": "verbose"
}
}
},
"flow": {
"data": {
"nodes": [
{
"width": 384,
"height": 316,
"id": "RetrievalQA-ypS4m",
"type": "genericNode",
"position": {
"x": 1928.7731793921107,
"y": -1748.5798539597242
},
"data": {
"id": "RetrievalQA-ypS4m",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags"
},
"_type": "RetrievalQA",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory"
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose"
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata"
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks"
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_key"
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "output_key"
},
"combine_documents_chain": {
"l2": false,
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"l2_name": "combine_documents_chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_source_documents"
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"RetrievalQA",
"BaseRetrievalQA",
"Chain",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1928.7731793921107,
"y": -1748.5798539597242
},
"dragging": false
},
{
"width": 384,
"height": 412,
"id": "CombineDocsChain-bMaXk",
"type": "genericNode",
"position": {
"x": 1321.387622087405,
"y": -2246.516142425817
},
"data": {
"id": "CombineDocsChain-bMaXk",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"prompt": {
"l2": false,
"info": "只对Stuff类型生效",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"l2_name": "prompt",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
},
"token_max": {
"l2": false,
"info": "只对Stuff类型生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": 7000,
"l2_name": "token_max",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "chain_type"
},
"document_prompt": {
"info": "",
"list": false,
"name": "document_prompt",
"type": "BasePromptTemplate",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "document_prompt"
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1321.387622087405,
"y": -2246.516142425817
},
"dragging": false
},
{
"width": 384,
"height": 454,
"id": "PromptTemplate-gM5zg",
"type": "genericNode",
"position": {
"x": 819.5843069233761,
"y": -2660.280062933134
},
"data": {
"id": "PromptTemplate-gM5zg",
"node": {
"name": "",
"template": {
"_type": "PromptTemplate",
"context": {
"info": "",
"list": false,
"name": "context",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "context"
},
"question": {
"info": "",
"list": false,
"name": "question",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "question"
},
"template": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "参考文本：\n{context}\n\n----------------------\n# Role：客服\n\n# Background :\n- 你是一名客服人员。你的日常工作是根据【参考文本】的内容回复用户的各类咨询问题。\n\n# Goals:\n- 以上【参考文本】中是从企业知识库中查找到的相关信息，只能结合以上信息进行回答，若以上内容为空或其中没有找到能回答【用户问题】的内容时，则回复“没有找到相关内容”，不能根据你自己的知识自己发挥。\n- 特别要注意区分【用户问题】与【参考文本】中不同的日期、人名、公司名这些关键信息。\n- 用中文回答问题，并且答案要严谨、专业、清晰、可读性好。\n- 拥有排版审美, 会利用序号, 缩进, 分隔线和换行符等等来美化信息排版。\n\n----------------------\n\n用户问题: {question}\n你的回答:",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "template"
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser"
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"context",
"question"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_variables"
},
"template_format": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "template_format"
},
"partial_variables": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "partial_variables"
},
"validate_template": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "validate_template"
}
},
"description": "Schema to represent a prompt for an LLM.",
"base_classes": [
"StringPromptTemplate",
"BasePromptTemplate",
"PromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"context",
"question"
],
"template": [
"context",
"question"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {
"formatters": {
"openai_api_key": {}
},
"base_formatters": {
"dict": {},
"list": {},
"show": {},
"union": {},
"kwargs": {},
"default": {},
"headers": {},
"optional": {},
"password": {},
"multiline": {},
"model_fields": {
"MODEL_DICT": {
"OpenAI": [
"text-davinci-003",
"text-davinci-002",
"text-curie-001",
"text-babbage-001",
"text-ada-001"
],
"Anthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
],
"ChatOpenAI": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k"
],
"ChatAnthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
]
}
},
"dict_code_file": {}
}
}
},
"type": "PromptTemplate",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 819.5843069233761,
"y": -2660.280062933134
},
"dragging": false
},
{
"width": 384,
"height": 664,
"id": "ElasticKeywordsSearch-dVhq9",
"type": "genericNode",
"position": {
"x": 831.1394852089952,
"y": -2032.1407340312012
},
"data": {
"id": "ElasticKeywordsSearch-dVhq9",
"node": {
"template": {
"ids": {
"l2": false,
"info": "",
"list": true,
"name": "ids",
"show": false,
"type": "str",
"l2_name": "ids",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ids"
},
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLLM",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "ElasticKeywordsSearch",
"prompt": {
"l2": false,
"info": "",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"l2_name": "prompt",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas"
},
"index_name": {
"l2": true,
"info": "",
"list": false,
"name": "index_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "知识库",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_name"
},
"ssl_verify": {
"l2": false,
"info": "",
"list": false,
"name": "ssl_verify",
"show": true,
"type": "str",
"value": "",
"l2_name": "ssl_verify",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ssl_verify"
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{\"k\":6,\"query_strategy\":\"match_phrase\",\"must_or_should\":\"must\"}",
"l2_name": "search_kwargs",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs"
},
"refresh_indices": {
"l2": false,
"info": "",
"list": false,
"name": "refresh_indices",
"show": false,
"type": "bool",
"value": true,
"l2_name": "refresh_indices",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "refresh_indices"
},
"elasticsearch_url": {
"l2": false,
"info": "",
"list": false,
"name": "elasticsearch_url",
"show": true,
"type": "str",
"value": "",
"l2_name": "elasticsearch_url",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ES_connection_url"
}
},
"description": "Construct ElasticKeywordsSearch wrapper from raw documents.",
"base_classes": [
"ElasticKeywordsSearch",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "ElasticKeywordsSearch",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "ElasticKeywordsSearch",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 831.1394852089952,
"y": -2032.1407340312012
},
"dragging": false
},
{
"width": 384,
"height": 372,
"id": "PromptTemplate-4iBFj",
"type": "genericNode",
"position": {
"x": 149.22093916394033,
"y": -1992.534049733149
},
"data": {
"id": "PromptTemplate-4iBFj",
"node": {
"name": "",
"template": {
"_type": "PromptTemplate",
"question": {
"info": "",
"list": false,
"name": "question",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "question"
},
"template": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "分析给定Question，提取Question中包含的KeyWords，输出列表形式\n\nExamples:\nQuestion: 数据项素2022年营收是多少？\nKeyWords: ['数据项素', '2022', '营收, '营业收入']\nQuestion: 深圳出差住宿标准是什么？\nKeyWords: ['深圳', '出差', '住宿标准', '报销标准']\n\n\n----------------\nQuestion: {question}\nKeyWords: ",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "template"
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser"
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"question"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_variables"
},
"template_format": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "template_format"
},
"partial_variables": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "partial_variables"
},
"validate_template": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "validate_template"
}
},
"description": "Schema to represent a prompt for an LLM.",
"base_classes": [
"StringPromptTemplate",
"BasePromptTemplate",
"PromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"question"
],
"template": [
"question"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {
"formatters": {
"openai_api_key": {}
},
"base_formatters": {
"dict": {},
"list": {},
"show": {},
"union": {},
"kwargs": {},
"default": {},
"headers": {},
"optional": {},
"password": {},
"multiline": {},
"model_fields": {
"MODEL_DICT": {
"OpenAI": [
"text-davinci-003",
"text-davinci-002",
"text-curie-001",
"text-babbage-001",
"text-ada-001"
],
"Anthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
],
"ChatOpenAI": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k"
],
"ChatAnthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
]
}
},
"dict_code_file": {}
}
}
},
"type": "PromptTemplate",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 149.22093916394033,
"y": -1992.534049733149
},
"dragging": false
},
{
"width": 384,
"height": 606,
"id": "ChatOpenAI-d9de6",
"type": "genericNode",
"position": {
"x": 181.27135225054826,
"y": -2719.2953758423564
},
"data": {
"type": "ChatOpenAI",
"node": {
"template": {
"callbacks": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "callbacks",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"list": true,
"file_path": "",
"display_name": "Callbacks"
},
"async_client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "async_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Async Client"
},
"cache": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": true,
"file_path": "",
"password": false,
"name": "cache",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"display_name": "Cache"
},
"client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Client"
},
"default_headers": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_headers",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Headers"
},
"default_query": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_query",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Query"
},
"http_client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "http_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Http Client"
},
"max_retries": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 2,
"password": false,
"name": "max_retries",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Retries"
},
"max_tokens": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": true,
"name": "max_tokens",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Tokens"
},
"metadata": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "metadata",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Metadata"
},
"model_kwargs": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "model_kwargs",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Model Kwargs"
},
"model_name": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "gpt-3.5-turbo-0613",
"password": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"name": "model_name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Model Name"
},
"n": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 1,
"password": false,
"name": "n",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "N"
},
"name": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Name"
},
"openai_api_base": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_api_base",
"display_name": "OpenAI API Base",
"advanced": false,
"dynamic": false,
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_api_key": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "",
"password": true,
"name": "openai_api_key",
"display_name": "OpenAI API Key",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_organization": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "openai_organization",
"display_name": "OpenAI Organization",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_proxy": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_proxy",
"display_name": "OpenAI Proxy",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"request_timeout": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "request_timeout",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Request Timeout"
},
"streaming": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": false,
"password": false,
"name": "streaming",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Streaming"
},
"tags": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tags",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Tags"
},
"temperature": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": 0.7,
"password": false,
"name": "temperature",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Temperature"
},
"tiktoken_model_name": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tiktoken_model_name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Tiktoken Model Name"
},
"verbose": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "verbose",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Verbose"
},
"_type": "ChatOpenAI"
},
"description": "[*Deprecated*] `OpenAI` Chat large language models API.",
"base_classes": [
"ChatOpenAI",
"BaseLanguageModel",
"BaseChatModel",
"BaseLLM"
],
"display_name": "ChatOpenAI",
"documentation": "https://python.langchain.com/docs/modules/model_io/models/chat/integrations/openai",
"custom_fields": {},
"output_types": [],
"field_formatters": {},
"beta": false,
"official": false
},
"id": "ChatOpenAI-d9de6",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 181.27135225054826,
"y": -2719.2953758423564
},
"dragging": false
}
],
"edges": [
{
"source": "CombineDocsChain-bMaXk",
"target": "RetrievalQA-ypS4m",
"sourceHandle": "CombineDocsChain|CombineDocsChain-bMaXk|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-ypS4m",
"id": "reactflow__edge-CombineDocsChain-bMaXkCombineDocsChain|CombineDocsChain-bMaXk|BaseCombineDocumentsChain|function-RetrievalQA-ypS4mBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-ypS4m",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "PromptTemplate-gM5zg",
"target": "CombineDocsChain-bMaXk",
"sourceHandle": "PromptTemplate|PromptTemplate-gM5zg|StringPromptTemplate|BasePromptTemplate|PromptTemplate",
"targetHandle": "BasePromptTemplate|prompt|CombineDocsChain-bMaXk",
"id": "reactflow__edge-PromptTemplate-gM5zgPromptTemplate|PromptTemplate-gM5zg|StringPromptTemplate|BasePromptTemplate|PromptTemplate-CombineDocsChain-bMaXkBasePromptTemplate|prompt|CombineDocsChain-bMaXk",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "ElasticKeywordsSearch-dVhq9",
"target": "RetrievalQA-ypS4m",
"sourceHandle": "ElasticKeywordsSearch|ElasticKeywordsSearch-dVhq9|ElasticKeywordsSearch|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-ypS4m",
"id": "reactflow__edge-ElasticKeywordsSearch-dVhq9ElasticKeywordsSearch|ElasticKeywordsSearch-dVhq9|ElasticKeywordsSearch|VectorStore|BaseRetriever|VectorStoreRetriever-RetrievalQA-ypS4mBaseRetriever|retriever|RetrievalQA-ypS4m",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "PromptTemplate-4iBFj",
"target": "ElasticKeywordsSearch-dVhq9",
"sourceHandle": "PromptTemplate|PromptTemplate-4iBFj|StringPromptTemplate|BasePromptTemplate|PromptTemplate",
"targetHandle": "BasePromptTemplate|prompt|ElasticKeywordsSearch-dVhq9",
"id": "reactflow__edge-PromptTemplate-4iBFjPromptTemplate|PromptTemplate-4iBFj|StringPromptTemplate|BasePromptTemplate|PromptTemplate-ElasticKeywordsSearch-dVhq9BasePromptTemplate|prompt|ElasticKeywordsSearch-dVhq9",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "ChatOpenAI-d9de6",
"sourceHandle": "ChatOpenAI|ChatOpenAI-d9de6|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM",
"target": "ElasticKeywordsSearch-dVhq9",
"targetHandle": "BaseLLM|llm|ElasticKeywordsSearch-dVhq9",
"style": {
"stroke": "#555"
},
"className": "stroke-foreground stroke-connection",
"animated": true,
"id": "reactflow__edge-ChatOpenAI-d9de6ChatOpenAI|ChatOpenAI-d9de6|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM-ElasticKeywordsSearch-dVhq9BaseLLM|llm|ElasticKeywordsSearch-dVhq9",
"selected": true
},
{
"source": "ChatOpenAI-d9de6",
"sourceHandle": "ChatOpenAI|ChatOpenAI-d9de6|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM",
"target": "CombineDocsChain-bMaXk",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-bMaXk",
"style": {
"stroke": "#555"
},
"className": "stroke-foreground stroke-connection",
"animated": true,
"id": "reactflow__edge-ChatOpenAI-d9de6ChatOpenAI|ChatOpenAI-d9de6|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM-CombineDocsChain-bMaXkBaseLanguageModel|llm|CombineDocsChain-bMaXk",
"selected": true
}
],
"viewport": {
"zoom": 1,
"x": 0,
"y": 0
}
},
"is_component": false,
"name": "",
"description": "",
"id": "cc01e",
"status": 0,
"write": false,
"guide_word": ""
}
}
}
},
{
"知识库问答_增强版": {
"id": "ConversationalRetrievalChain-4a1a1",
"type": "ConversationalRetrievalChain",
"node": {
"output_types": [],
"display_name": "知识库问答_增强版",
"documentation": "",
"base_classes": [
"BaseConversationalRetrievalChain",
"Chain",
"ConversationalRetrievalChain",
"function"
],
"description": "Convenience method to load chain from LLM and retriever.",
"template": {
"tags_MixEsVectorRetriever-J35CZ": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": true,
"type": "str",
"l2_name": "tags",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags",
"proxy": {
"id": "MixEsVectorRetriever-J35CZ",
"field": "tags"
}
},
"metadata_MixEsVectorRetriever-J35CZ": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata",
"proxy": {
"id": "MixEsVectorRetriever-J35CZ",
"field": "metadata"
}
},
"combine_strategy_MixEsVectorRetriever-J35CZ": {
"l2": false,
"info": "",
"list": true,
"name": "combine_strategy",
"show": true,
"type": "str",
"value": "mix",
"l2_name": "combine_strategy",
"options": [
"keyword_front",
"vector_front",
"mix"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "combine_strategy",
"proxy": {
"id": "MixEsVectorRetriever-J35CZ",
"field": "combine_strategy"
}
},
"drop_old_Milvus-cyR5W": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "drop_old",
"proxy": {
"id": "Milvus-cyR5W",
"field": "drop_old"
}
},
"documents_Milvus-cyR5W": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents",
"proxy": {
"id": "Milvus-cyR5W",
"field": "documents"
}
},
"embedding_Milvus-cyR5W": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding",
"proxy": {
"id": "Milvus-cyR5W",
"field": "embedding"
}
},
"metadatas_Milvus-cyR5W": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas",
"proxy": {
"id": "Milvus-cyR5W",
"field": "metadatas"
}
},
"index_params_Milvus-cyR5W": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_params",
"proxy": {
"id": "Milvus-cyR5W",
"field": "index_params"
}
},
"search_kwargs_Milvus-cyR5W": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": false,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs",
"proxy": {
"id": "Milvus-cyR5W",
"field": "search_kwargs"
}
},
"search_params_Milvus-cyR5W": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_params",
"proxy": {
"id": "Milvus-cyR5W",
"field": "search_params"
}
},
"collection_name_Milvus-cyR5W": {
"l2": true,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "collection_name",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "collection_name",
"proxy": {
"id": "Milvus-cyR5W",
"field": "collection_name"
}
},
"connection_args_Milvus-cyR5W": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": false,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "connection_args",
"proxy": {
"id": "Milvus-cyR5W",
"field": "connection_args"
}
},
"consistency_level_Milvus-cyR5W": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "consistency_level",
"proxy": {
"id": "Milvus-cyR5W",
"field": "consistency_level"
}
},
"ids_ElasticKeywordsSearch-31Et9": {
"l2": false,
"info": "",
"list": true,
"name": "ids",
"show": false,
"type": "str",
"l2_name": "ids",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ids",
"proxy": {
"id": "ElasticKeywordsSearch-31Et9",
"field": "ids"
}
},
"llm_ElasticKeywordsSearch-31Et9": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"type": "BaseLLM",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "LLM",
"proxy": {
"id": "ElasticKeywordsSearch-31Et9",
"field": "llm"
}
},
"prompt_ElasticKeywordsSearch-31Et9": {
"l2": false,
"info": "",
"list": false,
"name": "prompt",
"type": "BasePromptTemplate",
"l2_name": "prompt",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt",
"proxy": {
"id": "ElasticKeywordsSearch-31Et9",
"field": "prompt"
}
},
"documents_ElasticKeywordsSearch-31Et9": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents",
"proxy": {
"id": "ElasticKeywordsSearch-31Et9",
"field": "documents"
}
},
"embedding_ElasticKeywordsSearch-31Et9": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding",
"proxy": {
"id": "ElasticKeywordsSearch-31Et9",
"field": "embedding"
}
},
"metadatas_ElasticKeywordsSearch-31Et9": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas",
"proxy": {
"id": "ElasticKeywordsSearch-31Et9",
"field": "metadatas"
}
},
"index_name_ElasticKeywordsSearch-31Et9": {
"l2": true,
"info": "",
"list": false,
"name": "index_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "index_name",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_name",
"proxy": {
"id": "ElasticKeywordsSearch-31Et9",
"field": "index_name"
}
},
"ssl_verify_ElasticKeywordsSearch-31Et9": {
"l2": false,
"info": "",
"list": false,
"name": "ssl_verify",
"show": true,
"type": "str",
"value": "",
"l2_name": "ssl_verify",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ssl_verify",
"proxy": {
"id": "ElasticKeywordsSearch-31Et9",
"field": "ssl_verify"
}
},
"search_kwargs_ElasticKeywordsSearch-31Et9": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": false,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs",
"proxy": {
"id": "ElasticKeywordsSearch-31Et9",
"field": "search_kwargs"
}
},
"refresh_indices_ElasticKeywordsSearch-31Et9": {
"l2": false,
"info": "",
"list": false,
"name": "refresh_indices",
"show": false,
"type": "bool",
"value": true,
"l2_name": "refresh_indices",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "refresh_indices",
"proxy": {
"id": "ElasticKeywordsSearch-31Et9",
"field": "refresh_indices"
}
},
"elasticsearch_url_ElasticKeywordsSearch-31Et9": {
"l2": false,
"info": "",
"list": false,
"name": "elasticsearch_url",
"show": true,
"type": "str",
"value": "",
"l2_name": "elasticsearch_url",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ES_connection_url",
"proxy": {
"id": "ElasticKeywordsSearch-31Et9",
"field": "elasticsearch_url"
}
},
"verbose_ConversationalRetrievalChain-SmY3w": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose",
"proxy": {
"id": "ConversationalRetrievalChain-SmY3w",
"field": "verbose"
}
},
"callbacks_ConversationalRetrievalChain-SmY3w": {
"l2": false,
"info": "",
"list": false,
"name": "callbacks",
"show": false,
"type": "Callbacks",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks",
"proxy": {
"id": "ConversationalRetrievalChain-SmY3w",
"field": "callbacks"
}
},
"chain_type_ConversationalRetrievalChain-SmY3w": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "chain_type",
"proxy": {
"id": "ConversationalRetrievalChain-SmY3w",
"field": "chain_type"
}
},
"input_node_ConversationalRetrievalChain-SmY3w": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question",
"proxy": {
"id": "ConversationalRetrievalChain-SmY3w",
"field": "input_node"
}
},
"condense_question_llm_ConversationalRetrievalChain-SmY3w": {
"l2": false,
"info": "",
"list": false,
"name": "condense_question_llm",
"show": false,
"type": "BaseLanguageModel",
"l2_name": "condense_question_llm",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "condense_question_llm",
"proxy": {
"id": "ConversationalRetrievalChain-SmY3w",
"field": "condense_question_llm"
}
},
"return_source_documents_ConversationalRetrievalChain-SmY3w": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Return source documents",
"proxy": {
"id": "ConversationalRetrievalChain-SmY3w",
"field": "return_source_documents"
}
},
"condense_question_prompt_ConversationalRetrievalChain-SmY3w": {
"l2": false,
"info": "",
"list": false,
"name": "condense_question_prompt",
"show": true,
"type": "BasePromptTemplate",
"value": {
"_type": "prompt",
"template": "Given the following conversation and a follow up question, rephrase the follow up question to be a standalone question, in its original language.\n\nChat History:\n{chat_history}\nFollow Up Input: {question}\nStandalone question:",
"output_parser": null,
"input_variables": [
"chat_history",
"question"
],
"template_format": "f-string",
"partial_variables": {},
"validate_template": true
},
"l2_name": "condense_question_prompt",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "condense_question_prompt",
"proxy": {
"id": "ConversationalRetrievalChain-SmY3w",
"field": "condense_question_prompt"
}
},
"ai_prefix_ConversationBufferMemory-K1tss": {
"l2": false,
"info": "",
"list": false,
"name": "ai_prefix",
"show": false,
"type": "str",
"value": "AI",
"l2_name": "ai_prefix",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ai_prefix",
"proxy": {
"id": "ConversationBufferMemory-K1tss",
"field": "ai_prefix"
}
},
"input_key_ConversationBufferMemory-K1tss": {
"l2": false,
"info": "The variable to be used as Chat Input when more than one variable is available.",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "input_key",
"proxy": {
"id": "ConversationBufferMemory-K1tss",
"field": "input_key"
}
},
"memory_key_ConversationBufferMemory-K1tss": {
"l2": false,
"info": "",
"list": false,
"name": "memory_key",
"show": true,
"type": "str",
"value": "chat_history",
"l2_name": "memory_key",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory_key",
"proxy": {
"id": "ConversationBufferMemory-K1tss",
"field": "memory_key"
}
},
"output_key_ConversationBufferMemory-K1tss": {
"l2": false,
"info": "The variable to be used as Chat Output (e.g. answer in a ConversationalRetrievalChain)",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "answer",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_key",
"proxy": {
"id": "ConversationBufferMemory-K1tss",
"field": "output_key"
}
},
"chat_memory_ConversationBufferMemory-K1tss": {
"l2": false,
"info": "",
"list": false,
"name": "chat_memory",
"show": true,
"type": "BaseChatMessageHistory",
"l2_name": "chat_memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "chat_memory",
"proxy": {
"id": "ConversationBufferMemory-K1tss",
"field": "chat_memory"
}
},
"human_prefix_ConversationBufferMemory-K1tss": {
"l2": false,
"info": "",
"list": false,
"name": "human_prefix",
"show": false,
"type": "str",
"value": "Human",
"l2_name": "human_prefix",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "human_prefix",
"proxy": {
"id": "ConversationBufferMemory-K1tss",
"field": "human_prefix"
}
},
"return_messages_ConversationBufferMemory-K1tss": {
"l2": false,
"info": "",
"list": false,
"name": "return_messages",
"show": true,
"type": "bool",
"l2_name": "return_messages",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_messages",
"proxy": {
"id": "ConversationBufferMemory-K1tss",
"field": "return_messages"
}
},
"template_PromptTemplate-bs0vj": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "参考文本：\n{context}\n\n----------------------\n\n1.用中文回答用户问题，并且答案要严谨专业。\n2.你需要依据以上【参考文本】中的内容来回答，当【参考文本】中有明确与用户问题相关的内容时才进行回答，不可根据自己的知识来回答。\n3.由于【参考文本】可能包含多个来自不同信息源的信息，所以根据这些不同的信息源可能得出有差异甚至冲突的答案，当发现这种情况时，这些答案都列举出来；如果没有冲突或差异，则只需要给出一个最终结果。\n4.若【参考文本】中内容与用户问题不相关则回复“没有找到相关内容”。\n\n----------------------\n\n用户问题: {question}\n你的回答:",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "template",
"proxy": {
"id": "PromptTemplate-bs0vj",
"field": "template"
}
},
"output_parser_PromptTemplate-bs0vj": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser",
"proxy": {
"id": "PromptTemplate-bs0vj",
"field": "output_parser"
}
},
"input_variables_PromptTemplate-bs0vj": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"context",
"question"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_variables",
"proxy": {
"id": "PromptTemplate-bs0vj",
"field": "input_variables"
}
},
"template_format_PromptTemplate-bs0vj": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "template_format",
"proxy": {
"id": "PromptTemplate-bs0vj",
"field": "template_format"
}
},
"partial_variables_PromptTemplate-bs0vj": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "partial_variables",
"proxy": {
"id": "PromptTemplate-bs0vj",
"field": "partial_variables"
}
},
"validate_template_PromptTemplate-bs0vj": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "validate_template",
"proxy": {
"id": "PromptTemplate-bs0vj",
"field": "validate_template"
}
},
"callbacks_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "callbacks",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"list": true,
"file_path": "",
"display_name": "Callbacks",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "callbacks"
}
},
"async_client_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "async_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Async Client",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "async_client"
}
},
"cache_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": true,
"file_path": "",
"password": false,
"name": "cache",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"display_name": "Cache",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "cache"
}
},
"client_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Client",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "client"
}
},
"default_headers_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_headers",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Headers",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "default_headers"
}
},
"default_query_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_query",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Query",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "default_query"
}
},
"http_client_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "http_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Http Client",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "http_client"
}
},
"max_retries_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 2,
"password": false,
"name": "max_retries",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Retries",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "max_retries"
}
},
"max_tokens_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": true,
"name": "max_tokens",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Tokens",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "max_tokens"
},
"value": ""
},
"metadata_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "metadata",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Metadata",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "metadata"
}
},
"model_kwargs_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "model_kwargs",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Model Kwargs",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "model_kwargs"
}
},
"model_name_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "gpt-3.5-turbo-0613",
"password": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"name": "model_name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Model Name",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "model_name"
}
},
"n_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 1,
"password": false,
"name": "n",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "N",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "n"
}
},
"name_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Name",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "name"
}
},
"openai_api_base_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_api_base",
"display_name": "OpenAI API Base",
"advanced": true,
"dynamic": false,
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "openai_api_base"
}
},
"openai_api_key_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "",
"password": true,
"name": "openai_api_key",
"display_name": "OpenAI API Key",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "openai_api_key"
}
},
"openai_organization_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "openai_organization",
"display_name": "OpenAI Organization",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "openai_organization"
}
},
"openai_proxy_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_proxy",
"display_name": "OpenAI Proxy",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "openai_proxy"
}
},
"request_timeout_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "request_timeout",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Request Timeout",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "request_timeout"
}
},
"streaming_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": false,
"password": false,
"name": "streaming",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Streaming",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "streaming"
}
},
"tags_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tags",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Tags",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "tags"
}
},
"temperature_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": 0.7,
"password": false,
"name": "temperature",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Temperature",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "temperature"
}
},
"tiktoken_model_name_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tiktoken_model_name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Tiktoken Model Name",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "tiktoken_model_name"
}
},
"verbose_ChatOpenAI-029a9": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "verbose",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Verbose",
"proxy": {
"id": "ChatOpenAI-029a9",
"field": "verbose"
}
}
},
"flow": {
"data": {
"nodes": [
{
"width": 384,
"height": 350,
"id": "MixEsVectorRetriever-J35CZ",
"type": "genericNode",
"position": {
"x": 1146.2998776479994,
"y": -505.22140452358315
},
"data": {
"id": "MixEsVectorRetriever-J35CZ",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": true,
"type": "str",
"l2_name": "tags",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags"
},
"_type": "MixEsVectorRetriever",
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"l2_name": "metadata",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata"
},
"combine_strategy": {
"l2": false,
"info": "",
"list": true,
"name": "combine_strategy",
"show": true,
"type": "str",
"value": "mix",
"l2_name": "combine_strategy",
"options": [
"keyword_front",
"vector_front",
"mix"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "combine_strategy"
},
"vector_retriever": {
"l2": false,
"info": "",
"list": false,
"name": "vector_retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "vector_retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"keyword_retriever": {
"l2": false,
"info": "",
"list": false,
"name": "keyword_retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "keyword_retriever",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": "",
"value": ""
}
},
"description": "This class ensemble the results of es retriever and vector retriever.",
"base_classes": [
"BaseRetriever",
"MixEsVectorRetriever"
],
"display_name": "MixEsVectorRetriever",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "MixEsVectorRetriever",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1146.2998776479994,
"y": -505.22140452358315
},
"dragging": false
},
{
"width": 384,
"height": 428,
"id": "Milvus-cyR5W",
"type": "genericNode",
"position": {
"x": 543.3992621716293,
"y": -94.57897454061174
},
"data": {
"id": "Milvus-cyR5W",
"node": {
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "drop_old"
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas"
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_params"
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs"
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_params"
},
"collection_name": {
"l2": true,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "collection_name",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "collection_name"
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "connection_args"
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "consistency_level"
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"VectorStore",
"Milvus",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 543.3992621716293,
"y": -94.57897454061174
},
"dragging": false
},
{
"width": 384,
"height": 510,
"id": "ElasticKeywordsSearch-31Et9",
"type": "genericNode",
"position": {
"x": 545.7343868015423,
"y": -596.1314471305373
},
"data": {
"id": "ElasticKeywordsSearch-31Et9",
"node": {
"template": {
"ids": {
"l2": false,
"info": "",
"list": true,
"name": "ids",
"show": false,
"type": "str",
"l2_name": "ids",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ids"
},
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"type": "BaseLLM",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "ElasticKeywordsSearch",
"prompt": {
"l2": false,
"info": "",
"list": false,
"name": "prompt",
"type": "BasePromptTemplate",
"l2_name": "prompt",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas"
},
"index_name": {
"l2": true,
"info": "",
"list": false,
"name": "index_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "index_name",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_name"
},
"ssl_verify": {
"l2": false,
"info": "",
"list": false,
"name": "ssl_verify",
"show": true,
"type": "str",
"value": "",
"l2_name": "ssl_verify",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ssl_verify"
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs"
},
"refresh_indices": {
"l2": false,
"info": "",
"list": false,
"name": "refresh_indices",
"show": false,
"type": "bool",
"value": true,
"l2_name": "refresh_indices",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "refresh_indices"
},
"elasticsearch_url": {
"l2": false,
"info": "",
"list": false,
"name": "elasticsearch_url",
"show": true,
"type": "str",
"value": "",
"l2_name": "elasticsearch_url",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ES_connection_url"
}
},
"description": "Construct ElasticKeywordsSearch wrapper from raw documents.",
"base_classes": [
"VectorStore",
"ElasticKeywordsSearch",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "ElasticKeywordsSearch",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "ElasticKeywordsSearch",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 545.7343868015423,
"y": -596.1314471305373
},
"dragging": false
},
{
"width": 384,
"height": 494,
"id": "ConversationalRetrievalChain-SmY3w",
"type": "genericNode",
"position": {
"x": 2006.6489580098225,
"y": -918.5048523096177
},
"data": {
"id": "ConversationalRetrievalChain-SmY3w",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"_type": "ConversationalRetrievalChain",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseChatMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose"
},
"callbacks": {
"l2": false,
"info": "",
"list": false,
"name": "callbacks",
"show": false,
"type": "Callbacks",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks"
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "chain_type"
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"condense_question_llm": {
"l2": false,
"info": "",
"list": false,
"name": "condense_question_llm",
"show": false,
"type": "BaseLanguageModel",
"l2_name": "condense_question_llm",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "condense_question_llm"
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Return source documents"
},
"condense_question_prompt": {
"l2": false,
"info": "",
"list": false,
"name": "condense_question_prompt",
"show": true,
"type": "BasePromptTemplate",
"value": {
"_type": "prompt",
"template": "Given the following conversation and a follow up question, rephrase the follow up question to be a standalone question, in its original language.\n\nChat History:\n{chat_history}\nFollow Up Input: {question}\nStandalone question:",
"output_parser": null,
"input_variables": [
"chat_history",
"question"
],
"template_format": "f-string",
"partial_variables": {},
"validate_template": true
},
"l2_name": "condense_question_prompt",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "condense_question_prompt"
},
"combine_docs_chain_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "combine_docs_chain_kwargs",
"show": true,
"type": "BasePromptTemplate",
"l2_name": "combine_docs_chain_kwargs",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
}
},
"description": "Convenience method to load chain from LLM and retriever.",
"base_classes": [
"BaseConversationalRetrievalChain",
"Chain",
"ConversationalRetrievalChain",
"function"
],
"display_name": "ConversationalRetrievalChain",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/chat_vector_db"
},
"type": "ConversationalRetrievalChain",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 2006.6489580098225,
"y": -918.5048523096177
},
"dragging": false
},
{
"width": 384,
"height": 526,
"id": "ConversationBufferMemory-K1tss",
"type": "genericNode",
"position": {
"x": 1134.0608104034231,
"y": -1084.2478524136263
},
"data": {
"id": "ConversationBufferMemory-K1tss",
"node": {
"template": {
"_type": "ConversationBufferMemory",
"ai_prefix": {
"l2": false,
"info": "",
"list": false,
"name": "ai_prefix",
"show": false,
"type": "str",
"value": "AI",
"l2_name": "ai_prefix",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ai_prefix"
},
"input_key": {
"l2": false,
"info": "The variable to be used as Chat Input when more than one variable is available.",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "",
"l2_name": "input_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "input_key"
},
"memory_key": {
"l2": false,
"info": "",
"list": false,
"name": "memory_key",
"show": true,
"type": "str",
"value": "chat_history",
"l2_name": "memory_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory_key"
},
"output_key": {
"l2": false,
"info": "The variable to be used as Chat Output (e.g. answer in a ConversationalRetrievalChain)",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "answer",
"l2_name": "output_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_key"
},
"chat_memory": {
"l2": false,
"info": "",
"list": false,
"name": "chat_memory",
"show": true,
"type": "BaseChatMessageHistory",
"l2_name": "chat_memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "chat_memory"
},
"human_prefix": {
"l2": false,
"info": "",
"list": false,
"name": "human_prefix",
"show": false,
"type": "str",
"value": "Human",
"l2_name": "human_prefix",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "human_prefix"
},
"return_messages": {
"l2": false,
"info": "",
"list": false,
"name": "return_messages",
"show": true,
"type": "bool",
"l2_name": "return_messages",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_messages"
}
},
"description": "Buffer for storing conversation memory.",
"base_classes": [
"BaseMemory",
"ConversationBufferMemory",
"BaseChatMemory"
],
"display_name": "ConversationBufferMemory",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/memory/how_to/buffer"
},
"type": "ConversationBufferMemory",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1134.0608104034231,
"y": -1084.2478524136263
},
"dragging": false
},
{
"width": 384,
"height": 290,
"id": "PromptTemplate-bs0vj",
"type": "genericNode",
"position": {
"x": 1129.186016282717,
"y": -1612.3505488234584
},
"data": {
"id": "PromptTemplate-bs0vj",
"node": {
"name": "",
"template": {
"_type": "PromptTemplate",
"template": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "参考文本：\n{context}\n\n----------------------\n\n1.用中文回答用户问题，并且答案要严谨专业。\n2.你需要依据以上【参考文本】中的内容来回答，当【参考文本】中有明确与用户问题相关的内容时才进行回答，不可根据自己的知识来回答。\n3.由于【参考文本】可能包含多个来自不同信息源的信息，所以根据这些不同的信息源可能得出有差异甚至冲突的答案，当发现这种情况时，这些答案都列举出来；如果没有冲突或差异，则只需要给出一个最终结果。\n4.若【参考文本】中内容与用户问题不相关则回复“没有找到相关内容”。\n\n----------------------\n\n用户问题: {question}\n你的回答:",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "template"
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser"
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"context",
"question"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_variables"
},
"template_format": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "template_format"
},
"partial_variables": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "partial_variables"
},
"validate_template": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "validate_template"
}
},
"description": "Schema to represent a prompt for an LLM.",
"base_classes": [
"BasePromptTemplate",
"PromptTemplate",
"StringPromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"context",
"question"
],
"template": [
"context",
"question"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {
"formatters": {
"openai_api_key": {}
},
"base_formatters": {
"dict": {},
"list": {},
"show": {},
"union": {},
"kwargs": {},
"default": {},
"headers": {},
"optional": {},
"password": {},
"multiline": {},
"model_fields": {
"MODEL_DICT": {
"OpenAI": [
"text-davinci-003",
"text-davinci-002",
"text-curie-001",
"text-babbage-001",
"text-ada-001"
],
"Anthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
],
"ChatOpenAI": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k"
],
"ChatAnthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
]
}
},
"dict_code_file": {}
}
}
},
"type": "PromptTemplate",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1129.186016282717,
"y": -1612.3505488234584
},
"dragging": false
},
{
"width": 384,
"height": 606,
"id": "ChatOpenAI-029a9",
"type": "genericNode",
"position": {
"x": 520.7277052724213,
"y": -1262.5347130150906
},
"data": {
"type": "ChatOpenAI",
"node": {
"template": {
"callbacks": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "callbacks",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"list": true,
"file_path": "",
"display_name": "Callbacks"
},
"async_client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "async_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Async Client"
},
"cache": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": true,
"file_path": "",
"password": false,
"name": "cache",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"display_name": "Cache"
},
"client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Client"
},
"default_headers": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_headers",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Headers"
},
"default_query": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_query",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Query"
},
"http_client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "http_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Http Client"
},
"max_retries": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 2,
"password": false,
"name": "max_retries",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Retries"
},
"max_tokens": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": true,
"name": "max_tokens",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Tokens"
},
"metadata": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "metadata",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Metadata"
},
"model_kwargs": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "model_kwargs",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Model Kwargs"
},
"model_name": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "gpt-3.5-turbo-0613",
"password": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"name": "model_name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Model Name"
},
"n": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 1,
"password": false,
"name": "n",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "N"
},
"name": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Name"
},
"openai_api_base": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_api_base",
"display_name": "OpenAI API Base",
"advanced": false,
"dynamic": false,
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_api_key": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "",
"password": true,
"name": "openai_api_key",
"display_name": "OpenAI API Key",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_organization": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "openai_organization",
"display_name": "OpenAI Organization",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_proxy": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_proxy",
"display_name": "OpenAI Proxy",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"request_timeout": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "request_timeout",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Request Timeout"
},
"streaming": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": false,
"password": false,
"name": "streaming",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Streaming"
},
"tags": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tags",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Tags"
},
"temperature": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": 0.7,
"password": false,
"name": "temperature",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Temperature"
},
"tiktoken_model_name": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tiktoken_model_name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Tiktoken Model Name"
},
"verbose": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "verbose",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Verbose"
},
"_type": "ChatOpenAI"
},
"description": "[*Deprecated*] `OpenAI` Chat large language models API.",
"base_classes": [
"ChatOpenAI",
"BaseLanguageModel",
"BaseChatModel",
"BaseLLM"
],
"display_name": "ChatOpenAI",
"documentation": "https://python.langchain.com/docs/modules/model_io/models/chat/integrations/openai",
"custom_fields": {},
"output_types": [],
"field_formatters": {},
"beta": false
},
"id": "ChatOpenAI-029a9",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 520.7277052724213,
"y": -1262.5347130150906
},
"dragging": false
}
],
"edges": [
{
"source": "Milvus-cyR5W",
"target": "MixEsVectorRetriever-J35CZ",
"sourceHandle": "Milvus|Milvus-cyR5W|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|vector_retriever|MixEsVectorRetriever-J35CZ",
"id": "reactflow__edge-Milvus-cyR5WMilvus|Milvus-cyR5W|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever-MixEsVectorRetriever-J35CZBaseRetriever|vector_retriever|MixEsVectorRetriever-J35CZ",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "ElasticKeywordsSearch-31Et9",
"target": "MixEsVectorRetriever-J35CZ",
"sourceHandle": "ElasticKeywordsSearch|ElasticKeywordsSearch-31Et9|VectorStore|ElasticKeywordsSearch|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|keyword_retriever|MixEsVectorRetriever-J35CZ",
"id": "reactflow__edge-ElasticKeywordsSearch-31Et9ElasticKeywordsSearch|ElasticKeywordsSearch-31Et9|VectorStore|ElasticKeywordsSearch|BaseRetriever|VectorStoreRetriever-MixEsVectorRetriever-J35CZBaseRetriever|keyword_retriever|MixEsVectorRetriever-J35CZ",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "MixEsVectorRetriever-J35CZ",
"target": "ConversationalRetrievalChain-SmY3w",
"sourceHandle": "MixEsVectorRetriever|MixEsVectorRetriever-J35CZ|BaseRetriever|MixEsVectorRetriever",
"targetHandle": "BaseRetriever|retriever|ConversationalRetrievalChain-SmY3w",
"id": "reactflow__edge-MixEsVectorRetriever-J35CZMixEsVectorRetriever|MixEsVectorRetriever-J35CZ|BaseRetriever|MixEsVectorRetriever-ConversationalRetrievalChain-SmY3wBaseRetriever|retriever|ConversationalRetrievalChain-SmY3w",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "ConversationBufferMemory-K1tss",
"target": "ConversationalRetrievalChain-SmY3w",
"sourceHandle": "ConversationBufferMemory|ConversationBufferMemory-K1tss|BaseMemory|ConversationBufferMemory|BaseChatMemory",
"targetHandle": "BaseChatMemory|memory|ConversationalRetrievalChain-SmY3w",
"id": "reactflow__edge-ConversationBufferMemory-K1tssConversationBufferMemory|ConversationBufferMemory-K1tss|BaseMemory|ConversationBufferMemory|BaseChatMemory-ConversationalRetrievalChain-SmY3wBaseChatMemory|memory|ConversationalRetrievalChain-SmY3w",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "PromptTemplate-bs0vj",
"target": "ConversationalRetrievalChain-SmY3w",
"sourceHandle": "PromptTemplate|PromptTemplate-bs0vj|BasePromptTemplate|PromptTemplate|StringPromptTemplate",
"targetHandle": "BasePromptTemplate|combine_docs_chain_kwargs|ConversationalRetrievalChain-SmY3w",
"id": "reactflow__edge-PromptTemplate-bs0vjPromptTemplate|PromptTemplate-bs0vj|BasePromptTemplate|PromptTemplate|StringPromptTemplate-ConversationalRetrievalChain-SmY3wBasePromptTemplate|combine_docs_chain_kwargs|ConversationalRetrievalChain-SmY3w",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "ChatOpenAI-029a9",
"sourceHandle": "ChatOpenAI|ChatOpenAI-029a9|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM",
"target": "ConversationalRetrievalChain-SmY3w",
"targetHandle": "BaseLanguageModel|llm|ConversationalRetrievalChain-SmY3w",
"style": {
"stroke": "#555"
},
"className": "stroke-foreground stroke-connection",
"animated": true,
"id": "reactflow__edge-ChatOpenAI-029a9ChatOpenAI|ChatOpenAI-029a9|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM-ConversationalRetrievalChain-SmY3wBaseLanguageModel|llm|ConversationalRetrievalChain-SmY3w",
"selected": true
}
],
"viewport": {
"zoom": 1,
"x": 0,
"y": 0
}
},
"is_component": false,
"name": "",
"description": "",
"id": "24ac8",
"status": 0,
"write": false,
"guide_word": ""
}
}
}
},
{
"知识库问答": {
"id": "RetrievalQA-517cc",
"type": "RetrievalQA",
"node": {
"output_types": [],
"display_name": "知识库问答",
"documentation": "",
"base_classes": [
"RetrievalQA",
"Chain",
"BaseRetrievalQA",
"function"
],
"description": "Chain for question-answering against an index.",
"template": {
"drop_old_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "drop_old",
"proxy": {
"id": "Milvus-CWpae",
"field": "drop_old"
}
},
"documents_Milvus-CWpae": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents",
"proxy": {
"id": "Milvus-CWpae",
"field": "documents"
}
},
"embedding_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding",
"proxy": {
"id": "Milvus-CWpae",
"field": "embedding"
}
},
"metadatas_Milvus-CWpae": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas",
"proxy": {
"id": "Milvus-CWpae",
"field": "metadatas"
}
},
"index_params_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_params",
"proxy": {
"id": "Milvus-CWpae",
"field": "index_params"
}
},
"search_kwargs_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": false,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs",
"proxy": {
"id": "Milvus-CWpae",
"field": "search_kwargs"
}
},
"search_params_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_params",
"proxy": {
"id": "Milvus-CWpae",
"field": "search_params"
}
},
"collection_name_Milvus-CWpae": {
"l2": true,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "选择知识库",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "collection_name",
"proxy": {
"id": "Milvus-CWpae",
"field": "collection_name"
}
},
"connection_args_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": false,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "connection_args",
"proxy": {
"id": "Milvus-CWpae",
"field": "connection_args"
}
},
"consistency_level_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "consistency_level",
"proxy": {
"id": "Milvus-CWpae",
"field": "consistency_level"
}
},
"tags_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "tags"
}
},
"memory_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "memory"
}
},
"verbose_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "verbose"
}
},
"metadata_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "metadata"
}
},
"callbacks_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "callbacks"
}
},
"input_key_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_key",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "input_key"
}
},
"input_node_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "input_node"
}
},
"output_key_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "output_key",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "output_key"
}
},
"return_source_documents_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_source_documents",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "return_source_documents"
}
},
"token_max_CombineDocsChain-71a85": {
"info": "只对Stuff类型生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max",
"proxy": {
"id": "CombineDocsChain-71a85",
"field": "token_max"
}
},
"chain_type_CombineDocsChain-71a85": {
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "chain_type",
"proxy": {
"id": "CombineDocsChain-71a85",
"field": "chain_type"
}
},
"source_PromptTemplate-3c164": {
"info": "",
"list": false,
"name": "source",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "source",
"proxy": {
"id": "PromptTemplate-3c164",
"field": "source"
}
},
"template_PromptTemplate-3c164": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "title: {source}\n{page_content}\n----------------",
"l2_name": "提示词模板",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "template",
"proxy": {
"id": "PromptTemplate-3c164",
"field": "template"
}
},
"input_types_PromptTemplate-3c164": {
"info": "",
"list": false,
"name": "input_types",
"show": false,
"type": "code",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "input_types",
"proxy": {
"id": "PromptTemplate-3c164",
"field": "input_types"
}
},
"page_content_PromptTemplate-3c164": {
"info": "",
"list": false,
"name": "page_content",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "page_content",
"proxy": {
"id": "PromptTemplate-3c164",
"field": "page_content"
}
},
"output_parser_PromptTemplate-3c164": {
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser",
"proxy": {
"id": "PromptTemplate-3c164",
"field": "output_parser"
}
},
"input_variables_PromptTemplate-3c164": {
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"source",
"page_content"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_variables",
"proxy": {
"id": "PromptTemplate-3c164",
"field": "input_variables"
}
},
"template_format_PromptTemplate-3c164": {
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "Union[Literal['f-string'], Literal['jinja2']]",
"value": "f-string",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "template_format",
"proxy": {
"id": "PromptTemplate-3c164",
"field": "template_format"
}
},
"partial_variables_PromptTemplate-3c164": {
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "partial_variables",
"proxy": {
"id": "PromptTemplate-3c164",
"field": "partial_variables"
}
},
"validate_template_PromptTemplate-3c164": {
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": false,
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "validate_template",
"proxy": {
"id": "PromptTemplate-3c164",
"field": "validate_template"
}
},
"context_PromptTemplate-ced29": {
"info": "",
"list": false,
"name": "context",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "context",
"proxy": {
"id": "PromptTemplate-ced29",
"field": "context"
}
},
"question_PromptTemplate-ced29": {
"info": "",
"list": false,
"name": "question",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "question",
"proxy": {
"id": "PromptTemplate-ced29",
"field": "question"
}
},
"template_PromptTemplate-ced29": {
"l2": true,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "参考文本：\n{context}\n\n----------------------\n# Role：企业客服\n\n# Background :\n你是一家名叫XX公司的客服人员。你的日常工作是回复公司内用户的各类咨询问题，所以当用户提到“公司”时，指的是XX公司。\n\n# Goals:\n- 以上【参考文本】中是从企业知识库中查找到的相关信息，只能结合以上信息进行回答，若以上内容为空或其中没有找到能回答【用户问题】的内容时，则回复“没有找到相关内容”，不能根据你自己的知识自己发挥。\n- 特别要注意区分【用户问题】与【参考文本】中不同的日期、人名、公司名这些关键信息。\n- 用中文回答问题，并且答案要严谨、专业、清晰、可读性好。\n- 拥有排版审美, 会利用序号, 缩进, 分隔线和换行符等等来美化信息排版。\n\n----------------------\n\n用户问题: {question}\n你的回答:",
"l2_name": "提示词模板",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "template",
"proxy": {
"id": "PromptTemplate-ced29",
"field": "template"
}
},
"input_types_PromptTemplate-ced29": {
"info": "",
"list": false,
"name": "input_types",
"show": false,
"type": "code",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "input_types",
"proxy": {
"id": "PromptTemplate-ced29",
"field": "input_types"
}
},
"output_parser_PromptTemplate-ced29": {
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser",
"proxy": {
"id": "PromptTemplate-ced29",
"field": "output_parser"
}
},
"input_variables_PromptTemplate-ced29": {
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"context",
"question"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_variables",
"proxy": {
"id": "PromptTemplate-ced29",
"field": "input_variables"
}
},
"template_format_PromptTemplate-ced29": {
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "Union[Literal['f-string'], Literal['jinja2']]",
"value": "f-string",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "template_format",
"proxy": {
"id": "PromptTemplate-ced29",
"field": "template_format"
}
},
"partial_variables_PromptTemplate-ced29": {
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "partial_variables",
"proxy": {
"id": "PromptTemplate-ced29",
"field": "partial_variables"
}
},
"validate_template_PromptTemplate-ced29": {
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": false,
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "validate_template",
"proxy": {
"id": "PromptTemplate-ced29",
"field": "validate_template"
}
},
"callbacks_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "callbacks",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"list": true,
"file_path": "",
"display_name": "Callbacks",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "callbacks"
}
},
"async_client_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "async_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Async Client",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "async_client"
}
},
"cache_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": true,
"file_path": "",
"password": false,
"name": "cache",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"display_name": "Cache",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "cache"
}
},
"client_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Client",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "client"
}
},
"default_headers_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_headers",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Headers",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "default_headers"
}
},
"default_query_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_query",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Query",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "default_query"
}
},
"http_client_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "http_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Http Client",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "http_client"
}
},
"max_retries_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 2,
"password": false,
"name": "max_retries",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Retries",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "max_retries"
}
},
"max_tokens_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": true,
"name": "max_tokens",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Tokens",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "max_tokens"
},
"value": ""
},
"metadata_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "metadata",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Metadata",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "metadata"
}
},
"model_kwargs_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "model_kwargs",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Model Kwargs",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "model_kwargs"
}
},
"model_name_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "gpt-3.5-turbo-0613",
"password": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"name": "model_name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Model Name",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "model_name"
}
},
"n_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 1,
"password": false,
"name": "n",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "N",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "n"
}
},
"name_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Name",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "name"
}
},
"openai_api_base_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_api_base",
"display_name": "OpenAI API Base",
"advanced": true,
"dynamic": false,
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "openai_api_base"
}
},
"openai_api_key_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "",
"password": true,
"name": "openai_api_key",
"display_name": "OpenAI API Key",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "openai_api_key"
}
},
"openai_organization_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "openai_organization",
"display_name": "OpenAI Organization",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "openai_organization"
}
},
"openai_proxy_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_proxy",
"display_name": "OpenAI Proxy",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "openai_proxy"
}
},
"request_timeout_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "request_timeout",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Request Timeout",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "request_timeout"
}
},
"streaming_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": false,
"password": false,
"name": "streaming",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Streaming",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "streaming"
}
},
"tags_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tags",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Tags",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "tags"
}
},
"temperature_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": 0.7,
"password": false,
"name": "temperature",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Temperature",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "temperature"
}
},
"tiktoken_model_name_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tiktoken_model_name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Tiktoken Model Name",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "tiktoken_model_name"
}
},
"verbose_ChatOpenAI-7a446": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "verbose",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Verbose",
"proxy": {
"id": "ChatOpenAI-7a446",
"field": "verbose"
}
}
},
"flow": {
"data": {
"nodes": [
{
"width": 384,
"height": 346,
"id": "Milvus-CWpae",
"type": "genericNode",
"position": {
"x": 1146.2686160133917,
"y": -392.08641987984464
},
"data": {
"id": "Milvus-CWpae",
"node": {
"l2_name": "知识库",
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "drop_old"
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas"
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_params"
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs"
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_params"
},
"collection_name": {
"l2": true,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "选择知识库",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "collection_name"
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "connection_args"
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "consistency_level"
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"Milvus",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1146.2686160133917,
"y": -392.08641987984464
},
"dragging": false
},
{
"width": 384,
"height": 316,
"id": "RetrievalQA-x5nap",
"type": "genericNode",
"position": {
"x": 1809.8484174920595,
"y": -770.5069621032657
},
"data": {
"id": "RetrievalQA-x5nap",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags"
},
"_type": "RetrievalQA",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory"
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose"
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata"
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks"
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_key"
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "output_key"
},
"combine_documents_chain": {
"l2": false,
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"l2_name": "combine_documents_chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_source_documents"
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"RetrievalQA",
"Chain",
"BaseRetrievalQA",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1809.8484174920595,
"y": -770.5069621032657
}
},
{
"width": 384,
"height": 448,
"id": "CombineDocsChain-71a85",
"type": "genericNode",
"position": {
"x": 1135.690733661112,
"y": -1047.3523465428725
},
"data": {
"id": "CombineDocsChain-71a85",
"node": {
"template": {
"llm": {
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"prompt": {
"info": "只对Stuff类型生效",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "prompt"
},
"token_max": {
"info": "只对Stuff类型生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "chain_type"
},
"document_prompt": {
"info": "",
"list": false,
"name": "document_prompt",
"show": true,
"type": "BasePromptTemplate",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1135.690733661112,
"y": -1047.3523465428725
}
},
{
"width": 384,
"height": 454,
"id": "PromptTemplate-3c164",
"type": "genericNode",
"position": {
"x": 538.127581209942,
"y": -1530.8752798629753
},
"data": {
"id": "PromptTemplate-3c164",
"node": {
"name": "",
"l2_name": "召回段落拼接提示词",
"template": {
"_type": "PromptTemplate",
"source": {
"info": "",
"list": false,
"name": "source",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "source"
},
"template": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "title: {source}\n{page_content}\n----------------",
"l2_name": "提示词模板",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "template"
},
"input_types": {
"info": "",
"list": false,
"name": "input_types",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "input_types"
},
"page_content": {
"info": "",
"list": false,
"name": "page_content",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "page_content"
},
"output_parser": {
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser"
},
"input_variables": {
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"source",
"page_content"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_variables"
},
"template_format": {
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "Union[Literal['f-string'], Literal['jinja2']]",
"value": "f-string",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "template_format"
},
"partial_variables": {
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "partial_variables"
},
"validate_template": {
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "validate_template"
}
},
"description": "A prompt template for a language model.",
"base_classes": [
"Runnable",
"StringPromptTemplate",
"RunnableSerializable",
"PromptTemplate",
"Generic",
"BasePromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"source",
"page_content"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {}
},
"type": "PromptTemplate",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 538.127581209942,
"y": -1530.8752798629753
}
},
{
"width": 384,
"height": 454,
"id": "PromptTemplate-ced29",
"type": "genericNode",
"position": {
"x": 528.5375319975674,
"y": -1031.4149446082283
},
"data": {
"id": "PromptTemplate-ced29",
"node": {
"name": "",
"l2_name": "大模型问答提示词",
"template": {
"_type": "PromptTemplate",
"context": {
"info": "",
"list": false,
"name": "context",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "context"
},
"question": {
"info": "",
"list": false,
"name": "question",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser",
"VariableNode"
],
"placeholder": "",
"display_name": "question"
},
"template": {
"l2": true,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "参考文本：\n{context}\n\n----------------------\n# Role：企业客服\n\n# Background :\n你是一家名叫XX公司的客服人员。你的日常工作是回复公司内用户的各类咨询问题，所以当用户提到“公司”时，指的是XX公司。\n\n# Goals:\n- 以上【参考文本】中是从企业知识库中查找到的相关信息，只能结合以上信息进行回答，若以上内容为空或其中没有找到能回答【用户问题】的内容时，则回复“没有找到相关内容”，不能根据你自己的知识自己发挥。\n- 特别要注意区分【用户问题】与【参考文本】中不同的日期、人名、公司名这些关键信息。\n- 用中文回答问题，并且答案要严谨、专业、清晰、可读性好。\n- 拥有排版审美, 会利用序号, 缩进, 分隔线和换行符等等来美化信息排版。\n\n----------------------\n\n用户问题: {question}\n你的回答:",
"l2_name": "提示词模板",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "template"
},
"input_types": {
"info": "",
"list": false,
"name": "input_types",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "input_types"
},
"output_parser": {
"info": "",
"list": false,
"name": "output_parser",
"show": true,
"type": "BaseOutputParser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser"
},
"input_variables": {
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"context",
"question"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_variables"
},
"template_format": {
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "Union[Literal['f-string'], Literal['jinja2']]",
"value": "f-string",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "template_format"
},
"partial_variables": {
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "partial_variables"
},
"validate_template": {
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": false,
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "validate_template"
}
},
"description": "A prompt template for a language model.",
"base_classes": [
"Runnable",
"StringPromptTemplate",
"RunnableSerializable",
"PromptTemplate",
"Generic",
"BasePromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"": [
"context",
"question"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {}
},
"type": "PromptTemplate",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 528.5375319975674,
"y": -1031.4149446082283
}
},
{
"width": 384,
"height": 606,
"id": "ChatOpenAI-7a446",
"type": "genericNode",
"position": {
"x": 524.9021453356706,
"y": -519.7914736349932
},
"data": {
"type": "ChatOpenAI",
"node": {
"template": {
"callbacks": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "callbacks",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"list": true,
"file_path": "",
"display_name": "Callbacks"
},
"async_client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "async_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Async Client"
},
"cache": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": true,
"file_path": "",
"password": false,
"name": "cache",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"display_name": "Cache"
},
"client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Client"
},
"default_headers": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_headers",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Headers"
},
"default_query": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_query",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Query"
},
"http_client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "http_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Http Client"
},
"max_retries": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 2,
"password": false,
"name": "max_retries",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Retries"
},
"max_tokens": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": true,
"name": "max_tokens",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Tokens"
},
"metadata": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "metadata",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Metadata"
},
"model_kwargs": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "model_kwargs",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Model Kwargs"
},
"model_name": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "gpt-3.5-turbo-0613",
"password": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"name": "model_name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Model Name"
},
"n": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 1,
"password": false,
"name": "n",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "N"
},
"name": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Name"
},
"openai_api_base": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_api_base",
"display_name": "OpenAI API Base",
"advanced": false,
"dynamic": false,
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_api_key": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "",
"password": true,
"name": "openai_api_key",
"display_name": "OpenAI API Key",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_organization": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "openai_organization",
"display_name": "OpenAI Organization",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_proxy": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_proxy",
"display_name": "OpenAI Proxy",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"request_timeout": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "request_timeout",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Request Timeout"
},
"streaming": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": false,
"password": false,
"name": "streaming",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Streaming"
},
"tags": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tags",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Tags"
},
"temperature": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": 0.7,
"password": false,
"name": "temperature",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Temperature"
},
"tiktoken_model_name": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tiktoken_model_name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Tiktoken Model Name"
},
"verbose": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "verbose",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Verbose"
},
"_type": "ChatOpenAI"
},
"description": "[*Deprecated*] `OpenAI` Chat large language models API.",
"base_classes": [
"ChatOpenAI",
"BaseLanguageModel",
"BaseChatModel",
"BaseLLM"
],
"display_name": "ChatOpenAI",
"documentation": "https://python.langchain.com/docs/modules/model_io/models/chat/integrations/openai",
"custom_fields": {},
"output_types": [],
"field_formatters": {},
"beta": false
},
"id": "ChatOpenAI-7a446",
"value": null
},
"selected": true,
"dragging": false,
"positionAbsolute": {
"x": 524.9021453356706,
"y": -519.7914736349932
}
}
],
"edges": [
{
"source": "Milvus-CWpae",
"target": "RetrievalQA-x5nap",
"sourceHandle": "Milvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-x5nap",
"id": "reactflow__edge-Milvus-CWpaeMilvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever-RetrievalQA-x5napBaseRetriever|retriever|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "CombineDocsChain-71a85",
"target": "RetrievalQA-x5nap",
"sourceHandle": "CombineDocsChain|CombineDocsChain-71a85|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap",
"id": "reactflow__edge-CombineDocsChain-71a85CombineDocsChain|CombineDocsChain-71a85|BaseCombineDocumentsChain|function-RetrievalQA-x5napBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "PromptTemplate-3c164",
"target": "CombineDocsChain-71a85",
"sourceHandle": "PromptTemplate|PromptTemplate-3c164|Runnable|StringPromptTemplate|RunnableSerializable|PromptTemplate|Generic|BasePromptTemplate",
"targetHandle": "BasePromptTemplate|document_prompt|CombineDocsChain-71a85",
"id": "reactflow__edge-PromptTemplate-3c164PromptTemplate|PromptTemplate-3c164|Runnable|StringPromptTemplate|RunnableSerializable|PromptTemplate|Generic|BasePromptTemplate-CombineDocsChain-71a85BasePromptTemplate|document_prompt|CombineDocsChain-71a85",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "PromptTemplate-ced29",
"target": "CombineDocsChain-71a85",
"sourceHandle": "PromptTemplate|PromptTemplate-ced29|Runnable|StringPromptTemplate|RunnableSerializable|PromptTemplate|Generic|BasePromptTemplate",
"targetHandle": "BasePromptTemplate|prompt|CombineDocsChain-71a85",
"id": "reactflow__edge-PromptTemplate-ced29PromptTemplate|PromptTemplate-ced29|Runnable|StringPromptTemplate|RunnableSerializable|PromptTemplate|Generic|BasePromptTemplate-CombineDocsChain-71a85BasePromptTemplate|prompt|CombineDocsChain-71a85",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "ChatOpenAI-7a446",
"sourceHandle": "ChatOpenAI|ChatOpenAI-7a446|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM",
"target": "CombineDocsChain-71a85",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-71a85",
"style": {
"stroke": "#555"
},
"className": "stroke-foreground stroke-connection",
"animated": true,
"id": "reactflow__edge-ChatOpenAI-7a446ChatOpenAI|ChatOpenAI-7a446|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM-CombineDocsChain-71a85BaseLanguageModel|llm|CombineDocsChain-71a85",
"selected": true
}
],
"viewport": {
"zoom": 1,
"x": 0,
"y": 0
}
},
"is_component": false,
"name": "",
"description": "",
"id": "da453",
"status": 0,
"write": false,
"guide_word": ""
}
}
}
},
{
"表格数据问答": {
"id": "RetrievalQA-0a13b",
"type": "InputFileNode",
"node": {
"output_types": [],
"display_name": "表格数据问答",
"documentation": "",
"base_classes": [
"RetrievalQA",
"Chain",
"BaseRetrievalQA",
"function"
],
"description": "Chain for question-answering against an index.",
"template": {
"token_max_CombineDocsChain-RCPNA": {
"l2": false,
"info": "当前只对stuff 生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"l2_name": "token_max",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max",
"proxy": {
"id": "CombineDocsChain-RCPNA",
"field": "token_max"
}
},
"chain_type_CombineDocsChain-RCPNA": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "chain_type",
"proxy": {
"id": "CombineDocsChain-RCPNA",
"field": "chain_type"
}
},
"drop_old_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "drop_old",
"proxy": {
"id": "Milvus-CWpae",
"field": "drop_old"
}
},
"metadatas_Milvus-CWpae": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas",
"proxy": {
"id": "Milvus-CWpae",
"field": "metadatas"
}
},
"index_params_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_params",
"proxy": {
"id": "Milvus-CWpae",
"field": "index_params"
}
},
"search_kwargs_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": false,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs",
"proxy": {
"id": "Milvus-CWpae",
"field": "search_kwargs"
}
},
"search_params_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_params",
"proxy": {
"id": "Milvus-CWpae",
"field": "search_params"
}
},
"collection_name_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "选择知识库",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": "",
"display_name": "collection_name",
"proxy": {
"id": "Milvus-CWpae",
"field": "collection_name"
}
},
"connection_args_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": false,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "connection_args",
"proxy": {
"id": "Milvus-CWpae",
"field": "connection_args"
}
},
"consistency_level_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "consistency_level",
"proxy": {
"id": "Milvus-CWpae",
"field": "consistency_level"
}
},
"tags_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "tags"
}
},
"memory_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "memory"
}
},
"verbose_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "verbose"
}
},
"metadata_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "metadata"
}
},
"callbacks_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "callbacks"
}
},
"input_key_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_key",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "input_key"
}
},
"input_node_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "input_node"
}
},
"output_key_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "output_key",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "output_key"
}
},
"return_source_documents_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_source_documents",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "return_source_documents"
}
},
"proxy_url_OpenAIProxyEmbedding-1771b": {
"info": "",
"list": false,
"name": "proxy_url",
"show": false,
"type": "str",
"value": "",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "http://43.133.35.137:8080",
"display_name": "proxy_url",
"proxy": {
"id": "OpenAIProxyEmbedding-1771b",
"field": "proxy_url"
}
},
"metadata_CSVLoader-854c0": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"value": "{}",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Metadata",
"proxy": {
"id": "CSVLoader-854c0",
"field": "metadata"
}
},
"file_path_InputFileNode-e4363": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "file",
"value": "",
"advanced": true,
"password": false,
"required": false,
"suffixes": [
".csv"
],
"fileTypes": [
"csv"
],
"file_path": null,
"multiline": false,
"placeholder": "",
"display_name": "file_path",
"proxy": {
"id": "InputFileNode-e4363",
"field": "file_path"
}
},
"file_type_InputFileNode-e4363": {
"info": "Tips for which file should upload",
"list": false,
"name": "file_type",
"show": true,
"type": "str",
"value": "CSV文件",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "提示上传文件类型",
"display_name": "Name",
"proxy": {
"id": "InputFileNode-e4363",
"field": "file_type"
}
},
"callbacks_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "callbacks",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"list": true,
"file_path": "",
"display_name": "Callbacks",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "callbacks"
}
},
"async_client_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "async_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Async Client",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "async_client"
}
},
"cache_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": true,
"file_path": "",
"password": false,
"name": "cache",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"display_name": "Cache",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "cache"
}
},
"client_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Client",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "client"
}
},
"default_headers_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_headers",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Headers",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "default_headers"
}
},
"default_query_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_query",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Query",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "default_query"
}
},
"http_client_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "http_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Http Client",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "http_client"
}
},
"max_retries_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 2,
"password": false,
"name": "max_retries",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Retries",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "max_retries"
}
},
"max_tokens_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": true,
"name": "max_tokens",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Tokens",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "max_tokens"
},
"value": ""
},
"metadata_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "metadata",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Metadata",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "metadata"
}
},
"model_kwargs_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "model_kwargs",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Model Kwargs",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "model_kwargs"
}
},
"model_name_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "gpt-3.5-turbo-0613",
"password": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"name": "model_name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Model Name",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "model_name"
}
},
"n_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 1,
"password": false,
"name": "n",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "N",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "n"
}
},
"name_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Name",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "name"
}
},
"openai_api_base_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_api_base",
"display_name": "OpenAI API Base",
"advanced": true,
"dynamic": false,
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "openai_api_base"
}
},
"openai_api_key_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "",
"password": true,
"name": "openai_api_key",
"display_name": "OpenAI API Key",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "openai_api_key"
}
},
"openai_organization_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "openai_organization",
"display_name": "OpenAI Organization",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "openai_organization"
}
},
"openai_proxy_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_proxy",
"display_name": "OpenAI Proxy",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "openai_proxy"
}
},
"request_timeout_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "request_timeout",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Request Timeout",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "request_timeout"
}
},
"streaming_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": false,
"password": false,
"name": "streaming",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Streaming",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "streaming"
}
},
"tags_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tags",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Tags",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "tags"
}
},
"temperature_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": 0.7,
"password": false,
"name": "temperature",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Temperature",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "temperature"
}
},
"tiktoken_model_name_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tiktoken_model_name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Tiktoken Model Name",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "tiktoken_model_name"
}
},
"verbose_ChatOpenAI-c4ddb": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "verbose",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Verbose",
"proxy": {
"id": "ChatOpenAI-c4ddb",
"field": "verbose"
}
}
},
"flow": {
"data": {
"nodes": [
{
"width": 384,
"height": 376,
"id": "CombineDocsChain-RCPNA",
"type": "genericNode",
"position": {
"x": 991.5062827922865,
"y": -1765.3942107165124
},
"data": {
"id": "CombineDocsChain-RCPNA",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"token_max": {
"l2": false,
"info": "当前只对stuff 生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"l2_name": "token_max",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "chain_type"
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 991.5062827922865,
"y": -1765.3942107165124
},
"dragging": false
},
{
"width": 384,
"height": 346,
"id": "Milvus-CWpae",
"type": "genericNode",
"position": {
"x": 1004.1977914708131,
"y": -1222.8140072260721
},
"data": {
"id": "Milvus-CWpae",
"node": {
"l2_name": "知识库",
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "drop_old"
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas"
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_params"
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs"
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_params"
},
"collection_name": {
"l2": false,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "选择知识库",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": "",
"display_name": "collection_name"
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "connection_args"
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "consistency_level"
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"Milvus",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1004.1977914708131,
"y": -1222.8140072260721
},
"dragging": false
},
{
"width": 384,
"height": 316,
"id": "RetrievalQA-x5nap",
"type": "genericNode",
"position": {
"x": 1667.777592949481,
"y": -1602.5215637887113
},
"data": {
"id": "RetrievalQA-x5nap",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags"
},
"_type": "RetrievalQA",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory"
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose"
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata"
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks"
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_key"
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "output_key"
},
"combine_documents_chain": {
"l2": false,
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"l2_name": "combine_documents_chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_source_documents"
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"RetrievalQA",
"Chain",
"BaseRetrievalQA",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1667.777592949481,
"y": -1602.5215637887113
},
"dragging": false
},
{
"width": 384,
"height": 172,
"id": "OpenAIProxyEmbedding-1771b",
"type": "genericNode",
"position": {
"x": 492.2452900949522,
"y": -841.6318960897328
},
"data": {
"id": "OpenAIProxyEmbedding-1771b",
"node": {
"template": {
"_type": "proxy_embedding",
"proxy_url": {
"info": "",
"list": false,
"name": "proxy_url",
"show": false,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "http://43.133.35.137:8080",
"display_name": "proxy_url"
}
},
"description": " 使用自建的embedding服务使用openai进行embed ",
"base_classes": [
"Embeddings"
],
"display_name": "OpenAIProxyEmbedding",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "OpenAIProxyEmbedding",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 492.2452900949522,
"y": -841.6318960897328
},
"dragging": false
},
{
"width": 384,
"height": 290,
"id": "CSVLoader-854c0",
"type": "genericNode",
"position": {
"x": 473.20021062628746,
"y": -1217.8870866182915
},
"data": {
"id": "CSVLoader-854c0",
"node": {
"template": {
"_type": "CSVLoader",
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"value": "{}",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Metadata"
},
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "fileNode",
"value": "",
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".csv"
],
"fileTypes": [
"csv"
],
"multiline": false,
"placeholder": ""
}
},
"description": "Load a `CSV` file into a list of Documents.",
"base_classes": [
"Document"
],
"display_name": "CSVLoader",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/csv"
},
"type": "CSVLoader",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 473.20021062628746,
"y": -1217.8870866182915
},
"dragging": false
},
{
"width": 384,
"height": 336,
"id": "InputFileNode-e4363",
"type": "genericNode",
"position": {
"x": -29.907305337607568,
"y": -1246.4547058212886
},
"data": {
"id": "InputFileNode-e4363",
"node": {
"template": {
"_type": "InputFileNode",
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "file",
"value": "",
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".csv"
],
"fileTypes": [
"csv"
],
"file_path": null,
"multiline": false,
"placeholder": "",
"display_name": "file_path"
},
"file_type": {
"info": "Tips for which file should upload",
"list": false,
"name": "file_type",
"show": true,
"type": "str",
"value": "CSV文件",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "提示上传文件类型",
"display_name": "Name"
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"fileNode"
],
"display_name": "InputFileNode",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "InputFileNode",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": -29.907305337607568,
"y": -1246.4547058212886
},
"dragging": false
},
{
"width": 384,
"height": 606,
"id": "ChatOpenAI-c4ddb",
"type": "genericNode",
"position": {
"x": 492.0105811305291,
"y": -2015.8972405291959
},
"data": {
"type": "ChatOpenAI",
"node": {
"template": {
"callbacks": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "callbacks",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"list": true,
"file_path": "",
"display_name": "Callbacks"
},
"async_client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "async_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Async Client"
},
"cache": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": true,
"file_path": "",
"password": false,
"name": "cache",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"display_name": "Cache"
},
"client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Client"
},
"default_headers": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_headers",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Headers"
},
"default_query": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_query",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Query"
},
"http_client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "http_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Http Client"
},
"max_retries": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 2,
"password": false,
"name": "max_retries",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Retries"
},
"max_tokens": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": true,
"name": "max_tokens",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Tokens"
},
"metadata": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "metadata",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Metadata"
},
"model_kwargs": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "model_kwargs",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Model Kwargs"
},
"model_name": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "gpt-3.5-turbo-0613",
"password": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"name": "model_name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Model Name"
},
"n": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 1,
"password": false,
"name": "n",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "N"
},
"name": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Name"
},
"openai_api_base": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_api_base",
"display_name": "OpenAI API Base",
"advanced": false,
"dynamic": false,
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_api_key": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "",
"password": true,
"name": "openai_api_key",
"display_name": "OpenAI API Key",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_organization": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "openai_organization",
"display_name": "OpenAI Organization",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_proxy": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_proxy",
"display_name": "OpenAI Proxy",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"request_timeout": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "request_timeout",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Request Timeout"
},
"streaming": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": false,
"password": false,
"name": "streaming",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Streaming"
},
"tags": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tags",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Tags"
},
"temperature": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": 0.7,
"password": false,
"name": "temperature",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Temperature"
},
"tiktoken_model_name": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tiktoken_model_name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Tiktoken Model Name"
},
"verbose": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "verbose",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Verbose"
},
"_type": "ChatOpenAI"
},
"description": "[*Deprecated*] `OpenAI` Chat large language models API.",
"base_classes": [
"ChatOpenAI",
"BaseLanguageModel",
"BaseChatModel",
"BaseLLM"
],
"display_name": "ChatOpenAI",
"documentation": "https://python.langchain.com/docs/modules/model_io/models/chat/integrations/openai",
"custom_fields": {},
"output_types": [],
"field_formatters": {},
"beta": false,
"official": false
},
"id": "ChatOpenAI-c4ddb",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 492.0105811305291,
"y": -2015.8972405291959
},
"dragging": false
}
],
"edges": [
{
"source": "CombineDocsChain-RCPNA",
"target": "RetrievalQA-x5nap",
"sourceHandle": "CombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap",
"id": "reactflow__edge-CombineDocsChain-RCPNACombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function-RetrievalQA-x5napBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "Milvus-CWpae",
"target": "RetrievalQA-x5nap",
"sourceHandle": "Milvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-x5nap",
"id": "reactflow__edge-Milvus-CWpaeMilvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever-RetrievalQA-x5napBaseRetriever|retriever|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "OpenAIProxyEmbedding-1771b",
"target": "Milvus-CWpae",
"sourceHandle": "OpenAIProxyEmbedding|OpenAIProxyEmbedding-1771b|Embeddings",
"targetHandle": "Embeddings|embedding|Milvus-CWpae",
"id": "reactflow__edge-OpenAIProxyEmbedding-1771bOpenAIProxyEmbedding|OpenAIProxyEmbedding-1771b|Embeddings-Milvus-CWpaeEmbeddings|embedding|Milvus-CWpae",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "CSVLoader-854c0",
"target": "Milvus-CWpae",
"sourceHandle": "CSVLoader|CSVLoader-854c0|Document",
"targetHandle": "Document|documents|Milvus-CWpae",
"id": "reactflow__edge-CSVLoader-854c0CSVLoader|CSVLoader-854c0|Document-Milvus-CWpaeDocument|documents|Milvus-CWpae",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "InputFileNode-e4363",
"target": "CSVLoader-854c0",
"sourceHandle": "InputFileNode|InputFileNode-e4363|fileNode",
"targetHandle": "fileNode|file_path|CSVLoader-854c0",
"id": "reactflow__edge-InputFileNode-e4363InputFileNode|InputFileNode-e4363|fileNode-CSVLoader-854c0fileNode|file_path|CSVLoader-854c0",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "ChatOpenAI-c4ddb",
"sourceHandle": "ChatOpenAI|ChatOpenAI-c4ddb|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM",
"target": "CombineDocsChain-RCPNA",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-RCPNA",
"style": {
"stroke": "#555"
},
"className": "stroke-foreground stroke-connection",
"animated": true,
"id": "reactflow__edge-ChatOpenAI-c4ddbChatOpenAI|ChatOpenAI-c4ddb|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM-CombineDocsChain-RCPNABaseLanguageModel|llm|CombineDocsChain-RCPNA",
"selected": true
}
],
"viewport": {
"zoom": 1,
"x": 0,
"y": 0
}
},
"is_component": false,
"name": "",
"description": "",
"id": "dd09b",
"status": 0,
"write": false,
"guide_word": ""
}
}
}
},
{
"角色扮演": {
"id": "LLMChain-6df16",
"type": "LLMChain",
"node": {
"output_types": [],
"display_name": "角色扮演",
"documentation": "",
"base_classes": [
"Chain",
"LLMChain",
"function"
],
"description": "Chain to run queries against LLMs.",
"template": {
"tags_LLMChain-wRz0c": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags",
"proxy": {
"id": "LLMChain-wRz0c",
"field": "tags"
}
},
"verbose_LLMChain-wRz0c": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose",
"proxy": {
"id": "LLMChain-wRz0c",
"field": "verbose"
}
},
"metadata_LLMChain-wRz0c": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata",
"proxy": {
"id": "LLMChain-wRz0c",
"field": "metadata"
}
},
"callbacks_LLMChain-wRz0c": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks",
"proxy": {
"id": "LLMChain-wRz0c",
"field": "callbacks"
}
},
"input_node_LLMChain-wRz0c": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question",
"proxy": {
"id": "LLMChain-wRz0c",
"field": "input_node"
}
},
"llm_kwargs_LLMChain-wRz0c": {
"l2": false,
"info": "",
"list": false,
"name": "llm_kwargs",
"show": false,
"type": "code",
"l2_name": "llm_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "llm_kwargs",
"proxy": {
"id": "LLMChain-wRz0c",
"field": "llm_kwargs"
}
},
"output_key_LLMChain-wRz0c": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "text",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "output_key",
"proxy": {
"id": "LLMChain-wRz0c",
"field": "output_key"
}
},
"output_parser_LLMChain-wRz0c": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "BaseLLMOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser",
"proxy": {
"id": "LLMChain-wRz0c",
"field": "output_parser"
}
},
"return_final_only_LLMChain-wRz0c": {
"l2": false,
"info": "",
"list": false,
"name": "return_final_only",
"show": false,
"type": "bool",
"value": true,
"l2_name": "return_final_only",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_final_only",
"proxy": {
"id": "LLMChain-wRz0c",
"field": "return_final_only"
}
},
"input_PromptTemplate-ZoUVT": {
"info": "",
"list": false,
"name": "input",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser"
],
"placeholder": "",
"display_name": "input",
"proxy": {
"id": "PromptTemplate-ZoUVT",
"field": "input"
}
},
"template_PromptTemplate-ZoUVT": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "System：\n你将扮演一个科技公司的面试官，面试者是一名产品经理，提出 3 个犀利的问题。\n请注意：\n- 每次只问一个问题\n- 面试者回答问题后请简短点评一下，然后问下一个问题，不要试图纠正候选人的错误；\n- 如果你认为用户连续几次回答的都不对，就少问一点；\n- 问完最后一个问题后，你可以问这样一个问题：上一份工作为什么离职？用户回答该问题后，请表示理解与支持，并停止与面试者的对话。若面试者继续说话你可以礼貌拒绝。\n\n以下是你们的对话记录\n\n{chat_history}\n\n面试者：{input}\n\n你的回答：",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "template",
"proxy": {
"id": "PromptTemplate-ZoUVT",
"field": "template"
}
},
"chat_history_PromptTemplate-ZoUVT": {
"info": "",
"list": false,
"name": "chat_history",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser"
],
"placeholder": "",
"display_name": "chat_history",
"proxy": {
"id": "PromptTemplate-ZoUVT",
"field": "chat_history"
}
},
"output_parser_PromptTemplate-ZoUVT": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser",
"proxy": {
"id": "PromptTemplate-ZoUVT",
"field": "output_parser"
}
},
"input_variables_PromptTemplate-ZoUVT": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"chat_history",
"input"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_variables",
"proxy": {
"id": "PromptTemplate-ZoUVT",
"field": "input_variables"
}
},
"template_format_PromptTemplate-ZoUVT": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "template_format",
"proxy": {
"id": "PromptTemplate-ZoUVT",
"field": "template_format"
}
},
"partial_variables_PromptTemplate-ZoUVT": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "partial_variables",
"proxy": {
"id": "PromptTemplate-ZoUVT",
"field": "partial_variables"
}
},
"validate_template_PromptTemplate-ZoUVT": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "validate_template",
"proxy": {
"id": "PromptTemplate-ZoUVT",
"field": "validate_template"
}
},
"ai_prefix_ConversationBufferMemory-lI5Pt": {
"l2": false,
"info": "",
"list": false,
"name": "ai_prefix",
"show": false,
"type": "str",
"value": "AI",
"l2_name": "ai_prefix",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ai_prefix",
"proxy": {
"id": "ConversationBufferMemory-lI5Pt",
"field": "ai_prefix"
}
},
"input_key_ConversationBufferMemory-lI5Pt": {
"l2": false,
"info": "The variable to be used as Chat Input when more than one variable is available.",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "input_key",
"proxy": {
"id": "ConversationBufferMemory-lI5Pt",
"field": "input_key"
}
},
"memory_key_ConversationBufferMemory-lI5Pt": {
"l2": false,
"info": "",
"list": false,
"name": "memory_key",
"show": true,
"type": "str",
"value": "chat_history",
"l2_name": "memory_key",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory_key",
"proxy": {
"id": "ConversationBufferMemory-lI5Pt",
"field": "memory_key"
}
},
"output_key_ConversationBufferMemory-lI5Pt": {
"l2": false,
"info": "The variable to be used as Chat Output (e.g. answer in a ConversationalRetrievalChain)",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_key",
"proxy": {
"id": "ConversationBufferMemory-lI5Pt",
"field": "output_key"
}
},
"chat_memory_ConversationBufferMemory-lI5Pt": {
"l2": false,
"info": "",
"list": false,
"name": "chat_memory",
"show": true,
"type": "BaseChatMessageHistory",
"l2_name": "chat_memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "chat_memory",
"proxy": {
"id": "ConversationBufferMemory-lI5Pt",
"field": "chat_memory"
}
},
"human_prefix_ConversationBufferMemory-lI5Pt": {
"l2": false,
"info": "",
"list": false,
"name": "human_prefix",
"show": false,
"type": "str",
"value": "Human",
"l2_name": "human_prefix",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "human_prefix",
"proxy": {
"id": "ConversationBufferMemory-lI5Pt",
"field": "human_prefix"
}
},
"return_messages_ConversationBufferMemory-lI5Pt": {
"l2": false,
"info": "",
"list": false,
"name": "return_messages",
"show": true,
"type": "bool",
"l2_name": "return_messages",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_messages",
"proxy": {
"id": "ConversationBufferMemory-lI5Pt",
"field": "return_messages"
}
},
"callbacks_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "callbacks",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"list": true,
"file_path": "",
"display_name": "Callbacks",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "callbacks"
}
},
"async_client_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "async_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Async Client",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "async_client"
}
},
"cache_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": true,
"file_path": "",
"password": false,
"name": "cache",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"display_name": "Cache",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "cache"
}
},
"client_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Client",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "client"
}
},
"default_headers_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_headers",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Headers",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "default_headers"
}
},
"default_query_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_query",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Query",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "default_query"
}
},
"http_client_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "http_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Http Client",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "http_client"
}
},
"max_retries_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 2,
"password": false,
"name": "max_retries",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Retries",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "max_retries"
}
},
"max_tokens_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": true,
"name": "max_tokens",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Tokens",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "max_tokens"
},
"value": ""
},
"metadata_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "metadata",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Metadata",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "metadata"
}
},
"model_kwargs_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "model_kwargs",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Model Kwargs",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "model_kwargs"
}
},
"model_name_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "gpt-3.5-turbo-0613",
"password": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"name": "model_name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Model Name",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "model_name"
}
},
"n_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 1,
"password": false,
"name": "n",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "N",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "n"
}
},
"name_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Name",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "name"
}
},
"openai_api_base_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_api_base",
"display_name": "OpenAI API Base",
"advanced": true,
"dynamic": false,
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "openai_api_base"
}
},
"openai_api_key_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "",
"password": true,
"name": "openai_api_key",
"display_name": "OpenAI API Key",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "openai_api_key"
}
},
"openai_organization_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "openai_organization",
"display_name": "OpenAI Organization",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "openai_organization"
}
},
"openai_proxy_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_proxy",
"display_name": "OpenAI Proxy",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "openai_proxy"
}
},
"request_timeout_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "request_timeout",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Request Timeout",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "request_timeout"
}
},
"streaming_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": false,
"password": false,
"name": "streaming",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Streaming",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "streaming"
}
},
"tags_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tags",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Tags",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "tags"
}
},
"temperature_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": 0.7,
"password": false,
"name": "temperature",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Temperature",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "temperature"
}
},
"tiktoken_model_name_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tiktoken_model_name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Tiktoken Model Name",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "tiktoken_model_name"
}
},
"verbose_ChatOpenAI-1a993": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "verbose",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Verbose",
"proxy": {
"id": "ChatOpenAI-1a993",
"field": "verbose"
}
}
},
"flow": {
"data": {
"nodes": [
{
"width": 384,
"height": 316,
"id": "LLMChain-wRz0c",
"type": "genericNode",
"position": {
"x": 1452.9429776850677,
"y": -1823.729943530931
},
"data": {
"id": "LLMChain-wRz0c",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags"
},
"_type": "LLMChain",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": ""
},
"prompt": {
"l2": false,
"info": "",
"list": false,
"name": "prompt",
"show": true,
"type": "BasePromptTemplate",
"l2_name": "prompt",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"value": false,
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose"
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata"
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks"
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"llm_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "llm_kwargs",
"show": false,
"type": "code",
"l2_name": "llm_kwargs",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "llm_kwargs"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "text",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "output_key"
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "BaseLLMOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser"
},
"return_final_only": {
"l2": false,
"info": "",
"list": false,
"name": "return_final_only",
"show": false,
"type": "bool",
"value": true,
"l2_name": "return_final_only",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_final_only"
}
},
"description": "Chain to run queries against LLMs.",
"base_classes": [
"Chain",
"LLMChain",
"function"
],
"display_name": "LLMChain",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/foundational/llm_chain"
},
"type": "LLMChain",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1452.9429776850677,
"y": -1823.729943530931
},
"dragging": false
},
{
"width": 384,
"height": 418,
"id": "PromptTemplate-ZoUVT",
"type": "genericNode",
"position": {
"x": 679.2448826071954,
"y": -1404.485160984465
},
"data": {
"id": "PromptTemplate-ZoUVT",
"node": {
"name": "",
"template": {
"_type": "PromptTemplate",
"input": {
"info": "",
"list": false,
"name": "input",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser"
],
"placeholder": "",
"display_name": "input"
},
"template": {
"l2": false,
"info": "",
"list": false,
"name": "template",
"show": true,
"type": "prompt",
"value": "System：\n你将扮演一个科技公司的面试官，面试者是一名产品经理，提出 3 个犀利的问题。\n请注意：\n- 每次只问一个问题\n- 面试者回答问题后请简短点评一下，然后问下一个问题，不要试图纠正候选人的错误；\n- 如果你认为用户连续几次回答的都不对，就少问一点；\n- 问完最后一个问题后，你可以问这样一个问题：上一份工作为什么离职？用户回答该问题后，请表示理解与支持，并停止与面试者的对话。若面试者继续说话你可以礼貌拒绝。\n\n以下是你们的对话记录\n\n{chat_history}\n\n面试者：{input}\n\n你的回答：",
"l2_name": "template",
"advanced": false,
"password": false,
"required": true,
"multiline": true,
"placeholder": "",
"display_name": "template"
},
"chat_history": {
"info": "",
"list": false,
"name": "chat_history",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": true,
"input_types": [
"Document",
"BaseOutputParser"
],
"placeholder": "",
"display_name": "chat_history"
},
"output_parser": {
"l2": false,
"info": "",
"list": false,
"name": "output_parser",
"show": false,
"type": "BaseOutputParser",
"l2_name": "output_parser",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_parser"
},
"input_variables": {
"l2": false,
"info": "",
"list": true,
"name": "input_variables",
"show": false,
"type": "str",
"value": [
"chat_history",
"input"
],
"l2_name": "input_variables",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_variables"
},
"template_format": {
"l2": false,
"info": "",
"list": false,
"name": "template_format",
"show": false,
"type": "str",
"value": "f-string",
"l2_name": "template_format",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "template_format"
},
"partial_variables": {
"l2": false,
"info": "",
"list": false,
"name": "partial_variables",
"show": false,
"type": "code",
"l2_name": "partial_variables",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "partial_variables"
},
"validate_template": {
"l2": false,
"info": "",
"list": false,
"name": "validate_template",
"show": false,
"type": "bool",
"value": true,
"l2_name": "validate_template",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "validate_template"
}
},
"description": "Schema to represent a prompt for an LLM.",
"base_classes": [
"PromptTemplate",
"StringPromptTemplate",
"BasePromptTemplate"
],
"display_name": "PromptTemplate",
"output_types": [],
"custom_fields": {
"template": [
"chat_history",
"input"
]
},
"documentation": "https://python.langchain.com/docs/modules/model_io/prompts/prompt_templates/",
"field_formatters": {
"formatters": {
"openai_api_key": {}
},
"base_formatters": {
"dict": {},
"list": {},
"show": {},
"union": {},
"kwargs": {},
"default": {},
"headers": {},
"optional": {},
"password": {},
"multiline": {},
"model_fields": {
"MODEL_DICT": {
"OpenAI": [
"text-davinci-003",
"text-davinci-002",
"text-curie-001",
"text-babbage-001",
"text-ada-001"
],
"Anthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
],
"ChatOpenAI": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k"
],
"ChatAnthropic": [
"claude-v1",
"claude-v1-100k",
"claude-instant-v1",
"claude-instant-v1-100k",
"claude-v1.3",
"claude-v1.3-100k",
"claude-v1.2",
"claude-v1.0",
"claude-instant-v1.1",
"claude-instant-v1.1-100k",
"claude-instant-v1.0"
]
}
},
"dict_code_file": {}
}
}
},
"type": "PromptTemplate",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 679.2448826071954,
"y": -1404.485160984465
},
"dragging": false
},
{
"width": 384,
"height": 526,
"id": "ConversationBufferMemory-lI5Pt",
"type": "genericNode",
"position": {
"x": 275.26369845498505,
"y": -1967.2256382433638
},
"data": {
"id": "ConversationBufferMemory-lI5Pt",
"node": {
"template": {
"_type": "ConversationBufferMemory",
"ai_prefix": {
"l2": false,
"info": "",
"list": false,
"name": "ai_prefix",
"show": false,
"type": "str",
"value": "AI",
"l2_name": "ai_prefix",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "ai_prefix"
},
"input_key": {
"l2": false,
"info": "The variable to be used as Chat Input when more than one variable is available.",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "",
"l2_name": "input_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "input_key"
},
"memory_key": {
"l2": false,
"info": "",
"list": false,
"name": "memory_key",
"show": true,
"type": "str",
"value": "chat_history",
"l2_name": "memory_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory_key"
},
"output_key": {
"l2": false,
"info": "The variable to be used as Chat Output (e.g. answer in a ConversationalRetrievalChain)",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "",
"l2_name": "output_key",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "output_key"
},
"chat_memory": {
"l2": false,
"info": "",
"list": false,
"name": "chat_memory",
"show": true,
"type": "BaseChatMessageHistory",
"l2_name": "chat_memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "chat_memory"
},
"human_prefix": {
"l2": false,
"info": "",
"list": false,
"name": "human_prefix",
"show": false,
"type": "str",
"value": "Human",
"l2_name": "human_prefix",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "human_prefix"
},
"return_messages": {
"l2": false,
"info": "",
"list": false,
"name": "return_messages",
"show": true,
"type": "bool",
"l2_name": "return_messages",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_messages"
}
},
"description": "Buffer for storing conversation memory.",
"base_classes": [
"BaseChatMemory",
"ConversationBufferMemory",
"BaseMemory"
],
"display_name": "ConversationBufferMemory",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/memory/how_to/buffer"
},
"type": "ConversationBufferMemory",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 275.26369845498505,
"y": -1967.2256382433638
},
"dragging": false
},
{
"width": 384,
"height": 606,
"id": "ChatOpenAI-1a993",
"type": "genericNode",
"position": {
"x": 737.6920763029101,
"y": -2366.8833957378292
},
"data": {
"type": "ChatOpenAI",
"node": {
"template": {
"callbacks": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "callbacks",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"list": true,
"file_path": "",
"display_name": "Callbacks"
},
"async_client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "async_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Async Client"
},
"cache": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": true,
"file_path": "",
"password": false,
"name": "cache",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"display_name": "Cache"
},
"client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Client"
},
"default_headers": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_headers",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Headers"
},
"default_query": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_query",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Query"
},
"http_client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "http_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Http Client"
},
"max_retries": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 2,
"password": false,
"name": "max_retries",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Retries"
},
"max_tokens": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": true,
"name": "max_tokens",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Tokens"
},
"metadata": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "metadata",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Metadata"
},
"model_kwargs": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "model_kwargs",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Model Kwargs"
},
"model_name": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "gpt-3.5-turbo-0613",
"password": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"name": "model_name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Model Name"
},
"n": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 1,
"password": false,
"name": "n",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "N"
},
"name": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Name"
},
"openai_api_base": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_api_base",
"display_name": "OpenAI API Base",
"advanced": false,
"dynamic": false,
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_api_key": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "",
"password": true,
"name": "openai_api_key",
"display_name": "OpenAI API Key",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_organization": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "openai_organization",
"display_name": "OpenAI Organization",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_proxy": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_proxy",
"display_name": "OpenAI Proxy",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"request_timeout": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "request_timeout",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Request Timeout"
},
"streaming": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": false,
"password": false,
"name": "streaming",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Streaming"
},
"tags": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tags",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Tags"
},
"temperature": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": 0.7,
"password": false,
"name": "temperature",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Temperature"
},
"tiktoken_model_name": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tiktoken_model_name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Tiktoken Model Name"
},
"verbose": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "verbose",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Verbose"
},
"_type": "ChatOpenAI"
},
"description": "[*Deprecated*] `OpenAI` Chat large language models API.",
"base_classes": [
"ChatOpenAI",
"BaseLanguageModel",
"BaseChatModel",
"BaseLLM"
],
"display_name": "ChatOpenAI",
"documentation": "https://python.langchain.com/docs/modules/model_io/models/chat/integrations/openai",
"custom_fields": {},
"output_types": [],
"field_formatters": {},
"beta": false,
"official": false
},
"id": "ChatOpenAI-1a993",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 737.6920763029101,
"y": -2366.8833957378292
},
"dragging": false
}
],
"edges": [
{
"source": "PromptTemplate-ZoUVT",
"target": "LLMChain-wRz0c",
"sourceHandle": "PromptTemplate|PromptTemplate-ZoUVT|PromptTemplate|StringPromptTemplate|BasePromptTemplate",
"targetHandle": "BasePromptTemplate|prompt|LLMChain-wRz0c",
"id": "reactflow__edge-PromptTemplate-ZoUVTPromptTemplate|PromptTemplate-ZoUVT|PromptTemplate|StringPromptTemplate|BasePromptTemplate-LLMChain-wRz0cBasePromptTemplate|prompt|LLMChain-wRz0c",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "ConversationBufferMemory-lI5Pt",
"target": "LLMChain-wRz0c",
"sourceHandle": "ConversationBufferMemory|ConversationBufferMemory-lI5Pt|BaseChatMemory|ConversationBufferMemory|BaseMemory",
"targetHandle": "BaseMemory|memory|LLMChain-wRz0c",
"id": "reactflow__edge-ConversationBufferMemory-lI5PtConversationBufferMemory|ConversationBufferMemory-lI5Pt|BaseChatMemory|ConversationBufferMemory|BaseMemory-LLMChain-wRz0cBaseMemory|memory|LLMChain-wRz0c",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "ChatOpenAI-1a993",
"sourceHandle": "ChatOpenAI|ChatOpenAI-1a993|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM",
"target": "LLMChain-wRz0c",
"targetHandle": "BaseLanguageModel|llm|LLMChain-wRz0c",
"style": {
"stroke": "#555"
},
"className": "stroke-foreground stroke-connection",
"animated": true,
"id": "reactflow__edge-ChatOpenAI-1a993ChatOpenAI|ChatOpenAI-1a993|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM-LLMChain-wRz0cBaseLanguageModel|llm|LLMChain-wRz0c",
"selected": true
}
],
"viewport": {
"zoom": 1,
"x": 0,
"y": 0
}
},
"is_component": false,
"name": "",
"description": "",
"id": "5b80b",
"status": 0,
"write": false,
"guide_word": ""
}
}
}
},
{
"输入网址进行问答": {
"id": "RetrievalQA-222c1",
"type": "RetrievalQA",
"node": {
"output_types": [],
"display_name": "输入网址进行问答",
"documentation": "",
"base_classes": [
"RetrievalQA",
"Chain",
"BaseRetrievalQA",
"function"
],
"description": "Chain for question-answering against an index.",
"template": {
"token_max_CombineDocsChain-RCPNA": {
"l2": false,
"info": "当前只对stuff 生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"l2_name": "token_max",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max",
"proxy": {
"id": "CombineDocsChain-RCPNA",
"field": "token_max"
}
},
"chain_type_CombineDocsChain-RCPNA": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "chain_type",
"proxy": {
"id": "CombineDocsChain-RCPNA",
"field": "chain_type"
}
},
"drop_old_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "drop_old",
"proxy": {
"id": "Milvus-CWpae",
"field": "drop_old"
}
},
"metadatas_Milvus-CWpae": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas",
"proxy": {
"id": "Milvus-CWpae",
"field": "metadatas"
}
},
"index_params_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_params",
"proxy": {
"id": "Milvus-CWpae",
"field": "index_params"
}
},
"search_kwargs_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": false,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs",
"proxy": {
"id": "Milvus-CWpae",
"field": "search_kwargs"
}
},
"search_params_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_params",
"proxy": {
"id": "Milvus-CWpae",
"field": "search_params"
}
},
"collection_name_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "选择知识库",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": "",
"display_name": "collection_name",
"proxy": {
"id": "Milvus-CWpae",
"field": "collection_name"
}
},
"connection_args_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": false,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "connection_args",
"proxy": {
"id": "Milvus-CWpae",
"field": "connection_args"
}
},
"consistency_level_Milvus-CWpae": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "consistency_level",
"proxy": {
"id": "Milvus-CWpae",
"field": "consistency_level"
}
},
"tags_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "tags"
}
},
"memory_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "memory"
}
},
"verbose_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "verbose"
}
},
"metadata_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "metadata"
}
},
"callbacks_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "callbacks"
}
},
"input_key_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_key",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "input_key"
}
},
"input_node_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "input_node"
}
},
"output_key_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "output_key",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "output_key"
}
},
"return_source_documents_RetrievalQA-x5nap": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_source_documents",
"proxy": {
"id": "RetrievalQA-x5nap",
"field": "return_source_documents"
}
},
"metadata_WebBaseLoader-af6db": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"value": "{}",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Metadata",
"proxy": {
"id": "WebBaseLoader-af6db",
"field": "metadata"
}
},
"web_path_WebBaseLoader-af6db": {
"l2": true,
"info": "",
"list": false,
"name": "web_path",
"show": true,
"type": "str",
"value": "",
"l2_name": "网址",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Web Page",
"proxy": {
"id": "WebBaseLoader-af6db",
"field": "web_path"
}
},
"proxy_url_OpenAIProxyEmbedding-1771b": {
"info": "",
"list": false,
"name": "proxy_url",
"show": false,
"type": "str",
"value": "",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "http://43.133.35.137:8080",
"display_name": "proxy_url",
"proxy": {
"id": "OpenAIProxyEmbedding-1771b",
"field": "proxy_url"
}
},
"callbacks_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "callbacks",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"list": true,
"file_path": "",
"display_name": "Callbacks",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "callbacks"
}
},
"async_client_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "async_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Async Client",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "async_client"
}
},
"cache_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": true,
"file_path": "",
"password": false,
"name": "cache",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"display_name": "Cache",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "cache"
}
},
"client_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Client",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "client"
}
},
"default_headers_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_headers",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Headers",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "default_headers"
}
},
"default_query_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_query",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Query",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "default_query"
}
},
"http_client_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "http_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Http Client",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "http_client"
}
},
"max_retries_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 2,
"password": false,
"name": "max_retries",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Retries",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "max_retries"
}
},
"max_tokens_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": true,
"name": "max_tokens",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Tokens",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "max_tokens"
},
"value": ""
},
"metadata_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "metadata",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Metadata",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "metadata"
}
},
"model_kwargs_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "model_kwargs",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Model Kwargs",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "model_kwargs"
}
},
"model_name_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "gpt-3.5-turbo-0613",
"password": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"name": "model_name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Model Name",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "model_name"
}
},
"n_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 1,
"password": false,
"name": "n",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "N",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "n"
}
},
"name_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Name",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "name"
}
},
"openai_api_base_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_api_base",
"display_name": "OpenAI API Base",
"advanced": true,
"dynamic": false,
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "openai_api_base"
}
},
"openai_api_key_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "",
"password": true,
"name": "openai_api_key",
"display_name": "OpenAI API Key",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "openai_api_key"
}
},
"openai_organization_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "openai_organization",
"display_name": "OpenAI Organization",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "openai_organization"
}
},
"openai_proxy_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_proxy",
"display_name": "OpenAI Proxy",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "openai_proxy"
}
},
"request_timeout_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "request_timeout",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Request Timeout",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "request_timeout"
}
},
"streaming_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": false,
"password": false,
"name": "streaming",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Streaming",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "streaming"
}
},
"tags_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tags",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Tags",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "tags"
}
},
"temperature_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": 0.7,
"password": false,
"name": "temperature",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Temperature",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "temperature"
}
},
"tiktoken_model_name_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tiktoken_model_name",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Tiktoken Model Name",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "tiktoken_model_name"
}
},
"verbose_ChatOpenAI-04157": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "verbose",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Verbose",
"proxy": {
"id": "ChatOpenAI-04157",
"field": "verbose"
}
}
},
"flow": {
"data": {
"nodes": [
{
"width": 384,
"height": 376,
"id": "CombineDocsChain-RCPNA",
"type": "genericNode",
"position": {
"x": 1020.2314763495577,
"y": -2443.3442844414917
},
"data": {
"id": "CombineDocsChain-RCPNA",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"token_max": {
"l2": false,
"info": "当前只对stuff 生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"l2_name": "token_max",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "chain_type"
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1020.2314763495577,
"y": -2443.3442844414917
},
"dragging": false
},
{
"width": 384,
"height": 346,
"id": "Milvus-CWpae",
"type": "genericNode",
"position": {
"x": 1032.922985028084,
"y": -1900.7640809510517
},
"data": {
"id": "Milvus-CWpae",
"node": {
"l2_name": "知识库",
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "drop_old"
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas"
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_params"
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs"
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_params"
},
"collection_name": {
"l2": false,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "选择知识库",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"collection_id": "",
"display_name": "collection_name"
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "connection_args"
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "consistency_level"
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"Milvus",
"VectorStore",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1032.922985028084,
"y": -1900.7640809510517
},
"dragging": false
},
{
"width": 384,
"height": 316,
"id": "RetrievalQA-x5nap",
"type": "genericNode",
"position": {
"x": 1696.5027865067518,
"y": -2280.471637513691
},
"data": {
"id": "RetrievalQA-x5nap",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags"
},
"_type": "RetrievalQA",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory"
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose"
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata"
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks"
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_key"
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "output_key"
},
"combine_documents_chain": {
"l2": false,
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"l2_name": "combine_documents_chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_source_documents"
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"RetrievalQA",
"Chain",
"BaseRetrievalQA",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 1696.5027865067518,
"y": -2280.471637513691
},
"dragging": false
},
{
"width": 384,
"height": 356,
"id": "WebBaseLoader-af6db",
"type": "genericNode",
"position": {
"x": 508.27376400644664,
"y": -1916.3544587452286
},
"data": {
"id": "WebBaseLoader-af6db",
"node": {
"l2_name": "输入网址",
"template": {
"_type": "WebBaseLoader",
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"value": "{}",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Metadata"
},
"web_path": {
"l2": true,
"info": "",
"list": false,
"name": "web_path",
"show": true,
"type": "str",
"value": "",
"l2_name": "网址",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Web Page"
}
},
"description": "Load HTML pages using `urllib` and parse them with `BeautifulSoup'.",
"base_classes": [
"Document"
],
"display_name": "WebBaseLoader",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/data_connection/document_loaders/integrations/web_base"
},
"type": "WebBaseLoader",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 508.27376400644664,
"y": -1916.3544587452286
},
"dragging": false
},
{
"width": 384,
"height": 172,
"id": "OpenAIProxyEmbedding-1771b",
"type": "genericNode",
"position": {
"x": 520.9704836522232,
"y": -1519.5819698147122
},
"data": {
"id": "OpenAIProxyEmbedding-1771b",
"node": {
"template": {
"_type": "proxy_embedding",
"proxy_url": {
"info": "",
"list": false,
"name": "proxy_url",
"show": false,
"type": "str",
"value": "",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "http://43.133.35.137:8080",
"display_name": "proxy_url"
}
},
"description": " 使用自建的embedding服务使用openai进行embed ",
"base_classes": [
"Embeddings"
],
"display_name": "OpenAIProxyEmbedding",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "OpenAIProxyEmbedding",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 520.9704836522232,
"y": -1519.5819698147122
},
"dragging": false
},
{
"width": 384,
"height": 606,
"id": "ChatOpenAI-04157",
"type": "genericNode",
"position": {
"x": 487.6734875953049,
"y": -2569.9027779626213
},
"data": {
"type": "ChatOpenAI",
"node": {
"template": {
"callbacks": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "callbacks",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"list": true,
"file_path": "",
"display_name": "Callbacks"
},
"async_client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "async_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Async Client"
},
"cache": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": true,
"file_path": "",
"password": false,
"name": "cache",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"display_name": "Cache"
},
"client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Client"
},
"default_headers": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_headers",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Headers"
},
"default_query": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "default_query",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Default Query"
},
"http_client": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "http_client",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "Any",
"list": false,
"file_path": "",
"display_name": "Http Client"
},
"max_retries": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 2,
"password": false,
"name": "max_retries",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Retries"
},
"max_tokens": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": true,
"name": "max_tokens",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "Max Tokens"
},
"metadata": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "metadata",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Metadata"
},
"model_kwargs": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "model_kwargs",
"advanced": true,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "code",
"list": false,
"file_path": "",
"display_name": "Model Kwargs"
},
"model_name": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "gpt-3.5-turbo-0613",
"password": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"name": "model_name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Model Name"
},
"n": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": 1,
"password": false,
"name": "n",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "int",
"list": false,
"file_path": "",
"display_name": "N"
},
"name": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Name"
},
"openai_api_base": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_api_base",
"display_name": "OpenAI API Base",
"advanced": false,
"dynamic": false,
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_api_key": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": "",
"password": true,
"name": "openai_api_key",
"display_name": "OpenAI API Key",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_organization": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "openai_organization",
"display_name": "OpenAI Organization",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"openai_proxy": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"password": false,
"name": "openai_proxy",
"display_name": "OpenAI Proxy",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": ""
},
"request_timeout": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "request_timeout",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Request Timeout"
},
"streaming": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"value": false,
"password": false,
"name": "streaming",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Streaming"
},
"tags": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tags",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": true,
"file_path": "",
"display_name": "Tags"
},
"temperature": {
"required": false,
"placeholder": "",
"show": true,
"multiline": false,
"value": 0.7,
"password": false,
"name": "temperature",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "float",
"list": false,
"file_path": "",
"display_name": "Temperature"
},
"tiktoken_model_name": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "tiktoken_model_name",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "str",
"list": false,
"file_path": "",
"display_name": "Tiktoken Model Name"
},
"verbose": {
"required": false,
"placeholder": "",
"show": false,
"multiline": false,
"password": false,
"name": "verbose",
"advanced": false,
"dynamic": false,
"info": "",
"title_case": true,
"model_config": {},
"type": "bool",
"list": false,
"file_path": "",
"display_name": "Verbose"
},
"_type": "ChatOpenAI"
},
"description": "[*Deprecated*] `OpenAI` Chat large language models API.",
"base_classes": [
"ChatOpenAI",
"BaseLanguageModel",
"BaseChatModel",
"BaseLLM"
],
"display_name": "ChatOpenAI",
"documentation": "https://python.langchain.com/docs/modules/model_io/models/chat/integrations/openai",
"custom_fields": {},
"output_types": [],
"field_formatters": {},
"beta": false,
"official": false
},
"id": "ChatOpenAI-04157",
"value": null
},
"selected": true,
"positionAbsolute": {
"x": 487.6734875953049,
"y": -2569.9027779626213
},
"dragging": false
}
],
"edges": [
{
"source": "CombineDocsChain-RCPNA",
"target": "RetrievalQA-x5nap",
"sourceHandle": "CombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap",
"id": "reactflow__edge-CombineDocsChain-RCPNACombineDocsChain|CombineDocsChain-RCPNA|BaseCombineDocumentsChain|function-RetrievalQA-x5napBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "Milvus-CWpae",
"target": "RetrievalQA-x5nap",
"sourceHandle": "Milvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-x5nap",
"id": "reactflow__edge-Milvus-CWpaeMilvus|Milvus-CWpae|Milvus|VectorStore|BaseRetriever|VectorStoreRetriever-RetrievalQA-x5napBaseRetriever|retriever|RetrievalQA-x5nap",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "WebBaseLoader-af6db",
"target": "Milvus-CWpae",
"sourceHandle": "WebBaseLoader|WebBaseLoader-af6db|Document",
"targetHandle": "Document|documents|Milvus-CWpae",
"id": "reactflow__edge-WebBaseLoader-af6dbWebBaseLoader|WebBaseLoader-af6db|Document-Milvus-CWpaeDocument|documents|Milvus-CWpae",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "OpenAIProxyEmbedding-1771b",
"target": "Milvus-CWpae",
"sourceHandle": "OpenAIProxyEmbedding|OpenAIProxyEmbedding-1771b|Embeddings",
"targetHandle": "Embeddings|embedding|Milvus-CWpae",
"id": "reactflow__edge-OpenAIProxyEmbedding-1771bOpenAIProxyEmbedding|OpenAIProxyEmbedding-1771b|Embeddings-Milvus-CWpaeEmbeddings|embedding|Milvus-CWpae",
"style": {
"stroke": "#555"
},
"className": "stroke-gray-900 ",
"animated": false,
"selected": true
},
{
"source": "ChatOpenAI-04157",
"sourceHandle": "ChatOpenAI|ChatOpenAI-04157|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM",
"target": "CombineDocsChain-RCPNA",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-RCPNA",
"style": {
"stroke": "#555"
},
"className": "stroke-foreground stroke-connection",
"animated": true,
"id": "reactflow__edge-ChatOpenAI-04157ChatOpenAI|ChatOpenAI-04157|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM-CombineDocsChain-RCPNABaseLanguageModel|llm|CombineDocsChain-RCPNA",
"selected": true
}
],
"viewport": {
"zoom": 1,
"x": 0,
"y": 0
}
},
"is_component": false,
"name": "",
"description": "",
"id": "967b1",
"status": 0,
"write": false,
"guide_word": ""
}
}
}
},
{
"单文档问答": {
"id": "RetrievalQA-6a8ad",
"type": "InputFileNode",
"node": {
"output_types": [],
"display_name": "单文档问答",
"documentation": "",
"base_classes": [
"BaseRetrievalQA",
"RetrievalQA",
"Chain",
"function"
],
"description": "Chain for question-answering against an index.",
"template": {
"token_max_CombineDocsChain-Pud2p": {
"l2": false,
"info": "当前只对stuff 生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"l2_name": "token_max",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max",
"proxy": {
"id": "CombineDocsChain-Pud2p",
"field": "token_max"
}
},
"chain_type_CombineDocsChain-Pud2p": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "chain_type",
"proxy": {
"id": "CombineDocsChain-Pud2p",
"field": "chain_type"
}
},
"tags_RetrievalQA-qH6Mk": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags",
"proxy": {
"id": "RetrievalQA-qH6Mk",
"field": "tags"
}
},
"memory_RetrievalQA-qH6Mk": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory",
"proxy": {
"id": "RetrievalQA-qH6Mk",
"field": "memory"
}
},
"verbose_RetrievalQA-qH6Mk": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose",
"proxy": {
"id": "RetrievalQA-qH6Mk",
"field": "verbose"
}
},
"metadata_RetrievalQA-qH6Mk": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata",
"proxy": {
"id": "RetrievalQA-qH6Mk",
"field": "metadata"
}
},
"callbacks_RetrievalQA-qH6Mk": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks",
"proxy": {
"id": "RetrievalQA-qH6Mk",
"field": "callbacks"
}
},
"input_key_RetrievalQA-qH6Mk": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_key",
"proxy": {
"id": "RetrievalQA-qH6Mk",
"field": "input_key"
}
},
"input_node_RetrievalQA-qH6Mk": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question",
"proxy": {
"id": "RetrievalQA-qH6Mk",
"field": "input_node"
}
},
"output_key_RetrievalQA-qH6Mk": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "output_key",
"proxy": {
"id": "RetrievalQA-qH6Mk",
"field": "output_key"
}
},
"return_source_documents_RetrievalQA-qH6Mk": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_source_documents",
"proxy": {
"id": "RetrievalQA-qH6Mk",
"field": "return_source_documents"
}
},
"proxy_url_OpenAIProxyEmbedding-yvld7": {
"l2": false,
"info": "",
"list": false,
"name": "proxy_url",
"show": false,
"type": "str",
"value": "",
"l2_name": "proxy_url",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "http://43.133.35.137:8080",
"display_name": "proxy_url",
"proxy": {
"id": "OpenAIProxyEmbedding-yvld7",
"field": "proxy_url"
}
},
"file_path_InputFileNode-jdLkB": {
"l2": false,
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "file",
"value": "",
"l2_name": "file_path",
"advanced": true,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".jpg",
".png",
".jpeg",
".csv",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
".xlsx",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"jpg",
"png",
"jpeg",
"csv",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
"tiff",
"xlsx"
],
"file_path": null,
"multiline": false,
"placeholder": "",
"display_name": "输入内容",
"proxy": {
"id": "InputFileNode-jdLkB",
"field": "file_path"
}
},
"drop_old_Milvus-T3kRH": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "drop_old",
"proxy": {
"id": "Milvus-T3kRH",
"field": "drop_old"
}
},
"metadatas_Milvus-T3kRH": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas",
"proxy": {
"id": "Milvus-T3kRH",
"field": "metadatas"
}
},
"index_params_Milvus-T3kRH": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_params",
"proxy": {
"id": "Milvus-T3kRH",
"field": "index_params"
}
},
"search_kwargs_Milvus-T3kRH": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": false,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs",
"proxy": {
"id": "Milvus-T3kRH",
"field": "search_kwargs"
}
},
"search_params_Milvus-T3kRH": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_params",
"proxy": {
"id": "Milvus-T3kRH",
"field": "search_params"
}
},
"collection_name_Milvus-T3kRH": {
"l2": false,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "collection_name",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "collection_name",
"collection_id": "",
"proxy": {
"id": "Milvus-T3kRH",
"field": "collection_name"
}
},
"connection_args_Milvus-T3kRH": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": false,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "connection_args",
"proxy": {
"id": "Milvus-T3kRH",
"field": "connection_args"
}
},
"consistency_level_Milvus-T3kRH": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "consistency_level",
"proxy": {
"id": "Milvus-T3kRH",
"field": "consistency_level"
}
},
"metadata_ElemUnstructuredLoaderV0-LAGXM": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"value": "{}",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Metadata",
"proxy": {
"id": "ElemUnstructuredLoaderV0-LAGXM",
"field": "metadata"
}
},
"unstructured_api_url_ElemUnstructuredLoaderV0-LAGXM": {
"info": "",
"list": false,
"name": "unstructured_api_url",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "unstructured_api_url",
"proxy": {
"id": "ElemUnstructuredLoaderV0-LAGXM",
"field": "unstructured_api_url"
}
},
"chunk_size_RecursiveCharacterTextSplitter-a7fc9": {
"info": "",
"list": false,
"name": "chunk_size",
"show": true,
"type": "int",
"value": 1000,
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Chunk Size",
"proxy": {
"id": "RecursiveCharacterTextSplitter-a7fc9",
"field": "chunk_size"
}
},
"separators_RecursiveCharacterTextSplitter-a7fc9": {
"info": "",
"list": false,
"name": "separators",
"show": true,
"type": "str",
"value": "\\n\\n",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Separator",
"proxy": {
"id": "RecursiveCharacterTextSplitter-a7fc9",
"field": "separators"
}
},
"chunk_overlap_RecursiveCharacterTextSplitter-a7fc9": {
"info": "",
"list": false,
"name": "chunk_overlap",
"show": true,
"type": "int",
"value": 200,
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Chunk Overlap",
"proxy": {
"id": "RecursiveCharacterTextSplitter-a7fc9",
"field": "chunk_overlap"
}
},
"separator_type_RecursiveCharacterTextSplitter-a7fc9": {
"info": "",
"list": true,
"name": "separator_type",
"show": true,
"type": "str",
"value": "Text",
"options": [
"Text",
"cobol",
"cpp",
"csharp",
"go",
"html",
"java",
"js",
"kotlin",
"latex",
"markdown",
"php",
"proto",
"python",
"rst",
"ruby",
"rust",
"scala",
"sol",
"swift",
"ts"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Separator Type",
"proxy": {
"id": "RecursiveCharacterTextSplitter-a7fc9",
"field": "separator_type"
}
},
"n_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "n",
"show": false,
"type": "int",
"value": 1,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "N",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "n"
}
},
"name_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "name"
}
},
"tags_ChatOpenAI-1579b": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "tags"
}
},
"cache_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "cache",
"show": false,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "cache"
}
},
"client_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "client",
"show": false,
"type": "Any",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Client",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "client"
}
},
"verbose_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "verbose"
}
},
"metadata_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "metadata"
}
},
"callbacks_ChatOpenAI-1579b": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "callbacks"
}
},
"streaming_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "streaming",
"show": false,
"type": "bool",
"value": false,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "streaming"
}
},
"max_tokens_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "max_tokens",
"show": true,
"type": "int",
"dynamic": false,
"advanced": true,
"password": true,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Max Tokens",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "max_tokens"
},
"value": ""
},
"model_name_ChatOpenAI-1579b": {
"info": "",
"list": true,
"name": "model_name",
"show": true,
"type": "str",
"value": "gpt-3.5-turbo-0613",
"dynamic": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "model_name"
}
},
"http_client_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "http_client",
"show": false,
"type": "Any",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Http Client",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "http_client"
}
},
"max_retries_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "max_retries",
"show": false,
"type": "int",
"value": 2,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Max Retries",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "max_retries"
}
},
"temperature_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.7,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "temperature"
}
},
"async_client_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "async_client",
"show": false,
"type": "Any",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Async Client",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "async_client"
}
},
"model_kwargs_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "model_kwargs",
"show": false,
"type": "code",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Kwargs",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "model_kwargs"
}
},
"openai_proxy_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "openai_proxy",
"show": true,
"type": "str",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "OpenAI Proxy",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "openai_proxy"
}
},
"default_query_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "default_query",
"show": false,
"type": "code",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Default Query",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "default_query"
}
},
"openai_api_key_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "openai_api_key",
"show": true,
"type": "str",
"value": "",
"dynamic": false,
"advanced": true,
"password": true,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "OpenAI API Key",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "openai_api_key"
}
},
"default_headers_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "default_headers",
"show": false,
"type": "code",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Default Headers",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "default_headers"
}
},
"openai_api_base_ChatOpenAI-1579b": {
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"list": false,
"name": "openai_api_base",
"show": true,
"type": "str",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "OpenAI API Base",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "openai_api_base"
}
},
"request_timeout_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "request_timeout",
"show": false,
"type": "float",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Request Timeout",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "request_timeout"
}
},
"openai_organization_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "openai_organization",
"show": false,
"type": "str",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "OpenAI Organization",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "openai_organization"
}
},
"tiktoken_model_name_ChatOpenAI-1579b": {
"info": "",
"list": false,
"name": "tiktoken_model_name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tiktoken Model Name",
"model_config": {},
"proxy": {
"id": "ChatOpenAI-1579b",
"field": "tiktoken_model_name"
}
}
},
"flow": {
"data": {
"nodes": [
{
"width": 384,
"height": 376,
"id": "CombineDocsChain-Pud2p",
"data": {
"id": "CombineDocsChain-Pud2p",
"node": {
"template": {
"llm": {
"l2": false,
"info": "",
"list": false,
"name": "llm",
"show": true,
"type": "BaseLanguageModel",
"l2_name": "llm",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "LLM"
},
"_type": "load_qa_chain",
"token_max": {
"l2": false,
"info": "当前只对stuff 生效",
"list": false,
"name": "token_max",
"show": true,
"type": "int",
"value": -1,
"l2_name": "token_max",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "token_max"
},
"chain_type": {
"l2": false,
"info": "",
"list": true,
"name": "chain_type",
"show": true,
"type": "str",
"value": "stuff",
"l2_name": "chain_type",
"options": [
"stuff",
"map_reduce",
"map_rerank",
"refine"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "chain_type"
}
},
"description": "Load question answering chain.",
"base_classes": [
"BaseCombineDocumentsChain",
"function"
],
"display_name": "CombineDocsChain",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "CombineDocsChain",
"value": null
},
"type": "genericNode",
"dragging": false,
"position": {
"x": 1196.5168428585357,
"y": -1921.6230996659683
},
"selected": true,
"positionAbsolute": {
"x": 1196.5168428585357,
"y": -1921.6230996659683
}
},
{
"width": 384,
"height": 316,
"id": "RetrievalQA-qH6Mk",
"data": {
"id": "RetrievalQA-qH6Mk",
"node": {
"template": {
"tags": {
"l2": false,
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"l2_name": "tags",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "tags"
},
"_type": "RetrievalQA",
"memory": {
"l2": false,
"info": "",
"list": false,
"name": "memory",
"show": true,
"type": "BaseMemory",
"l2_name": "memory",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "memory"
},
"verbose": {
"l2": false,
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"l2_name": "verbose",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "verbose"
},
"metadata": {
"l2": false,
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"l2_name": "metadata",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadata"
},
"callbacks": {
"l2": false,
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain.callbacks.base.BaseCallbackHandler",
"l2_name": "callbacks",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "callbacks"
},
"input_key": {
"l2": false,
"info": "",
"list": false,
"name": "input_key",
"show": true,
"type": "str",
"value": "query",
"l2_name": "input_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "input_key"
},
"retriever": {
"l2": false,
"info": "",
"list": false,
"name": "retriever",
"show": true,
"type": "BaseRetriever",
"l2_name": "retriever",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"input_node": {
"l2": false,
"info": "",
"list": false,
"name": "input_node",
"show": true,
"type": "input",
"l2_name": "input_node",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Preset Question"
},
"output_key": {
"l2": false,
"info": "",
"list": false,
"name": "output_key",
"show": true,
"type": "str",
"value": "result",
"l2_name": "output_key",
"advanced": true,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "output_key"
},
"combine_documents_chain": {
"l2": false,
"info": "",
"list": false,
"name": "combine_documents_chain",
"show": true,
"type": "BaseCombineDocumentsChain",
"l2_name": "combine_documents_chain",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"return_source_documents": {
"l2": false,
"info": "",
"list": false,
"name": "return_source_documents",
"show": true,
"type": "bool",
"value": true,
"l2_name": "return_source_documents",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "return_source_documents"
}
},
"description": "Chain for question-answering against an index.",
"base_classes": [
"BaseRetrievalQA",
"RetrievalQA",
"Chain",
"function"
],
"display_name": "RetrievalQA",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/chains/popular/vector_db_qa"
},
"type": "RetrievalQA",
"value": null
},
"type": "genericNode",
"dragging": false,
"position": {
"x": 1872.7881530157297,
"y": -1758.750452738167
},
"selected": true,
"positionAbsolute": {
"x": 1872.7881530157297,
"y": -1758.750452738167
}
},
{
"width": 384,
"height": 172,
"id": "OpenAIProxyEmbedding-yvld7",
"data": {
"id": "OpenAIProxyEmbedding-yvld7",
"node": {
"template": {
"_type": "proxy_embedding",
"proxy_url": {
"l2": false,
"info": "",
"list": false,
"name": "proxy_url",
"show": false,
"type": "str",
"value": "",
"l2_name": "proxy_url",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "http://43.133.35.137:8080",
"display_name": "proxy_url"
}
},
"description": " 使用自建的embedding服务使用openai进行embed ",
"base_classes": [
"Embeddings"
],
"display_name": "OpenAIProxyEmbedding",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "OpenAIProxyEmbedding",
"value": null
},
"type": "genericNode",
"dragging": false,
"position": {
"x": 559.1495072972474,
"y": -790.6410754898252
},
"selected": true,
"positionAbsolute": {
"x": 559.1495072972474,
"y": -790.6410754898252
}
},
{
"width": 384,
"height": 254,
"id": "InputFileNode-jdLkB",
"data": {
"id": "InputFileNode-jdLkB",
"node": {
"template": {
"_type": "InputFileNode",
"file_path": {
"l2": false,
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "file",
"value": "",
"l2_name": "file_path",
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".jpg",
".png",
".jpeg",
".csv",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
".xlsx",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"jpg",
"png",
"jpeg",
"csv",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
"tiff",
"xlsx"
],
"file_path": null,
"multiline": false,
"placeholder": "",
"display_name": "输入内容"
}
},
"description": "输入节点，用来自动对接输入",
"base_classes": [
"fileNode"
],
"display_name": "InputFileNode",
"output_types": [],
"custom_fields": {},
"documentation": ""
},
"type": "InputFileNode",
"value": null
},
"type": "genericNode",
"dragging": false,
"position": {
"x": -93.84570908264766,
"y": -1503.8309966323186
},
"selected": true,
"positionAbsolute": {
"x": -93.84570908264766,
"y": -1503.8309966323186
}
},
{
"width": 384,
"height": 346,
"id": "Milvus-T3kRH",
"data": {
"id": "Milvus-T3kRH",
"node": {
"template": {
"_type": "Milvus",
"drop_old": {
"l2": false,
"info": "",
"list": false,
"name": "drop_old",
"show": false,
"type": "bool",
"value": false,
"l2_name": "drop_old",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "drop_old"
},
"documents": {
"l2": false,
"info": "",
"list": true,
"name": "documents",
"show": true,
"type": "Document",
"l2_name": "documents",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "Documents"
},
"embedding": {
"l2": false,
"info": "",
"list": false,
"name": "embedding",
"show": true,
"type": "Embeddings",
"l2_name": "embedding",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Embedding"
},
"metadatas": {
"l2": false,
"info": "",
"list": true,
"name": "metadatas",
"show": false,
"type": "code",
"l2_name": "metadatas",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "metadatas"
},
"index_params": {
"l2": false,
"info": "",
"list": false,
"name": "index_params",
"show": false,
"type": "code",
"l2_name": "index_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "index_params"
},
"search_kwargs": {
"l2": false,
"info": "",
"list": false,
"name": "search_kwargs",
"show": true,
"type": "code",
"value": "{}",
"l2_name": "search_kwargs",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_kwargs"
},
"search_params": {
"l2": false,
"info": "",
"list": false,
"name": "search_params",
"show": false,
"type": "code",
"l2_name": "search_params",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "search_params"
},
"collection_name": {
"l2": false,
"info": "",
"list": false,
"name": "collection_name",
"show": true,
"type": "str",
"value": "",
"l2_name": "collection_name",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "collection_name",
"collection_id": ""
},
"connection_args": {
"l2": false,
"info": "",
"list": false,
"name": "connection_args",
"show": true,
"type": "code",
"value": "",
"l2_name": "connection_args",
"advanced": true,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "connection_args"
},
"consistency_level": {
"l2": false,
"info": "",
"list": false,
"name": "consistency_level",
"show": false,
"type": "str",
"value": "Session",
"l2_name": "consistency_level",
"advanced": false,
"password": false,
"required": false,
"multiline": false,
"placeholder": "",
"display_name": "consistency_level"
}
},
"description": "Create a Milvus collection, indexes it with HNSW, and insert data.",
"base_classes": [
"VectorStore",
"Milvus",
"BaseRetriever",
"VectorStoreRetriever"
],
"display_name": "Milvus",
"output_types": [],
"custom_fields": {},
"documentation": "http://192.168.106.116:8030"
},
"type": "Milvus",
"value": null
},
"type": "genericNode",
"dragging": false,
"position": {
"x": 1329.9391213745507,
"y": -1099.88185650176
},
"selected": true,
"positionAbsolute": {
"x": 1329.9391213745507,
"y": -1099.88185650176
}
},
{
"width": 384,
"height": 392,
"id": "ElemUnstructuredLoaderV0-LAGXM",
"data": {
"id": "ElemUnstructuredLoaderV0-LAGXM",
"node": {
"template": {
"_type": "ElemUnstructuredLoaderV0",
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": true,
"type": "code",
"value": "{}",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Metadata"
},
"file_path": {
"info": "",
"list": false,
"name": "file_path",
"show": true,
"type": "fileNode",
"value": "",
"advanced": false,
"password": false,
"required": false,
"suffixes": [
".html",
".md",
".txt",
".jpg",
".png",
".jpeg",
".csv",
".doc",
".docx",
".pdf",
".ppt",
".pptx",
".xlsx",
".tiff"
],
"fileTypes": [
"html",
"md",
"txt",
"jpg",
"png",
"jpeg",
"csv",
"doc",
"docx",
"pdf",
"ppt",
"pptx",
"tiff",
"xlsx"
],
"multiline": false,
"placeholder": ""
},
"unstructured_api_url": {
"info": "",
"list": false,
"name": "unstructured_api_url",
"show": true,
"type": "str",
"value": "",
"advanced": false,
"password": true,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "unstructured_api_url"
}
},
"description": "Loads a PDF with pypdf and chunks at character level. dummy version",
"base_classes": [
"Document"
],
"display_name": "ElemUnstructuredLoaderV0",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": ""
},
"type": "ElemUnstructuredLoaderV0",
"value": null
},
"type": "genericNode",
"dragging": false,
"position": {
"x": 344.1888604075044,
"y": -1515.628333041525
},
"selected": true,
"positionAbsolute": {
"x": 344.1888604075044,
"y": -1515.628333041525
}
},
{
"width": 384,
"height": 540,
"id": "RecursiveCharacterTextSplitter-a7fc9",
"data": {
"id": "RecursiveCharacterTextSplitter-a7fc9",
"node": {
"template": {
"_type": "RecursiveCharacterTextSplitter",
"documents": {
"info": "",
"list": false,
"name": "documents",
"show": true,
"type": "Document",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": ""
},
"chunk_size": {
"info": "",
"list": false,
"name": "chunk_size",
"show": true,
"type": "int",
"value": 1000,
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Chunk Size"
},
"separators": {
"info": "",
"list": false,
"name": "separators",
"show": true,
"type": "str",
"value": "\\n\\n",
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Separator"
},
"chunk_overlap": {
"info": "",
"list": false,
"name": "chunk_overlap",
"show": true,
"type": "int",
"value": 200,
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Chunk Overlap"
},
"separator_type": {
"info": "",
"list": true,
"name": "separator_type",
"show": true,
"type": "str",
"value": "Text",
"options": [
"Text",
"cobol",
"cpp",
"csharp",
"go",
"html",
"java",
"js",
"kotlin",
"latex",
"markdown",
"php",
"proto",
"python",
"rst",
"ruby",
"rust",
"scala",
"sol",
"swift",
"ts"
],
"advanced": false,
"password": false,
"required": true,
"multiline": false,
"placeholder": "",
"display_name": "Separator Type"
}
},
"description": "Splitting text by recursively look at characters.",
"base_classes": [
"Document"
],
"display_name": "RecursiveCharacterTextSplitter",
"output_types": [
"Document"
],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/data_connection/document_transformers/text_splitters/recursive_text_splitter"
},
"type": "RecursiveCharacterTextSplitter",
"value": null
},
"type": "genericNode",
"dragging": false,
"position": {
"x": 813.860681674371,
"y": -1470.0417352310492
},
"selected": true,
"positionAbsolute": {
"x": 813.860681674371,
"y": -1470.0417352310492
}
},
{
"width": 384,
"height": 606,
"id": "ChatOpenAI-1579b",
"data": {
"id": "ChatOpenAI-1579b",
"node": {
"beta": false,
"official": false,
"template": {
"n": {
"info": "",
"list": false,
"name": "n",
"show": false,
"type": "int",
"value": 1,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "N",
"model_config": {}
},
"name": {
"info": "",
"list": false,
"name": "name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Name",
"model_config": {}
},
"tags": {
"info": "",
"list": true,
"name": "tags",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tags",
"model_config": {}
},
"_type": "ChatOpenAI",
"cache": {
"info": "",
"list": false,
"name": "cache",
"show": false,
"type": "bool",
"value": true,
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Cache",
"model_config": {}
},
"client": {
"info": "",
"list": false,
"name": "client",
"show": false,
"type": "Any",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Client",
"model_config": {}
},
"verbose": {
"info": "",
"list": false,
"name": "verbose",
"show": false,
"type": "bool",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Verbose",
"model_config": {}
},
"metadata": {
"info": "",
"list": false,
"name": "metadata",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Metadata",
"model_config": {}
},
"callbacks": {
"info": "",
"list": true,
"name": "callbacks",
"show": false,
"type": "langchain_core.callbacks.base.BaseCallbackHandler",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Callbacks",
"model_config": {}
},
"streaming": {
"info": "",
"list": false,
"name": "streaming",
"show": false,
"type": "bool",
"value": false,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Streaming",
"model_config": {}
},
"max_tokens": {
"info": "",
"list": false,
"name": "max_tokens",
"show": true,
"type": "int",
"dynamic": false,
"advanced": true,
"password": true,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Max Tokens",
"model_config": {}
},
"model_name": {
"info": "",
"list": true,
"name": "model_name",
"show": true,
"type": "str",
"value": "gpt-3.5-turbo-0613",
"dynamic": false,
"options": [
"gpt-3.5-turbo-0613",
"gpt-3.5-turbo",
"gpt-3.5-turbo-16k-0613",
"gpt-3.5-turbo-16k",
"gpt-4-0613",
"gpt-4-32k-0613",
"gpt-4",
"gpt-4-32k",
"gpt-4-1106-preview"
],
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Name",
"model_config": {}
},
"http_client": {
"info": "",
"list": false,
"name": "http_client",
"show": false,
"type": "Any",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Http Client",
"model_config": {}
},
"max_retries": {
"info": "",
"list": false,
"name": "max_retries",
"show": false,
"type": "int",
"value": 2,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Max Retries",
"model_config": {}
},
"temperature": {
"info": "",
"list": false,
"name": "temperature",
"show": true,
"type": "float",
"value": 0.7,
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Temperature",
"model_config": {}
},
"async_client": {
"info": "",
"list": false,
"name": "async_client",
"show": false,
"type": "Any",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Async Client",
"model_config": {}
},
"model_kwargs": {
"info": "",
"list": false,
"name": "model_kwargs",
"show": true,
"type": "code",
"dynamic": false,
"advanced": true,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Model Kwargs",
"model_config": {}
},
"openai_proxy": {
"info": "",
"list": false,
"name": "openai_proxy",
"show": true,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "OpenAI Proxy",
"model_config": {}
},
"default_query": {
"info": "",
"list": false,
"name": "default_query",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Default Query",
"model_config": {}
},
"openai_api_key": {
"info": "",
"list": false,
"name": "openai_api_key",
"show": true,
"type": "str",
"value": "",
"dynamic": false,
"advanced": false,
"password": true,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "OpenAI API Key",
"model_config": {}
},
"default_headers": {
"info": "",
"list": false,
"name": "default_headers",
"show": false,
"type": "code",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Default Headers",
"model_config": {}
},
"openai_api_base": {
"info": "\nThe base URL of the OpenAI API. Defaults to https://api.openai.com/v1.\n\nYou can change this to use other APIs like JinaChat, LocalAI and Prem.\n",
"list": false,
"name": "openai_api_base",
"show": true,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "OpenAI API Base",
"model_config": {}
},
"request_timeout": {
"info": "",
"list": false,
"name": "request_timeout",
"show": false,
"type": "float",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Request Timeout",
"model_config": {}
},
"openai_organization": {
"info": "",
"list": false,
"name": "openai_organization",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "OpenAI Organization",
"model_config": {}
},
"tiktoken_model_name": {
"info": "",
"list": false,
"name": "tiktoken_model_name",
"show": false,
"type": "str",
"dynamic": false,
"advanced": false,
"password": false,
"required": false,
"file_path": "",
"multiline": false,
"title_case": true,
"placeholder": "",
"display_name": "Tiktoken Model Name",
"model_config": {}
}
},
"description": "[*Deprecated*] `OpenAI` Chat large language models API.",
"base_classes": [
"ChatOpenAI",
"BaseLanguageModel",
"BaseChatModel",
"BaseLLM"
],
"display_name": "ChatOpenAI",
"output_types": [],
"custom_fields": {},
"documentation": "https://python.langchain.com/docs/modules/model_io/models/chat/integrations/openai",
"field_formatters": {}
},
"type": "ChatOpenAI",
"value": null
},
"type": "genericNode",
"dragging": false,
"position": {
"x": 611.2229342454906,
"y": -2218.8216244014975
},
"selected": true,
"positionAbsolute": {
"x": 611.2229342454906,
"y": -2218.8216244014975
}
}
],
"edges": [
{
"id": "reactflow__edge-CombineDocsChain-Pud2pCombineDocsChain|CombineDocsChain-Pud2p|BaseCombineDocumentsChain|function-RetrievalQA-qH6MkBaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-qH6Mk",
"style": {
"stroke": "#555"
},
"source": "CombineDocsChain-Pud2p",
"target": "RetrievalQA-qH6Mk",
"animated": false,
"selected": true,
"className": "stroke-gray-900 ",
"sourceHandle": "CombineDocsChain|CombineDocsChain-Pud2p|BaseCombineDocumentsChain|function",
"targetHandle": "BaseCombineDocumentsChain|combine_documents_chain|RetrievalQA-qH6Mk"
},
{
"id": "reactflow__edge-Milvus-T3kRHMilvus|Milvus-T3kRH|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever-RetrievalQA-qH6MkBaseRetriever|retriever|RetrievalQA-qH6Mk",
"style": {
"stroke": "#555"
},
"source": "Milvus-T3kRH",
"target": "RetrievalQA-qH6Mk",
"animated": false,
"selected": true,
"className": "stroke-gray-900 ",
"sourceHandle": "Milvus|Milvus-T3kRH|VectorStore|Milvus|BaseRetriever|VectorStoreRetriever",
"targetHandle": "BaseRetriever|retriever|RetrievalQA-qH6Mk"
},
{
"id": "reactflow__edge-OpenAIProxyEmbedding-yvld7OpenAIProxyEmbedding|OpenAIProxyEmbedding-yvld7|Embeddings-Milvus-T3kRHEmbeddings|embedding|Milvus-T3kRH",
"style": {
"stroke": "#555"
},
"source": "OpenAIProxyEmbedding-yvld7",
"target": "Milvus-T3kRH",
"animated": false,
"selected": true,
"className": "stroke-gray-900 ",
"sourceHandle": "OpenAIProxyEmbedding|OpenAIProxyEmbedding-yvld7|Embeddings",
"targetHandle": "Embeddings|embedding|Milvus-T3kRH"
},
{
"id": "reactflow__edge-InputFileNode-jdLkBInputFileNode|InputFileNode-jdLkB|fileNode-ElemUnstructuredLoaderV0-LAGXMfileNode|file_path|ElemUnstructuredLoaderV0-LAGXM",
"style": {
"stroke": "#555"
},
"source": "InputFileNode-jdLkB",
"target": "ElemUnstructuredLoaderV0-LAGXM",
"animated": false,
"selected": true,
"className": "stroke-gray-900 ",
"sourceHandle": "InputFileNode|InputFileNode-jdLkB|fileNode",
"targetHandle": "fileNode|file_path|ElemUnstructuredLoaderV0-LAGXM"
},
{
"id": "reactflow__edge-ElemUnstructuredLoaderV0-LAGXMElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-LAGXM|Document-RecursiveCharacterTextSplitter-a7fc9Document|documents|RecursiveCharacterTextSplitter-a7fc9",
"style": {
"stroke": "#555"
},
"source": "ElemUnstructuredLoaderV0-LAGXM",
"target": "RecursiveCharacterTextSplitter-a7fc9",
"animated": false,
"selected": true,
"className": "stroke-gray-900 ",
"sourceHandle": "ElemUnstructuredLoaderV0|ElemUnstructuredLoaderV0-LAGXM|Document",
"targetHandle": "Document|documents|RecursiveCharacterTextSplitter-a7fc9"
},
{
"id": "reactflow__edge-RecursiveCharacterTextSplitter-a7fc9RecursiveCharacterTextSplitter|RecursiveCharacterTextSplitter-a7fc9|Document-Milvus-T3kRHDocument|documents|Milvus-T3kRH",
"style": {
"stroke": "#555"
},
"source": "RecursiveCharacterTextSplitter-a7fc9",
"target": "Milvus-T3kRH",
"animated": false,
"selected": true,
"className": "stroke-gray-900 ",
"sourceHandle": "RecursiveCharacterTextSplitter|RecursiveCharacterTextSplitter-a7fc9|Document",
"targetHandle": "Document|documents|Milvus-T3kRH"
},
{
"id": "reactflow__edge-ChatOpenAI-1579bChatOpenAI|ChatOpenAI-1579b|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM-CombineDocsChain-Pud2pBaseLanguageModel|llm|CombineDocsChain-Pud2p",
"style": {
"stroke": "#555"
},
"source": "ChatOpenAI-1579b",
"target": "CombineDocsChain-Pud2p",
"animated": true,
"className": "stroke-foreground stroke-connection",
"sourceHandle": "ChatOpenAI|ChatOpenAI-1579b|ChatOpenAI|BaseLanguageModel|BaseChatModel|BaseLLM",
"targetHandle": "BaseLanguageModel|llm|CombineDocsChain-Pud2p",
"selected": true
}
],
"viewport": {
"zoom": 1,
"x": 0,
"y": 0
}
},
"is_component": false,
"name": "",
"description": "",
"id": "4d071",
"status": 0,
"write": false,
"guide_word": ""
}
}
}
}
]
================================================
FILE: src/backend/bisheng/telemetry/README.md
================================================
# 数据看板模块
================================================
FILE: src/backend/bisheng/telemetry/__init__.py
================================================
================================================
FILE: src/backend/bisheng/telemetry/domain/__init__.py
================================================
================================================
FILE: src/backend/bisheng/telemetry/domain/mid_table/__init__.py
================================================
================================================
FILE: src/backend/bisheng/telemetry/domain/mid_table/app_increment.py
================================================
from bisheng.common.constants.enums.telemetry import ApplicationTypeEnum
from bisheng.telemetry.domain.mid_table.base import BaseMidTable, BaseRecord
class AppIncrementRecord(BaseRecord):
app_id: str
app_name: str
app_type: ApplicationTypeEnum
class AppIncrement(BaseMidTable):
_index_name = 'mid_app_increment'
_mappings = {
"app_id": {"type": "keyword", "fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
"app_name": {"type": "keyword", "fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
"app_type": {"type": "keyword", "fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
}
================================================
FILE: src/backend/bisheng/telemetry/domain/mid_table/base.py
================================================
from typing import Dict, Any, List, Optional
from elasticsearch import Elasticsearch, AsyncElasticsearch, exceptions as es_exceptions, helpers
from loguru import logger
from pydantic import BaseModel, Field
from bisheng.common.schemas.telemetry.base_telemetry_schema import UserContext
from bisheng.common.services import telemetry_service
from bisheng.core.search.elasticsearch.manager import get_es_connection, get_es_connection_sync
common_properties = {
"user_id": {"type": "keyword",
"fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
"user_name": {"type": "keyword",
"fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
"user_group_infos": {
"type": "nested",
"properties": {
"user_group_id": {"type": "keyword",
"fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
"user_group_name": {"type": "keyword",
"fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}}
}
},
"user_role_infos": {
"type": "nested",
"properties": {
"role_id": {"type": "keyword",
"fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
"role_name": {"type": "keyword", "fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
"group_id": {"type": "keyword",
"fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
}
},
"timestamp": {"type": "date", "format": "strict_date_optional_time||epoch_second"},
}
common_settings = {
"analysis": {
"tokenizer": {
"single_char_tokenizer": {
"type": "ngram",
"min_gram": 1,
"max_gram": 1,
"token_chars": [
"letter",
"digit",
"punctuation",
"symbol"
]
}
},
"analyzer": {
"single_char_analyzer": {
"type": "custom",
"tokenizer": "single_char_tokenizer"
}
}
}
}
class BaseRecord(UserContext):
timestamp: int
es_id: Optional[str] = Field(default=None)
class BaseMidTable(BaseModel):
_index_name: str = ""
_mappings: Dict[str, Any] = {}
_es_client: AsyncElasticsearch = None
_es_client_sync: Elasticsearch = None
def __init__(self, **kwargs: Any):
super().__init__(**kwargs)
self.ensure_index_exists_sync()
async def ensure_index_exists(self) -> None:
if not self._index_name:
return None
if not self._es_client:
self._es_client = await get_es_connection()
mappings = self._mappings | common_properties
try:
exists = await self._es_client.indices.exists(index=self._index_name)
if not exists:
# Incoming body Applications Mapping
await self._es_client.indices.create(index=self._index_name,
body={"settings": common_settings,
"mappings": {"properties": mappings}})
except es_exceptions.RequestError as e:
# Ignore on concurrency creation "resource_already_exists_exception"
if "resource_already_exists_exception" not in str(e):
logger.error(f"Failed to create ES index: {e}")
raise e
return None
def ensure_index_exists_sync(self) -> None:
if not self._index_name:
return None
if not self._es_client_sync:
self._es_client_sync = get_es_connection_sync()
mappings = self._mappings | common_properties
try:
exists = self._es_client_sync.indices.exists(index=self._index_name)
if not exists:
# Incoming body Applications Mapping
self._es_client_sync.indices.create(index=self._index_name,
body={"settings": common_settings,
"mappings": {"properties": mappings}})
except es_exceptions.RequestError as e:
# Ignore on concurrency creation "resource_already_exists_exception"
if "resource_already_exists_exception" not in str(e):
logger.error(f"Failed to create ES index: {e}")
raise e
return None
def get_latest_record_time_sync(self) -> int | None:
""" Time to fetch the last record """
query = {
"size": 1,
"sort": [{"timestamp": {"order": "desc"}}],
"_source": ["timestamp"]
}
response = self._es_client_sync.search(index=self._index_name, body=query)
hits = response.get('hits', {}).get('hits', [])
if hits:
latest_time = hits[0]['_source']['timestamp']
return latest_time
return None
def insert_records_sync(self, records: List[BaseRecord]) -> None:
""" Batch Insert Record """
actions = []
for rec in records:
action = {
"_index": self._index_name,
"_source": rec.model_dump(exclude={"es_id"}),
}
if rec.es_id is not None:
action["_id"] = rec.es_id
actions.append(action)
helpers.bulk(self._es_client_sync, actions)
def search_from_base_sync(self, **kwargs) -> List[Dict[str, Any]]:
""" Synchronize search methods """
response = self._es_client_sync.search(index=telemetry_service.index_name, **kwargs)
return response.get('hits', {}).get('hits', [])
================================================
FILE: src/backend/bisheng/telemetry/domain/mid_table/knowledge_increment.py
================================================
from bisheng.telemetry.domain.mid_table.base import BaseMidTable, BaseRecord
class KnowledgeIncrementRecord(BaseRecord):
knowledge_id: int
knowledge_name: str
knowledge_type: int
class KnowledgeIncrement(BaseMidTable):
_index_name = 'mid_knowledge_increment'
_mappings = {
"knowledge_id": {"type": "keyword", "fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
"knowledge_name": {"type": "keyword", "fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
"knowledge_type": {"type": "keyword", "fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
}
================================================
FILE: src/backend/bisheng/telemetry/domain/mid_table/user_increment.py
================================================
from typing import Dict, Any
from .base import BaseMidTable, BaseRecord
class UserIncrementRecord(BaseRecord):
pass
class UserIncrement(BaseMidTable):
_index_name: str = 'mid_user_increment'
_mappings: Dict[str, Any] = {}
================================================
FILE: src/backend/bisheng/telemetry/domain/mid_table/user_interact.py
================================================
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum
from bisheng.telemetry.domain.mid_table.base import BaseMidTable, BaseRecord
class UserInteractRecord(BaseRecord):
message_id: int
event_id: str
interact_type: str
app_id: str
app_name: str
class UserInteract(BaseMidTable):
_index_name = 'mid_user_interact_dtl'
_mappings = {
"message_id": {"type": "keyword", "fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
"event_id": {"type": "keyword", "fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
"interact_type": {"type": "keyword", "fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
"app_id": {"type": "keyword", "fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
"app_name": {"type": "keyword", "fields": {"text": {"type": "text", "analyzer": "single_char_analyzer"}}},
}
def get_records_by_time_range_sync(self, start_time: int, end_time: int, page: int = 1, page_size: int = 1000):
query = {
"bool": {
"filter": [
{"term": {"event_type": {"value": BaseTelemetryTypeEnum.MESSAGE_FEEDBACK.value}}},
{"range": {"timestamp": {"gte": start_time * 1000, "lt": end_time * 1000}}}
]
}
}
return self.search_from_base_sync(body={"query": query, "from": (page - 1) * page_size, "size": page_size})
================================================
FILE: src/backend/bisheng/template/__init__.py
================================================
================================================
FILE: src/backend/bisheng/template/field/__init__.py
================================================
================================================
FILE: src/backend/bisheng/template/field/base.py
================================================
from typing import Any, Callable, Optional, Union
from bisheng.field_typing.range_spec import RangeSpec
from pydantic import BaseModel, ConfigDict, Field
class TemplateField(BaseModel):
model_config = ConfigDict()
field_type: str = Field(default='str', serialization_alias='type')
"""The type of field this is. Default is a string."""
required: bool = False
"""Specifies if the field is required. Defaults to False."""
placeholder: str = ''
"""A placeholder string for the field. Default is an empty string."""
is_list: bool = Field(default=False, serialization_alias='list')
"""Defines if the field is a list. Default is False."""
show: bool = True
"""Should the field be shown. Defaults to True."""
multiline: bool = False
"""Defines if the field will allow the user to open a text editor. Default is False."""
value: Any = ''
"""The value of the field. Default is None."""
suffixes: list[str] = []
fileTypes: list[str] = []
file_types: list[str] = Field(default_factory=list, serialization_alias='fileTypes')
"""List of file types associated with the field. Default is an empty list. (duplicate)"""
file_path: Optional[str] = ''
"""The file path of the field if it is a file. Defaults to None."""
password: bool = False
"""Specifies if the field is a password. Defaults to False."""
options: Optional[Union[list[str], Callable]] = None
"""List of options for the field. Only used when is_list=True. Default is an empty list."""
name: Optional[str] = None
"""Name of the field. Default is an empty string."""
display_name: Optional[str] = None
"""Display name of the field. Defaults to None."""
advanced: bool = False
"""Specifies if the field will an advanced parameter (hidden). Defaults to False."""
input_types: Optional[list[str]] = None
"""List of input types for the handle when the field has more than one type. Default is an empty list."""
dynamic: bool = False
"""Specifies if the field is dynamic. Defaults to False."""
info: Optional[str] = ''
"""Additional information about the field to be shown in the tooltip. Defaults to an empty string."""
refresh: Optional[bool] = None
"""Specifies if the field should be refreshed. Defaults to False."""
range_spec: Optional[RangeSpec] = Field(default=None, serialization_alias='rangeSpec')
"""Range specification for the field. Defaults to None."""
title_case: bool = True
"""Specifies if the field should be displayed in title case. Defaults to True."""
def to_dict(self):
result = self.dict()
for key in list(result.keys()):
if result[key] is None or result[key] == []:
del result[key]
result['type'] = result.pop('field_type')
result['list'] = result.pop('is_list')
if result.get('field_type') == 'float' and self.range_spec is None:
self.range_spec = RangeSpec()
if result.get('file_types'):
result['fileTypes'] = result.pop('file_types')
if self.field_type == 'file':
result['file_path'] = self.file_path
else:
result['file_path'] = ''
if result.get('display_name') is None:
value = self.name.replace('_', ' ')
if self.title_case:
value = value.title()
result['display_name'] = value
return result
================================================
FILE: src/backend/bisheng/template/frontend_node/__init__.py
================================================
from bisheng.template.frontend_node import (agents, chains, documentloaders, embeddings,
input_output, llms, memories, prompts, textsplitters,
tools, vectorstores)
__all__ = [
'agents',
'chains',
'embeddings',
'memories',
'tools',
'llms',
'prompts',
'vectorstores',
'documentloaders',
'textsplitters',
'input_output',
# 'output_parsers'
]
================================================
FILE: src/backend/bisheng/template/frontend_node/agents.py
================================================
from typing import Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
from bisheng.template.template.base import Template
from langchain.agents import types
NON_CHAT_AGENTS = {
agent_type: agent_class
for agent_type, agent_class in types.AGENT_TO_CLASS.items() if 'chat' not in agent_type.value
}
class AgentFrontendNode(FrontendNode):
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
if field.name in ['suffix', 'prefix', 'format_instructions', 'input_variables']:
field.show = False if name == 'ChatglmFunctionsAgent' else True
if field.name == 'Tools' and name == 'ZeroShotAgent':
field.field_type = 'BaseTool'
field.is_list = True
if 'path' == field.name and name == 'CSVAgent':
field.field_type = 'file'
field.required = True
field.show = True
field.value = ''
field.suffixes = ['.csv']
field.fileTypes = ['csv']
if name == 'OpenAIToolsAgent' and field.name in ['human_prompt', 'system_message']:
field.show = True
def add_extra_fields(self) -> None:
self.template.fields.append(
TemplateField(
field_type='BaseChatMemory',
required=False,
show=True,
name='memory',
advanced=False,
))
class SQLAgentNode(FrontendNode):
name: str = 'SQLAgent'
template: Template = Template(
type_name='sql_agent',
fields=[
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=False,
value='',
name='database_uri',
),
TemplateField(
field_type='BaseLanguageModel',
required=True,
show=True,
name='llm',
display_name='LLM',
),
],
)
description: str = """Construct an SQL agent from an LLM and tools."""
base_classes: list[str] = ['AgentExecutor', 'function']
def to_dict(self):
return super().to_dict()
class VectorStoreRouterAgentNode(FrontendNode):
name: str = 'VectorStoreRouterAgent'
template: Template = Template(
type_name='vectorstorerouter_agent',
fields=[
TemplateField(
field_type='VectorStoreRouterToolkit',
required=True,
show=True,
name='vectorstoreroutertoolkit',
display_name='Vector Store Router Toolkit',
),
TemplateField(
field_type='BaseLanguageModel',
required=True,
show=True,
name='llm',
display_name='LLM',
),
],
)
description: str = """Construct an agent from a Vector Store Router."""
base_classes: list[str] = ['AgentExecutor', 'function']
def to_dict(self):
return super().to_dict()
class VectorStoreAgentNode(FrontendNode):
name: str = 'VectorStoreAgent'
template: Template = Template(
type_name='vectorstore_agent',
fields=[
TemplateField(
field_type='VectorStoreInfo',
required=True,
show=True,
name='vectorstoreinfo',
display_name='Vector Store Info',
),
TemplateField(
field_type='BaseLanguageModel',
required=True,
show=True,
name='llm',
display_name='LLM',
),
],
)
description: str = """Construct an agent from a Vector Store."""
base_classes: list[str] = ['AgentExecutor', 'function']
def to_dict(self):
return super().to_dict()
class SQLDatabaseNode(FrontendNode):
name: str = 'SQLDatabase'
template: Template = Template(
type_name='sql_database',
fields=[
TemplateField(
field_type='str',
required=True,
is_list=False,
show=True,
multiline=False,
value='',
name='uri',
),
],
)
description: str = """SQLAlchemy wrapper around a database."""
base_classes: list[str] = ['SQLDatabase']
def to_dict(self):
return super().to_dict()
class InitializeAgentNode(FrontendNode):
name: str = 'AgentInitializer'
display_name: str = 'AgentInitializer'
template: Template = Template(
type_name='initialize_agent',
fields=[
TemplateField(
field_type='str',
required=True,
is_list=True,
show=True,
multiline=False,
options=list(NON_CHAT_AGENTS.keys()),
value=list(NON_CHAT_AGENTS.keys())[0],
name='agent',
advanced=False,
),
TemplateField(
field_type='BaseChatMemory',
required=False,
show=True,
name='memory',
advanced=False,
),
TemplateField(
field_type='Tool',
required=False,
show=True,
name='tools',
is_list=True,
advanced=False,
),
TemplateField(
field_type='BaseLanguageModel',
required=True,
show=True,
name='llm',
display_name='LLM',
advanced=False,
),
],
)
description: str = """Construct a zero shot agent from an LLM and tools."""
base_classes: list[str] = ['AgentExecutor', 'function', 'Chain']
def to_dict(self):
return super().to_dict()
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
# do nothing and don't return anything
pass
class JsonAgentNode(FrontendNode):
name: str = 'JsonAgent'
template: Template = Template(
type_name='json_agent',
fields=[
TemplateField(
field_type='BaseToolkit',
required=True,
show=True,
name='toolkit',
),
TemplateField(
field_type='BaseLanguageModel',
required=True,
show=True,
name='llm',
display_name='LLM',
),
],
)
description: str = """Construct a json agent from an LLM and tools."""
base_classes: list[str] = ['AgentExecutor', 'function']
def to_dict(self):
return super().to_dict()
================================================
FILE: src/backend/bisheng/template/frontend_node/autogenrole.py
================================================
from typing import Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
class AutogenRoleFrontNode(FrontendNode):
def add_extra_base_classes(self) -> None:
self.base_classes.append('ConversableAgent')
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
FrontendNode.format_field(field, name)
field.show = True
def add_extra_fields(self) -> None:
if self.name in {
'AutoGenAssistant',
'AutoGenGroupChatManager',
}:
self.template.add_field(
TemplateField(field_type='BaseLanguageModel',
required=True,
show=True,
name='llm',
advanced=False))
if self.name == 'AutoGenGroupChatManager':
self.template.add_field(
TemplateField(
field_type='ConversableAgent',
is_list=True,
required=False,
show=True,
name='agents',
advanced=False,
))
self.template.add_field(
TemplateField(field_type='int',
required=False,
show=True,
name='max_round',
advanced=False,
value=50))
self.template.add_field(
TemplateField(field_type='str',
required=False,
show=True,
name='system_message',
advanced=False,
value='Group chat manager.'))
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='name',
value='chat_manage',
advanced=False,
))
else:
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='name',
value='',
advanced=False,
))
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='system_message',
advanced=False,
))
if self.name == 'AutoGenUser':
self.template.add_field(
TemplateField(field_type='int',
required=False,
show=True,
name='max_consecutive_auto_reply',
advanced=False,
value=10))
self.template.add_field(
TemplateField(field_type='str',
required=True,
show=True,
name='human_input_mode',
advanced=False,
value='ALWAYS'))
if self.name == 'AutoGenCustomRole':
self.template.add_field(
TemplateField(
field_type='function',
required=True,
show=True,
name='func',
advanced=False,
))
================================================
FILE: src/backend/bisheng/template/frontend_node/base.py
================================================
import re
from collections import defaultdict
from typing import ClassVar, Dict, List, Optional, Union
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.constants import CLASSES_TO_REMOVE, FORCE_SHOW_FIELDS
from bisheng.template.frontend_node.formatter import field_formatters
from bisheng.template.template.base import Template
from bisheng.utils import constants
from pydantic import BaseModel, Field
class FieldFormatters(BaseModel):
formatters: ClassVar[Dict] = {
'openai_api_key': field_formatters.OpenAIAPIKeyFormatter(),
}
base_formatters: ClassVar[Dict] = {
'kwargs': field_formatters.KwargsFormatter(),
'optional': field_formatters.RemoveOptionalFormatter(),
'list': field_formatters.ListTypeFormatter(),
'dict': field_formatters.DictTypeFormatter(),
'union': field_formatters.UnionTypeFormatter(),
'multiline': field_formatters.MultilineFieldFormatter(),
'show': field_formatters.ShowFieldFormatter(),
'password': field_formatters.PasswordFieldFormatter(),
'default': field_formatters.DefaultValueFormatter(),
'headers': field_formatters.HeadersDefaultValueFormatter(),
'dict_code_file': field_formatters.DictCodeFileFormatter(),
'model_fields': field_formatters.ModelSpecificFieldFormatter(),
}
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
for key, formatter in self.base_formatters.items():
formatter.format(field, name)
for key, formatter in self.formatters.items():
if key == field.name:
formatter.format(field, name)
class FrontendNode(BaseModel):
_format_template: bool = True
template: Template
description: Optional[str] = None
icon: Optional[str] = None
base_classes: List[str]
name: str = ''
display_name: Optional[str] = ''
documentation: str = ''
custom_fields: Optional[Dict] = defaultdict(list)
output_types: List[str] = []
full_path: Optional[str] = None
field_formatters: FieldFormatters = Field(default_factory=FieldFormatters)
beta: bool = False
error: Optional[str] = None
# field formatters is an instance attribute but it is not used in the class
# so we need to create a method to get it
@staticmethod
def get_field_formatters() -> FieldFormatters:
return FieldFormatters()
def set_documentation(self, documentation: str) -> None:
"""Sets the documentation of the frontend node."""
self.documentation = documentation
# @field_serializer("base_classes")
# def process_base_classes(self, base_classes: List[str]) -> List[str]:
# """Removes unwanted base classes from the list of base classes."""
#
# return [base_class for base_class in base_classes if base_class not in CLASSES_TO_REMOVE]
# @field_serializer("display_name")
# def process_display_name(self, display_name: str) -> str:
# """Sets the display name of the frontend node."""
#
# return display_name or self.name
# For backwards compatibility
def to_dict(self, add_name=True) -> dict:
"""Returns a dict representation of the frontend node."""
result = self.dict(by_alias=True, exclude_none=True)
if hasattr(self, 'template') and hasattr(self.template, 'to_dict'):
format_func = self.format_field if self._format_template else None
result['template'] = self.template.to_dict(format_func)
name = result.pop('name')
result['display_name'] = result.get('display_name') if result.get(
'display_name') else self.name
result['base_classes'] = [
base_class for base_class in result['base_classes']
if base_class not in CLASSES_TO_REMOVE
]
dump = {name: result}
if not add_name:
return dump.pop(self.name)
return dump
def add_extra_fields(self) -> None:
pass
def add_extra_base_classes(self) -> None:
pass
def add_base_class(self, base_class: Union[str, List[str]]) -> None:
"""Adds a base class to the frontend node."""
if isinstance(base_class, str):
self.base_classes.append(base_class)
elif isinstance(base_class, list):
self.base_classes.extend(base_class)
def add_output_type(self, output_type: Union[str, List[str]]) -> None:
"""Adds an output type to the frontend node."""
if isinstance(output_type, str):
self.output_types.append(output_type)
elif isinstance(output_type, list):
self.output_types.extend(output_type)
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
"""Formats a given field based on its attributes and value."""
FrontendNode.get_field_formatters().format(field, name)
@staticmethod
def remove_optional(_type: str) -> str:
"""Removes 'Optional' wrapper from the type if present."""
return re.sub(r'Optional\[(.*)\]', r'\1', _type)
@staticmethod
def check_for_list_type(_type: str) -> tuple:
"""Checks for list type and returns the modified type and a boolean indicating if it's a list."""
is_list = 'List' in _type or 'Sequence' in _type
if is_list:
_type = re.sub(r'(List|Sequence)\[(.*)\]', r'\2', _type)
return _type, is_list
@staticmethod
def replace_mapping_with_dict(_type: str) -> str:
"""Replaces 'Mapping' with 'dict'."""
return _type.replace('Mapping', 'dict')
@staticmethod
def handle_union_type(_type: str) -> str:
"""Simplifies the 'Union' type to the first type in the Union."""
if 'Union' in _type:
_type = _type.replace('Union[', '')[:-1]
_type = _type.split(',')[0]
_type = _type.replace(']', '').replace('[', '')
return _type
@staticmethod
def handle_special_field(field, key: str, _type: str, SPECIAL_FIELD_HANDLERS) -> str:
"""Handles special field by using the respective handler if present."""
handler = SPECIAL_FIELD_HANDLERS.get(key)
return handler(field) if handler else _type
@staticmethod
def handle_dict_type(field: TemplateField, _type: str) -> str:
"""Handles 'dict' type by replacing it with 'code' or 'file' based on the field name."""
if 'dict' in _type.lower() and field.name == 'dict_':
field.field_type = 'file'
field.file_types = ['.json', '.yaml', '.yml']
elif _type.startswith('Dict') or _type.startswith('Mapping') or _type.startswith('dict'):
field.field_type = 'dict'
return _type
@staticmethod
def replace_default_value(field: TemplateField, value: dict) -> None:
"""Replaces default value with actual value if 'default' is present in value."""
if 'default' in value:
field.value = value['default']
@staticmethod
def handle_specific_field_values(field: TemplateField,
key: str,
name: Optional[str] = None) -> None:
"""Handles specific field values for certain fields."""
if key == 'headers':
field.value = """{"Authorization": "Bearer "}"""
FrontendNode._handle_model_specific_field_values(field, key, name)
FrontendNode._handle_api_key_specific_field_values(field, key, name)
@staticmethod
def _handle_model_specific_field_values(field: TemplateField,
key: str,
name: Optional[str] = None) -> None:
"""Handles specific field values related to models."""
model_dict = {
'OpenAI': constants.OPENAI_MODELS,
'ChatOpenAI': constants.CHAT_OPENAI_MODELS,
'Anthropic': constants.ANTHROPIC_MODELS,
'ChatAnthropic': constants.ANTHROPIC_MODELS,
}
if name in model_dict and key == 'model_name':
field.options = model_dict[name]
field.is_list = True
@staticmethod
def _handle_api_key_specific_field_values(field: TemplateField,
key: str,
name: Optional[str] = None) -> None:
"""Handles specific field values related to API keys."""
if 'api_key' in key and 'OpenAI' in str(name):
field.display_name = 'OpenAI API Key'
field.required = False
if field.value is None:
field.value = ''
@staticmethod
def handle_kwargs_field(field: TemplateField) -> None:
"""Handles kwargs field by setting certain attributes."""
if 'kwargs' in (field.name or '').lower():
field.advanced = True
field.required = False
field.show = False
@staticmethod
def handle_api_key_field(field: TemplateField, key: str) -> None:
"""Handles api key field by setting certain attributes."""
if 'api' in key.lower() and 'key' in key.lower():
field.required = False
field.advanced = False
field.display_name = key.replace('_', ' ').title()
field.display_name = field.display_name.replace('Api', 'API')
@staticmethod
def should_show_field(key: str, required: bool) -> bool:
"""Determines whether the field should be shown."""
return ((required and key not in ['input_variables']) or key in FORCE_SHOW_FIELDS
or 'api' in key or ('key' in key and 'input' not in key and 'output' not in key))
@staticmethod
def should_be_password(key: str, show: bool) -> bool:
"""Determines whether the field should be a password field."""
return any(text in key.lower() for text in {'password', 'token', 'api', 'key'}) and show
@staticmethod
def should_be_multiline(key: str) -> bool:
"""Determines whether the field should be multiline."""
return key in {
'suffix',
'prefix',
'template',
'examples',
'code',
'headers',
'description',
}
@staticmethod
def set_field_default_value(field: TemplateField, value: dict, key: str) -> None:
"""Sets the field value with the default value if present."""
if 'default' in value:
field.value = value['default']
if key == 'headers':
field.value = """{"Authorization": "Bearer "}"""
================================================
FILE: src/backend/bisheng/template/frontend_node/chains.py
================================================
from typing import Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
from bisheng.template.frontend_node.constants import QA_CHAIN_TYPES, SUMMARIZE_CHAIN_TYPES
from bisheng.template.template.base import Template
class ChainFrontendNode(FrontendNode):
def add_extra_fields(self) -> None:
if self.template.type_name == 'ConversationalRetrievalChain':
# add memory
self.template.add_field(
TemplateField(
field_type='BaseChatMemory',
required=True,
show=True,
name='memory',
advanced=False,
))
# add return_source_documents
self.template.add_field(
TemplateField(
field_type='bool',
required=False,
show=True,
name='return_source_documents',
advanced=False,
value=True,
display_name='Return source documents',
))
self.template.add_field(
TemplateField(
field_type='str',
required=True,
is_list=True,
show=True,
multiline=False,
options=QA_CHAIN_TYPES,
value=QA_CHAIN_TYPES[0],
name='chain_type',
advanced=False,
))
self.template.add_field(
TemplateField(
field_type='BasePromptTemplate',
show=True,
name='document_prompt',
advanced=False,
))
elif self.template.type_name == 'SequentialChain':
self.template.add_field(
TemplateField(field_type='str',
required=True,
show=True,
name='chain_order',
advanced=False,
value='[]'))
elif self.template.type_name in {'MultiPromptChain', 'MultiRuleChain'}:
self.template.add_field(
TemplateField(field_type='Chain',
required=True,
show=True,
is_list=True,
name='LLMChains',
advanced=False,
value='[]'))
self.template.add_field(
TemplateField(field_type='NestedDict',
required=True,
show=True,
is_list=True,
name='destination_chain_name',
advanced=False,
info='{chain_id: name}',
value='{}'))
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
FrontendNode.format_field(field, name)
if name == 'RuleBasedRouter' and field.name == 'rule_function':
field.field_type = 'function'
if name == 'RuleBasedRouter' and field.name == 'input_variables':
field.show = True
if name == 'LoaderOutputChain' and field.name == 'documents':
field.is_list = False
if name == 'RetrievalQA' and field.name == 'memory':
field.show = False
field.required = False
if name == 'APIChain' and field.name == 'limit_to_domains':
field.show = True
field.required = False
field.value = None
if name == 'BishengRetrievalQA' and field.name == 'max_content':
field.show = True
field.required = False
field.advanced = False
if 'key' in field.name and field.name != 'keyword_store': # Under special treatmentbishengRetrievalQAComponent
field.password = False
field.show = False
if field.name in ['input_key', 'output_key']:
field.required = True
field.show = True
field.advanced = False
# We should think of a way to deal with this later
# if field.field_type == "PromptTemplate":
# field.field_type = "str"
# field.multiline = True
# field.show = True
# field.advanced = False
# field.value = field.value.template
# Separated for possible future changes
if field.name == 'prompt' and field.value is None:
field.required = True
field.show = True
field.advanced = False
if field.name == 'prompt' and name == 'DalleGeneratorChain':
field.show = True
field.required = False
if field.name == 'condense_question_prompt':
field.required = False
field.show = True
if field.name in {'memory', 'document_prompt'}:
# field.required = False
field.show = True
field.advanced = False
if field.name == 'verbose':
field.required = False
field.show = False
field.advanced = True
if field.name == 'llm':
field.required = True
field.show = True
field.advanced = False
field.field_type = 'BaseLanguageModel' # temporary fix
field.is_list = False
if field.name == 'llm' and name == 'DalleGeneratorChain':
field.required = False
if field.name == 'return_source_documents':
field.required = False
field.show = True
field.advanced = True
field.value = True
if field.name == 'combine_docs_chain_kwargs':
field.show = True
field.field_type = 'BasePromptTemplate'
field.display_name = 'prompt'
if field.name == 'recipient':
field.display_name = 'AutogenRole'
if field.name == 'destination_chains':
field.show = False
if name == 'TransformChain' and field.name == 'input_variables':
field.show = True
if name == 'TransformChain' and field.name == 'transform_cb':
field.show = True
field.field_type = 'function'
field.name = 'transform'
if name == 'TransformChain' and field.name == 'atransform_cb':
field.show = True
field.field_type = 'function'
field.name = 'atransform'
if name in ['QAGenerationChain', 'QAGenerationChainV2']:
if field.name in ['k', 'chunk_size', 'question_prompt', 'answer_prompt', 'filter_lowquality_context',
'filter_lowquality_question']:
field.show = True
elif field.name == 'input_node':
field.show = False
class SeriesCharacterChainNode(FrontendNode):
name: str = 'SeriesCharacterChain'
template: Template = Template(
type_name='SeriesCharacterChain',
fields=[
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
advanced=False,
multiline=False,
name='character',
),
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
advanced=False,
multiline=False,
name='series',
),
TemplateField(
field_type='BaseLanguageModel',
required=True,
placeholder='',
is_list=False,
show=True,
advanced=False,
multiline=False,
name='llm',
display_name='LLM',
),
],
)
description: str = 'SeriesCharacterChain is a chain you can use to have a conversation with a character from a series.' # noqa
base_classes: list[str] = [
'LLMChain',
'BaseCustomChain',
'Chain',
'ConversationChain',
'SeriesCharacterChain',
'function',
]
class TimeTravelGuideChainNode(FrontendNode):
name: str = 'TimeTravelGuideChain'
template: Template = Template(
type_name='TimeTravelGuideChain',
fields=[
TemplateField(
field_type='BaseLanguageModel',
required=True,
placeholder='',
is_list=False,
show=True,
advanced=False,
multiline=False,
name='llm',
display_name='LLM',
),
TemplateField(
field_type='BaseChatMemory',
required=False,
show=True,
name='memory',
advanced=False,
),
],
)
description: str = 'Time travel guide chain.'
base_classes: list[str] = [
'LLMChain',
'BaseCustomChain',
'TimeTravelGuideChain',
'Chain',
'ConversationChain',
]
class MidJourneyPromptChainNode(FrontendNode):
name: str = 'MidJourneyPromptChain'
template: Template = Template(
type_name='MidJourneyPromptChain',
fields=[
TemplateField(
field_type='BaseLanguageModel',
required=True,
placeholder='',
is_list=False,
show=True,
advanced=False,
multiline=False,
name='llm',
display_name='LLM',
),
TemplateField(
field_type='BaseChatMemory',
required=False,
show=True,
name='memory',
advanced=False,
),
],
)
description: str = 'MidJourneyPromptChain is a chain you can use to generate new MidJourney prompts.'
base_classes: list[str] = [
'LLMChain',
'BaseCustomChain',
'Chain',
'ConversationChain',
'MidJourneyPromptChain',
]
class CombineDocsChainNode(FrontendNode):
name: str = 'CombineDocsChain'
template: Template = Template(
type_name='load_qa_chain',
fields=[
TemplateField(
field_type='str',
required=True,
is_list=True,
show=True,
multiline=False,
options=QA_CHAIN_TYPES,
value=QA_CHAIN_TYPES[0],
name='chain_type',
advanced=False,
),
TemplateField(
field_type='BaseLanguageModel',
required=True,
show=True,
name='llm',
display_name='LLM',
advanced=False,
),
TemplateField(
field_type='int',
required=False,
show=True,
name='token_max',
display_name='token_max',
advanced=False,
info='Hide Content Only ForStuffType Effective',
value=-1,
),
TemplateField(field_type='BasePromptTemplate',
required=False,
show=True,
name='prompt',
display_name='prompt',
advanced=False,
info='Hide Content Only ForStuffType Effective'),
TemplateField(
field_type='BasePromptTemplate',
required=False,
show=True,
name='document_prompt',
advanced=False,
)
],
)
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
pass
description: str = """Load question answering chain."""
base_classes: list[str] = ['BaseCombineDocumentsChain', 'function']
class SummarizeDocsChain(FrontendNode):
name: str = 'SummarizeDocsChain'
template: Template = Template(
type_name='load_summarize_chain',
fields=[
TemplateField(
field_type='str',
required=True,
is_list=True,
show=True,
multiline=False,
options=SUMMARIZE_CHAIN_TYPES,
value=SUMMARIZE_CHAIN_TYPES[0],
name='chain_type',
advanced=False,
),
TemplateField(
field_type='BaseLanguageModel',
required=True,
show=True,
name='llm',
display_name='LLM',
advanced=False,
),
TemplateField(
field_type='int',
required=False,
show=True,
name='token_max',
display_name='token_max',
advanced=False,
info='Currently only forstuff ENTRY INTO FORCE',
value=-1,
),
TemplateField(field_type='BasePromptTemplate',
required=False,
show=True,
name='prompt',
display_name='prompt',
advanced=False,
info='Hide Content Only ForStuffType Effective')
],
)
description: str = """Load summarize chain."""
base_classes: list[str] = ['BaseCombineDocumentsChain', 'function']
================================================
FILE: src/backend/bisheng/template/frontend_node/constants.py
================================================
FORCE_SHOW_FIELDS = [
'allowed_tools',
'memory',
'prefix',
'examples',
'temperature',
'model_name',
'headers',
'max_value_length',
'max_tokens',
'file_path',
'input_node',
'output_node',
]
DEFAULT_PROMPT = """
I want you to act as a naming consultant for new companies.
Here are some examples of good company names:
- search engine, Google
- social media, Facebook
- video sharing, YouTube
The name should be short, catchy and easy to remember.
What is a good name for a company that makes {product}?
"""
SYSTEM_PROMPT = """
You are a helpful assistant that talks casually about life in general.
You are a good listener and you can talk about anything.
"""
HUMAN_PROMPT = '{input}'
QA_CHAIN_TYPES = ['stuff', 'map_reduce', 'map_rerank', 'refine']
SUMMARIZE_CHAIN_TYPES = ['stuff', 'map_reduce', 'refine']
CTRANSFORMERS_DEFAULT_CONFIG = {
'top_k': 40,
'top_p': 0.95,
'temperature': 0.8,
'repetition_penalty': 1.1,
'last_n_tokens': 64,
'seed': -1,
'max_new_tokens': 256,
'stop': None,
'stream': False,
'reset': True,
'batch_size': 8,
'threads': -1,
'context_length': -1,
'gpu_layers': 0,
}
# This variable is used to tell the user
# that it can be changed to use other APIs
# like Prem and LocalAI
OPENAI_API_BASE_INFO = """
The base URL of the OpenAI API. Defaults to https://api.openai.com/v1.
You can change this to use other APIs like JinaChat, LocalAI and Prem.
"""
INPUT_KEY_INFO = """The variable to be used as Chat Input when more than one variable is available."""
OUTPUT_KEY_INFO = """The variable to be used as Chat Output (e.g. answer in a ConversationalRetrievalChain)"""
CLASSES_TO_REMOVE = [
'RunnableSerializable',
'Serializable',
'BaseModel',
'object',
'Runnable',
'Generic',
]
================================================
FILE: src/backend/bisheng/template/frontend_node/custom_components.py
================================================
from typing import Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
from bisheng.template.template.base import Template
DEFAULT_CUSTOM_COMPONENT_CODE = """from bisheng import CustomComponent
from typing import Optional, List, Dict, Union
from bisheng.field_typing import (
AgentExecutor,
BaseChatMemory,
BaseLanguageModel,
BaseLLM,
BaseLoader,
BaseMemory,
BaseOutputParser,
BasePromptTemplate,
BaseRetriever,
Callable,
Chain,
ChatPromptTemplate,
Data,
Document,
Embeddings,
NestedDict,
Object,
PromptTemplate,
TextSplitter,
Tool,
VectorStore,
)
class Component(CustomComponent):
display_name: str = "Custom Component"
description: str = "Create any custom component you want!"
def build_config(self):
return {"param": {"display_name": "Parameter"}}
def build(self, param: Data) -> Data:
return param
"""
class CustomComponentFrontendNode(FrontendNode):
_format_template: bool = False
name: str = 'CustomComponent'
display_name: Optional[str] = 'CustomComponent'
beta: bool = True
template: Template = Template(
type_name='CustomComponent',
fields=[
TemplateField(
field_type='code',
required=True,
placeholder='',
is_list=False,
show=True,
value=DEFAULT_CUSTOM_COMPONENT_CODE,
name='code',
advanced=False,
dynamic=True,
)
],
)
description: Optional[str] = None
base_classes: list[str] = []
================================================
FILE: src/backend/bisheng/template/frontend_node/documentloaders.py
================================================
from typing import ClassVar, Dict, Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
def build_file_field(suffixes: list,
fileTypes: list,
name: str = 'file_path',
fieldType='fileNode') -> TemplateField:
"""Build a template field for a document loader."""
return TemplateField(
field_type=fieldType,
required=True,
show=True,
name=name,
value='',
suffixes=suffixes,
fileTypes=fileTypes,
)
class DocumentLoaderFrontNode(FrontendNode):
def add_extra_base_classes(self) -> None:
self.base_classes = ['Document']
self.output_types = ['Document']
file_path_templates: ClassVar[Dict] = {
'ElemUnstructuredLoaderV0':
build_file_field(
suffixes=[
'.html', '.md', '.txt', '.bmp', '.jpg', '.png', '.jpeg', '.doc', '.docx', '.pdf',
'.ppt', '.pptx', '.xls', '.xlsx', '.tsv', '.csv', '.tiff'
],
fileTypes=[
'html',
'md',
'txt',
'bmp',
'jpg',
'png',
'jpeg',
'doc',
'docx',
'pdf',
'ppt',
'pptx',
'.xls',
'xlsx',
'tsv',
'csv',
'tiff',
],
),
'AirbyteJSONLoader':
build_file_field(suffixes=['.json'], fileTypes=['json']),
'CoNLLULoader':
build_file_field(suffixes=['.csv'], fileTypes=['csv']),
'CSVLoader':
build_file_field(suffixes=['.csv'], fileTypes=['csv']),
'UnstructuredEmailLoader':
build_file_field(suffixes=['.eml'], fileTypes=['eml']),
'SlackDirectoryLoader':
build_file_field(suffixes=['.zip'], fileTypes=['zip']),
'EverNoteLoader':
build_file_field(suffixes=['.xml'], fileTypes=['xml']),
'FacebookChatLoader':
build_file_field(suffixes=['.json'], fileTypes=['json']),
'BSHTMLLoader':
build_file_field(suffixes=['.html'], fileTypes=['html']),
'UnstructuredHTMLLoader':
build_file_field(suffixes=['.html'], fileTypes=['html']),
'UnstructuredImageLoader':
build_file_field(
suffixes=['.jpg', '.jpeg', '.png', '.gif', '.bmp'],
fileTypes=['jpg', 'jpeg', 'png', 'gif', 'bmp'],
),
'UnstructuredMarkdownLoader':
build_file_field(suffixes=['.md'], fileTypes=['md']),
'PyPDFLoader':
build_file_field(suffixes=['.pdf'], fileTypes=['pdf'], fieldType='fileNode'),
'UnstructuredPowerPointLoader':
build_file_field(suffixes=['.pptx', '.ppt'], fileTypes=['pptx', 'ppt']),
'SRTLoader':
build_file_field(suffixes=['.srt'], fileTypes=['srt']),
'TelegramChatLoader':
build_file_field(suffixes=['.json'], fileTypes=['json']),
'TextLoader':
build_file_field(suffixes=['.txt'], fileTypes=['txt']),
'UnstructuredWordDocumentLoader':
build_file_field(suffixes=['.docx', '.doc'], fileTypes=['docx', 'doc']),
'PDFWithSemanticLoader':
build_file_field(suffixes=['.pdf'], fileTypes=['pdf']),
'UniversalKVLoader':
build_file_field(
suffixes=['.jpg', '.png', '.jpeg', '.bmp', '.pdf'],
fileTypes=['jpg', 'png', 'jpeg', 'bmp', 'pdf'],
),
'CustomKVLoader':
build_file_field(
suffixes=[
'.jpg', '.png', '.jpeg', '.pdf', '.txt', '.docx', '.doc', '.bmp', '.tif', '.tiff',
'.xls', '.xlsx'
],
fileTypes=[
'jpg', 'png', 'jpeg', 'pdf', 'txt', 'docx', 'doc', 'bmp', 'tif', 'tiff', 'xls',
'xlsx'
],
),
}
def add_extra_fields(self) -> None:
name = None
display_name = 'Web Page'
if self.template.type_name in {'PDFWithSemanticLoader'}:
for field in build_pdf_semantic_loader_fields():
self.template.add_field(field)
if self.template.type_name in {'GitLoader'}:
# Add fields repo_path, clone_url, branch and file_filter
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='repo_path',
value='',
display_name='Path to repository',
advanced=False,
))
self.template.add_field(
TemplateField(
field_type='str',
required=False,
show=True,
name='clone_url',
value='',
display_name='Clone URL',
advanced=False,
))
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='branch',
value='',
display_name='Branch',
advanced=False,
))
self.template.add_field(
TemplateField(
field_type='str',
required=False,
show=True,
name='file_filter',
value='',
display_name='File extensions (comma-separated)',
advanced=False,
))
elif self.template.type_name in {'ElemUnstructuredLoaderV0'}:
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='unstructured_api_url',
value='',
display_name='unstructured_api_url',
advanced=False,
))
self.template.add_field(
TemplateField(
field_type='dict',
required=True,
show=True,
name='kwargs',
display_name='kwargs',
advanced=False,
))
self.template.add_field(self.file_path_templates[self.template.type_name])
elif self.template.type_name in {'UniversalKVLoader'}:
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='ellm_model_url',
value='',
display_name='ellm_model_url',
advanced=False,
))
self.template.add_field(
TemplateField(
field_type='int',
required=False,
show=True,
name='max_pages',
value=30,
display_name='max_pages',
advanced=False,
))
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='schema',
value='',
display_name='schema',
advanced=False,
))
self.template.add_field(self.file_path_templates[self.template.type_name])
elif self.template.type_name in {'CustomKVLoader'}:
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='schemas',
advanced=False,
))
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='elm_api_base_url',
advanced=False,
))
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='elm_api_key',
advanced=False,
))
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='elem_server_id',
advanced=False,
))
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='task_type',
advanced=False,
))
self.template.add_field(
TemplateField(field_type='int',
required=True,
show=True,
name='request_timeout',
advanced=True,
value=600))
self.template.add_field(self.file_path_templates[self.template.type_name])
elif self.template.type_name in self.file_path_templates:
self.template.add_field(self.file_path_templates[self.template.type_name])
elif self.template.type_name in {'FireCrawlLoader'}:
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='url',
value='',
display_name='URL',
advanced=False,
))
self.template.add_field(
TemplateField(field_type='dict',
multiline=True,
required=False,
show=True,
name='params'))
self.template.add_field(
TemplateField(field_type='str',
multiline=True,
required=False,
show=True,
name='api_key'))
self.template.add_field(
TemplateField(field_type='str',
multiline=True,
required=True,
value='crawl',
options=['crawl', 'scrape'],
show=True,
name='mode'))
elif self.template.type_name in {
'WebBaseLoader',
'AZLyricsLoader',
'CollegeConfidentialLoader',
'HNLoader',
'IFixitLoader',
'IMSDbLoader',
'GutenbergLoader',
}:
name = 'web_path'
elif self.template.type_name in {'GutenbergLoader'}:
name = 'file_path'
elif self.template.type_name in {'GitbookLoader'}:
name = 'web_page'
elif self.template.type_name in {
'DirectoryLoader',
'ReadTheDocsLoader',
'NotionDirectoryLoader',
'PyPDFDirectoryLoader',
}:
name = 'path'
display_name = 'Local directory'
if self.template.type_name == 'WebBaseLoader':
self.template.add_field(
TemplateField(
field_type='code',
required=True,
show=True,
name='requests_kwargs',
value='{"timeout": 30}',
display_name='RequestsKwargs',
multiline=False,
))
if name:
if self.template.type_name in {'DirectoryLoader'}:
for field in build_directory_loader_fields():
self.template.add_field(field)
else:
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name=name,
value='',
display_name=display_name,
))
# add a metadata field of type dict
self.template.add_field(
TemplateField(
field_type='code',
required=True,
show=True,
name='metadata',
value='{}',
display_name='Metadata',
multiline=False,
))
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
FrontendNode.format_field(field, name)
if field.name == 'metadata':
field.show = True
field.advanced = False
field.show = True
if field.name == 'unstructured_api_url':
field.show = True
field.advanced = False
if name == 'CustomKVLoader' and field.name == 'task_type':
field.options = ['extraction-job', 'logic-job']
field.value = 'logic-job'
if name == 'CustomKVLoader' and field.name == 'schemas':
field.field_type = 'str'
field.info = "please use '|' seperate"
field.is_list = False
if name == 'ElemUnstructuredLoaderV0' and field.name == 'kwargs':
field.show = True
field.advanced = False
def build_pdf_semantic_loader_fields():
# file_path: str,
# password: Optional[Union[str, bytes]] = None,
# layout_api_key: str = None,
# layout_api_url: str = None,
# is_join_table: bool = True,
# with_columns: bool = False,
# support_rotate: bool = False,
# text_elem_sep: str = '\n',
# start: int = 0,
# n: int = None,
# verbose: bool = False
file_path = TemplateField(field_type='file',
required=True,
show=True,
name='file_path',
value='',
display_name='pdfDoc.')
password = TemplateField(field_type='str',
required=True,
show=True,
advanced=True,
name='password',
value=None,
display_name='password')
layout_api_key = TemplateField(field_type='str',
required=False,
show=True,
name='layout_api_key',
value=None,
display_name='layout_api_key')
layout_api_url = TemplateField(field_type='str',
required=False,
show=True,
name='layout_api_url',
value=None,
display_name='layout_api_url')
is_join_table = TemplateField(field_type='bool',
required=True,
show=True,
advanced=True,
name='is_join_table',
value='True',
display_name='is_join_table')
with_columns = TemplateField(field_type='bool',
required=True,
show=True,
advanced=True,
name='with_columns',
value='False',
display_name='with_columns')
support_rotate = TemplateField(field_type='bool',
required=True,
show=True,
advanced=True,
name='support_rotate',
value='False',
display_name='support_rotate')
text_elem_sep = TemplateField(field_type='str',
required=True,
show=True,
advanced=True,
name='text_elem_sep',
value='\\n',
display_name='text_elem_sep')
start = TemplateField(field_type='int',
required=True,
show=True,
advanced=True,
name='start',
value=0,
display_name='start')
n = TemplateField(field_type='int',
required=False,
show=True,
advanced=True,
name='n',
value=None,
display_name='n')
verbose = TemplateField(field_type='bool',
required=True,
show=True,
advanced=True,
name='verbose',
value='False',
display_name='verbose')
return (file_path, password, layout_api_key, layout_api_url, n, verbose, is_join_table,
with_columns, support_rotate, text_elem_sep, start)
def build_directory_loader_fields():
# if loader_kwargs is None:
# loader_kwargs = {}
# self.path = path
# self.glob = glob
# self.load_hidden = load_hidden
# self.loader_cls = loader_cls
# self.loader_kwargs = loader_kwargs
# self.silent_errors = silent_errors
# self.recursive = recursive
# self.show_progress = show_progress
# self.use_multithreading = use_multithreading
# self.max_concurrency = max_concurrency
# Based on the above fields, we can build the following fields:
# path, glob, load_hidden, silent_errors, recursive, show_progress, use_multithreading, max_concurrency
# path
path = TemplateField(
field_type='str',
required=True,
show=True,
name='path',
value='',
display_name='Local directory',
advanced=False,
)
# glob
glob = TemplateField(
field_type='str',
required=True,
show=True,
name='glob',
value='**/*.txt',
display_name='glob',
advanced=False,
)
# load_hidden
load_hidden = TemplateField(
field_type='bool',
required=False,
show=True,
name='load_hidden',
value='False',
display_name='Load hidden files',
advanced=True,
)
# silent_errors
silent_errors = TemplateField(
field_type='bool',
required=False,
show=True,
name='silent_errors',
value='False',
display_name='Silent errors',
advanced=True,
)
# recursive
recursive = TemplateField(
field_type='bool',
required=False,
show=True,
name='recursive',
value='True',
display_name='Recursive',
advanced=True,
)
# use_multithreading
use_multithreading = TemplateField(
field_type='bool',
required=False,
show=True,
name='use_multithreading',
value='True',
display_name='Use multithreading',
advanced=True,
)
# max_concurrency
max_concurrency = TemplateField(
field_type='int',
required=False,
show=True,
name='max_concurrency',
value=10,
display_name='Max concurrency',
advanced=True,
)
return (
path,
glob,
load_hidden,
silent_errors,
recursive,
use_multithreading,
max_concurrency,
)
================================================
FILE: src/backend/bisheng/template/frontend_node/embeddings.py
================================================
from typing import Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
from bisheng.template.template.base import Template
class EmbeddingFrontendNode(FrontendNode):
@staticmethod
def format_jina_fields(field: TemplateField):
if 'jina' in field.name:
field.show = True
field.advanced = False
if 'auth' in field.name or 'token' in field.name:
field.password = True
field.show = True
field.advanced = False
if field.name == 'jina_api_url':
field.show = True
field.advanced = True
field.display_name = 'Jina API URL'
field.password = False
@staticmethod
def format_openai_fields(field: TemplateField):
if 'openai' in field.name:
field.show = True
field.advanced = True
split_name = field.name.split('_')
title_name = ' '.join([s.capitalize() for s in split_name])
field.display_name = title_name.replace('Openai', 'OpenAI').replace('Api', 'API')
if 'api_key' in field.name:
field.password = True
field.show = True
field.advanced = False
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
FrontendNode.format_field(field, name)
field.advanced = not field.required
field.show = True
if field.name == 'headers':
field.show = False
field.value = ''
if field.name == 'host_base_url':
field.show = True
field.advanced = False
if field.name == 'model':
field.show = True
field.advanced = False
if name == 'BishengEmbedding':
if field.name == 'model_id':
field.show = True
field.display_name = 'Model Name'
field.field_type = 'bisheng_embedding'
field.advanced = False
elif field.name in ['model', 'llm_node_type', "app_type", "app_id", "app_name", "user_id"]:
field.show = False
# Format Jina fields
EmbeddingFrontendNode.format_jina_fields(field)
EmbeddingFrontendNode.format_openai_fields(field)
class OpenAIProxyEmbedding(FrontendNode):
name: str = 'OpenAIProxyEmbedding'
description: str = """ Use self-builtembeddingService UsageopenaiTo be performedembed """
base_classes: list[str] = ['Embeddings']
template: Template = Template(type_name='proxy_embedding',
fields=[
TemplateField(
field_type='str',
required=False,
placeholder='http://43.133.35.137:8080',
is_list=False,
show=True,
multiline=False,
value='',
name='proxy_url',
),
])
def to_dict(self):
return super().to_dict()
================================================
FILE: src/backend/bisheng/template/frontend_node/formatter/__init__.py
================================================
================================================
FILE: src/backend/bisheng/template/frontend_node/formatter/base.py
================================================
from abc import ABC, abstractmethod
from typing import Optional
from bisheng.template.field.base import TemplateField
from pydantic import BaseModel
class FieldFormatter(BaseModel, ABC):
@abstractmethod
def format(self, field: TemplateField, name: Optional[str]) -> None:
pass
================================================
FILE: src/backend/bisheng/template/frontend_node/formatter/field_formatters.py
================================================
import re
from typing import ClassVar, Dict, Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.constants import FORCE_SHOW_FIELDS
from bisheng.template.frontend_node.formatter.base import FieldFormatter
from bisheng.utils.constants import ANTHROPIC_MODELS, CHAT_OPENAI_MODELS, OPENAI_MODELS
class OpenAIAPIKeyFormatter(FieldFormatter):
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
if 'api_key' in field.name and 'OpenAI' in str(name):
field.display_name = 'OpenAI API Key'
field.required = False
if field.value is None:
field.value = ''
class ModelSpecificFieldFormatter(FieldFormatter):
MODEL_DICT: ClassVar[Dict] = {
'OpenAI': OPENAI_MODELS,
'ChatOpenAI': CHAT_OPENAI_MODELS,
'Anthropic': ANTHROPIC_MODELS,
'ChatAnthropic': ANTHROPIC_MODELS,
}
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
if name in self.MODEL_DICT and field.name == 'model_name':
field.options = self.MODEL_DICT[name]
field.is_list = True
class KwargsFormatter(FieldFormatter):
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
if 'kwargs' in field.name.lower():
field.advanced = True
field.required = False
field.show = False
class APIKeyFormatter(FieldFormatter):
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
if 'api' in field.name.lower() and 'key' in field.name.lower():
field.required = False
field.advanced = False
field.display_name = field.name.replace('_', ' ').title()
field.display_name = field.display_name.replace('Api', 'API')
class RemoveOptionalFormatter(FieldFormatter):
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
_type = field.field_type
field.field_type = re.sub(r'Optional\[(.*)\]', r'\1', _type)
class ListTypeFormatter(FieldFormatter):
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
_type = field.field_type
is_list = 'List' in _type or 'Sequence' in _type
if is_list:
_type = re.sub(r'(List|Sequence)\[(.*)\]', r'\2', _type)
field.is_list = True
field.field_type = _type
class DictTypeFormatter(FieldFormatter):
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
_type = field.field_type
_type = _type.replace('Mapping', 'dict')
field.field_type = _type
class UnionTypeFormatter(FieldFormatter):
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
_type = field.field_type
if 'Union' in _type:
_type = _type.replace('Union[', '')[:-1]
_type = _type.split(',')[0]
_type = _type.replace(']', '').replace('[', '')
field.field_type = _type
class SpecialFieldFormatter(FieldFormatter):
SPECIAL_FIELD_HANDLERS: ClassVar[Dict] = {
'allowed_tools': lambda field: 'Tool',
'max_value_length': lambda field: 'int',
}
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
handler = self.SPECIAL_FIELD_HANDLERS.get(field.name)
field.field_type = handler(field) if handler else field.field_type
class ShowFieldFormatter(FieldFormatter):
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
key = field.name
required = field.required
field.show = ((required and (key not in ['input_variables'] or name == 'SequentialChain'))
or key in FORCE_SHOW_FIELDS or 'api' in key
or ('key' in key and 'input' not in key and 'output' not in key))
class PasswordFieldFormatter(FieldFormatter):
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
key = field.name
show = field.show
if (any(text in key.lower() for text in {'password', 'token', 'api', 'key'}) and show):
field.password = True
class MultilineFieldFormatter(FieldFormatter):
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
key = field.name
if key in {
'suffix',
'prefix',
'template',
'examples',
'code',
'headers',
'description',
}:
field.multiline = True
class DefaultValueFormatter(FieldFormatter):
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
value = field.to_dict()
if 'default' in value:
field.value = value['default']
class HeadersDefaultValueFormatter(FieldFormatter):
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
key = field.name
if key == 'headers':
field.value = """{'Authorization': 'Bearer '}"""
class DictCodeFileFormatter(FieldFormatter):
def format(self, field: TemplateField, name: Optional[str] = None) -> None:
key = field.name
value = field.to_dict()
_type = value['type']
if 'dict' in _type.lower() and key == 'dict_':
field.field_type = 'file'
field.file_types = ['.json', '.yaml', '.yml']
elif _type.startswith('Dict') or _type.startswith('Mapping') or _type.startswith('dict'):
field.field_type = 'dict'
================================================
FILE: src/backend/bisheng/template/frontend_node/input_output.py
================================================
from typing import Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
from bisheng.template.template.base import Template
def build_file_field(suffixes: list,
fileTypes: list,
name: str = 'file_path',
fieldType='fileNode') -> TemplateField:
"""Build a template field for a document loader."""
return TemplateField(
field_type=fieldType,
required=True,
show=True,
name=name,
value='',
suffixes=suffixes,
fileTypes=fileTypes,
)
class InputOutputNode(FrontendNode):
name: str = 'InputOutputNode'
base_classes: list[str] = ['input', 'output']
def add_extra_base_classes(self) -> None:
if self.name in {'AudioInputNode', 'FileInputNode'}:
self.base_classes.append('input')
def add_extra_fields(self) -> None:
pass
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
FrontendNode.format_field(field, name)
if name == 'Report':
if field.name == 'memory':
field.show = False
elif field.name == 'input_node':
field.show = False
elif field.name == 'chains':
field.show = True
field.field_type = 'function'
field.display_name = 'functions'
elif field.name == 'report_name':
field.show = True
field.display_name = 'Report Name'
field.info = 'the file name we generate'
elif field.name == 'variables':
field.show = True
field.field_type = 'VariableNode'
if name == 'VariableNode':
if field.name == 'variables':
field.show = True
field.field_type = 'variable'
field.required = True
if name == 'AudioInputNode':
if field.name == 'file_path':
field.show = True
field.field_type = 'fileNode'
field.required = True
field.suffixes = ['.mp3']
field.fileTypes = ['mp3']
elif field.name == 'openai_proxy':
field.show = True
if name == 'FileInputNode':
if field.name == 'file_path':
field.show = True
field.field_type = 'fileNode'
field.required = True
field.suffixes = ['.jpg', '.png', '.jpeg']
field.fileTypes = ['jpg', 'png', 'jpeg']
elif field.name == 'openai_proxy':
field.show = True
class InputNode(FrontendNode):
name: str = 'InputNode'
template: Template = Template(
type_name='input',
fields=[
TemplateField(
field_type='str',
is_list=True,
multiline=True,
required=True,
show=True,
name='input',
display_name='输入内容',
),
],
)
description: str = """输入节点，用来自动对接输入"""
base_classes: list[str] = ['input']
def to_dict(self):
return super().to_dict()
class InputFileNode(FrontendNode):
name: str = 'InputFileNode'
template: Template = Template(
type_name='InputFileNode',
fields=[
TemplateField(
field_type='file',
show=True,
name='file_path',
value='',
),
TemplateField(field_type='str',
show=True,
name='file_type',
placeholder='提示上传文件类型',
display_name='Name',
info='Tips for which file should upload'),
],
)
description: str = """输入节点，用来自动对接输入"""
base_classes: list[str] = ['fileNode']
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
FrontendNode.format_field(field, name)
field.show = True
def to_dict(self):
return super().to_dict()
class OutputNode(FrontendNode):
name: str = 'OutputNode'
template: Template = Template(
type_name='output',
fields=[
TemplateField(
field_type='str',
list=False,
multiline=True,
required=True,
show=True,
name='output',
display_name='展示输出内容',
),
],
)
description: str = """输出节点，用来表示输出"""
base_classes: list[str] = ['output']
def to_dict(self):
return super().to_dict()
================================================
FILE: src/backend/bisheng/template/frontend_node/llms.py
================================================
import json
from typing import Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
from bisheng.template.frontend_node.constants import (CTRANSFORMERS_DEFAULT_CONFIG,
OPENAI_API_BASE_INFO)
class LLMFrontendNode(FrontendNode):
def add_extra_fields(self) -> None:
self.template.add_field(
TemplateField(field_type='bool',
required=False,
show=True,
advanced=True,
name='cache',
value=True))
if 'VertexAI' in self.template.type_name:
# Add credentials field which should of type file.
self.template.add_field(
TemplateField(
field_type='file',
required=False,
show=True,
name='credentials',
value='',
suffixes=['.json'],
fileTypes=['json'],
))
@staticmethod
def format_vertex_field(field: TemplateField, name: str):
if 'VertexAI' in name:
advanced_fields = [
'tuned_model_name',
'verbose',
'top_p',
'top_k',
'max_output_tokens',
]
if field.name in advanced_fields:
field.advanced = True
show_fields = [
'tuned_model_name',
'verbose',
'project',
'location',
'credentials',
'max_output_tokens',
'model_name',
'temperature',
'top_p',
'top_k',
]
if field.name in show_fields:
field.show = True
@staticmethod
def format_openai_field(field: TemplateField):
if 'openai' in field.name.lower():
field.display_name = (field.name.title().replace('Openai',
'OpenAI').replace('_', ' ')).replace(
'Api', 'API')
if 'key' not in field.name.lower() and 'token' not in field.name.lower():
field.password = False
if field.name == 'openai_api_base':
field.info = OPENAI_API_BASE_INFO
if field.name == 'openai_proxy':
field.show = True
def add_extra_base_classes(self) -> None:
if 'BaseLLM' not in self.base_classes:
self.base_classes.append('BaseLLM')
@staticmethod
def format_azure_field(field: TemplateField):
if field.name == 'model_name':
field.show = False # Azure uses deployment_name instead of model_name.
elif field.name == 'openai_api_base':
# openai < 1.0.0
field.show = False
elif field.name == 'openai_api_type':
field.show = False
elif field.name == 'openai_api_version':
field.show = True
field.password = False
elif field.name == 'azure_endpoint':
field.show = True
elif field.name == 'openai_api_key':
field.show = True
field.advanced = False
elif field.name == 'deployment_name':
field.show = True
field.value = 'chatgpt'
elif field.name == 'azure_ad_token_provider':
field.show = False
elif field.name == 'openai_proxy':
field.advanced = True
@staticmethod
def format_contribute_field(field: TemplateField):
advanced_fields = [
'top_p',
'top_k',
'max_tokens',
]
if field.name in advanced_fields:
field.advanced = True
if field.name == 'headers':
field.show = True
field.advanced = True
field.value = ''
show_fields = [
'model_name',
'temperature',
'top_p',
'top_k',
'max_tokens',
]
if field.name in show_fields:
field.show = True
if 'api' in field.name.lower() or 'id' in field.name.lower() or 'key' in field.name.lower(
) or 'base' in field.name.lower():
field.show = True
field.advanced = False
field.field_type = 'str'
@staticmethod
def format_llama_field(field: TemplateField):
field.show = True
field.advanced = not field.required
@staticmethod
def format_ctransformers_field(field: TemplateField):
if field.name == 'config':
field.show = True
field.advanced = True
field.value = json.dumps(CTRANSFORMERS_DEFAULT_CONFIG, indent=2)
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
display_names_dict = {
'huggingfacehub_api_token': 'HuggingFace Hub API Token',
}
FrontendNode.format_field(field, name)
LLMFrontendNode.format_openai_field(field)
LLMFrontendNode.format_ctransformers_field(field)
LLMFrontendNode.format_contribute_field(field)
if name and 'llama' in name.lower() and 'host' not in name.lower():
LLMFrontendNode.format_llama_field(field)
if name and 'vertex' in name.lower():
LLMFrontendNode.format_vertex_field(field, name)
SHOW_FIELDS = ['repo_id']
HIDDEN_FIELDS = ['custom_get_token_ids']
if field.name in SHOW_FIELDS:
field.show = True
if field.name in HIDDEN_FIELDS:
field.show = False
if 'api' in field.name and ('key' in field.name or
('token' in field.name and 'tokens' not in field.name)):
field.password = True
field.show = True
# Required should be False to support
# loading the API key from environment variables
field.required = False
field.advanced = False
if field.name == 'task':
field.required = True
field.show = True
field.is_list = True
field.options = ['text-generation', 'text2text-generation', 'summarization']
field.value = field.options[0]
field.advanced = True
if display_name := display_names_dict.get(field.name):
field.display_name = display_name
if field.name == 'model_kwargs':
field.field_type = 'code'
field.advanced = True
field.show = True
elif field.name in [
'model_name',
'temperature',
'model_file',
'model_type',
'deployment_name',
'credentials',
'openai_proxy',
]:
field.advanced = False
field.show = True
if field.name == 'credentials':
field.field_type = 'file'
if name == 'VertexAI' and field.name not in [
'callbacks',
'client',
'stop',
'tags',
'cache',
]:
field.show = True
if field.name in ['cache']:
field.show = True
if name and 'azure' in name.lower():
LLMFrontendNode.format_azure_field(field)
# Release the streaming configuration of the component and cancel the global streaming control
if field.name == 'streaming':
field.show = True
if name == 'BishengLLM':
if field.name in ["temperature", "top_p", "cache"]:
field.show = True
if field.name == 'model_id':
field.field_type = "bisheng_model"
field.display_name = "Model Name"
elif field.name in ['model_name', "app_type", "app_id", "app_name", "user_id"]:
field.show = False
================================================
FILE: src/backend/bisheng/template/frontend_node/memories.py
================================================
from typing import Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
from bisheng.template.frontend_node.constants import INPUT_KEY_INFO, OUTPUT_KEY_INFO
from bisheng.template.template.base import Template
from langchain_community.chat_message_histories.mongodb import (DEFAULT_COLLECTION_NAME,
DEFAULT_DBNAME)
from langchain_community.chat_message_histories.postgres import DEFAULT_CONNECTION_STRING
class MemoryFrontendNode(FrontendNode):
# ! Needs testing
def add_extra_fields(self) -> None:
# chat history should have another way to add common field?
# prevent adding incorect field in ChatMessageHistory
base_message_classes = ['BaseEntityStore', 'BaseChatMessageHistory']
if any(base_class in self.base_classes for base_class in base_message_classes):
return
# add return_messages field
self.template.add_field(
TemplateField(
field_type='bool',
required=False,
show=True,
name='return_messages',
advanced=False,
value=False,
))
# add input_key and output_key str fields
self.template.add_field(
TemplateField(
field_type='str',
required=False,
show=True,
name='input_key',
advanced=True,
value='',
))
if self.template.type_name not in {'VectorStoreRetrieverMemory'}:
self.template.add_field(
TemplateField(
field_type='str',
required=False,
show=True,
name='output_key',
advanced=True,
value='',
))
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
FrontendNode.format_field(field, name)
if not isinstance(field.value, str):
field.value = None
if field.name == 'k':
field.required = True
field.show = True
field.field_type = 'int'
field.value = 10
field.display_name = 'Memory Size'
field.password = False
if field.name == 'return_messages':
field.required = False
field.show = True
field.advanced = False
if field.name in {'input_key', 'output_key'}:
field.required = False
field.show = True
field.advanced = False
field.value = 'answer' if field.name == 'output_key' else ''
field.info = (INPUT_KEY_INFO if field.name == 'input_key' else OUTPUT_KEY_INFO)
if field.name == 'memory_key':
field.value = 'chat_history'
if field.name == 'chat_memory':
field.show = True
field.advanced = False
field.required = False
if field.name == 'url':
field.show = True
if field.name == 'entity_store':
field.show = False
if name == 'ConversationEntityMemory' and field.name == 'memory_key':
field.show = False
field.required = False
class PostgresChatMessageHistoryFrontendNode(MemoryFrontendNode):
name: str = 'PostgresChatMessageHistory'
template: Template = Template(
type_name='PostgresChatMessageHistory',
fields=[
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=False,
name='session_id',
),
TemplateField(
field_type='str',
required=True,
show=True,
name='connection_string',
value=DEFAULT_CONNECTION_STRING,
),
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=False,
value='message_store',
name='table_name',
),
],
)
description: str = 'Memory store with Postgres'
base_classes: list[str] = ['PostgresChatMessageHistory', 'BaseChatMessageHistory']
class MongoDBChatMessageHistoryFrontendNode(MemoryFrontendNode):
name: str = 'MongoDBChatMessageHistory'
template: Template = Template(
# langchain/memory/chat_message_histories/mongodb.py
# connection_string: str,
# session_id: str,
# database_name: str = DEFAULT_DBNAME,
# collection_name: str = DEFAULT_COLLECTION_NAME,
type_name='MongoDBChatMessageHistory',
fields=[
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=False,
name='session_id',
),
TemplateField(
field_type='str',
required=True,
show=True,
name='connection_string',
value='',
info='MongoDB connection string (e.g mongodb://mongo_user:password123@mongo:27017)',
),
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=False,
value=DEFAULT_DBNAME,
name='database_name',
),
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=False,
value=DEFAULT_COLLECTION_NAME,
name='collection_name',
),
],
)
description: str = 'Memory store with MongoDB'
base_classes: list[str] = ['MongoDBChatMessageHistory', 'BaseChatMessageHistory']
================================================
FILE: src/backend/bisheng/template/frontend_node/output_parsers.py
================================================
from typing import Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
class OutputParserFrontendNode(FrontendNode):
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
FrontendNode.format_field(field, name)
field.show = True
# class RouterOutputParserFrontendNode(FrontendNode):
# @staticmethod
# def format_field(field: TemplateField, name: Optional[str] = None) -> None:
# FrontendNode.format_field(field, name)
# field.show = True
================================================
FILE: src/backend/bisheng/template/frontend_node/prompts.py
================================================
from typing import Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
from bisheng.template.frontend_node.constants import DEFAULT_PROMPT, HUMAN_PROMPT, SYSTEM_PROMPT
from bisheng.template.template.base import Template
from langchain.agents.mrkl import prompt
class PromptFrontendNode(FrontendNode):
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
# if field.field_type == "StringPromptTemplate"
# change it to str
PROMPT_FIELDS = [
'template',
'suffix',
'prefix',
'examples',
'format_instructions',
]
if field.field_type == 'StringPromptTemplate' and 'Message' in str(name):
field.field_type = 'prompt'
field.multiline = True
field.value = HUMAN_PROMPT if 'Human' in field.name else SYSTEM_PROMPT
if field.name == 'template' and field.value == '':
field.value = DEFAULT_PROMPT
if field.name == 'output_parser':
field.show = True
if field.name in PROMPT_FIELDS:
field.field_type = 'prompt'
field.advanced = False
if 'Union' in field.field_type:
if 'BaseMessagePromptTemplate' in field.field_type:
field.field_type = 'BaseMessagePromptTemplate'
elif 'StringPromptTemplate' in field.field_type:
field.field_type = 'StringPromptTemplate'
# All prompt fields should be password=False
field.password = False
class PromptTemplateNode(FrontendNode):
name: str = 'PromptTemplate'
template: Template
description: str
base_classes: list[str] = ['BasePromptTemplate']
def to_dict(self):
return super().to_dict()
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
FrontendNode.format_field(field, name)
if field.name == 'examples':
field.advanced = False
class BasePromptFrontendNode(FrontendNode):
name: str
template: Template
description: str
base_classes: list[str]
def to_dict(self):
return super().to_dict()
class ZeroShotPromptNode(BasePromptFrontendNode):
name: str = 'ZeroShotPrompt'
template: Template = Template(
type_name='ZeroShotPrompt',
fields=[
TemplateField(
field_type='str',
required=False,
placeholder='',
is_list=False,
show=True,
multiline=True,
value=prompt.PREFIX,
name='prefix',
),
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=True,
value=prompt.FORMAT_INSTRUCTIONS,
name='format_instructions',
),
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=True,
value=prompt.SUFFIX,
name='suffix',
),
],
)
description: str = 'Prompt template for Zero Shot Agent.'
base_classes: list[str] = ['BasePromptTemplate']
def to_dict(self):
return super().to_dict()
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
PromptFrontendNode.format_field(field, name)
================================================
FILE: src/backend/bisheng/template/frontend_node/retrievers.py
================================================
from typing import Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
STRATEGY_TYPES = ['keyword_front', 'vector_front', 'mix']
class RetrieverFrontendNode(FrontendNode):
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
FrontendNode.format_field(field, name)
# Define common field attributes
field.show = True
if field.name == 'parser_key':
field.display_name = 'Parser Key'
field.password = False
field.advanced = True
field.show = False
if field.name == 'combine_strategy' and name == 'MixEsVectorRetriever':
field.field_type = 'str'
field.required = True
field.is_list = True
field.show = True
field.multiline = False
field.options = STRATEGY_TYPES
field.value = STRATEGY_TYPES[0]
field.name = 'combine_strategy'
field.advanced = False
if field.name in {'metadata', 'tags'} and name == 'MixEsVectorRetriever':
field.show = True
field.advanced = True
================================================
FILE: src/backend/bisheng/template/frontend_node/textsplitters.py
================================================
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
from langchain.text_splitter import Language
class TextSplittersFrontendNode(FrontendNode):
def add_extra_base_classes(self) -> None:
self.base_classes = ['Document']
self.output_types = ['Document']
def add_extra_fields(self) -> None:
self.template.add_field(
TemplateField(
field_type='Document',
required=True,
show=True,
name='documents',
)
)
name = 'separator'
if self.template.type_name == 'CharacterTextSplitter':
name = 'separator'
elif self.template.type_name == 'RecursiveCharacterTextSplitter':
name = 'separators'
# Add a field for type of separator
# which will have Text or any value from the
# Language enum
options = [x.value for x in Language] + ['Text']
options.sort()
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
name='separator_type',
advanced=False,
is_list=True,
options=options,
value='Text',
display_name='Separator Type',
)
)
self.template.add_field(
TemplateField(
field_type='str',
required=True,
show=True,
value='\\n',
name=name,
display_name='Separator',
)
)
self.template.add_field(
TemplateField(
field_type='int',
required=True,
show=True,
value=1000,
name='chunk_size',
display_name='Chunk Size',
)
)
self.template.add_field(
TemplateField(
field_type='int',
required=True,
show=True,
value=200,
name='chunk_overlap',
display_name='Chunk Overlap',
)
)
================================================
FILE: src/backend/bisheng/template/frontend_node/tools.py
================================================
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
from bisheng.template.template.base import Template
from bisheng.utils.constants import DEFAULT_PYTHON_FUNCTION
class ToolNode(FrontendNode):
name: str = 'Tool'
template: Template = Template(
type_name='Tool',
fields=[
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=True,
value='',
name='name',
advanced=False,
),
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=True,
value='',
name='description',
advanced=False,
),
TemplateField(
name='func',
field_type='function',
required=True,
is_list=False,
show=True,
multiline=True,
advanced=False,
),
TemplateField(
field_type='bool',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=False,
value=False,
name='return_direct',
),
TemplateField(
field_type='NestedDict',
required=True,
placeholder='',
show=True,
multiline=True,
value='{"arg1": {"type": "string"}}',
name='args_schema',
),
],
)
description: str = 'Converts a chain, agent or function into a tool.'
base_classes: list[str] = ['Tool', 'BaseTool']
def to_dict(self):
return super().to_dict()
class PythonFunctionToolNode(FrontendNode):
name: str = 'PythonFunctionTool'
template: Template = Template(
type_name='PythonFunctionTool',
fields=[
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=False,
value='',
name='name',
advanced=False,
),
TemplateField(
field_type='str',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=False,
value='',
name='description',
advanced=False,
),
TemplateField(
field_type='code',
required=True,
placeholder='',
is_list=False,
show=True,
value=DEFAULT_PYTHON_FUNCTION,
name='code',
advanced=False,
),
TemplateField(
field_type='bool',
required=True,
placeholder='',
is_list=False,
show=True,
multiline=False,
value=False,
name='return_direct',
),
],
)
description: str = 'Python function to be executed.'
base_classes: list[str] = ['BaseTool', 'Tool']
def to_dict(self):
return super().to_dict()
class PythonFunctionNode(FrontendNode):
name: str = 'PythonFunction'
template: Template = Template(
type_name='PythonFunction',
fields=[
TemplateField(
field_type='code',
required=True,
placeholder='',
is_list=False,
show=True,
value=DEFAULT_PYTHON_FUNCTION,
name='code',
advanced=False,
)
],
)
description: str = 'Python function to be executed.'
base_classes: list[str] = ['function']
def to_dict(self):
return super().to_dict()
================================================
FILE: src/backend/bisheng/template/frontend_node/utilities.py
================================================
import ast
import json
from typing import Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
class UtilitiesFrontendNode(FrontendNode):
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
FrontendNode.format_field(field, name)
# field.field_type could be "Literal['news', 'search', 'places', 'images']
# we need to convert it to a list
if 'Literal' in field.field_type:
field.options = ast.literal_eval(field.field_type.replace('Literal', ''))
field.is_list = True
field.field_type = 'str'
if isinstance(field.value, dict):
field.field_type = 'code'
field.value = json.dumps(field.value, indent=4)
================================================
FILE: src/backend/bisheng/template/frontend_node/vectorstores.py
================================================
from typing import List, Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
class VectorStoreFrontendNode(FrontendNode):
def add_extra_fields(self) -> None:
extra_fields: List[TemplateField] = []
# Add search_kwargs field
extra_field = TemplateField(
name='search_kwargs',
field_type='code',
required=False,
placeholder='',
show=True,
advanced=True,
multiline=False,
value='{}',
)
extra_field2 = TemplateField(
name='search_type',
field_type='str',
required=False,
placeholder='',
value='similarity',
)
extra_fields.append(extra_field2)
extra_fields.append(extra_field)
if self.template.type_name == 'Weaviate':
extra_field = TemplateField(
name='weaviate_url',
field_type='str',
required=True,
placeholder='http://localhost:8080',
show=True,
advanced=False,
multiline=False,
value='http://localhost:8080',
)
# Add client_kwargs field
extra_field2 = TemplateField(
name='client_kwargs',
field_type='code',
required=False,
placeholder='',
show=True,
advanced=True,
multiline=False,
value='{}',
)
extra_fields.extend((extra_field, extra_field2))
elif self.template.type_name == 'Chroma':
# New bool field for persist parameter
extra_field = TemplateField(
name='persist',
field_type='bool',
required=False,
show=True,
advanced=False,
value=False,
display_name='Persist',
)
extra_fields.append(extra_field)
elif self.template.type_name == 'Pinecone':
# add pinecone_api_key and pinecone_env
extra_field = TemplateField(
name='pinecone_api_key',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=True,
multiline=False,
password=True,
value='',
)
extra_field2 = TemplateField(
name='pinecone_env',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=True,
multiline=False,
value='',
)
extra_fields.extend((extra_field, extra_field2))
elif self.template.type_name == 'FAISS':
extra_field = TemplateField(
name='folder_path',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=True,
multiline=False,
display_name='Local Path',
value='',
)
extra_field2 = TemplateField(
name='index_name',
field_type='str',
required=False,
show=True,
advanced=False,
value='',
display_name='Index Name',
)
extra_fields.extend((extra_field, extra_field2))
elif self.template.type_name == 'SupabaseVectorStore':
self.display_name = 'Supabase'
# Add table_name and query_name
extra_field = TemplateField(
name='table_name',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=True,
multiline=False,
value='',
)
extra_field2 = TemplateField(
name='query_name',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=True,
multiline=False,
value='',
)
# Add supabase_url and supabase_service_key
extra_field3 = TemplateField(
name='supabase_url',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=True,
multiline=False,
value='',
)
extra_field4 = TemplateField(
name='supabase_service_key',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=True,
multiline=False,
password=True,
value='',
)
extra_fields.extend((extra_field, extra_field2, extra_field3, extra_field4))
elif self.template.type_name == 'MongoDBAtlasVectorSearch':
self.display_name = 'MongoDB Atlas'
extra_field = TemplateField(
name='mongodb_atlas_cluster_uri',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=True,
multiline=False,
display_name='MongoDB Atlas Cluster URI',
value='',
)
extra_field2 = TemplateField(
name='collection_name',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=True,
multiline=False,
display_name='Collection Name',
value='',
)
extra_field3 = TemplateField(
name='db_name',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=True,
multiline=False,
display_name='Database Name',
value='',
)
extra_field4 = TemplateField(
name='index_name',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=True,
multiline=False,
display_name='Index Name',
value='',
)
extra_fields.extend((extra_field, extra_field2, extra_field3, extra_field4))
elif self.template.type_name == 'ElasticKeywordsSearch':
extra_field = TemplateField(
name='elasticsearch_url',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=False,
multiline=False,
display_name='ES_connection_url',
value='',
)
extra_field2 = TemplateField(
name='ssl_verify',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=False,
multiline=False,
display_name='ssl_verify',
value='',
)
extra_field3 = TemplateField(
name='llm',
field_type='BaseLLM',
required=False,
show=True,
advanced=False,
display_name='LLM',
)
extra_field4 = TemplateField(
name='prompt',
field_type='BasePromptTemplate',
required=False,
show=True,
advanced=False,
display_name='prompt',
)
extra_fields.extend((extra_field, extra_field2, extra_field3, extra_field4))
elif self.template.type_name == 'ElasticsearchStore':
extra_fields.append(
TemplateField(
name='embedding',
field_type='str',
required=False,
placeholder='',
show=True,
advanced=False,
multiline=False,
value='',
))
extra_fields.append(
TemplateField(
name='connect_kwargs',
field_type='dict',
required=True,
show=True,
advanced=False,
multiline=False,
value={
'es_url': 'http://bisheng-es:9200',
'es_user': 'elastic',
'es_password': ''
},
))
extra_fields.append(
TemplateField(name='index_name',
field_type='str',
required=False,
show=True,
advanced=False,
multiline=False))
if extra_fields:
for field in extra_fields:
self.template.add_field(field)
def add_extra_base_classes(self) -> None:
self.base_classes.extend(('BaseRetriever', 'VectorStoreRetriever'))
if self.name == 'ElasticsearchWithPermissionCheck':
self.base_classes.append('ElasticKeywordsSearch')
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
FrontendNode.format_field(field, name)
# Define common field attributes
basic_fields = [
'work_dir', 'collection_name', 'api_key', 'location', 'persist_directory', 'persist',
'weaviate_url', 'index_name', 'namespace', 'folder_path', 'table_name', 'query_name',
'supabase_url', 'supabase_service_key', 'mongodb_atlas_cluster_uri', 'collection_name',
'db_name', 'ssl_verify', 'elasticsearch_url', 'llm', 'prompt', 'connect_kwargs'
]
advanced_fields = [
'n_dim', 'key', 'prefix', 'distance_func', 'content_payload_key',
'metadata_payload_key', 'timeout', 'host', 'path', 'url', 'port', 'https',
'prefer_grpc', 'grpc_port', 'pinecone_api_key', 'pinecone_env', 'client_kwargs',
'search_kwargs', 'search_type'
]
# Check and set field attributes
if field.name == 'texts':
# if field.name is "texts" it has to be replaced
# when instantiating the vectorstores
field.name = 'documents'
field.field_type = 'Document'
field.display_name = 'Documents'
field.required = False
field.show = True
field.advanced = False
if name == 'MilvusWithPermissionCheck' or name == 'ElasticsearchWithPermissionCheck':
field.show = False
field.advanced = True
elif 'embedding' in field.name:
# for backwards compatibility
field.name = 'embedding'
field.required = True
field.show = True
field.advanced = False
field.display_name = 'Embedding'
field.field_type = 'Embeddings'
if name == 'ElasticKeywordsSearch':
field.show = False
field.required = False
elif name in ['MilvusWithPermissionCheck', 'ElasticsearchWithPermissionCheck']:
field.advanced = True
field.show = False
field.required = False
elif field.name == 'collection_name':
field.show = True
field.advanced = False
field.value = ''
field.field_type = 'knowledge_one' # Knowledge base radio type, front-end rendering radio list
if name == 'MilvusWithPermissionCheck':
field.is_list = True
field.field_type = 'knowledge_list' # Knowledge base multi-select type, front-end rendering multi-select list
field.required = True
elif field.name == 'index_name':
field.show = True
field.advanced = False
field.value = ''
field.field_type = 'knowledge_one'
if name == 'ElasticsearchWithPermissionCheck':
field.is_list = True
field.field_type = 'knowledge_list' # Knowledge base multi-select type, front-end rendering multi-select list
field.required = True
elif field.name in basic_fields:
field.show = True
field.advanced = False
if field.name == 'api_key':
field.display_name = 'API Key'
field.password = True
elif field.name == 'location':
field.value = ':memory:'
field.placeholder = ':memory:'
elif field.name in advanced_fields:
field.show = True
field.advanced = True
if 'key' in field.name:
field.password = False
elif field.name == 'text_key':
field.show = False
elif field.name == 'connection_args':
field.show = True
field.advanced = False
field.value = ''
if name in ['MilvusWithPermissionCheck', 'ElasticsearchWithPermissionCheck']:
field.show = False
field.advanced = True
================================================
FILE: src/backend/bisheng/template/frontend_node/wrappers.py
================================================
from typing import Optional
from bisheng.template.field.base import TemplateField
from bisheng.template.frontend_node.base import FrontendNode
from sqlmodel import true
class WrappersFrontendNode(FrontendNode):
name: str = 'WrappersFrontendNode'
@staticmethod
def format_field(field: TemplateField, name: Optional[str] = None) -> None:
FrontendNode.format_field(field, name)
if name == 'DallEAPIWrapper':
if field.name == 'model_name':
field.show = True
field.name = 'model'
field.value = 'dall-e-3'
elif field.name == 'openai_api_key':
field.show = true
field.name = 'api_key'
field.display_name = 'openai_api_key'
field.password = true
elif field.name == 'openai_api_base':
field.show = True
field.name = 'api_base'
field.display_name = 'openai_api_base'
elif field.name == 'openai_proxy':
field.show = True
elif field.name == 'n':
field.show = True
field.field_type = 'int'
field.value = 1
field.info = '生成图片数量，使用dalle3 只能填1'
field.advanced = True
elif field.name == 'size':
field.show = True
field.is_list = True
field.field_type = 'str'
field.options = ['256x256', '512x512', '1024x1024', '1024x1792', '1792x1024']
field.value = field.options[2]
field.advanced = True
elif field.name == 'quality':
field.show = True
field.field_type = 'str'
field.advanced = True
field.value = 'standard'
elif field.name == 'request_timeout':
field.show = True
field.name = 'timeout'
field.value = 30
================================================
FILE: src/backend/bisheng/template/template/__init__.py
================================================
================================================
FILE: src/backend/bisheng/template/template/base.py
================================================
from typing import Callable, Optional, Union
from bisheng.template.field.base import TemplateField
from bisheng.utils.constants import DIRECT_TYPES
from pydantic import BaseModel
class Template(BaseModel):
type_name: str
fields: list[TemplateField]
def process_fields(
self,
name: Optional[str] = None,
format_field_func: Union[Callable, None] = None,
):
if format_field_func:
for field in self.fields:
format_field_func(field, name)
def sort_fields(self):
# first sort alphabetically
# then sort fields so that fields that have .field_type in DIRECT_TYPES are first
self.fields.sort(key=lambda x: x.name)
self.fields.sort(key=lambda x: x.field_type in DIRECT_TYPES, reverse=False)
def to_dict(self, format_field_func=None):
self.process_fields(self.type_name, format_field_func)
self.sort_fields()
result = {field.name: field.to_dict() for field in self.fields}
result['_type'] = self.type_name # type: ignore
return result
def add_field(self, field: TemplateField) -> None:
self.fields.append(field)
def get_field(self, field_name: str) -> TemplateField:
"""Returns the field with the given name."""
field = next((field for field in self.fields if field.name == field_name), None)
if field is None:
raise ValueError(f'Field {field_name} not found in template {self.type_name}')
return field
def update_field(self, field_name: str, field: TemplateField) -> None:
"""Updates the field with the given name."""
for idx, template_field in enumerate(self.fields):
if template_field.name == field_name:
self.fields[idx] = field
return
raise ValueError(f'Field {field_name} not found in template {self.type_name}')
def upsert_field(self, field_name: str, field: TemplateField) -> None:
"""Updates the field with the given name or adds it if it doesn't exist."""
try:
self.update_field(field_name, field)
except ValueError:
self.add_field(field)
================================================
FILE: src/backend/bisheng/tool/__init__.py
================================================
================================================
FILE: src/backend/bisheng/tool/api/__init__.py
================================================
================================================
FILE: src/backend/bisheng/tool/api/tool.py
================================================
import json
from typing import Optional, Dict
from fastapi import APIRouter, Depends, Body, Request
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.schemas.api import UnifiedResponseModel, resp_200
from bisheng.mcp_manage.langchain.tool import McpTool
from bisheng.mcp_manage.manager import ClientManager
from bisheng.tool.domain.models.gpts_tools import GptsToolsTypeRead
from bisheng.tool.domain.schemas import TestToolReq
from bisheng.tool.domain.services.openapi import OpenApiSchema
from bisheng.tool.domain.services.tool import ToolServices
from bisheng_langchain.gpts.tools.api_tools.openapi import OpenApiTools
router = APIRouter(prefix='/tool', tags=['Tool'])
@router.get('')
async def get_tool_list(*,
is_preset: Optional[int] = None,
login_user: UserPayload = Depends(UserPayload.get_login_user)):
"""Query all visibletool Vertical"""
res = await ToolServices(login_user=login_user).get_tool_list(is_preset)
return resp_200(data=res)
@router.post('')
async def add_tool_type(request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user),
req: Dict = Body(default={})):
""" Add customizationtool """
req = GptsToolsTypeRead(**req)
services = ToolServices(request=request, login_user=login_user)
return resp_200(data=await services.add_tools(req))
@router.put('')
async def update_tool_type(request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user),
req: Dict = Body(default={})):
""" Custom field updated.tool """
req = GptsToolsTypeRead(**req)
services = ToolServices(request=request, login_user=login_user)
return resp_200(data=await services.update_tools(req))
@router.delete('')
async def delete_tool_type(*, request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user),
tool_type_id: int = Body(..., embed=True)):
""" Remove Customizer """
services = ToolServices(request=request, login_user=login_user)
await services.delete_tools(tool_type_id)
return resp_200()
@router.post('/config')
async def update_tool_config(*,
login_user: UserPayload = Depends(UserPayload.get_admin_user),
tool_id: int = Body(description='Tool Category UniqueID'),
extra: Dict = Body(..., description='Configuration information for the tool')):
""" Update the configuration of the tool """
data = await ToolServices(login_user=login_user).update_tool_config(tool_id, extra)
return resp_200(data=data)
@router.post('/schema')
async def get_tool_schema(request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user),
download_url: Optional[str] = Body(default=None,
description='MengunduhurlIf it is not empty, download it firsturl'),
file_content: Optional[str] = Body(default=None, description='files uploaded')):
""" Download or parseopenapi schemaThe contents of the Convert to Assistant Customizer Format """
services = ToolServices(request=request, login_user=login_user)
tool_type = await services.parse_openapi_schema(download_url, file_content)
return resp_200(data=tool_type)
@router.post('/mcp/schema')
async def get_mcp_tool_schema(request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user),
file_content: Optional[str] = Body(default=None, embed=True,
description='mcpService Configuration Content')):
""" analyzingmcpTool Profile """
services = ToolServices(request=request, login_user=login_user)
tool_type = await services.parse_mcp_schema(file_content)
return resp_200(data=tool_type)
@router.post('/test')
async def tool_run(*, login_user: UserPayload = Depends(UserPayload.get_login_user), req: TestToolReq):
""" Test custom tool """
extra = json.loads(req.extra)
extra.update({'api_location': req.api_location, 'parameter_name': req.parameter_name})
tool_params = OpenApiSchema.parse_openapi_tool_params('test', 'test', json.dumps(extra),
req.server_host, req.auth_method,
req.auth_type, req.api_key)
openapi_tool = OpenApiTools.get_api_tool('test', **tool_params)
resp = await openapi_tool.arun(req.request_params)
return resp_200(data=resp)
@router.post('/mcp/test')
async def mcp_tool_run(login_user: UserPayload = Depends(UserPayload.get_login_user),
req: TestToolReq = None):
""" TestmcpTools for Services """
# Instantiatemcpservice object, getting a list of tools
client = await ClientManager.connect_mcp_from_json(req.openapi_schema)
extra = json.loads(req.extra)
tool_name = extra.get('name')
mcp_tool = McpTool.get_mcp_tool(name=tool_name, description=extra.get("description"), mcp_client=client,
mcp_tool_name=tool_name, arg_schema=extra.get('inputSchema', {}))
resp = await mcp_tool.arun(req.request_params)
return resp_200(data=resp)
@router.post('/mcp/refresh')
async def refresh_all_mcp_tools(request: Request, login_user: UserPayload = Depends(UserPayload.get_login_user)):
""" Refresh all of the user's currentmcpTools List """
services = ToolServices(request=request, login_user=login_user)
error_name = await services.refresh_all_mcp()
return resp_200(data=error_name)
@router.get("/linsight/preset", summary="Get a list of Ideas presets", response_model=UnifiedResponseModel)
async def get_linsight_tools():
"""
Get a list of Ideas presets
"""
tools = await ToolServices.get_linsight_tools()
return resp_200(data=tools)
================================================
FILE: src/backend/bisheng/tool/domain/__init__.py
================================================
================================================
FILE: src/backend/bisheng/tool/domain/const.py
================================================
from enum import Enum
class ToolPresetType(Enum):
PRESET = 1 # Provisioning Tools
API = 0 # CustomizableAPITools
MCP = 2 # mcpTypes of Tools
class AuthMethod(Enum):
NO = 0
API_KEY = 1
class AuthType(Enum):
BASIC = "basic"
BEARER = "bearer"
CUSTOM = "custom"
================================================
FILE: src/backend/bisheng/tool/domain/langchain/__init__.py
================================================
================================================
FILE: src/backend/bisheng/tool/domain/langchain/knowledge.py
================================================
from typing import Any, Type, List, Optional
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain_core.documents import Document, BaseDocumentCompressor
from langchain_core.language_models import BaseChatModel
from langchain_core.prompts import SystemMessagePromptTemplate, HumanMessagePromptTemplate, ChatPromptTemplate
from langchain_core.retrievers import BaseRetriever
from langchain_core.tools import BaseTool
from pydantic import BaseModel, Field
from bisheng.core.ai.rerank.rrf_rerank import RRFRerank
system_template = """# 任务
你是一位知识库问答助手，遵守以下规则回答问题：
1. 严谨、专业地回答用户的问题。
2. 回答时须严格基于【参考文本】中的内容：
- 如果【参考文本】中有明确与用户问题相关的文字内容，请依据相关内容进行回答；如果【参考文本】中没有任何与用户问题相关的内容，则直接回复：“没有找到相关内容”。
- 如果相关内容中包含 markdown 格式的图片（例如 ），必须严格保留其原始 markdown 格式，不得添加引号、代码块（`或```）或其他特殊符号，也不得修改图片路径，保证可以正常渲染 markdown 图片。
3. 当【参考文本】中的内容来源于多个不同的信息源时，若相关内容存在明显差异或冲突，请分别列出这些差异或冲突的答案；若无差异或冲突，只给出一个统一的回答即可。
# 参考文本
{context}"""
messages = [
SystemMessagePromptTemplate.from_template(system_template),
HumanMessagePromptTemplate.from_template("{question}"),
]
CHAT_PROMPT = ChatPromptTemplate.from_messages(messages)
class ToolInputSchema(BaseModel):
query: str = Field(description='question asked by the user.')
class KnowledgeRetrieverTool(BaseTool):
name: str = "knowledge_retriever_tool"
description: str = "在知识库中检索与查询相关的文档内容。"
args_schema: Type[BaseModel] = ToolInputSchema
vector_retriever: Optional[BaseRetriever] = None
elastic_retriever: Optional[BaseRetriever] = None
rerank: Optional[BaseDocumentCompressor] = None
max_content: int = Field(default=15000, description='The max length of the combined document content.')
sort_by_source_and_index: bool = Field(default=False, description='Sort by document name & chunk index.')
rrf_weights: List[float] = Field(default=None)
rrf_remove_zero_score: bool = Field(default=False)
def _run(self, query: str, **kwargs: Any) -> List[Document]:
milvus_docs, es_docs = [], []
if self.vector_retriever:
milvus_docs = self.vector_retriever.invoke(query)
if self.elastic_retriever:
es_docs = self.elastic_retriever.invoke(query)
finally_docs = self._rrf_rerank(milvus_docs, es_docs, query)
if self.rerank:
finally_docs = self.rerank.compress_documents(finally_docs, query)
return finally_docs
async def _arun(self, query: str, **kwargs: Any) -> List[Document]:
milvus_docs, es_docs = [], []
if self.vector_retriever:
milvus_docs = await self.vector_retriever.ainvoke(query)
if self.elastic_retriever:
es_docs = await self.elastic_retriever.ainvoke(query)
finally_docs = self._rrf_rerank(milvus_docs, es_docs, query)
if self.rerank:
finally_docs = await self.rerank.acompress_documents(finally_docs, query)
return finally_docs
def _rrf_rerank(self, milvus_docs: List[Document], es_docs: List[Document], query: str) -> List[Document]:
if not milvus_docs and not es_docs:
return []
rrf_rerank = RRFRerank(retrievers=[self.vector_retriever, self.elastic_retriever],
weights=self.rrf_weights,
remove_zero_score=self.rrf_remove_zero_score)
finally_docs = rrf_rerank.compress_documents(documents=[es_docs, milvus_docs], query=query)
# limit by max_chunk_size
doc_num, doc_content_sum = 0, 0
same_file_id = set()
for doc in finally_docs:
if doc_content_sum > self.max_content:
break
doc_content_sum += len(doc.page_content)
same_file_id.add((doc.metadata.get('document_id'), doc.metadata.get('document_name')))
doc_num += 1
finally_docs = finally_docs[:doc_num]
# sort by source and index if only one file
if self.sort_by_source_and_index and len(same_file_id) == 1:
finally_docs = sorted(finally_docs,
key=lambda x: (x.metadata.get('document_name', ""), x.metadata.get('chunk_index', 0)))
return finally_docs
class KnowledgeRagTool(BaseTool):
name: str
description: str
args_schema: Type[BaseModel] = ToolInputSchema
llm: BaseChatModel
chat_prompt: Optional[ChatPromptTemplate] = CHAT_PROMPT
vector_retriever: Optional[BaseRetriever] = None
elastic_retriever: Optional[BaseRetriever] = None
max_content: int = Field(default=15000, description='The max length of the combined document content.')
sort_by_source_and_index: bool = Field(default=False, description='Sort by document name & chunk index.')
rrf_weights: List[float] = Field(default=None)
rrf_remove_zero_score: bool = Field(default=False)
knowledge_retriever_tool: KnowledgeRetrieverTool = None
@classmethod
def init_knowledge_rag_tool(cls, name: str, description: str, **kwargs) -> BaseTool:
llm = kwargs.pop('llm')
chat_prompt = kwargs.pop('chat_prompt', CHAT_PROMPT)
# cancel assistant deep callback
kwargs.pop("callbacks", None)
knowledge_retriever_tool = KnowledgeRetrieverTool(**kwargs)
return cls(name=name,
description=description,
args_schema=ToolInputSchema,
llm=llm,
chat_prompt=chat_prompt,
knowledge_retriever_tool=knowledge_retriever_tool)
def _run(self, query: str) -> Any:
# 1. retrieve documents
finally_docs = self.knowledge_retriever_tool.invoke({"query": query})
llm_inputs = self._get_llm_inputs(query, finally_docs)
qa_chain = create_stuff_documents_chain(llm=self.llm, prompt=self.chat_prompt)
return qa_chain.invoke(llm_inputs)
async def _arun(self, query: str) -> Any:
finally_docs = await self.knowledge_retriever_tool.ainvoke({"query": query})
llm_inputs = self._get_llm_inputs(query, finally_docs)
qa_chain = create_stuff_documents_chain(llm=self.llm, prompt=self.chat_prompt)
return await qa_chain.ainvoke(llm_inputs)
def _get_llm_inputs(self, query: str, finally_docs: List[Document]) -> Any:
inputs = {
"context": finally_docs,
}
if "question" in self.chat_prompt.input_variables:
inputs["question"] = query
return inputs
================================================
FILE: src/backend/bisheng/tool/domain/langchain/linsight_knowledge.py
================================================
import json
from typing import Optional, Type
from langchain_core.tools import BaseTool
from pydantic import BaseModel, Field
from bisheng.api.services.knowledge_imp import decide_vectorstores
from bisheng.linsight.domain.models.linsight_session_version import LinsightSessionVersionDao
from bisheng.interface.importing.utils import import_vectorstore
from bisheng.interface.initialize.loading import instantiate_vectorstore
from bisheng.knowledge.domain.models.knowledge import KnowledgeDao
from bisheng.llm.domain import LLMService
from bisheng.llm.domain.models import LLMDao
class ToolInput(BaseModel):
query: str = Field(..., description='需要检索的关键词')
knowledge_id: str = Field(..., description='语义检索库id')
limit: int = Field(default=2, description='返回结果的最大数量')
call_reason: str = Field(default='', description='调用该工具的原因，原因中不要使用id来描述文件或知识库')
class SearchKnowledgeBase(BaseTool):
name: str = "search_knowledge_base"
description: str = """在语义检索库中搜索相关内容。
用法:在你需要在知识库中进行语义搜索时,调用此工具。
Args:
query: 需要检索的关键词
knowledge_id: 语义检索库id
limit: 返回结果的最大数量，默认为2
Returns:
包含搜索结果（chunk的列表）的字典"""
args_schema: Type[BaseModel] = ToolInput
def _run(self, query: str, knowledge_id: Optional[str] = None,
**kwargs) -> str:
"""Use the tool."""
return "not supported in sync mode, please use async version"
async def _arun(self, query: str, knowledge_id: Optional[str] = None,
**kwargs) -> str:
limit = kwargs.get('limit', None) or 2
if not query:
raise ValueError("query Parameter shall not be empty")
try:
knowledge_id = int(knowledge_id)
return await self.search_knowledge(query, knowledge_id, limit)
except ValueError:
return await self.search_linsight_file(query, knowledge_id, limit)
async def base_search(self, vector_client, query: str, k: int):
documents = await vector_client.asimilarity_search(query, k=k)
if not documents:
# "没有找到相关的知识内容"
return '{"状态": "无结果", "错误信息":"没有找到相关的知识内容"}'
result = {
"状态": "成功",
"结果": [one.page_content for one in documents]
}
result = json.dumps(result, ensure_ascii=False, indent=2)
return result
async def search_linsight_file(self, query: str, file_id: str, limit: int) -> str:
"""检索Linsight用户上传的文件"""
session_info = await LinsightSessionVersionDao.get_session_version_by_file_id(file_id=file_id)
if not session_info:
raise Exception("文件不存在或已被删除")
files = session_info.files
file_info = None
for one in files:
if one.get("file_id") == file_id:
file_info = one
break
if not file_info:
raise Exception("File does not exist or has been deleted")
class_obj = import_vectorstore('Milvus')
embeddings = await LLMService.get_bisheng_linsight_embedding(session_info.user_id,
file_info.get("embedding_model_id"))
params = {
'collection_name': file_info.get("collection_name"),
'embedding': embeddings,
'metadata_expr': f'file_id in {[file_id]}'
}
milvus_client = instantiate_vectorstore('Milvus', class_object=class_obj, params=params)
return await self.base_search(milvus_client, query, limit)
async def search_knowledge(self, query: str, knowledge_id: int, limit: int) -> str:
knowledge_info = KnowledgeDao.query_by_id(knowledge_id)
if not knowledge_info:
raise Exception("Knowledgebase does not exist or has been deleted")
if not knowledge_info.model:
# "Knowledge Base Not ConfiguredembeddingModels"
raise Exception("Knowledge Base Not ConfiguredembeddingModels")
embed_info = LLMDao.get_model_by_id(int(knowledge_info.model))
if not embed_info:
# "Configured by the Knowledge BaseembeddingModel does not exist or has been deleted"
raise Exception("Configured by the Knowledge BaseembeddingModel does not exist or has been deleted")
embeddings = await LLMService.get_bisheng_knowledge_embedding(0, model_id=int(knowledge_info.model))
milvus_client = decide_vectorstores(
knowledge_info.collection_name, "Milvus", embeddings, knowledge_id=knowledge_id
)
return await self.base_search(milvus_client, query, limit)
================================================
FILE: src/backend/bisheng/tool/domain/models/__init__.py
================================================
================================================
FILE: src/backend/bisheng/tool/domain/models/gpts_tools.py
================================================
import json
from datetime import datetime
from typing import Dict, List, Optional
from pydantic import model_validator
from sqlalchemy import JSON, Column, DateTime, String, text, func
from sqlmodel import Field, or_, select, Text, update, col
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session, get_async_db_session
from bisheng.utils import md5_hash, generate_uuid
from bisheng.utils.mask_data import JsonFieldMasker
from ..const import AuthType, ToolPresetType
class GptsToolsBase(SQLModelSerializable):
name: str = Field(sa_column=Column(String(length=125), index=True))
logo: Optional[str] = Field(default=None, sa_column=Column(String(length=512), index=False))
desc: Optional[str] = Field(default=None, sa_column=Column(String(length=2048), index=False))
tool_key: str = Field(sa_column=Column(String(length=125), index=False))
type: int = Field(default=0, description='of the category to which they belongID')
is_preset: int = Field(default=ToolPresetType.API.value, description="The category of the tool, the historical reason field is not renamed")
is_delete: int = Field(default=0, description='1 Indicates logical deletion')
api_params: Optional[List[Dict]] = Field(default=None, sa_column=Column(JSON), description='Used to storeapiParameter and other information')
user_id: Optional[int] = Field(default=None, index=True, description='Create UserID， nullIndicates system creation')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class GptsToolsTypeBase(SQLModelSerializable):
id: Optional[int] = Field(default=None, index=True, primary_key=True)
name: str = Field(default='', sa_column=Column(String(length=1024)), description="Tool Category Name")
logo: Optional[str] = Field(default='', description="of the tool categorylogoFile URL")
extra: Optional[str] = Field(default='{}', sa_column=Column(Text),
description="Configuration information for the tool category to store the configuration information required for the tool category")
description: str = Field(default='', description="Description of the tool category")
server_host: Optional[str] = Field(default='', description="The access root address of the custom tool, which must behttporhttpsWhat/the beginning?")
auth_method: Optional[int] = Field(default=0, description="Authentication method of tool category")
api_key: Optional[str] = Field(default='', description="Tool Authenticationapi_key", sa_column=Column(String(length=2048)),
max_length=1000)
auth_type: Optional[str] = Field(default=AuthType.BASIC.value, description="Authentication method of tool authentication")
is_preset: Optional[int] = Field(default=ToolPresetType.API.value, description="The category of the tool, the historical reason field is not renamed")
user_id: Optional[int] = Field(default=None, index=True, description='Create UserID， nullIndicates system creation')
is_delete: int = Field(default=0, description='1 Indicates logical deletion')
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class GptsTools(GptsToolsBase, table=True):
__tablename__ = 't_gpts_tools'
extra: Optional[str | dict] = Field(default=None, sa_column=Column(Text, index=False),
description='Used to store additional information, such as parameter requirements, including &initdb_conf_key Data field'
'Indicates that the configuration information is obtained from the system configuration,For multi-level use.with ')
id: Optional[int] = Field(default=None, primary_key=True)
class GptsToolsType(GptsToolsTypeBase, table=True):
__tablename__ = 't_gpts_tools_type'
openapi_schema: str = Field(default="", sa_column=Column(Text),
description="of the tool categoryschemaContent, complies withopenapiSpecified Data")
class GptsToolsTypeRead(GptsToolsTypeBase):
openapi_schema: Optional[str] = Field(default="", description="of the tool categoryschemaContent, complies withopenapiSpecified Data")
children: Optional[List[GptsTools]] = Field(default_factory=list, description="List of tools under the Tools category")
parameter_name: Optional[str] = Field(default="", description="Custom request header parameter name")
api_location: Optional[str] = Field(default="", description="Custom Request Header Parameter Position header or query")
write: Optional[bool] = Field(default=False, description="Do you have write access")
@model_validator(mode="after")
def validate(self):
# Needs to be populated when echoingapi_locationAndparameter_nameData field
if self.extra and not self.api_location:
result = json.loads(self.extra)
self.api_location = result.get('api_location')
self.parameter_name = result.get('parameter_name')
def mask_sensitive_data(self):
json_masker = JsonFieldMasker()
# The provisioning tool needs to be desensitizedextraData field
if self.extra and self.is_preset == ToolPresetType.PRESET.value:
extra_json = json.loads(self.extra)
extra_json = json_masker.mask_json(extra_json)
self.extra = json.dumps(extra_json, ensure_ascii=False)
return self
class GptsToolsRead(GptsToolsBase):
id: int
class GptsToolsDao(GptsToolsBase):
@classmethod
def insert(cls, obj: GptsTools):
with get_sync_db_session() as session:
session.add(obj)
session.commit()
session.refresh(obj)
return obj
@classmethod
def query_by_name(cls, name: str) -> List[GptsTools]:
with get_sync_db_session() as session:
statement = select(GptsTools).where(GptsTools.name.like(f'%{name}%'))
return session.exec(statement).all()
@classmethod
def update_tools(cls, data: GptsTools) -> GptsTools:
with get_sync_db_session() as session:
session.add(data)
session.commit()
session.refresh(data)
return data
@classmethod
def update_tool_list(cls, data: List[GptsTools]) -> List[GptsTools]:
with get_sync_db_session() as session:
for one in data:
session.add(one)
session.commit()
return data
@classmethod
def delete_tool(cls, data: GptsTools) -> GptsTools:
data.is_delete = 1
return cls.update_tools(data)
@classmethod
def delete_tool_by_ids(cls, tool_ids: List[int]) -> None:
with get_sync_db_session() as session:
statement = update(GptsTools).where(GptsTools.id.in_(tool_ids)).values(is_delete=1)
session.exec(statement)
session.commit()
@classmethod
def get_one_tool(cls, tool_id: int) -> GptsTools:
with get_sync_db_session() as session:
statement = select(GptsTools).where(GptsTools.id == tool_id)
return session.exec(statement).first()
@classmethod
async def aget_one_tool(cls, tool_id: int) -> GptsTools:
statement = select(GptsTools).where(GptsTools.id == tool_id)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.first()
@classmethod
def get_list_by_ids(cls, tool_ids: List[int]) -> List[GptsTools]:
statement = select(GptsTools).where(col(GptsTools.id).in_(tool_ids)).where(GptsTools.is_delete == 0)
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
async def aget_list_by_ids(cls, tool_ids: List[int]) -> List[GptsTools]:
statement = select(GptsTools).where(col(GptsTools.id).in_(tool_ids)).where(GptsTools.is_delete == 0)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def get_list_by_user(cls, user_id: int, page: int = 0, page_size: int = 0) -> List[GptsTools]:
"""
Get all the tools available to your users
"""
with get_sync_db_session() as session:
statement = select(GptsTools).where(
or_(GptsTools.user_id == user_id,
GptsTools.is_preset == ToolPresetType.PRESET.value)).where(GptsTools.is_delete == 0)
if page and page_size:
statement = statement.offset((page - 1) * page_size).limit(page_size)
statement = statement.order_by(GptsTools.create_time.desc())
list_tools = session.exec(statement).all()
return list_tools
@classmethod
def get_list_by_type(cls, tool_type_ids: List[int]) -> List[GptsTools]:
"""
Get all the tools under the Tools category
"""
with get_sync_db_session() as session:
statement = select(GptsTools).where(
GptsTools.type.in_(tool_type_ids)).where(
GptsTools.is_delete == 0).order_by(GptsTools.create_time.desc())
return session.exec(statement).all()
@classmethod
async def aget_list_by_type(cls, tool_type_ids: List[int]) -> List[GptsTools]:
"""
Get all the tools under the Tools category asynchronously
"""
statement = select(GptsTools).where(
GptsTools.type.in_(tool_type_ids)).where(
GptsTools.is_delete == 0).order_by(GptsTools.create_time.desc())
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def get_all_tool_type(cls, tool_type_ids: List[int]) -> List[GptsToolsType]:
"""
Get all tool categories
"""
with get_sync_db_session() as session:
statement = select(GptsToolsType).filter(
GptsToolsType.is_delete == 0,
GptsToolsType.id.in_(tool_type_ids)
)
return session.exec(statement).all()
@classmethod
async def aget_all_tool_type(cls, tool_type_ids: List[int]) -> List[GptsToolsType]:
""" get tool types by tool ids """
statement = select(GptsToolsType).filter(
col(GptsToolsType.is_delete) == 0,
col(GptsToolsType.id).in_(tool_type_ids)
)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def get_preset_tool_type(cls) -> List[GptsToolsType]:
"""
Get all preset tool categories
"""
with get_sync_db_session() as session:
statement = select(GptsToolsType).where(GptsToolsType.is_preset == ToolPresetType.PRESET.value,
GptsToolsType.is_delete == 0)
statement = statement.order_by(GptsToolsType.update_time.desc())
return session.exec(statement).all()
@classmethod
async def aget_preset_tool_type(cls) -> List[GptsToolsType]:
"""
Get all preset tool categories asynchronously
"""
statement = select(GptsToolsType).where(GptsToolsType.is_preset == ToolPresetType.PRESET.value,
GptsToolsType.is_delete == 0)
statement = statement.order_by(GptsToolsType.update_time.desc())
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def _get_user_tool_type_statement(cls, user_id: int, extra_tool_type_ids: List[int] = None,
include_preset: bool = True,
is_preset: ToolPresetType = None):
"""
Get the value of all tool categories visible to the userstatement
"""
statement = select(GptsToolsType).where(GptsToolsType.is_delete == 0)
filters = []
if extra_tool_type_ids:
filters.append(or_(
GptsToolsType.id.in_(extra_tool_type_ids),
GptsToolsType.user_id == user_id
))
else:
filters.append(GptsToolsType.user_id == user_id)
if include_preset:
filters.append(GptsToolsType.is_preset == ToolPresetType.PRESET.value)
if is_preset is not None:
statement = statement.where(GptsToolsType.is_preset == is_preset.value)
statement = statement.where(or_(*filters))
statement = statement.order_by(func.field(GptsToolsType.is_preset,
ToolPresetType.PRESET.value).desc(),
GptsToolsType.update_time.desc())
return statement
@classmethod
def get_user_tool_type(cls, user_id: int, extra_tool_type_ids: List[int] = None, include_preset: bool = True,
is_preset: ToolPresetType = None) -> List[GptsToolsType]:
"""
Get all tool categories visible to the user
"""
statement = cls._get_user_tool_type_statement(user_id, extra_tool_type_ids, include_preset, is_preset)
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
async def aget_user_tool_type(cls, user_id: int, extra_tool_type_ids: List[int] = None, include_preset: bool = True,
is_preset: ToolPresetType = None) -> List[GptsToolsType]:
"""
Get all tool categories visible to the user
"""
statement = cls._get_user_tool_type_statement(user_id, extra_tool_type_ids, include_preset, is_preset)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def filter_tool_types_by_ids(cls, tool_type_ids: List[int], keyword: Optional[str] = None, page: int = 0,
limit: int = 0, include_preset: bool = False) -> (List[GptsToolsType], int):
"""
By Tool CategoryidFilter Category
"""
statement = select(GptsToolsType).where(GptsToolsType.is_delete == 0)
count_statement = select(func.count(GptsToolsType.id)).where(GptsToolsType.is_delete == 0)
if not include_preset:
statement = statement.where(GptsToolsType.is_preset != ToolPresetType.PRESET.value)
count_statement = count_statement.where(GptsToolsType.is_preset != ToolPresetType.PRESET.value)
if tool_type_ids:
statement = statement.where(GptsToolsType.id.in_(tool_type_ids))
count_statement = count_statement.where(GptsToolsType.id.in_(tool_type_ids))
if keyword:
statement = statement.where(or_(
GptsToolsType.name.like(f'%{keyword}%'),
GptsToolsType.description.like(f'%{keyword}%')
))
count_statement = count_statement.where(or_(
GptsToolsType.name.like(f'%{keyword}%'),
GptsToolsType.description.like(f'%{keyword}%')
))
if limit and page:
statement = statement.offset(
(page - 1) * limit
).limit(limit).order_by(GptsToolsType.update_time.desc())
with get_sync_db_session() as session:
return session.exec(statement).all(), session.scalar(count_statement)
@classmethod
def get_one_tool_type(cls, tool_type_id: int) -> GptsToolsType:
"""
Get details about a category, includingopenapiright of privacyschemaAgreement Wording
"""
with get_sync_db_session() as session:
statement = select(GptsToolsType).where(GptsToolsType.id == tool_type_id)
return session.exec(statement).first()
@classmethod
async def aget_one_tool_type(cls, tool_type_id: int) -> GptsToolsType:
"""
Get details about a category asynchronously, includingopenapiright of privacyschemaAgreement Wording
"""
statement = select(GptsToolsType).where(GptsToolsType.id == tool_type_id)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.first()
@classmethod
async def get_one_tool_type_by_name(cls, user_id: int, tool_type_name: str) -> GptsToolsType:
"""
Get the details of a tool category asynchronously
"""
statement = select(GptsToolsType).filter(
col(GptsToolsType.name) == tool_type_name,
col(GptsToolsType.user_id) == user_id,
col(GptsToolsType.is_delete) == 0
)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.first()
@classmethod
async def insert_tool_type(cls, data: GptsToolsTypeRead) -> GptsToolsTypeRead:
"""
Add Tool Category and the corresponding list of tools
"""
children = data.children
gpts_tool_type = GptsToolsType(**data.model_dump(exclude={'children'}))
# Insert Tool Category
async with get_async_db_session() as session:
session.add(gpts_tool_type)
await session.commit()
await session.refresh(gpts_tool_type)
if children:
# Insert Tools List
for one in children:
one.type = gpts_tool_type.id
one.tool_key = cls.get_tool_key(gpts_tool_type.id, one)
session.add_all(children)
await session.commit()
res = GptsToolsTypeRead(**gpts_tool_type.model_dump(), children=children)
return res
@classmethod
async def update_tool_type(cls, data: GptsToolsType,
del_tool_ids: List[int],
add_tool_list: List[GptsTools],
update_tool_list: List[GptsTools]):
"""
Update Tool Category Information
param data: GptsToolsType
param del_tool_ids: Tools to removeid
param add_tool_list: List of tools that need to be added
param update_tool_list: List of tools that need to be updated
"""
finally_children = []
async with get_async_db_session() as session:
# Update tool category data
session.add(data)
# Delete a list of tools that don't exist
delete_old_tools = update(GptsTools).where(GptsTools.id.in_(del_tool_ids)).values(is_delete=1)
await session.exec(delete_old_tools)
# Add Tool List
for one in add_tool_list:
one.type = data.id
one.tool_key = cls.get_tool_key(data.id, one)
session.add(one)
finally_children.append(one)
# Update Tool List
for one in update_tool_list:
session.add(one)
finally_children.append(one)
await session.commit()
await session.refresh(data)
@classmethod
async def delete_tool_type(cls, tool_type_id: int) -> None:
"""
Delete Tool Category
"""
statement = update(GptsToolsType).where(col(GptsToolsType.id) == tool_type_id,
col(GptsToolsType.is_preset) != ToolPresetType.PRESET.value).values(
is_delete=1)
tool_statement = update(GptsTools).where(col(GptsTools.type) == tool_type_id,
col(GptsTools.is_preset) != ToolPresetType.PRESET.value).values(
is_delete=1)
async with get_async_db_session() as session:
await session.exec(statement)
await session.exec(tool_statement)
await session.commit()
@classmethod
def get_tool_key(cls, tool_type_id: int, gpt_tool: GptsTools) -> str:
"""
of stitching custom toolstool_key
"""
if gpt_tool.is_preset == ToolPresetType.MCP.value:
return f"{gpt_tool.name}_{generate_uuid()[:8]}"
return f"tool_type_{tool_type_id}_{md5_hash(gpt_tool.name)}"
@classmethod
async def update_tools_extra(cls, tool_type_id: int, extra: str) -> bool:
async with get_async_db_session() as session:
statement = update(GptsToolsType).where(col(GptsToolsType.id) == tool_type_id).values(extra=extra)
await session.exec(statement)
statement = update(GptsTools).where(col(GptsTools.type) == tool_type_id).values(extra=text('NULL'))
await session.exec(statement)
await session.commit()
return True
@classmethod
def get_tool_by_tool_key(cls, tool_key: str) -> GptsTools:
with get_sync_db_session() as session:
statement = select(GptsTools).where(GptsTools.tool_key == tool_key)
return session.exec(statement).first()
@classmethod
async def aget_tool_by_tool_key(cls, tool_key: str) -> GptsTools:
statement = select(GptsTools).where(GptsTools.tool_key == tool_key)
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.first()
================================================
FILE: src/backend/bisheng/tool/domain/schemas.py
================================================
from typing import Optional, Dict
from pydantic import BaseModel, Field
from bisheng.tool.domain.const import AuthMethod, AuthType
class TestToolReq(BaseModel):
server_host: str = Field(default='', description='Root address of the service')
openapi_schema: Optional[str] = Field(default='', description='openapi schema')
extra: str = Field(default='', description='Api After the object is parsedextraData field')
auth_method: int = Field(default=AuthMethod.NO.value, description='Certification Type')
auth_type: Optional[str] = Field(default=AuthType.BASIC.value, description='Auth Type')
api_key: Optional[str] = Field(default='', description='api key')
api_location: Optional[str] = Field(default='', description='api location')
parameter_name: Optional[str] = Field(default='', description='parameter_name')
request_params: Dict = Field(default=None, description='User Filled Request Parameters')
================================================
FILE: src/backend/bisheng/tool/domain/services/__init__.py
================================================
================================================
FILE: src/backend/bisheng/tool/domain/services/executor.py
================================================
import functools
import json
from inspect import signature
from typing import Any, Annotated, Optional, Dict, List
from langchain_core.callbacks import CallbackManagerForToolRun, Callbacks
from langchain_core.runnables import RunnableConfig
from langchain_core.tools import BaseTool, ArgsSchema
from pydantic import Field, SkipValidation
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum, StatusEnum, ApplicationTypeEnum
from bisheng.common.schemas.telemetry.event_data_schema import ToolInvocationEventData
from bisheng.common.services import telemetry_service
from bisheng.common.services.config_service import settings
from bisheng.core.logger import trace_id_var
from bisheng.knowledge.domain.knowledge_rag import KnowledgeRag
from bisheng.knowledge.domain.models.knowledge import KnowledgeDao, Knowledge
from bisheng.mcp_manage.langchain.tool import McpTool
from bisheng.mcp_manage.manager import ClientManager
from bisheng.tool.domain.const import ToolPresetType
from bisheng.tool.domain.langchain.knowledge import KnowledgeRagTool
from bisheng.tool.domain.models.gpts_tools import GptsToolsDao, GptsTools, GptsToolsType
from bisheng.tool.domain.services.openapi import OpenApiSchema
from bisheng_langchain.gpts.load_tools import load_tools
from bisheng_langchain.gpts.tools.api_tools.openapi import OpenApiTools
def wrapper_tool(func):
@functools.wraps(func)
async def inner(*args, **kwargs):
# Here you can add logic to record tool usage
self: ToolExecutor = args[0]
status = StatusEnum.SUCCESS
try:
return await func(*args, **kwargs)
except Exception as e:
status = StatusEnum.FAILED
raise e
finally:
await telemetry_service.log_event(**self.get_invoke_log_data(status))
return inner
def wrapper_tool_sync(func):
@functools.wraps(func)
def inner(*args, **kwargs):
# Here you can add logic to record tool usage
self: ToolExecutor = args[0]
status = StatusEnum.SUCCESS
try:
return func(*args, **kwargs)
except Exception as e:
status = StatusEnum.FAILED
raise e
finally:
# RecordTelemetryJournal
telemetry_service.log_event_sync(**self.get_invoke_log_data(status))
return inner
class ToolExecutor(BaseTool):
# must provide fields for telemetry logging
app_id: str = Field(..., description="Application Identifier")
app_name: str = Field(..., description="Application Name")
app_type: ApplicationTypeEnum = Field(..., description="Application Type")
user_id: int = Field(..., description="Invoke User ID")
# bisheng tool fields
tool_id: int = Field(..., description="Tool ID")
tool_is_preset: int = Field(..., description="kind of the Tool, api、mcp、preset")
tool_name: str = Field(..., description="Tool Name")
# langchain tool fields
name: str = Field(..., description="Tool Name for llm")
description: str = Field(..., description="Tool Description for llm")
args_schema: Annotated[Optional[ArgsSchema], SkipValidation()] = Field(
default=None, description="The tool schema."
)
tool_instance: BaseTool = Field(..., description="Langchain Tool Instance")
@classmethod
def init_by_tool_instance(cls, base_tool: BaseTool, tool: GptsTools, *, app_id: str, app_name: str,
app_type: ApplicationTypeEnum, user_id: int) -> BaseTool:
return cls(name=base_tool.name, description=base_tool.description, args_schema=base_tool.args_schema,
tool_instance=base_tool,
app_id=app_id, app_name=app_name, app_type=app_type, user_id=user_id,
tool_id=tool.id, tool_is_preset=tool.is_preset, tool_name=tool.name)
@staticmethod
def parse_preset_tool_params(tool: GptsTools, tool_type: GptsToolsType) -> Dict:
""" parse tool init params """
# get tool config from tool.extra first
params = {}
if tool.extra:
if isinstance(tool.extra, str):
params = json.loads(tool.extra)
elif isinstance(tool.extra, dict):
params = tool.extra
elif tool_type.extra:
if isinstance(tool_type.extra, str):
params = json.loads(tool_type.extra)
elif isinstance(tool_type.extra, dict):
params = tool_type.extra
# special handling for bisheng_code_interpreter, because it needs minio config
if tool.tool_key == 'bisheng_code_interpreter':
return {'minio': settings.get_minio_conf().model_dump(), **params}
return params
@staticmethod
def parse_api_tool_params(tool: GptsTools, tool_type: GptsToolsType, **kwargs) -> Dict:
""" parse tool init params """
extra_json = json.loads(tool.extra) if tool.extra else {}
extra_json.update(json.loads(tool_type.extra) if tool_type.extra else {})
params = OpenApiSchema.parse_openapi_tool_params(tool.name, tool.desc, extra_json,
tool_type.server_host,
tool_type.auth_method,
tool_type.auth_type,
tool_type.api_key)
if "llm" in kwargs:
kwargs.pop("llm")
params.update(kwargs)
return params
@classmethod
def _init_preset_tool(cls, tool: GptsTools, tool_type: GptsToolsType, **kwargs) -> BaseTool:
tool_name_param = {
tool.tool_key: cls.parse_preset_tool_params(tool, tool_type)
}
tool_langchain = load_tools(tool_params=tool_name_param, **kwargs)
if not tool_langchain:
raise ValueError(f"Failed to load preset tool: {tool.tool_key}")
return tool_langchain[0]
@classmethod
def _init_api_tool(cls, tool: GptsTools, tool_type: GptsToolsType, callbacks: Callbacks = None,
**kwargs) -> BaseTool:
tool_params = cls.parse_api_tool_params(tool, tool_type, **kwargs)
tool = OpenApiTools.get_api_tool(tool.tool_key, **tool_params)
tool.callbacks = callbacks
return tool
@classmethod
def _init_mcp_tool(cls, tool: GptsTools, tool_type: GptsToolsType, **kwargs) -> BaseTool:
mcp_client = ClientManager.sync_connect_mcp_from_json(tool_type.openapi_schema)
input_schema = json.loads(tool.extra)
return McpTool.get_mcp_tool(name=tool.tool_key, description=tool.desc, mcp_client=mcp_client,
mcp_tool_name=tool.name, arg_schema=input_schema['inputSchema'],
**kwargs)
@classmethod
def _init_by_tool_and_type(cls, tool: GptsTools, tool_type: GptsToolsType, *, app_id: str, app_name: str,
app_type: ApplicationTypeEnum, user_id: int, **kwargs) -> BaseTool:
if tool.is_preset == ToolPresetType.PRESET.value:
tool_instance = cls._init_preset_tool(tool, tool_type, **kwargs)
elif tool.is_preset == ToolPresetType.API.value:
tool_instance = cls._init_api_tool(tool, tool_type, **kwargs)
elif tool.is_preset == ToolPresetType.MCP.value:
tool_instance = cls._init_mcp_tool(tool, tool_type, **kwargs)
else:
raise ValueError(f"Unsupported tool preset type: {tool.is_preset}")
return cls.init_by_tool_instance(base_tool=tool_instance, tool=tool, app_id=app_id,
app_name=app_name, app_type=app_type, user_id=user_id)
@classmethod
async def init_by_tool_id(cls, tool_id: int = None, tool: GptsTools = None, *, app_id: str, app_name: str,
app_type: ApplicationTypeEnum, user_id: int, **kwargs) -> BaseTool:
if not tool_id and not tool:
raise ValueError("Either tool_id or tool must be provided.")
if not tool:
tool = await GptsToolsDao.aget_one_tool(tool_id=tool_id)
if not tool:
raise ValueError(f"Tool with id {tool_id} not found.")
tool_type = await GptsToolsDao.aget_one_tool_type(tool_type_id=tool.type)
if not tool_type:
raise ValueError(f"Tool type with id {tool.type} not found.")
return cls._init_by_tool_and_type(tool=tool, tool_type=tool_type, app_id=app_id, app_name=app_name,
app_type=app_type, user_id=user_id, **kwargs)
@classmethod
def _init_tools(cls, tools: List[GptsTools], tool_types_map: Dict[int, GptsToolsType], *,
app_id: str, app_name: str, app_type: ApplicationTypeEnum,
user_id: int, **kwargs) -> List[BaseTool]:
result = []
for tool in tools:
tool_type = tool_types_map.get(tool.type)
if not tool_type:
raise ValueError(f"Tool type with id {tool.type} not found.")
result.append(
cls._init_by_tool_and_type(tool=tool, tool_type=tool_type, app_id=app_id,
app_name=app_name, app_type=app_type, user_id=user_id, **kwargs)
)
return result
@classmethod
async def init_by_tool_ids(cls, tool_ids: list[int], *, app_id: str, app_name: str, app_type: ApplicationTypeEnum,
user_id: int, **kwargs) -> List[BaseTool]:
tools = await GptsToolsDao.aget_list_by_ids(tool_ids)
tool_type_ids = [tool.type for tool in tools]
tool_types = await GptsToolsDao.aget_all_tool_type(list(set(tool_type_ids)))
tool_type_map = {tool_type.id: tool_type for tool_type in tool_types}
return cls._init_tools(tools, tool_type_map, app_id=app_id, app_name=app_name, app_type=app_type,
user_id=user_id, **kwargs)
@classmethod
def init_by_tool_ids_sync(cls, tool_ids: list[int], app_id: str, app_name: str, app_type: ApplicationTypeEnum,
user_id: int, **kwargs) -> List[BaseTool]:
tools = GptsToolsDao.get_list_by_ids(tool_ids)
tool_type_ids = [tool.type for tool in tools]
tool_types = GptsToolsDao.get_all_tool_type(list(set(tool_type_ids)))
tool_type_map = {tool_type.id: tool_type for tool_type in tool_types}
return cls._init_tools(tools, tool_type_map, app_id=app_id, app_name=app_name, app_type=app_type,
user_id=user_id, **kwargs)
@classmethod
def init_by_tool_id_sync(cls, tool_id: int, *, app_id: str, app_name: str, app_type: ApplicationTypeEnum,
user_id: int, **kwargs) -> BaseTool:
tool = GptsToolsDao.get_one_tool(tool_id=tool_id)
if not tool:
raise ValueError(f"Tool with id {tool_id} not found.")
tool_type = GptsToolsDao.get_one_tool_type(tool_type_id=tool.type)
if not tool_type:
raise ValueError(f"Tool type with id {tool.type} not found.")
return cls._init_by_tool_and_type(tool=tool, tool_type=tool_type, app_id=app_id, app_name=app_name,
app_type=app_type, user_id=user_id, **kwargs)
@classmethod
def _init_knowledge_rag_tool(cls, knowledge: Knowledge, **kwargs) -> BaseTool:
return KnowledgeRagTool.init_knowledge_rag_tool(name=f'knowledge_{knowledge.id}',
description=f'{knowledge.name}:{knowledge.description}',
**kwargs)
@classmethod
async def init_knowledge_tool(cls, invoke_user_id: int, knowledge_id: int, **kwargs) -> BaseTool:
knowledge = await KnowledgeDao.aquery_by_id(knowledge_id=knowledge_id)
if not knowledge:
raise ValueError(f"Knowledge with id {knowledge_id} not found.")
vector_client = await KnowledgeRag.init_knowledge_milvus_vectorstore(invoke_user_id, knowledge)
es_client = await KnowledgeRag.init_knowledge_es_vectorstore(knowledge)
return cls._init_knowledge_rag_tool(knowledge=knowledge, vector_retriever=vector_client.as_retriever(),
elastic_retriever=es_client.as_retriever(), **kwargs)
@classmethod
def init_knowledge_tool_sync(cls, invoke_user_id: int, knowledge_id: int, **kwargs) -> BaseTool:
knowledge = KnowledgeDao.query_by_id(knowledge_id=knowledge_id)
if not knowledge:
raise ValueError(f"Knowledge with id {knowledge_id} not found.")
vector_client = KnowledgeRag.init_knowledge_milvus_vectorstore_sync(invoke_user_id, knowledge)
es_client = KnowledgeRag.init_knowledge_es_vectorstore_sync(knowledge)
return cls._init_knowledge_rag_tool(knowledge, vector_retriever=vector_client.as_retriever(),
elastic_retriever=es_client.as_retriever(), **kwargs)
@classmethod
def init_tmp_knowledge_tool_sync(cls, **kwargs) -> BaseTool:
return KnowledgeRagTool.init_knowledge_rag_tool(**kwargs)
@wrapper_tool_sync
def _run(self,
*args: Any,
config: RunnableConfig,
run_manager: Optional[CallbackManagerForToolRun] = None,
**kwargs) -> Any:
if signature(self.tool_instance._run).parameters.get("config"):
kwargs["config"] = config
if signature(self.tool_instance._run).parameters.get("run_manager"):
kwargs["run_manager"] = run_manager
return self.tool_instance._run(*args, **kwargs)
@wrapper_tool
async def _arun(self,
*args: Any,
config: RunnableConfig,
run_manager: Optional[CallbackManagerForToolRun] = None,
**kwargs) -> Any:
if signature(self.tool_instance._run).parameters.get("config"):
kwargs["config"] = config
if signature(self.tool_instance._run).parameters.get("run_manager"):
kwargs["run_manager"] = run_manager
return await self.tool_instance._arun(*args, **kwargs)
def __getattr__(self, item):
return getattr(self.tool_instance, item)
def get_invoke_log_data(self, status: StatusEnum):
# RecordTelemetryJournal
return {
"user_id": self.user_id,
"event_type": BaseTelemetryTypeEnum.TOOL_INVOKE,
"trace_id": trace_id_var.get(),
"event_data": ToolInvocationEventData(
app_id=self.app_id,
app_name=self.app_name,
app_type=self.app_type,
tool_id=self.tool_id,
tool_name=self.tool_name,
tool_type=self.tool_is_preset,
status=status
)}
================================================
FILE: src/backend/bisheng/tool/domain/services/openapi.py
================================================
import json
from typing import Dict
from bisheng.tool.domain.const import AuthMethod, AuthType
class OpenApiSchema:
def __init__(self, contents: dict):
self.parameter_name = None
self.api_location = None
self.contents = contents
self.version = contents['openapi']
self.info = contents['info']
self.title = self.info['title']
self.auth_type = 'basic'
self.auth_method = 0
self.description = self.info.get('description', '')
self.default_server = ''
self.apis = []
def parse_server(self) -> str:
"""
get the default server url
"""
if self.contents.get('servers') is None:
raise Exception('openapi schema must have servers')
servers = self.contents['servers']
if isinstance(servers, list):
self.default_server = servers[0]['url']
else:
self.default_server = servers['url']
security_schemes = self.contents.get('components', {}).get('securitySchemes', {})
api_key_auth = security_schemes.get('ApiKeyAuth', {})
# Get certification type
auth_type = api_key_auth.get('type')
if auth_type == 'apiKey':
self.auth_type = 'custom'
elif auth_type == 'http':
self.auth_type = api_key_auth.get('schema')
else:
self.auth_type = 'basic'
# Set authentication method
self.auth_method = 1 if auth_type in ('apiKey', 'http') else 0
# Dapatkan API Location and parameter name
self.api_location = api_key_auth.get('in')
self.parameter_name = api_key_auth.get('name')
return self.default_server
def parse_paths(self) -> list[dict]:
paths = self.contents['paths']
self.apis = []
for path, path_info in paths.items():
for method, method_info in path_info.items():
one_api_info = {
'path': path,
'method': method,
'description': method_info.get('description', '')
or method_info.get('summary', ''),
'operationId': method_info['operationId'],
'parameters': [],
}
if method not in ['get', 'post', 'put', 'delete']:
continue
if 'requestBody' in method_info:
for _, content in method_info['requestBody']['content'].items():
if '$ref' in content['schema']:
schema_ref = content['schema']['$ref']
schema_name = schema_ref.split('/')[-1]
schema = self.contents['components']['schemas'][schema_name]
else:
schema = content['schema']
if 'properties' in schema:
for param_name, param_info in schema['properties'].items():
param = {
'name': param_name,
'description': param_info.get('description', ''),
'in': 'body',
'required': param_name in schema.get('required', []),
'schema': {
'type': param_info.get('type', 'string'),
'title': param_info.get('title', param_name),
'properties': param_info.get('properties', {})
},
}
one_api_info['parameters'].append(param)
else:
# no request body get parameters
one_api_info['parameters'].extend(method_info.get('parameters', []))
self.apis.append(one_api_info)
return self.apis
@staticmethod
def parse_openapi_tool_params(name: str,
description: str,
extra: str | Dict,
server_host: str,
auth_method: int,
auth_type: str = None,
api_key: str = None):
if isinstance(extra, str):
extra = json.loads(extra)
# Stitching request header
headers = {}
if auth_method == AuthMethod.API_KEY.value:
if auth_type == AuthType.CUSTOM.value:
location = extra["api_location"]
parameter_name = extra["parameter_name"]
if location == "header":
headers = {parameter_name: api_key}
elif auth_type == AuthType.BASIC.value:
headers = {'Authorization': f'Basic {api_key}'}
elif auth_type == AuthType.BEARER.value:
headers = {'Authorization': f'Bearer {api_key}'}
# Back to initialization openapiRequired input parameters
params = {
'params': extra,
'headers': headers,
'api_key': api_key,
'url': server_host,
'description': name + description if description else name
}
return params
================================================
FILE: src/backend/bisheng/tool/domain/services/tool.py
================================================
import asyncio
import json
from typing import Optional, List
import yaml
from fastapi import Request
from langchain_core.tools import BaseTool
from loguru import logger
from pydantic import BaseModel, ConfigDict
from bisheng.api.services.audit_log import AuditLogService
from bisheng.api.utils import get_url_content
from bisheng.common.dependencies.user_deps import UserPayload
from bisheng.common.errcode import BaseErrorCode
from bisheng.common.errcode.http_error import UnAuthorizedError, NotFoundError
from bisheng.common.errcode.tool import ToolTypeNotExistsError, ToolTypeRepeatError, ToolTypeNameError, \
ToolTypeIsPresetError, ToolSchemaDownloadError, ToolSchemaEmptyError, ToolSchemaParseError, ToolSchemaServerError, \
ToolMcpSchemaError
from bisheng.database.models.group_resource import GroupResourceDao, ResourceTypeEnum, GroupResource
from bisheng.database.models.role_access import AccessType
from bisheng.database.models.user_group import UserGroupDao
from bisheng.mcp_manage.manager import ClientManager
from bisheng.tool.domain.const import ToolPresetType
from bisheng.tool.domain.langchain.linsight_knowledge import SearchKnowledgeBase
from bisheng.tool.domain.models.gpts_tools import GptsToolsDao, GptsTools, GptsToolsType, GptsToolsTypeRead
from bisheng.tool.domain.services.openapi import OpenApiSchema
from bisheng.utils import md5_hash, get_request_ip
from bisheng.utils.mask_data import JsonFieldMasker
from bisheng_langchain.gpts.load_tools import load_tools
class ToolServices(BaseModel):
""" Tool service class """
model_config = ConfigDict(arbitrary_types_allowed=True)
request: Optional[Request] = None
login_user: Optional[UserPayload] = None
async def get_tool_list(self, is_preset: Optional[int] = None) -> List[GptsToolsTypeRead]:
""" Get a list of tools visible to users """
# Get Tool Categories Visible to Users
tool_type_ids_extra = []
if is_preset != ToolPresetType.PRESET.value:
# When getting a list of custom tools, you need to include a list of tools available to the user
access_resources = await self.login_user.aget_user_access_resource_ids([AccessType.GPTS_TOOL_READ])
if access_resources:
tool_type_ids_extra = [int(access) for access in access_resources]
if is_preset is None:
# Get a list of all tools visible to the user
all_tool_type = await GptsToolsDao.aget_user_tool_type(self.login_user.user_id, tool_type_ids_extra)
elif is_preset == ToolPresetType.PRESET.value:
# Get a list of preset tools
all_tool_type = await GptsToolsDao.aget_preset_tool_type()
else:
# Get a list of custom tools visible to users
all_tool_type = await GptsToolsDao.aget_user_tool_type(self.login_user.user_id, tool_type_ids_extra, False,
ToolPresetType(is_preset))
tool_type_id = [one.id for one in all_tool_type]
res: List[GptsToolsTypeRead] = []
tool_type_children = {}
for one in all_tool_type:
tool_type_id.append(one.id)
tool_type_children[one.id] = []
res.append(GptsToolsTypeRead.model_validate(one))
# Get the list of tools under the corresponding category
tool_list = await GptsToolsDao.aget_list_by_type(tool_type_id)
for one in tool_list:
tool_type_children[one.type].append(one)
# check write permission
write_tool_type = None
for one in res:
if self.login_user.is_admin() or one.user_id == self.login_user.user_id:
one.write = True
else:
if write_tool_type is None:
write_resources = await self.login_user.aget_user_access_resource_ids([AccessType.GPTS_TOOL_WRITE])
write_tool_type = {int(x): True for x in write_resources}
one.write = write_tool_type.get(one.id, False)
one.children = tool_type_children.get(one.id, [])
# Data desensitization
one.mask_sensitive_data()
return res
async def add_tools(self, req: GptsToolsTypeRead) -> GptsToolsTypeRead:
""" Add custom tool """
# Try to parse theopenapi schemaSee if it can be parsed normally, Save if not possible Do not allow to save
if req.is_preset == ToolPresetType.API.value:
await self.parse_openapi_schema('', req.openapi_schema)
elif req.is_preset == ToolPresetType.MCP.value:
await self.parse_mcp_schema(req.openapi_schema)
req.id = None
if req.name.__len__() > 1000 or req.name.__len__() == 0:
raise ToolTypeNameError()
# Determine if the category already exists
tool_type = await GptsToolsDao.get_one_tool_type_by_name(self.login_user.user_id, req.name)
if tool_type:
raise ToolTypeRepeatError()
req.user_id = self.login_user.user_id
for one in req.children:
one.id = None
one.user_id = self.login_user.user_id
one.is_delete = 0
one.is_preset = req.is_preset
tool_extra = {"api_location": req.api_location, "parameter_name": req.parameter_name}
req.extra = json.dumps(tool_extra, ensure_ascii=False)
# Add Tool Category and Corresponding Tools List
res = await GptsToolsDao.insert_tool_type(req)
self.add_gpts_tools_hook(self.request, self.login_user, res)
return res
@classmethod
def add_gpts_tools_hook(cls, request: Request, user: UserPayload, gpts_tool_type: GptsToolsTypeRead) -> bool:
""" After adding custom toolshookFunction """
# Query the user group the user belongs to under
user_group = UserGroupDao.get_user_group(user.user_id)
group_ids = []
if user_group:
# Batch Insert Custom Tools into Correlation Table
batch_resource = []
for one in user_group:
group_ids.append(one.group_id)
batch_resource.append(GroupResource(
group_id=one.group_id,
third_id=gpts_tool_type.id,
type=ResourceTypeEnum.GPTS_TOOL.value))
GroupResourceDao.insert_group_batch(batch_resource)
AuditLogService.create_tool(user, get_request_ip(request), group_ids, gpts_tool_type)
return True
async def update_tool_config(self, tool_type_id: int, extra: dict) -> bool:
# Get Tool Categories
tool_type = await GptsToolsDao.aget_one_tool_type(tool_type_id)
if not tool_type or tool_type.is_preset != ToolPresetType.PRESET.value:
raise NotFoundError()
if not await self.login_user.async_access_check(tool_type.user_id, str(tool_type.id),
AccessType.GPTS_TOOL_WRITE):
raise UnAuthorizedError()
if tool_type.extra is None:
tool_type.extra = '{}'
json_masker = JsonFieldMasker()
old_config = json.loads(tool_type.extra)
# special handle dall-e config update。 Wait until you unify this configuration with the configuration logic of other tools (the configuration parameters remain unchanged), and then delete this special processing logic
if tool_type.name == "Dalle3- painting;":
# Instructions not toggledtab. Just changed the configuration
if ("azure_endpoint" in old_config and "azure_endpoint" in extra) or (
"azure_endpoint" not in old_config and "azure_endpoint" not in extra
):
# Update the configuration of all tools under the Tools category
merge_extra = json_masker.update_json_with_masked(old_config, extra)
merge_extra = json.dumps(merge_extra, ensure_ascii=False)
elif not old_config:
# The description has not been configured before, it is now configured
merge_extra = json.dumps(extra, ensure_ascii=False)
else:
# Instructions switchedtab, you need to overwrite the previous configuration
merge_extra = json.dumps(extra, ensure_ascii=False)
else:
# Update the configuration of all tools under the Tools category
merge_extra = json_masker.update_json_with_masked(old_config, extra)
merge_extra = json.dumps(merge_extra, ensure_ascii=False)
await GptsToolsDao.update_tools_extra(tool_type_id, merge_extra)
return True
@staticmethod
async def parse_openapi_schema(download_url: str, file_content: str) -> GptsToolsTypeRead:
if download_url:
try:
file_content = await get_url_content(download_url)
except Exception as e:
logger.exception(f'file {download_url} download error')
raise ToolSchemaDownloadError(exception=e)
if not file_content:
raise ToolSchemaEmptyError()
# Depending on the content of the document, is it possible to`{`At the beginning, what analytical method is used to determine
try:
if file_content.startswith('{'):
res = json.loads(file_content)
else:
res = yaml.safe_load(file_content)
except Exception as e:
logger.exception(f'openapi schema parse error {e}')
raise ToolSchemaParseError(exception=e)
# analyzingopenapi schemaConvert to Helper Tool Format
try:
schema = OpenApiSchema(res)
schema.parse_server()
if not schema.default_server.startswith(('http', 'https')):
raise ToolSchemaServerError(data={"url": schema.default_server})
tool_type = GptsToolsTypeRead(name=schema.title,
description=schema.description,
is_preset=ToolPresetType.API.value,
server_host=schema.default_server,
openapi_schema=file_content,
api_location=schema.api_location,
parameter_name=schema.parameter_name,
auth_type=schema.auth_type,
auth_method=schema.auth_method,
children=[])
# Parsing to get all theapi
schema.parse_paths()
for one in schema.apis:
tool_type.children.append(
GptsTools(
name=one['operationId'],
desc=one['description'],
tool_key=md5_hash(one['operationId']),
is_preset=0,
is_delete=0,
api_params=one['parameters'],
extra=json.dumps(one, ensure_ascii=False),
))
return tool_type
except BaseErrorCode as e:
raise e
except Exception as e:
logger.exception(f'openapi schema parse error {e}')
raise ToolSchemaParseError(exception=e)
@staticmethod
async def parse_mcp_schema(file_content: str) -> GptsToolsTypeRead:
try:
result = json.loads(file_content)
mcp_servers = result['mcpServers']
except Exception as e:
logger.exception(f'mcp tool schema parse error {e}')
raise ToolMcpSchemaError(exception=e)
tool_type = None
for key, value in mcp_servers.items():
# analyzingmcpService Config
tool_type = GptsToolsTypeRead(name=value.get('name', ''),
server_host=value.get('url', ''),
description=value.get('description', ''),
is_preset=ToolPresetType.MCP.value,
openapi_schema=file_content,
children=[])
# Instantiatemcpservice object, getting a list of tools
client = await ClientManager.connect_mcp_from_json(result)
tools = await client.list_tools()
for one in tools:
tool_type.children.append(GptsTools(
name=one.name,
desc=one.description,
tool_key=md5_hash(one.name),
is_preset=ToolPresetType.MCP.value,
api_params=ToolServices.convert_input_schema(one.inputSchema),
extra=one.model_dump_json(),
))
break
if tool_type is None:
raise ToolMcpSchemaError()
return tool_type
@classmethod
async def _update_gpts_tools(cls, exist_tool_type: GptsToolsType, req: GptsToolsTypeRead) -> GptsToolsTypeRead:
exist_tool_type.name = req.name
exist_tool_type.logo = req.logo
exist_tool_type.description = req.description
exist_tool_type.server_host = req.server_host
exist_tool_type.auth_method = req.auth_method
exist_tool_type.api_key = req.api_key
exist_tool_type.auth_type = req.auth_type
exist_tool_type.openapi_schema = req.openapi_schema
tool_extra = {"api_location": req.api_location, "parameter_name": req.parameter_name}
exist_tool_type.extra = json.dumps(tool_extra, ensure_ascii=False)
children_map = {}
for one in req.children:
children_map[one.name] = one
# Get old under this categoryAPIVertical
old_tool_list = await GptsToolsDao.aget_list_by_type([exist_tool_type.id])
# List of tools that need to be removed
delete_tool_id_list = []
# List of tools that need to be updated
update_tool_list = []
for one in old_tool_list:
# Explain this tool Removal required
if children_map.get(one.name) is None:
delete_tool_id_list.append(one.id)
else:
# Explain that this tool needs to be updated
new_tool_info = children_map.pop(one.name)
one.name = new_tool_info.name
one.desc = new_tool_info.desc
one.extra = new_tool_info.extra
one.api_params = new_tool_info.api_params
update_tool_list.append(one)
add_children = []
for one in children_map.values():
one.id = None
one.user_id = exist_tool_type.user_id
one.is_preset = exist_tool_type.is_preset
one.is_delete = 0
add_children.append(one)
await GptsToolsDao.update_tool_type(exist_tool_type, delete_tool_id_list,
add_children, update_tool_list)
children = await GptsToolsDao.aget_list_by_type([exist_tool_type.id])
return GptsToolsTypeRead(**exist_tool_type.model_dump(), children=children)
async def update_tools(self, req: GptsToolsTypeRead) -> GptsToolsTypeRead:
"""
Updating tool categories, including updating tool category names and deletions, adding tool categoryAPI
"""
# Try to parse theopenapi schemaSee if it can be parsed normally, Save if not possible Do not allow to save
if req.is_preset == ToolPresetType.API.value:
await self.parse_openapi_schema('', req.openapi_schema)
elif req.is_preset == ToolPresetType.MCP.value:
await self.parse_mcp_schema(req.openapi_schema)
exist_tool_type = await GptsToolsDao.aget_one_tool_type(req.id)
if not exist_tool_type:
raise ToolTypeNotExistsError()
if req.name.__len__() > 1000 or req.name.__len__() == 0:
raise ToolTypeNameError()
# Determine if the tool category name is a duplicate
tool_type = await GptsToolsDao.get_one_tool_type_by_name(self.login_user.user_id, req.name)
if tool_type and tool_type.id != exist_tool_type.id:
raise ToolTypeRepeatError()
# Determine if there are update permissions
if not await self.login_user.async_access_check(exist_tool_type.user_id, str(exist_tool_type.id),
AccessType.GPTS_TOOL_WRITE):
raise UnAuthorizedError()
res = await self._update_gpts_tools(exist_tool_type, req)
await self.update_tool_hook(self.request, self.login_user, exist_tool_type)
return res
@classmethod
async def update_tool_hook(cls, request: Request, user: UserPayload, exist_tool_type):
groups = await GroupResourceDao.aget_resource_group(ResourceTypeEnum.GPTS_TOOL, exist_tool_type.id)
group_ids = [int(one.group_id) for one in groups]
await asyncio.to_thread(AuditLogService.update_tool, user, get_request_ip(request), group_ids, exist_tool_type)
async def delete_tools(self, tool_type_id: int) -> bool:
""" Delete Tool Category """
exist_tool_type = await GptsToolsDao.aget_one_tool_type(tool_type_id)
if not exist_tool_type:
return True
if exist_tool_type.is_preset == ToolPresetType.PRESET.value:
raise ToolTypeIsPresetError()
# Determine if there are update permissions
if not await self.login_user.async_access_check(exist_tool_type.user_id, str(exist_tool_type.id),
AccessType.GPTS_TOOL_WRITE):
raise UnAuthorizedError()
await GptsToolsDao.delete_tool_type(tool_type_id)
await asyncio.to_thread(self.delete_tool_hook, self.request, self.login_user, exist_tool_type)
return True
@classmethod
def delete_tool_hook(cls, request, user: UserPayload, gpts_tool_type) -> bool:
""" After deleting the customizerhookFunction """
logger.info(f"delete_gpts_tool_hook id: {gpts_tool_type.id}, user: {user.user_id}")
GroupResourceDao.delete_group_resource_by_third_id(gpts_tool_type.id, ResourceTypeEnum.GPTS_TOOL)
groups = GroupResourceDao.get_resource_group(ResourceTypeEnum.GPTS_TOOL, gpts_tool_type.id)
group_ids = [int(one.group_id) for one in groups]
AuditLogService.delete_tool(user, get_request_ip(request), group_ids, gpts_tool_type)
return True
async def refresh_all_mcp(self) -> list[str]:
""" return mcp server error msg """
# get user all mcp tool
tool_types = await GptsToolsDao.aget_user_tool_type(self.login_user.user_id, is_preset=ToolPresetType.MCP)
if not tool_types:
return []
tools = await GptsToolsDao.aget_list_by_type(tool_type_ids=[one.id for one in tool_types])
tools_map = {}
for one in tools:
if one.type not in tools_map:
tools_map[one.type] = []
tools_map[one.type].append(one)
error_name = []
for one in tool_types:
try:
await self.refresh_mcp_tools(one, tools_map.get(one.id, []))
except Exception as e:
logger.exception(f'{one.name} tool refresh failed')
error_name.append(one.name)
return error_name
async def refresh_mcp_tools(self, tool_type: GptsToolsType, old_tools: list[GptsTools]):
""" refresh mcp tools """
# 1. get all new tools
# Instantiatemcpservice object, getting a list of tools
client = await ClientManager.connect_mcp_from_json(tool_type.openapi_schema)
tools = await client.list_tools()
children = []
for one in tools:
children.append(GptsTools(
name=one.name,
desc=one.description,
is_preset=ToolPresetType.MCP.value,
api_params=self.convert_input_schema(one.inputSchema),
extra=one.model_dump_json(),
type=tool_type.id,
))
req = GptsToolsTypeRead(**tool_type.model_dump(), children=children)
await self._update_gpts_tools(tool_type, req)
@classmethod
def convert_input_schema(cls, input_schema: dict):
""" TukarmcpInput parameters for the tool Formatting for custom tools"""
required = input_schema.get('required', [])
properties = input_schema.get('properties', {})
res = []
for filed, field_info in properties.items():
res.append({
'in': "query",
'name': filed,
'description': field_info.get('description'),
'required': filed in required,
'schema': {
'type': field_info.get('type'),
}
})
return res
@classmethod
async def init_linsight_tools(cls, root_path: str) -> List[BaseTool]:
""" InisialisasiLinsight Default Tools, The special point is that the parameters initialized by the local file tool are not fixed, but are determined during rerun """
# Tools for loading local file operations
local_file_tools = load_tools({
"list_files": {"root_path": root_path},
"get_file_details": {"root_path": root_path},
"search_files": {"root_path": root_path},
# "search_text_in_file": {"root_path": root_path},
"read_text_file": {"root_path": root_path},
"add_text_to_file": {"root_path": root_path},
"replace_file_lines": {"root_path": root_path},
})
knowledge_tools = [SearchKnowledgeBase()]
return knowledge_tools + local_file_tools
@classmethod
async def get_linsight_tools(cls) -> list[GptsToolsTypeRead]:
return [
GptsToolsTypeRead(
id=100000,
name="知识库和文件内容检索",
description="检索组织知识库、个人知识库以及本地上传文件的内容",
children=[
GptsTools(
id=100001,
name="知识库和文件内容检索",
desc="检索组织知识库、个人知识库以及本地上传文件的内容。",
tool_key="search_knowledge_base",
)
]
),
GptsToolsTypeRead(
id=200000,
name="文件操作",
description="本地文件系统的浏览、搜索与编辑工具集",
children=[
GptsTools(
id=200001,
name="获取所有文件和目录",
desc="列出指定目录下的所有文件和子目录。",
tool_key="list_files"
),
GptsTools(
id=200002,
name="获取文件详细信息",
desc="获取指定文件的文件名、文件大小、文件地址、字数、行数等详细信息。",
tool_key="get_file_details"
),
GptsTools(
id=200003,
name="搜索文件",
desc="在指定目录中搜索文件和子目录。",
tool_key="search_files"
),
GptsTools(
id=200004,
name="读取文件内容",
desc="读取本地文本文件的内容。",
tool_key="read_text_file"
),
GptsTools(
id=200005,
name="写入文件内容",
desc="将文本内容追加到文本文件，如果文件不存在，则创建文件",
tool_key="add_text_to_file"
),
GptsTools(
id=200006,
name="替换文件指定行范围内容",
desc="替换文件中的指定行范围。",
tool_key="replace_file_lines"
),
]
)
]
================================================
FILE: src/backend/bisheng/user/__init__.py
================================================
================================================
FILE: src/backend/bisheng/user/api/__init__.py
================================================
================================================
FILE: src/backend/bisheng/user/api/role.py
================================================
================================================
FILE: src/backend/bisheng/user/api/user.py
================================================
import hashlib
import random
from base64 import b64encode
from datetime import datetime
from io import BytesIO
from typing import Annotated, Dict, List, Optional
import rsa
from captcha.image import ImageCaptcha
from fastapi import APIRouter, Depends, HTTPException, Query, Body, Request
from fastapi.security import OAuth2PasswordBearer
from loguru import logger
from sqlmodel import select
from bisheng.api.services.audit_log import AuditLogService
from bisheng.api.v1.schemas import resp_200, CreateUserReq
from bisheng.common.errcode.http_error import UnAuthorizedError, NotFoundError
from bisheng.common.errcode.user import (UserNotPasswordError, UserValidateError, UserPasswordError, UserForbiddenError)
from bisheng.common.services.config_service import settings
from bisheng.core.cache.redis_manager import get_redis_client, get_redis_client_sync
from bisheng.core.database import get_sync_db_session
from bisheng.database.constants import AdminRole, DefaultRole
from bisheng.database.models.group import GroupDao
from bisheng.database.models.mark_task import MarkTaskDao
from bisheng.database.models.role import Role, RoleCreate, RoleDao, RoleUpdate
from bisheng.database.models.role_access import RoleRefresh, RoleAccessDao, AccessType
from bisheng.database.models.user_group import UserGroupDao
from bisheng.utils import generate_uuid
from bisheng.utils import get_request_ip
from bisheng.utils.constants import CAPTCHA_PREFIX, RSA_KEY, USER_PASSWORD_ERROR, USER_CURRENT_SESSION
from ..domain.models.user import User, UserCreate, UserDao, UserLogin, UserRead, UserUpdate
from ..domain.models.user_role import UserRole, UserRoleCreate, UserRoleDao
from ..domain.services.auth import AuthJwt, LoginUser
from ..domain.services.user import UserService
from ...common.constants.enums.telemetry import BaseTelemetryTypeEnum
from ...common.schemas.telemetry.event_data_schema import UserLoginEventData
from ...common.services import telemetry_service
from ...core.logger import trace_id_var
# build router
router = APIRouter(prefix='', tags=['User'])
oauth2_scheme = OAuth2PasswordBearer(tokenUrl='token')
@router.post('/user/regist')
async def regist(*, user: UserCreate):
# Captcha Verification
db_user = await UserService.user_register(user)
return resp_200(db_user)
@router.post('/user/sso')
async def sso(*, request: Request, user: UserCreate, auth_jwt: AuthJwt = Depends()):
""" Login interface for closed source gateways """
if settings.get_system_login_method().bisheng_pro: # Judgingsso Open or not
account_name = user.user_name
user_exist = UserDao.get_unique_user_by_name(account_name)
if not user_exist:
# Determine if there is a user under the platform
user_all = UserDao.get_all_users(page=1, limit=1)
# Automatically create users
user_exist = User.model_validate(user)
logger.info('act=create_user account={}', account_name)
default_admin = settings.get_system_login_method().admin_username
# Insert as Super Admin if there is no user on the platform or if the username matches the configured admin username
if len(user_all) == 0 or (default_admin and default_admin == account_name):
# Create as Super Admin
user_exist = await UserDao.add_user_and_admin_role(user_exist)
else:
# Create as Normal User
user_exist = await UserDao.add_user_and_default_role(user_exist)
await UserGroupDao.add_default_user_group(user_exist.user_id)
if 1 == user_exist.delete:
raise UserForbiddenError.http_exception()
access_token = LoginUser.create_access_token(user_exist, auth_jwt=auth_jwt)
# Set the logged in user's currentcookie, .jwtValid for an additional hour
redis_client = await get_redis_client()
await redis_client.aset(USER_CURRENT_SESSION.format(user_exist.user_id), access_token,
settings.cookie_conf.jwt_token_expire_time + 3600)
# Log Audit Logs
login_user = await LoginUser.init_login_user(user_id=user_exist.user_id, user_name=user_exist.user_name)
AuditLogService.user_login(login_user, get_request_ip(request))
# RecordTelemetryJournal
await telemetry_service.log_event(user_id=login_user.user_id, event_type=BaseTelemetryTypeEnum.USER_LOGIN,
trace_id=trace_id_var.get(),
event_data=UserLoginEventData(method="oss"))
return resp_200({'access_token': access_token, 'refresh_token': access_token})
else:
raise ValueError('Interface not supported')
def get_error_password_key(username: str):
return USER_PASSWORD_ERROR.format(username)
def clear_error_password_key(username: str):
# Count of cleanup password errors
error_key = get_error_password_key(username)
get_redis_client_sync().delete(error_key)
@router.post('/user/login')
async def login(*, request: Request, user: UserLogin, auth_jwt: AuthJwt = Depends()):
return await UserService.user_login(request, user=user, auth_jwt=auth_jwt)
@router.get('/user/admin')
async def get_admins(login_user: LoginUser = Depends(LoginUser.get_login_user)):
"""
Get all Super Admin accounts
"""
# check if user already exist
if not login_user.is_admin():
raise HTTPException(status_code=500, detail="Quit that! You don't have rights to view this.")
try:
# Get all Super Admin accounts
admins = UserRoleDao.get_admins_user()
admins_ids = [admin.user_id for admin in admins]
admin_users = UserDao.get_user_by_ids(admins_ids)
res = [UserRead(**one.__dict__) for one in admin_users]
return resp_200(res)
except Exception:
raise HTTPException(status_code=500, detail='User information failed')
@router.get('/user/info')
async def get_info(login_user: LoginUser = Depends(LoginUser.get_login_user)):
user_id = login_user.user_id
db_user = await UserDao.aget_user(user_id)
if not db_user:
raise NotFoundError()
role, web_menu = await login_user.get_roles_web_menu(db_user)
admin_group = await UserGroupDao.aget_user_admin_group(user_id)
admin_group = [one.group_id for one in admin_group]
return resp_200(UserRead(role=str(role), web_menu=web_menu, admin_groups=admin_group, **db_user.__dict__))
@router.post('/user/logout', status_code=201)
async def logout(auth_jwt: AuthJwt = Depends()):
auth_jwt.unset_access_token()
return resp_200()
@router.get('/user/list', status_code=201)
async def list_user(*,
name: Optional[str] = None,
page_size: Optional[int] = 10,
page_num: Optional[int] = 1,
group_id: Annotated[List[int], Query()] = None,
role_id: Annotated[List[int], Query()] = None,
login_user: LoginUser = Depends(LoginUser.get_login_user)):
groups = group_id
roles = role_id
user_admin_groups = []
if not login_user.is_admin():
# Query if you are an administrator of another user group under
user_admin_groups = UserGroupDao.get_user_admin_group(login_user.user_id)
user_admin_groups = [one.group_id for one in user_admin_groups]
groups = user_admin_groups
# Not an administrator of any user group does not have permission to view
if not groups:
raise HTTPException(status_code=500, detail="Quit that! You don't have rights to view this.")
# Filter bygroup_idand administrator permissionsgroupsDoing Intersections
if group_id:
groups = list(set(groups) & set(group_id))
if not groups:
raise HTTPException(status_code=500, detail="Quit that! You don't have rights to view this.")
# Query roles under user groups, Intersect with the role filter to get the role that really needs to be queriedID
group_roles = RoleDao.get_role_by_groups(groups, None, 0, 0)
if role_id:
roles = list(set(role_id) & set([one.id for one in group_roles]))
# Users filtered by user groups and rolesid
user_ids = []
if groups:
# Query users under user groupsID
groups_user_ids = UserGroupDao.get_groups_user(groups)
if not groups_user_ids:
return resp_200({'data': [], 'total': 0})
user_ids = list(set([one.user_id for one in groups_user_ids]))
if roles:
roles_user_ids = UserRoleDao.get_roles_user(roles)
if not roles_user_ids:
return resp_200({'data': [], 'total': 0})
roles_user_ids = [one.user_id for one in roles_user_ids]
# Automatically close purchase order afteruser_idsis not empty, the description isgroupsDo intersection screening together, otherwise only do role screening
if user_ids:
user_ids = list(set(user_ids) & set(roles_user_ids))
if not user_ids:
return resp_200({'data': [], 'total': 0})
else:
user_ids = list(set(roles_user_ids))
users, total_count = UserDao.filter_users(user_ids, name, page_num, page_size)
res = []
role_dict = {}
group_dict = {}
for one in users:
one_data = one.model_dump()
user_roles = get_user_roles(one, role_dict)
user_groups = get_user_groups(one, group_dict)
# If not hyper-managed, data needs to be filtered, Cannot see the list of roles and user groups within a user group not managed by him
if user_admin_groups:
for i in range(len(user_roles) - 1, -1, -1):
if user_roles[i]["group_id"] not in user_admin_groups:
del user_roles[i]
for i in range(len(user_groups) - 1, -1, -1):
if user_groups[i]["id"] not in user_admin_groups:
del user_groups[i]
one_data["roles"] = user_roles
one_data["groups"] = user_groups
res.append(one_data)
return resp_200({'data': res, 'total': total_count})
def get_user_roles(user: User, role_cache: Dict) -> List[Dict]:
# Query a list of roles for a user
user_roles = UserRoleDao.get_user_roles(user.user_id)
user_role_ids: List[int] = [one_role.role_id for one_role in user_roles]
res = []
for i in range(len(user_role_ids) - 1, -1, -1):
if role_cache.get(user_role_ids[i]):
res.append(role_cache.get(user_role_ids[i]))
del user_role_ids[i]
# Query database for role information without caching
if user_role_ids:
role_list = RoleDao.get_role_by_ids(user_role_ids)
for role_info in role_list:
role_cache[role_info.id] = {
"id": role_info.id,
"group_id": role_info.group_id,
"name": role_info.role_name
}
res.append(role_cache.get(role_info.id))
return res
def get_user_groups(user: User, group_cache: Dict) -> List[Dict]:
# Query a list of roles for a user
user_groups = UserGroupDao.get_user_group(user.user_id)
user_group_ids: List[int] = [one_group.group_id for one_group in user_groups]
res = []
for i in range(len(user_group_ids) - 1, -1, -1):
if group_cache.get(user_group_ids[i]):
res.append(group_cache.get(user_group_ids[i]))
del user_group_ids[i]
# Query database for role information without caching
if user_group_ids:
group_list = GroupDao.get_group_by_ids(user_group_ids)
for group_info in group_list:
group_cache[group_info.id] = {'id': group_info.id, 'name': group_info.group_name}
res.append(group_cache.get(group_info.id))
return res
@router.post('/user/update', status_code=201)
async def update(*,
request: Request,
user: UserUpdate,
login_user: LoginUser = Depends(LoginUser.get_login_user)):
db_user = UserDao.get_user(user.user_id)
if not db_user:
raise HTTPException(status_code=500, detail='Pengguna tidak ada')
if not login_user.is_admin():
# Check if is an administrator of a user group under
user_group = UserGroupDao.get_user_group(db_user.user_id)
user_group = [one.group_id for one in user_group]
if not login_user.check_groups_admin(user_group):
raise HTTPException(status_code=500, detail="Quit that! You don't have rights to view this.")
# check if user already exist
if db_user and user.delete is not None:
# Determine if it's an admin
with get_sync_db_session() as session:
admin = session.exec(
select(UserRole).where(UserRole.role_id == 1,
UserRole.user_id == user.user_id)).first()
if admin:
raise HTTPException(status_code=500, detail='Cannot operate admin')
if user.delete == db_user.delete:
return resp_200()
db_user.delete = user.delete
if db_user.delete == 0: # Enable User
# Count of cleanup password errors
clear_error_password_key(db_user.user_name)
with get_sync_db_session() as session:
session.add(db_user)
session.commit()
session.refresh(db_user)
update_user_delete_hook(request, login_user, db_user)
return resp_200()
def update_user_delete_hook(request: Request, login_user: LoginUser, user: User) -> bool:
logger.info(f'update_user_delete_hook: {request}, user={user}')
if user.delete == 0: # Enable User
AuditLogService.recover_user(login_user, get_request_ip(request), user)
elif user.delete == 1: # Disabled User
AuditLogService.forbid_user(login_user, get_request_ip(request), user)
return True
@router.post('/role/add', status_code=201)
async def create_role(*,
request: Request,
role: RoleCreate,
login_user: LoginUser = Depends(LoginUser.get_login_user)):
if not role.group_id:
raise HTTPException(status_code=500, detail='User GroupsIDTidak boleh kosong.')
if not role.role_name:
raise HTTPException(status_code=500, detail='msg.role_name_not_be_empty')
if not login_user.check_group_admin(role.group_id):
return UnAuthorizedError.return_resp()
db_role = Role.model_validate(role)
try:
with get_sync_db_session() as session:
session.add(db_role)
session.commit()
session.refresh(db_role)
create_role_hook(request, login_user, db_role)
return resp_200(db_role)
except Exception:
logger.exception('add role error')
raise HTTPException(status_code=500, detail='Failed to add, check if it is added repeatedly')
def create_role_hook(request: Request, login_user: LoginUser, db_role: Role) -> bool:
logger.info(f'create_role_hook: {login_user.user_name}, role={db_role}')
AuditLogService.create_role(login_user, get_request_ip(request), db_role)
@router.patch('/role/{role_id}', status_code=201)
async def update_role(*,
request: Request,
role_id: int,
role: RoleUpdate,
login_user: LoginUser = Depends(LoginUser.get_login_user)):
db_role = RoleDao.get_role_by_id(role_id)
if not db_role:
raise HTTPException(status_code=404, detail='This character does not exist')
if not login_user.check_group_admin(db_role.group_id):
return UnAuthorizedError.return_resp()
try:
if role.role_name:
db_role.role_name = role.role_name
if role.remark:
db_role.remark = role.remark
with get_sync_db_session() as session:
session.add(db_role)
session.commit()
session.refresh(db_role)
update_role_hook(request, login_user, db_role)
return resp_200(db_role)
except Exception:
logger.exception(f'update_role')
raise HTTPException(status_code=500, detail='Update failed, server side exception')
def update_role_hook(request: Request, login_user: LoginUser, db_role: Role) -> bool:
logger.info(f'update_role_hook: {login_user.user_name}, role={db_role}')
AuditLogService.update_role(login_user, get_request_ip(request), db_role)
@router.get('/role/list', status_code=200)
async def get_role(*,
role_name: str = None,
page: int = 0,
limit: int = 0,
login_user: LoginUser = Depends(LoginUser.get_login_user)):
"""
Get a list of roles visible to the user, Return different data according to different user permissions
"""
# Parameter Processing
if role_name:
role_name = role_name.strip()
# Determine if it's a Super Admin
if login_user.is_admin():
# Is Super Admin Get All
group_ids = []
else:
# Query if you are an administrator of another user group under
user_groups = UserGroupDao.get_user_admin_group(login_user.user_id)
group_ids = [one.group_id for one in user_groups if one.is_group_admin]
if not group_ids:
raise HTTPException(status_code=500, detail="Quit that! You don't have rights to view this.")
# Query a list of all roles
res = RoleDao.get_role_by_groups(group_ids, role_name, page, limit)
total = RoleDao.count_role_by_groups(group_ids, role_name)
return resp_200(data={"data": res, "total": total})
@router.delete('/role/{role_id}', status_code=200)
async def delete_role(*,
request: Request,
role_id: int, login_user: LoginUser = Depends(LoginUser.get_login_user)):
db_role = RoleDao.get_role_by_id(role_id)
if not db_role:
return resp_200()
if not login_user.check_group_admin(db_role.group_id):
return UnAuthorizedError.return_resp()
if db_role.id == AdminRole or db_role.id == DefaultRole:
raise HTTPException(status_code=500, detail='Built-in roles cannot be deleted')
# DeleteroleRelated data
try:
RoleDao.delete_role(role_id)
except Exception as e:
logger.exception(e)
raise HTTPException(status_code=500, detail='Failed to delete role')
AuditLogService.delete_role(login_user, get_request_ip(request), db_role)
return resp_200()
@router.post('/user/role_add', status_code=200)
async def user_addrole(*,
request: Request,
user_role: UserRoleCreate,
login_user: LoginUser = Depends(LoginUser.get_login_user)):
"""
Resets the role of the user. The scope of the data varies depending on the permissions
"""
# Get a list of the user's previous roles
old_roles = UserRoleDao.get_user_roles(user_role.user_id)
old_roles = [one.role_id for one in old_roles]
# Determine if the role being edited is Super Admin, Super Admin does not allow editing
user_role_list = UserRoleDao.get_user_roles(user_role.user_id)
if any(one.role_id == AdminRole for one in user_role_list):
raise HTTPException(status_code=500, detail='Editing is not allowed by the system administrator')
if any(one == AdminRole for one in user_role.role_id):
raise HTTPException(status_code=500, detail='Setting as system administrator is not allowed')
if not login_user.is_admin():
# Determine which user groups you have administrative access to
admin_group = UserGroupDao.get_user_admin_group(login_user.user_id)
admin_group = [one.group_id for one in admin_group]
if not admin_group:
raise HTTPException(status_code=500, detail='No rights')
# Get a list of all roles under an admin group
admin_roles = RoleDao.get_role_by_groups(admin_group, '', 0, 0)
admin_roles = [one.id for one in admin_roles]
# Do the intersection to get the list of roles visible to the user group administrator
for i in range(len(old_roles) - 1, -1, -1):
if old_roles[i] not in admin_roles:
del old_roles[i]
# Determine if the reset role list is in Under the name of the user group administrator
for i in range(len(user_role.role_id) - 1, -1, -1):
if user_role.role_id[i] not in admin_roles:
raise HTTPException(status_code=500, detail=f'No permission to add roles{user_role.role_id[i]}')
need_add_role = []
need_delete_role = old_roles.copy()
for one in user_role.role_id:
if one not in old_roles:
# Role needs to be added
need_add_role.append(one)
else:
# All that remains is the list of roles that need to be deleted
need_delete_role.remove(one)
if need_add_role:
UserRoleDao.add_user_roles(user_role.user_id, need_add_role)
if need_delete_role:
# Delete the corresponding role list
UserRoleDao.delete_user_roles(user_role.user_id, need_delete_role)
update_user_role_hook(request, login_user, user_role.user_id, old_roles, user_role.role_id)
return resp_200()
def update_user_role_hook(request: Request, login_user: LoginUser, user_id: int,
old_roles: List[int], new_roles: List[int]):
logger.info(f'update_user_role_hook, user_id: {user_id}, old_roles: {old_roles}, new_roles: {new_roles}')
# Write Audit Log
role_info = RoleDao.get_role_by_ids(old_roles + new_roles)
group_ids = list(set([role.group_id for role in role_info]))
role_dict = {one.id: one.role_name for one in role_info}
note = "Pre-edit role:"
for one in old_roles:
note += role_dict[one] + "、"
note = note.rstrip("、")
note += "Post-edited roles:"
for one in new_roles:
note += role_dict[one] + "、"
note = note.rstrip("、")
AuditLogService.update_user(login_user, get_request_ip(request), user_id, group_ids, note)
@router.post('/role_access/refresh', status_code=200)
async def access_refresh(*, request: Request, data: RoleRefresh,
login_user: LoginUser = Depends(LoginUser.get_login_user)):
db_role = await RoleDao.aget_role_by_id(data.role_id)
if not db_role:
raise NotFoundError().http_exception()
if db_role.id == AdminRole:
raise UnAuthorizedError.http_exception()
if not await login_user.async_check_group_admin(db_role.group_id):
raise UnAuthorizedError.http_exception()
role_id = data.role_id
access_type = data.type
access_id = data.access_id
await RoleAccessDao.update_role_access_all(role_id, AccessType(access_type), access_id)
update_role_hook(request, login_user, db_role)
return resp_200()
@router.get('/role_access/list', status_code=200)
async def access_list(*, role_id: int, access_type: Optional[int] = Query(default=None, alias="type"),
login_user: LoginUser = Depends(LoginUser.get_login_user)):
db_role = await RoleDao.aget_role_by_id(role_id)
if not db_role:
raise NotFoundError().http_exception()
if not await login_user.async_check_group_admin(db_role.group_id):
return UnAuthorizedError.return_resp()
access_type = None
if access_type:
access_type = AccessType(access_type)
res = await RoleAccessDao.aget_role_access([role_id], access_type)
return resp_200({
'data': res,
'total': len(res)
})
@router.get('/user/get_captcha', status_code=200)
async def get_captcha():
# generate captcha
chr_all = "abcdefghjkmnpqrstuvwxyABCDEFGHJKMNPQRSTUVWXY3456789"
chr_4 = ''.join(random.sample(chr_all, 4))
image = ImageCaptcha().generate_image(chr_4)
# Right.image To be performedbase 64 Code
buffered = BytesIO()
image.save(buffered, format='PNG')
capthca_b64 = b64encode(buffered.getvalue()).decode()
logger.info('get_captcha captcha_char={}', chr_4)
# generate key, Generate Simple Uniqueid，
key = CAPTCHA_PREFIX + generate_uuid()[:8]
redis_client = await get_redis_client()
await redis_client.aset(key, chr_4, expiration=300)
# Add configuration, whether the verification code must be used
return resp_200({
'captcha_key': key,
'captcha': capthca_b64,
'user_capthca': settings.get_from_db('use_captcha') or False
})
@router.get('/user/public_key', status_code=200)
async def get_rsa_publish_key():
# redis Storage
key = RSA_KEY
redis_client = await get_redis_client()
# redis lock
if await redis_client.asetNx(key, 1):
# Generate a key pair
(pubkey, privkey) = rsa.newkeys(512)
# Save the keys to strings
await redis_client.aset(key, (pubkey, privkey), 3600)
else:
pubkey, privkey = await redis_client.aget(key)
pubkey_str = pubkey.save_pkcs1().decode()
return resp_200({'public_key': pubkey_str})
@router.post('/user/reset_password', status_code=200)
async def reset_password(
*,
user_id: int = Body(embed=True),
password: str = Body(embed=True),
login_user: LoginUser = Depends(LoginUser.get_login_user),
):
"""
Admin Reset User Password
"""
# Get user information to change password
user_info = UserDao.get_user(user_id)
if not user_info:
raise HTTPException(status_code=404, detail='Pengguna tidak ada')
user_payload = LoginUser(**{
'user_id': user_info.user_id,
'user_name': user_info.user_name,
'role': ''
})
# If the user being modified is a system administrator, Need to determine if it's me
if user_payload.is_admin() and login_user.user_id != user_id:
raise HTTPException(status_code=500, detail='System administrators can only reset passwords themselves')
# Query the user group the user belongs to
user_groups = UserGroupDao.get_user_group(user_info.user_id)
user_group_ids = [one.group_id for one in user_groups]
# Check if there are administrative permissions for the group
if not login_user.check_groups_admin(user_group_ids):
raise HTTPException(status_code=403, detail='No permission to reset password')
user_info.password = UserService.decrypt_md5_password(password)
user_info.password_update_time = datetime.now()
UserDao.update_user(user_info)
clear_error_password_key(user_info.user_name)
return resp_200()
@router.post('/user/change_password', status_code=200)
async def change_password(*,
password: str = Body(embed=True),
new_password: str = Body(embed=True),
login_user: LoginUser = Depends(LoginUser.get_login_user)):
"""
Login user Change my password
"""
user_info = UserDao.get_user(login_user.user_id)
if not user_info.password:
return UserNotPasswordError.return_resp()
password = UserService.decrypt_md5_password(password)
# Logged in user told it was the wrong password
if user_info.password != password:
return UserPasswordError.return_resp()
user_info.password = UserService.decrypt_md5_password(new_password)
user_info.password_update_time = datetime.now()
UserDao.update_user(user_info)
clear_error_password_key(user_info.user_name)
return resp_200()
@router.post('/user/change_password_public', status_code=200)
async def change_password_public(*,
username: str = Body(embed=True),
password: str = Body(embed=True),
new_password: str = Body(embed=True)):
"""
Not Logged-In Users Change my password
"""
user_info = UserDao.get_user_by_username(username)
if not user_info.password:
return UserValidateError.return_resp()
if user_info.password != UserService.decrypt_md5_password(password):
return UserValidateError.return_resp()
user_info.password = UserService.decrypt_md5_password(new_password)
user_info.password_update_time = datetime.now()
UserDao.update_user(user_info)
clear_error_password_key(username)
return resp_200()
@router.get('/user/mark', status_code=200)
async def has_mark_access(*, request: Request, login_user: LoginUser = Depends(LoginUser.get_login_user)):
"""
Get whether the current user has annotation permission,Determine if the current user isadmin Or a user group administrator
"""
user_groups = UserGroupDao.get_user_group(login_user.user_id)
user_group_ids = [one.group_id for one in user_groups]
has_mark_access = False
# Check if there are administrative permissions for the group
task = MarkTaskDao.get_task(login_user.user_id)
if task:
has_mark_access = True
return resp_200(data=has_mark_access)
@router.post('/user/create', status_code=200)
async def create_user(*,
request: Request,
admin_user: LoginUser = Depends(LoginUser.get_admin_user),
req: CreateUserReq):
"""
Super Admin Create User
"""
logger.info(f'create_user username={admin_user.user_name}, username={req.user_name}')
data = UserService.create_user(request, admin_user, req)
return resp_200(data=data)
def md5_hash(string):
md5 = hashlib.md5()
md5.update(string.encode('utf-8'))
return md5.hexdigest()
================================================
FILE: src/backend/bisheng/user/domain/__init__.py
================================================
================================================
FILE: src/backend/bisheng/user/domain/const.py
================================================
# Stores the number of times a user's password is incorrect.keyareusername
USER_PASSWORD_ERROR = 'user_password_error:{}'
# Store current user logged incookie, keyIs Userid
USER_CURRENT_SESSION = 'user_current_session:{}'
================================================
FILE: src/backend/bisheng/user/domain/models/__init__.py
================================================
================================================
FILE: src/backend/bisheng/user/domain/models/user.py
================================================
from datetime import datetime
from typing import List, Optional
from pydantic import field_validator
from sqlalchemy import Column, DateTime, func, text
from sqlalchemy.orm import selectinload
from sqlmodel import Field, select, Relationship, col
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session, get_async_db_session
from bisheng.database.constants import AdminRole, DefaultRole
from bisheng.database.models.group import Group
from bisheng.database.models.role import Role
from bisheng.database.models.user_group import UserGroup
from bisheng.user.domain.models.user_role import UserRole
class UserBase(SQLModelSerializable):
user_name: str = Field(index=True, unique=True)
email: Optional[str] = Field(default=None, index=True)
phone_number: Optional[str] = Field(default=None, index=True)
dept_id: Optional[str] = Field(default=None, index=True)
remark: Optional[str] = Field(default=None, index=False)
delete: int = Field(default=0, index=False)
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, index=True, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
@field_validator('user_name')
@classmethod
def validate_str(cls, v):
# dict_keys(['description', 'name', 'id', 'data'])
if not v:
raise ValueError('user_name Tidak boleh kosong.')
return v
class User(UserBase, table=True):
user_id: Optional[int] = Field(default=None, primary_key=True)
password: str = Field(index=False)
password_update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')), description='Password Last Modified')
# DefinitiongroupsAndrolesQuery Relationships for
groups: List["Group"] = Relationship(link_model=UserGroup)
roles: List["Role"] = Relationship(link_model=UserRole)
__tablename__ = "user"
class UserRead(UserBase):
user_id: Optional[int] = None
role: Optional[str] = None # admin / group_admin
access_token: Optional[str] = None
web_menu: Optional[List[str]] = None
admin_groups: Optional[List[int]] = None # Managed User GroupsIDVertical
class UserQuery(UserBase):
pass
class UserLogin(UserBase):
password: str
captcha_key: Optional[str] = None
captcha: Optional[str] = None
class UserCreate(UserBase):
password: Optional[str] = Field(default='')
captcha_key: Optional[str] = None
captcha: Optional[str] = None
class UserUpdate(SQLModelSerializable):
user_id: int
delete: Optional[int] = 0
class UserDao(UserBase):
@classmethod
def get_user(cls, user_id: int) -> User | None:
with get_sync_db_session() as session:
statement = select(User).where(User.user_id == user_id)
return session.exec(statement).first()
@classmethod
async def aget_user(cls, user_id: int) -> User | None:
async with get_async_db_session() as session:
statement = select(User).where(User.user_id == user_id)
result = await session.exec(statement)
return result.first()
@classmethod
def get_user_by_ids(cls, user_ids: List[int]) -> List[User] | None:
with get_sync_db_session() as session:
statement = select(User).where(User.user_id.in_(user_ids))
return session.exec(statement).all()
@classmethod
async def aget_user_by_ids(cls, user_ids: List[int]) -> List[User] | None:
async with get_async_db_session() as session:
statement = select(User).where(User.user_id.in_(user_ids))
result = await session.exec(statement)
return result.all()
@classmethod
def get_user_by_username(cls, username: str) -> User | None:
with get_sync_db_session() as session:
statement = select(User).where(User.user_name == username)
return session.exec(statement).first()
@classmethod
async def aget_user_by_username(cls, username: str) -> User | None:
async with get_async_db_session() as session:
statement = select(User).where(User.user_name == username)
result = await session.exec(statement)
return result.first()
@classmethod
def update_user(cls, user: User) -> User:
with get_sync_db_session() as session:
session.add(user)
session.commit()
session.refresh(user)
return user
@classmethod
async def aupdate_user(cls, user: User) -> User:
async with get_async_db_session() as session:
session.add(user)
await session.commit()
await session.refresh(user)
return user
@classmethod
def _filter_users_statement(cls,
statement,
user_ids: List[int],
keyword: str = None):
if user_ids:
statement = statement.where(User.user_id.in_(user_ids))
if keyword:
statement = statement.where(User.user_name.like(f'%{keyword}%'))
return statement.order_by(User.user_id.desc())
@classmethod
def filter_users(cls,
user_ids: List[int],
keyword: str = None,
page: int = 0,
limit: int = 0) -> (List[User], int):
statement = select(User)
statement = cls._filter_users_statement(statement, user_ids, keyword)
count_statement = select(func.count(User.user_id))
count_statement = cls._filter_users_statement(count_statement, user_ids, keyword)
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
statement = statement.order_by(User.user_id.desc())
with get_sync_db_session() as session:
return session.exec(statement).all(), session.scalar(count_statement)
@classmethod
async def afilter_users(cls,
user_ids: List[int],
keyword: str = None,
page: int = 0,
limit: int = 0) -> List[User]:
statement = select(User)
statement = cls._filter_users_statement(statement, user_ids, keyword)
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
statement = statement.order_by(User.user_id.desc())
async with get_async_db_session() as session:
result = await session.exec(statement)
return result.all()
@classmethod
def get_unique_user_by_name(cls, user_name: str) -> User | None:
with get_sync_db_session() as session:
statement = select(User).where(User.user_name == user_name)
return session.exec(statement).first()
@classmethod
def search_user_by_name(cls, user_name: str) -> List[User] | None:
with get_sync_db_session() as session:
statement = select(User).where(User.user_name.like('%{}%'.format(user_name)))
return session.exec(statement).all()
@classmethod
def create_user(cls, db_user: User) -> User:
with get_sync_db_session() as session:
session.add(db_user)
session.commit()
session.refresh(db_user)
return db_user
@classmethod
async def add_user_and_default_role(cls, user: User) -> User:
"""
Add users and add default roles
"""
async with get_async_db_session() as session:
session.add(user)
await session.commit()
await session.refresh(user)
db_user_role = UserRole(user_id=user.user_id, role_id=DefaultRole)
session.add(db_user_role)
await session.commit()
await session.refresh(user)
return user
@classmethod
async def add_user_and_admin_role(cls, user: User) -> User:
"""
Add users and add super admin roles
"""
async with get_async_db_session() as session:
session.add(user)
await session.commit()
await session.refresh(user)
db_user_role = UserRole(user_id=user.user_id, role_id=AdminRole)
session.add(db_user_role)
await session.commit()
await session.refresh(user)
return user
@classmethod
def add_user_with_groups_and_roles(cls, user: User, group_ids: List[int],
role_ids: List[int]) -> User:
with get_sync_db_session() as session:
session.add(user)
session.flush()
for group_id in group_ids:
db_user_group = UserGroup(user_id=user.user_id, group_id=group_id)
session.add(db_user_group)
for role_id in role_ids:
db_user_role = UserRole(user_id=user.user_id, role_id=role_id)
session.add(db_user_role)
session.commit()
session.refresh(user)
return user
@classmethod
def get_all_users(cls, page: int = 0, limit: int = 0) -> List[User]:
"""
Pagination Get All Users
"""
statement = select(User)
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
def get_user_with_group_role(cls, *, start_time: datetime = None, end_time: datetime = None,
user_ids: List[int] = None, page: int = 0, page_size: int = 0) -> List[User]:
statement = select(User)
if start_time and end_time:
statement = statement.where(User.create_time >= start_time, User.create_time < end_time)
if page and page_size:
statement = statement.offset((page - 1) * page_size).limit(page_size)
if user_ids:
statement = statement.where(col(User.user_id).in_(user_ids))
statement = statement.order_by(User.user_id)
statement = statement.options(
selectinload(User.groups), # type: ignore
selectinload(User.roles) # type: ignore
)
with get_sync_db_session() as session:
return session.exec(statement).all()
@classmethod
def get_first_user(cls) -> User | None:
statement = select(User).order_by(col(User.user_id).asc()).limit(1)
with get_sync_db_session() as session:
return session.exec(statement).first()
================================================
FILE: src/backend/bisheng/user/domain/models/user_role.py
================================================
from datetime import datetime
from typing import List, Optional
from pydantic import BaseModel
from sqlalchemy import Column, DateTime, text, delete, INT
from sqlmodel import Field, select
from bisheng.common.models.base import SQLModelSerializable
from bisheng.core.database import get_sync_db_session, get_async_db_session
from bisheng.database.constants import AdminRole
class UserRoleBase(SQLModelSerializable):
user_id: Optional[int] = Field(
default=None,
foreign_key="user.user_id",
primary_key=True,
ondelete="CASCADE"
)
role_id: int = Field(
default=None,
foreign_key="role.id",
primary_key=True,
ondelete="CASCADE"
)
create_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP')))
update_time: Optional[datetime] = Field(default=None, sa_column=Column(
DateTime, nullable=False, server_default=text('CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP')))
class UserRole(UserRoleBase, table=True):
id: Optional[int] = Field(default=None, sa_column=Column(INT, primary_key=True, autoincrement=True))
class UserRoleRead(UserRoleBase):
id: Optional[int] = None
class UserRoleCreate(BaseModel):
user_id: int
role_id: list[int]
class UserRoleDao(UserRoleBase):
@classmethod
def get_user_roles(cls, user_id: int) -> List[UserRole]:
with get_sync_db_session() as session:
return session.exec(select(UserRole).where(UserRole.user_id == user_id)).all()
@classmethod
async def aget_user_roles(cls, user_id: int) -> List[UserRole]:
async with get_async_db_session() as session:
result = await session.exec(select(UserRole).where(UserRole.user_id == user_id))
return result.all()
@classmethod
def get_roles_user(cls, role_ids: List[int], page: int = 0, limit: int = 0) -> List[UserRole]:
"""
Get the user for the role
"""
with get_sync_db_session() as session:
statement = select(UserRole).where(UserRole.role_id.in_(role_ids))
if page and limit:
statement = statement.offset((page - 1) * limit).limit(limit)
return session.exec(statement).all()
@classmethod
def get_admins_user(cls) -> List[UserRole]:
"""
Get all super-managed accounts
"""
with get_sync_db_session() as session:
statement = select(UserRole).where(UserRole.role_id == AdminRole)
return session.exec(statement).all()
@classmethod
async def set_admin_user(cls, user_id: int) -> UserRole:
"""
Set user as Super Admin
"""
async with get_async_db_session() as session:
user_role = UserRole(user_id=user_id, role_id=AdminRole)
session.add(user_role)
await session.commit()
await session.refresh(user_role)
return user_role
@classmethod
def add_user_roles(cls, user_id: int, role_ids: List[int]) -> List[UserRole]:
"""
Add roles to users in bulk
"""
with get_sync_db_session() as session:
user_roles = [UserRole(user_id=user_id, role_id=role_id) for role_id in role_ids]
session.add_all(user_roles)
session.commit()
return user_roles
@classmethod
def delete_user_roles(cls, user_id: int, role_ids: List[int]) -> None:
"""
Remove users from certain roles
"""
with get_sync_db_session() as session:
statement = delete(UserRole).where(UserRole.user_id == user_id).where(UserRole.role_id.in_(role_ids))
session.exec(statement)
session.commit()
================================================
FILE: src/backend/bisheng/user/domain/repositories/__init__.py
================================================
================================================
FILE: src/backend/bisheng/user/domain/repositories/implementations/__init__.py
================================================
================================================
FILE: src/backend/bisheng/user/domain/repositories/implementations/user_repository_impl.py
================================================
from typing import Union
from sqlalchemy.orm import selectinload
from sqlmodel import select, Session
from sqlmodel.ext.asyncio.session import AsyncSession
from bisheng.common.repositories.implementations.base_repository_impl import BaseRepositoryImpl
from bisheng.user.domain.models.user import User
from bisheng.user.domain.repositories.interfaces.user_repository import UserRepository
class UserRepositoryImpl(BaseRepositoryImpl[User, int], UserRepository):
"""Shared link repository implementation"""
def __init__(self, session: Union[AsyncSession, Session]):
super().__init__(session, User)
# accordinguser_idget user info user、user_groups、roles
async def get_user_with_groups_and_roles_by_user_id(self, user_id: int) -> User | None:
statement = (
select(User).where(User.user_id == user_id).options(
selectinload(User.groups), # type: ignore
selectinload(User.roles) # type: ignore
)
)
result = await self.session.exec(statement)
return result.first()
def get_user_with_groups_and_roles_by_user_id_sync(self, user_id: int) -> User | None:
statement = (
select(User).where(User.user_id == user_id).options(
selectinload(User.groups), # type: ignore
selectinload(User.roles) # type: ignore
)
)
result = self.session.exec(statement)
return result.first()
================================================
FILE: src/backend/bisheng/user/domain/repositories/interfaces/__init__.py
================================================
================================================
FILE: src/backend/bisheng/user/domain/repositories/interfaces/user_repository.py
================================================
from abc import ABC
from bisheng.common.repositories.interfaces.base_repository import BaseRepository
from bisheng.user.domain.models.user import User, UserQuery
class UserRepository(BaseRepository[User, int], ABC):
"""User Repository Interface Class"""
# accordinguser_idget user info user、user_groups、roles
async def get_user_with_groups_and_roles_by_user_id(self, user_id: int) -> UserQuery:
pass
def get_user_with_groups_and_roles_by_user_id_sync(self, user_id: int) -> UserQuery:
pass
================================================
FILE: src/backend/bisheng/user/domain/services/__init__.py
================================================
================================================
FILE: src/backend/bisheng/user/domain/services/auth.py
================================================
import functools
import json
from datetime import datetime, timezone
from functools import cached_property
from typing import List, Dict, Any, Optional
import jwt
from fastapi import Request, Response, Depends
from pydantic import BaseModel, Field
from starlette.websockets import WebSocket
from typing_extensions import Self
from bisheng.common.errcode.http_error import UnAuthorizedError
from bisheng.common.exceptions.auth import JWTDecodeError
from bisheng.common.services.config_service import settings
from bisheng.database.constants import AdminRole
from bisheng.database.models.group import GroupDao
from bisheng.database.models.role_access import AccessType, RoleAccessDao, WebMenuResource
from bisheng.database.models.user_group import UserGroupDao
from ..models.user import User
from ..models.user_role import UserRoleDao
class AuthJwt:
def __init__(self, req: Request = None, res: Response = None):
self.req = req
self.res = res
self.jwt_secret = settings.jwt_secret
self.cookie_conf = settings.cookie_conf
self._access_cookie_key = "access_token_cookie"
self._encode_algorithm = "HS256"
self._decode_algorithms = [self._encode_algorithm]
def create_access_token(self, subject: dict) -> str:
""" create jwt token """
if isinstance(subject, dict):
subject = json.dumps(subject)
payload = {
'sub': subject,
'exp': int(datetime.now(timezone.utc).timestamp()) + self.cookie_conf.jwt_token_expire_time,
'iss': self.cookie_conf.jwt_iss
}
token = jwt.encode(payload, self.jwt_secret, algorithm=self._encode_algorithm)
return token
def set_access_token(self, token: str, response: Response = None, max_age: int = None) -> None:
""" set jwt token to cookie """
response = response or self.res
response.set_cookie(
self._access_cookie_key,
token,
max_age=max_age or self.cookie_conf.max_age,
path=self.cookie_conf.path,
domain=self.cookie_conf.domain,
secure=self.cookie_conf.secure,
httponly=self.cookie_conf.httponly,
samesite=self.cookie_conf.samesite
)
def unset_access_token(self) -> None:
self.res.delete_cookie(
self._access_cookie_key,
path=self.cookie_conf.path,
domain=self.cookie_conf.domain
)
def get_subject(self,
auth_from: str = "request",
token: Optional[str] = None,
websocket: Optional[WebSocket] = None) -> Dict:
""" decode jwt token """
if auth_from == "request":
if not token:
token = self.req.cookies.get(self._access_cookie_key)
elif auth_from == "websocket":
if websocket:
token = websocket.cookies.get(self._access_cookie_key)
elif auth_from == "headers":
if not token:
token = self.req.headers.get("Authorization").split(" ")[-1]
else:
raise ValueError("unsupported auth_from value")
return self.decode_jwt_token(token)
def decode_jwt_token(self, token: str) -> Dict:
""" decode jwt token """
try:
payload = jwt.decode(token, self.jwt_secret, issuer=self.cookie_conf.jwt_iss,
algorithms=self._decode_algorithms)
return json.loads(payload.get('sub'))
except Exception as e:
raise JWTDecodeError(status_code=422, message=str(e))
class LoginUser(BaseModel):
user_id: int
user_name: str = Field(default="")
user_role: List[int] = Field(default_factory=list, description="Users GroupsIDVertical")
group_cache: Dict[int, Any] = Field(default_factory=dict, description="User Group Cache")
def __init__(self, **kwargs):
super().__init__(**kwargs)
self.user_id = kwargs.get('user_id')
self.user_name = kwargs.get('user_name')
self.user_role = kwargs.get('user_role')
self.group_cache = kwargs.get('group_cache', {})
if not self.user_role:
self.user_role = []
user_role = UserRoleDao.get_user_roles(self.user_id)
self.user_role = [user_role.role_id for user_role in user_role]
@cached_property
def _check_admin(self):
if isinstance(self.user_role, list):
for one in self.user_role:
if one == AdminRole:
return True
return False
def is_admin(self):
return self._check_admin
@staticmethod
def wrapper_access_check(func):
"""
Decorator for permissions check
ifadminThe user does not perform subsequent specific check logic
"""
@functools.wraps(func)
def wrapper(*args, **kwargs):
if args[0].is_admin():
return True
return func(*args, **kwargs)
return wrapper
@staticmethod
def async_wrapper_access_check(func):
"""
Decorator for asynchronous permission checking
ifadminThe user does not perform subsequent specific check logic
"""
@functools.wraps(func)
async def wrapper(*args, **kwargs):
if args[0].is_admin():
return True
return await func(*args, **kwargs)
return wrapper
@wrapper_access_check
def access_check(self, owner_user_id: int, target_id: str, access_type: AccessType) -> bool:
"""
Check if the user has permission to a resource
"""
# Determine if it belongs to my resource
if self.user_id == owner_user_id:
return True
# Judgment Authorization
if RoleAccessDao.judge_role_access(self.user_role, target_id, access_type):
return True
return False
@async_wrapper_access_check
async def async_access_check(self, owner_user_id: int, target_id: str, access_type: AccessType) -> bool:
if self.user_id == owner_user_id:
return True
flag = await RoleAccessDao.ajudge_role_access(self.user_role, target_id, access_type)
return True if flag else False
@wrapper_access_check
def copiable_check(self, owner_user_id: int) -> bool:
"""
Check if the user has permission to copy a resource
"""
# Determine if it belongs to my resource
if self.user_id == owner_user_id:
return True
return False
@wrapper_access_check
def check_group_admin(self, group_id: int) -> bool:
"""
Check if the user is an administrator of a group
"""
# Determine if you are an administrator of a user group
user_group = UserGroupDao.get_user_admin_group(self.user_id)
if not user_group:
return False
for one in user_group:
if one.group_id == group_id:
return True
return False
@async_wrapper_access_check
async def async_check_group_admin(self, group_id: int) -> bool:
"""
Asynchronously check if the user is an administrator of a group
"""
# Determine if you are an administrator of a user group
user_group = await UserGroupDao.aget_user_admin_group(self.user_id, group_id)
if not user_group:
return False
for one in user_group:
if one.group_id == group_id:
return True
return False
@wrapper_access_check
def check_groups_admin(self, group_ids: List[int]) -> bool:
"""
Check if the user is an administrator in the user group list, one of which istrue
"""
user_groups = UserGroupDao.get_user_admin_group(self.user_id)
for one in user_groups:
if one.is_group_admin and one.group_id in group_ids:
return True
return False
async def get_user_groups(self, user_id: int) -> List[Dict]:
""" Query a list of roles for a user """
user_groups = await UserGroupDao.aget_user_group(user_id)
user_group_ids: List[int] = [one_group.group_id for one_group in user_groups]
res = []
for i in range(len(user_group_ids) - 1, -1, -1):
if self.group_cache.get(user_group_ids[i]):
res.append(self.group_cache.get(user_group_ids[i]))
del user_group_ids[i]
# Query database for role information without caching
if user_group_ids:
group_list = await GroupDao.aget_group_by_ids(user_group_ids)
for group_info in group_list:
self.group_cache[group_info.id] = {'id': group_info.id, 'name': group_info.group_name}
res.append(self.group_cache.get(group_info.id))
return res
async def get_user_group_ids(self, user_id: int = None):
if user_id is None:
user_id = self.user_id
user_groups = await UserGroupDao.aget_user_group(user_id)
return [one_group.group_id for one_group in user_groups]
def get_user_access_resource_ids(self, access_types: List[AccessType]) -> List[str]:
""" Query resources for which the user has the corresponding permissionsIDVertical """
role_access = RoleAccessDao.get_role_access_batch(self.user_role, access_types)
return list(set([one.third_id for one in role_access]))
async def aget_user_access_resource_ids(self, access_types: List[AccessType]) -> List[str]:
""" Resources with corresponding permissions for asynchronous query usersIDVertical """
role_access = await RoleAccessDao.aget_role_access_batch(self.user_role, access_types)
return list(set([one.third_id for one in role_access]))
# some methods related to AuthJwt
@classmethod
def create_access_token(cls, user: User, auth_jwt: AuthJwt) -> str:
""" Create access for userstoken """
payload = {
'user_id': user.user_id,
'user_name': user.user_name
}
token = auth_jwt.create_access_token(subject=payload)
return token
@classmethod
def set_access_cookies(cls, token: str, auth_jwt: AuthJwt, **kwargs) -> None:
""" set access token into cookie """
auth_jwt.set_access_token(token, **kwargs)
@classmethod
def unset_access_cookies(cls, auth_jwt: AuthJwt) -> None:
auth_jwt.unset_access_token()
@classmethod
async def init_login_user(cls, user_id: int, user_name: str) -> Self:
user_roles = await UserRoleDao.aget_user_roles(user_id)
role_ids = [user_role.role_id for user_role in user_roles]
login_user = cls(user_id=user_id, user_name=user_name, user_role=role_ids)
return login_user
@classmethod
def init_login_user_sync(cls, user_id: int, user_name: str) -> Self:
user_roles = UserRoleDao.get_user_roles(user_id)
role_ids = [user_role.role_id for user_role in user_roles]
login_user = cls(user_id=user_id, user_name=user_name, user_role=role_ids)
return login_user
@classmethod
async def get_login_user(cls, auth_jwt: AuthJwt = Depends()) -> Self:
subject = auth_jwt.get_subject()
return await cls.init_login_user(user_id=subject['user_id'], user_name=subject['user_name'])
@classmethod
async def get_admin_user(cls, auth_jwt: AuthJwt = Depends()) -> Self:
login_user = await cls.get_login_user(auth_jwt)
if not login_user.is_admin():
raise UnAuthorizedError.http_exception()
return login_user
@classmethod
async def get_login_user_from_ws(cls, websocket: WebSocket, auth_jwt: AuthJwt = Depends(), t: str = None) -> Self:
subject = auth_jwt.get_subject(auth_from="websocket", websocket=websocket, token=t)
return await cls.init_login_user(user_id=subject['user_id'], user_name=subject['user_name'])
@classmethod
async def get_admin_user_from_ws(cls, websocket: WebSocket, auth_jwt: AuthJwt = Depends(), t: str = None) -> Self:
login_user = await cls.get_login_user_from_ws(websocket, auth_jwt, t)
if not login_user.is_admin():
raise UnAuthorizedError.http_exception()
return login_user
@classmethod
async def get_roles_web_menu(cls, user: User) -> (List[int] | str, List[str]):
""" get user roles and web menu """
db_user_role = await UserRoleDao.aget_user_roles(user.user_id)
role = ''
role_ids = []
for user_role in db_user_role:
if user_role.role_id == AdminRole:
role = 'admin'
else:
role_ids.append(user_role.role_id)
if role != 'admin':
# is user group admin ?
db_user_groups = await UserGroupDao.aget_user_admin_group(user.user_id)
if len(db_user_groups) > 0:
role = 'group_admin'
else:
role = role_ids
# Get a list of a user's menu bar permissions
web_menu = await RoleAccessDao.aget_role_access(role_ids, AccessType.WEB_MENU)
web_menu = list(set([one.third_id for one in web_menu]))
else:
web_menu = [one.value for one in WebMenuResource]
return role, web_menu
================================================
FILE: src/backend/bisheng/user/domain/services/captcha.py
================================================
from bisheng.core.cache.redis_manager import get_redis_client
async def verify_captcha(captcha: str, captcha_key: str):
# check captcha
redis_client = await get_redis_client()
captcha_value = await redis_client.aget(captcha_key)
if captcha_value:
await redis_client.adelete(captcha_key)
return captcha_value.lower() == captcha.lower()
else:
return False
================================================
FILE: src/backend/bisheng/user/domain/services/user.py
================================================
from base64 import b64decode
from datetime import datetime
from typing import List
import rsa
from fastapi import Request, Depends
from bisheng.api.services.audit_log import AuditLogService
from bisheng.api.v1.schemas import CreateUserReq
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum
from bisheng.common.errcode.user import (UserNameAlreadyExistError,
UserNeedGroupAndRoleError, UserForbiddenError, CaptchaError, UserValidateError,
UserPasswordMaxTryError, UserPasswordExpireError, UserNameTooLongError)
from bisheng.common.schemas.api import resp_200
from bisheng.common.schemas.telemetry.event_data_schema import UserLoginEventData
from bisheng.common.services import telemetry_service
from bisheng.common.services.config_service import settings
from bisheng.core.cache.redis_manager import get_redis_client_sync, get_redis_client
from bisheng.core.logger import trace_id_var
from bisheng.database.models.user_group import UserGroupDao
from bisheng.user.domain.models.user import User, UserDao, UserLogin, UserRead, UserCreate
from bisheng.utils import md5_hash, get_request_ip
from bisheng.utils.constants import RSA_KEY
from .auth import LoginUser, AuthJwt
from .captcha import verify_captcha
from ..const import USER_PASSWORD_ERROR, USER_CURRENT_SESSION
class UserService:
@classmethod
def decrypt_md5_password(cls, password: str):
if value := get_redis_client_sync().get(RSA_KEY):
private_key = value[1]
password = md5_hash(rsa.decrypt(b64decode(password), private_key).decode('utf-8'))
else:
password = md5_hash(password)
return password
@classmethod
def create_user(cls, request: Request, login_user: LoginUser, req_data: CreateUserReq):
"""
Create User
"""
exists_user = UserDao.get_user_by_username(req_data.user_name)
if exists_user:
# Throwing an exception?
raise UserNameAlreadyExistError.http_exception()
user = User(
user_name=req_data.user_name,
password=cls.decrypt_md5_password(req_data.password),
)
group_ids = []
role_ids = []
for one in req_data.group_roles:
group_ids.append(one.group_id)
role_ids.extend(one.role_ids)
if not group_ids or not role_ids:
raise UserNeedGroupAndRoleError.http_exception()
user = UserDao.add_user_with_groups_and_roles(user, group_ids, role_ids)
return user
@staticmethod
def get_error_password_key(username: str):
return USER_PASSWORD_ERROR.format(username)
@classmethod
async def clear_error_password_key(cls, username: str):
# Count of cleanup password errors
error_key = cls.get_error_password_key(username)
(await get_redis_client()).delete(error_key)
@classmethod
async def judge_user_password(cls, db_user: User, password: str) -> None:
redis_client = await get_redis_client()
password_conf = await settings.get_password_conf()
if not db_user.password:
raise UserValidateError()
if db_user.password == password:
# Determine if the password has not been changed for a long time
if password_conf.password_valid_period and password_conf.password_valid_period > 0:
if (datetime.now() - db_user.password_update_time).days >= password_conf.password_valid_period:
raise UserPasswordExpireError()
return
# Determine if the number of errors needs to be logged
if not password_conf.login_error_time_window or not password_conf.max_error_times:
raise UserValidateError()
# Number of errors plus1
error_key = cls.get_error_password_key(db_user.user_name)
error_num = await redis_client.aincr(error_key)
if error_num == 1:
# First time setupkeyExpiration date
await redis_client.aexpire_key(error_key, password_conf.login_error_time_window * 60)
if error_num and int(error_num) >= password_conf.max_error_times:
# Maximum number of errors reached, account banned
db_user.delete = 1
await UserDao.aupdate_user(db_user)
raise UserPasswordMaxTryError()
raise UserValidateError()
@classmethod
async def user_register(cls, user: UserCreate):
# Captcha Verification
if settings.get_from_db('use_captcha'):
if not user.captcha_key or not await verify_captcha(user.captcha, user.captcha_key):
raise CaptchaError()
db_user = User.model_validate(user)
# check if user already exist
user_exists = await UserDao.aget_user_by_username(db_user.user_name)
if user_exists:
raise UserNameAlreadyExistError()
if len(db_user.user_name) > 30:
raise UserNameTooLongError()
db_user.password = cls.decrypt_md5_password(user.password)
# Under JudgmentadminDoes the user exist
admin = await UserDao.aget_user(1)
if admin:
db_user = await UserDao.add_user_and_default_role(db_user)
else:
db_user.user_id = 1
db_user = await UserDao.add_user_and_admin_role(db_user)
# Write users to the default user group
await UserGroupDao.add_default_user_group(db_user.user_id)
return db_user
@classmethod
async def user_login(cls, request: Request, user: UserLogin, auth_jwt: AuthJwt = Depends()):
if await settings.aget_from_db('use_captcha'):
if not user.captcha_key or not await verify_captcha(user.captcha, user.captcha_key):
raise CaptchaError()
# get user info
db_user = await UserDao.aget_user_by_username(user.user_name)
# verify user exists
if not db_user:
return UserValidateError.return_resp()
if db_user.delete == 1:
raise UserForbiddenError()
# verify password
password = cls.decrypt_md5_password(user.password)
await cls.judge_user_password(db_user, password)
# gen jwt token
access_token = LoginUser.create_access_token(user=db_user, auth_jwt=auth_jwt)
# set cookies
LoginUser.set_access_cookies(access_token, auth_jwt=auth_jwt)
# Set the logged in user's currentcookie, .jwtValid for an additional hour
redis_client = await get_redis_client()
await redis_client.aset(USER_CURRENT_SESSION.format(db_user.user_id), access_token,
auth_jwt.cookie_conf.jwt_token_expire_time + 3600)
# Log Audit Logs
login_user = await LoginUser.init_login_user(db_user.user_id, db_user.user_name)
AuditLogService.user_login(login_user, get_request_ip(request))
# RecordTelemetryJournal
await telemetry_service.log_event(user_id=db_user.user_id, event_type=BaseTelemetryTypeEnum.USER_LOGIN,
trace_id=trace_id_var.get(),
event_data=UserLoginEventData(method="password"))
return resp_200(UserRead(access_token=access_token, **db_user.__dict__))
@classmethod
def get_user_all_info(cls, *, start_time: datetime = None, end_time: datetime = None, user_ids: List[int] = None,
page: int = 1, page_size: int = 100) -> List[User]:
""" Get user information, including user group and role information """
return UserDao.get_user_with_group_role(page=page, page_size=page_size, user_ids=user_ids,
start_time=start_time, end_time=end_time)
@classmethod
def get_first_user(cls) -> User | None:
""" Get the first user """
return UserDao.get_first_user()
@classmethod
async def get_user_by_id(cls, user_id: int) -> User | None:
""" Get user by username """
return await UserDao.aget_user(user_id)
================================================
FILE: src/backend/bisheng/utils/__init__.py
================================================
import hashlib
import io
import time
import uuid
from typing import BinaryIO, Union, IO
from fastapi import Request, WebSocket
def generate_uuid() -> str:
""" generate uuid4 string """
return uuid.uuid4().hex
def md5_hash(original_string: str):
""" generate md5 hash string """
md5 = hashlib.md5()
md5.update(original_string.encode('utf-8'))
return md5.hexdigest()
def get_request_ip(request: Request | WebSocket) -> str:
""" get client real ip address """
x_forwarded_for = request.headers.get('X-Forwarded-For')
if x_forwarded_for:
return x_forwarded_for.split(',')[0]
ip = request.headers.get('X-Real-IP')
if ip:
return ip
return request.client.host
def generate_knowledge_index_name() -> str:
""" generate knowledge index name """
return f"col_{int(time.time())}_{generate_uuid()[:8]}"
def calc_data_sha256(data: Union[str, bytes, IO[bytes], None]) -> Union[str, None]:
"""
calculate sha256 hash of data
:param data: str, bytes, or a file-like object (with read() method)
:return: sha256 hex digest string or None
"""
if data is None:
return None
hasher = hashlib.sha256()
# Handle str and bytes directly
if isinstance(data, (str, bytes)):
if isinstance(data, str):
data = data.encode('utf-8')
hasher.update(data)
return hasher.hexdigest()
if hasattr(data, 'read'):
# Assume it's a file-like object
try:
current_pos = data.tell()
except Exception:
current_pos = 0
data.seek(0)
chunk_size = 65536 # 64KB per chunk
while True:
chunk = data.read(chunk_size)
if not chunk:
break
hasher.update(chunk)
# Reset the file pointer to its original position
data.seek(current_pos)
return hasher.hexdigest()
return None
================================================
FILE: src/backend/bisheng/utils/consisten_hash.py
================================================
import bisect
import hashlib
from typing import Dict, List, Optional, Any
class ConsistentHash:
def __init__(self, nodes: Optional[List[str]] = None,
virtual_replicas: int = 100,
hash_fn: Optional[Any] = None):
self.virtual_replicas = virtual_replicas
self.hash_fn = hash_fn or (lambda x: int(hashlib.md5(x.encode()).hexdigest(), 16))
# save hash ring: hash_value -> node
self.ring: Dict[int, str] = {}
# save sorted hash values for find
self.sorted_hashes: List[int] = []
# save all nodes
self.nodes = set()
# init nodes
if nodes:
for node in nodes:
self.add_node(node)
def add_node(self, node: str) -> None:
if node in self.nodes:
return
self.nodes.add(node)
# create virtual nodes
for i in range(self.virtual_replicas):
# virtual node name: node#0, node#1, ...
virtual_node = f"{node}#{i}"
# calc
hash_value = self.hash_fn(virtual_node)
# Avoid hash collisions
while hash_value in self.ring:
virtual_node = f"{virtual_node}#"
hash_value = self.hash_fn(virtual_node)
# 添加到环中
self.ring[hash_value] = node
# resort hash values
self._update_sorted_hashes()
def remove_node(self, node: str) -> None:
if node not in self.nodes:
return
self.nodes.remove(node)
# Remove all virtual nodes of this node.
hashes_to_remove = [
hash_val for hash_val, n in self.ring.items()
if n == node
]
for hash_val in hashes_to_remove:
del self.ring[hash_val]
self._update_sorted_hashes()
def _update_sorted_hashes(self) -> None:
"""Update the sorted hash value list"""
self.sorted_hashes = sorted(self.ring.keys())
def find_node(self, key: str) -> Optional[str]:
"""
Retrieve the corresponding node based on the key
Args:
key: hash key
Returns:
node name or None if no nodes exist
"""
if not self.ring:
return None
key_hash = self.hash_fn(key)
# Use binary search to find the first node that is greater than or equal to key_hash.
idx = bisect.bisect_left(self.sorted_hashes, key_hash)
# If idx is out of range, return to the beginning of the ring.
if idx == len(self.sorted_hashes):
idx = 0
# return node name
return self.ring[self.sorted_hashes[idx]]
def find_nodes(self, key: str, count: int = 1) -> List[str]:
"""
Retrieves multiple nodes for a given key (in clockwise order).
Args:
key: hash key
count: Number of nodes to be returned
Returns:
Returns a list of nodes that processed the key.
"""
if not self.ring or count <= 0:
return []
if count > len(self.nodes):
count = len(self.nodes)
key_hash = self.hash_fn(key)
idx = bisect.bisect_left(self.sorted_hashes, key_hash)
if idx == len(self.sorted_hashes):
idx = 0
nodes = []
seen_nodes = set()
while len(nodes) < count:
node = self.ring[self.sorted_hashes[idx]]
if node not in seen_nodes:
seen_nodes.add(node)
nodes.append(node)
idx += 1
if idx == len(self.sorted_hashes):
idx = 0
return nodes
def get_ring_size(self) -> int:
return len(self.ring)
def get_node_count(self) -> int:
return len(self.nodes)
def get_all_nodes(self) -> List[str]:
return list(self.nodes)
if __name__ == "__main__":
# 创建一致性哈希环
ch = ConsistentHash(virtual_replicas=10)
# 测试数据分布
test_keys = ["key1", "key2", "key3", "key4", "key5", "key6", "key7", "key8", "key9", "key10"]
print("初始节点分布:")
distribution = {}
for key in test_keys:
node = ch.find_node(key)
distribution[node] = distribution.get(node, 0) + 1
print(f" {key} -> {node}")
print("\n数据分布统计:")
for node, count in distribution.items():
print(f" {node}: {count} 个键")
# 添加新节点
print("\n添加新节点 server4")
ch.add_node("server4")
# 查看添加节点后的数据分布变化
moved_keys = []
for key in test_keys:
new_node = ch.find_node(key)
# 这里简化比较，实际应该记录之前分配的节点
print(f" {key} -> {new_node}")
# 获取多个节点
print("\n获取处理 'key1' 的3个节点:")
nodes = ch.find_nodes("key1", 3)
for i, node in enumerate(nodes, 1):
print(f" 第{i}候选: {node}")
# 移除节点
print("\n移除节点 server2")
ch.remove_node("server2")
print(f"当前物理节点数: {ch.get_node_count()}")
print(f"当前虚拟节点数: {ch.get_ring_size()}")
================================================
FILE: src/backend/bisheng/utils/constants.py
================================================
from typing import Any, Dict, List
OPENAI_MODELS = [
'text-davinci-003',
'text-davinci-002',
'text-curie-001',
'text-babbage-001',
'text-ada-001',
]
CHAT_OPENAI_MODELS = [
'gpt-3.5-turbo-0613', 'gpt-3.5-turbo', 'gpt-3.5-turbo-16k-0613', 'gpt-3.5-turbo-16k',
'gpt-4-0613', 'gpt-4-32k-0613', 'gpt-4', 'gpt-4-32k', 'gpt-4-1106-preview', 'gpt-4o'
]
ANTHROPIC_MODELS = [
'claude-v1', # largest model, ideal for a wide range of more complex tasks.
'claude-v1-100k', # An enhanced version of claude-v1 with a 100,000 token (roughly 75,000 word) context window.
'claude-instant-v1', # A smaller model with far lower latency, sampling at roughly 40 words/sec!
'claude-instant-v1-100k', # Like claude-instant-v1 with a 100,000 token context window but retains its performance.
# Specific sub-versions of the above models:
'claude-v1.3', # Vs claude-v1.2: better instruction-following, code, and non-English dialogue and writing.
'claude-v1.3-100k', # An enhanced version of claude-v1.3 with a 100,000 token (roughly 75,000 word) context window.
'claude-v1.2', # Vs claude-v1.1: small adv in general helpfulness, instruction following, coding, and other tasks.
'claude-v1.0', # An earlier version of claude-v1.
'claude-instant-v1.1', # Latest version of claude-instant-v1. Better than claude-instant-v1.0 at most tasks.
'claude-instant-v1.1-100k', # Version of claude-instant-v1.1 with a 100K token context window.
'claude-instant-v1.0', # An earlier version of claude-instant-v1.
]
DEFAULT_PYTHON_FUNCTION = """
def python_function(text: str) -> str:
\"\"\"This is a default python function that returns the input text\"\"\"
return text
"""
PYTHON_BASIC_TYPES = [str, bool, int, float, tuple, list, dict, set]
DIRECT_TYPES = [
'str', 'bool', 'dict', 'int', 'float', 'Any', 'prompt', 'code', 'NestedDict', 'variable', 'knowledge_list',
'knowledge_one', 'bisheng_model', 'bisheng_embedding'
]
# Added to recordnode_id And of the relationship between objectskey
NODE_ID_DICT = 'node_id_dict'
# Used to record preset questions, dict: key is the object of the actionid
PRESET_QUESTION = 'preset_question'
# redis key
CAPTCHA_PREFIX = 'cap_'
RSA_KEY = 'rsa_'
# Stores the number of times a user's password is incorrect.keyareusername
USER_PASSWORD_ERROR = 'user_password_error:{}'
# Store current user logged incookie, keyIs Userid
USER_CURRENT_SESSION = 'user_current_session:{}'
LOADERS_INFO: List[Dict[str, Any]] = [
{
'loader': 'AirbyteJSONLoader',
'name': 'Airbyte JSON (.jsonl)',
'import': 'langchain.document_loaders.AirbyteJSONLoader',
'defaultFor': ['jsonl'],
'allowdTypes': ['jsonl'],
},
{
'loader': 'JSONLoader',
'name': 'JSON (.json)',
'import': 'langchain.document_loaders.JSONLoader',
'defaultFor': ['json'],
'allowdTypes': ['json'],
},
{
'loader': 'BSHTMLLoader',
'name': 'BeautifulSoup4 HTML (.html, .htm)',
'import': 'langchain.document_loaders.BSHTMLLoader',
'allowdTypes': ['html', 'htm'],
},
{
'loader': 'CSVLoader',
'name': 'CSV (.csv)',
'import': 'langchain.document_loaders.CSVLoader',
'defaultFor': ['csv'],
'allowdTypes': ['csv'],
},
{
'loader': 'CoNLLULoader',
'name': 'CoNLL-U (.conllu)',
'import': 'langchain.document_loaders.CoNLLULoader',
'defaultFor': ['conllu'],
'allowdTypes': ['conllu'],
},
{
'loader': 'EverNoteLoader',
'name': 'EverNote (.enex)',
'import': 'langchain.document_loaders.EverNoteLoader',
'defaultFor': ['enex'],
'allowdTypes': ['enex'],
},
{
'loader': 'FacebookChatLoader',
'name': 'Facebook Chat (.json)',
'import': 'langchain.document_loaders.FacebookChatLoader',
'allowdTypes': ['json'],
},
{
'loader': 'OutlookMessageLoader',
'name': 'Outlook Message (.msg)',
'import': 'langchain.document_loaders.OutlookMessageLoader',
'defaultFor': ['msg'],
'allowdTypes': ['msg'],
},
{
'loader': 'PyPDFLoader',
'name': 'PyPDF (.pdf)',
'import': 'langchain.document_loaders.PyPDFLoader',
'defaultFor': ['pdf'],
'allowdTypes': ['pdf'],
},
{
'loader': 'STRLoader',
'name': 'Subtitle (.str)',
'import': 'langchain.document_loaders.STRLoader',
'defaultFor': ['str'],
'allowdTypes': ['str'],
},
{
'loader': 'TextLoader',
'name': 'Text (.txt)',
'import': 'langchain.document_loaders.TextLoader',
'defaultFor': ['txt'],
'allowdTypes': ['txt'],
},
{
'loader': 'UnstructuredEmailLoader',
'name': 'Unstructured Email (.eml)',
'import': 'langchain.document_loaders.UnstructuredEmailLoader',
'defaultFor': ['eml'],
'allowdTypes': ['eml'],
},
{
'loader': 'UnstructuredHTMLLoader',
'name': 'Unstructured HTML (.html, .htm)',
'import': 'langchain.document_loaders.UnstructuredHTMLLoader',
'defaultFor': ['html', 'htm'],
'allowdTypes': ['html', 'htm'],
},
{
'loader': 'UnstructuredMarkdownLoader',
'name': 'Unstructured Markdown (.md)',
'import': 'langchain.document_loaders.UnstructuredMarkdownLoader',
'defaultFor': ['md'],
'allowdTypes': ['md'],
},
{
'loader': 'UnstructuredPowerPointLoader',
'name': 'Unstructured PowerPoint (.pptx)',
'import': 'langchain.document_loaders.UnstructuredPowerPointLoader',
'defaultFor': ['pptx'],
'allowdTypes': ['pptx'],
},
{
'loader': 'UnstructuredWordLoader',
'name': 'Unstructured Word (.docx)',
'import': 'langchain.document_loaders.UnstructuredWordLoader',
'defaultFor': ['docx'],
'allowdTypes': ['docx'],
},
]
================================================
FILE: src/backend/bisheng/utils/docx_temp.py
================================================
import os
import tempfile
from pathlib import Path
from typing import IO, Dict, List, Any, Tuple
from urllib.parse import unquote, urlparse
import pandas as pd
import requests
from docx import Document
from docx.shared import Inches
from loguru import logger
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync
from bisheng.utils.util import _is_valid_url
def find_lcs(str1, str2):
lstr1 = len(str1)
lstr2 = len(str2)
record = [[0 for i in range(lstr2 + 1)] for j in range(lstr1 + 1)] # One more person
maxNum = 0
p = 0
for i in range(lstr1):
for j in range(lstr2):
if str1[i] == str2[j]:
record[i + 1][j + 1] = record[i][j] + 1
if record[i + 1][j + 1] > maxNum:
maxNum = record[i + 1][j + 1]
p = i + 1
return str1[p - maxNum: p], maxNum
class DocxTemplateRender(object):
def __init__(self, filepath: str = None, file_content: IO[bytes] = None):
self.filepath = filepath
self.file_content = file_content
if self.filepath:
self.doc = Document(self.filepath)
else:
self.doc = Document(self.file_content)
def _insert_image(self, paragraph, image_path: str, alt_text: str = "Images"):
"""
Insert image in paragraph
Args:
paragraph: WordParagraph object
image_path: Image file path
alt_text: Set Alt Text(s)
"""
logger.debug(f"[Simple Illustration] Start Inserting Pictures: {image_path}")
try:
if os.path.exists(image_path):
# Check file size
file_size = os.path.getsize(image_path)
logger.debug(f"[Simple Illustration] Image file exists: {image_path}, size: {file_size}byte")
# Insert an image with a maximum width of6Inch
run = paragraph.runs[0] if paragraph.runs else paragraph.add_run()
logger.debug(f"[Simple Illustration] Ready atrunInsert a picture in therunQuantity: {len(paragraph.runs)}")
run.add_picture(image_path, width=Inches(6))
logger.info(f"[Simple Illustration] ✅ Successfully inserted image: {image_path}, size: {file_size}byte")
else:
# Image file does not exist, use original path
logger.error(f"[Simple Illustration] ❌ Image file does not exist: {image_path}")
if paragraph.runs:
paragraph.runs[0].text = image_path
else:
paragraph.add_run(image_path)
except Exception as e:
# Inserting image failed, show originalURL
logger.error(f"[Simple Illustration] ❌ Failed to insert image: {image_path}, Error type: {type(e).__name__}, Error-free: {str(e)}")
if paragraph.runs:
paragraph.runs[0].text = image_path
else:
paragraph.add_run(image_path)
def _replace_placeholder_with_image(self, paragraph, placeholder, image_path, alt_text):
"""
Precisely replace the placeholder in the paragraph with the picture, correctly handle the spanrunPlaceholder for
Args:
paragraph: WordParagraph object
placeholder: Placeholder to replace
image_path: Image file path
alt_text: Image Alt Text
"""
logger.debug(f"[Image Replacement] Start replacing placeholders: {placeholder} -> {image_path}")
# Get the full text of the paragraph
paragraph_text = paragraph.text
# Find the location of the placeholder
placeholder_start = paragraph_text.find(placeholder)
if placeholder_start == -1:
logger.warning(f"[Image Replacement] Placeholder not found: {placeholder}")
return # Placeholder does not exist
placeholder_end = placeholder_start + len(placeholder)
# Anchor placeholder inrunsPosition in
current_pos = 0
start_run_index = -1
start_run_pos = 0
end_run_index = -1
end_run_pos = 0
for i, run in enumerate(paragraph.runs):
run_len = len(run.text)
# Find the placeholder start position
if start_run_index == -1 and current_pos + run_len > placeholder_start:
start_run_index = i
start_run_pos = placeholder_start - current_pos
# Find the placeholder end position
if current_pos + run_len >= placeholder_end:
end_run_index = i
end_run_pos = placeholder_end - current_pos
break
current_pos += run_len
# Clear placeholder text
if start_run_index == end_run_index:
# Placeholder is the samerunand within
run = paragraph.runs[start_run_index]
run.text = run.text[:start_run_pos] + run.text[end_run_pos:]
else:
# Placeholders span multipleruns
# Clear startrunSections in
start_run = paragraph.runs[start_run_index]
start_run.text = start_run.text[:start_run_pos]
# Clear EndrunSections in
end_run = paragraph.runs[end_run_index]
end_run.text = end_run.text[end_run_pos:]
# Clear middleruns
for i in range(end_run_index - 1, start_run_index, -1):
paragraph.runs[i].text = ""
# Insert image at placeholder position
# Find a suitable insertion location (first non-empty after cleaningrunafterwards
insert_run = None
for i in range(start_run_index, len(paragraph.runs)):
if paragraph.runs[i].text or i == start_run_index:
insert_run = paragraph.runs[i]
break
if insert_run is not None:
# In therunInsert image after
logger.debug(f"[Image Replacement] InsiderunPosition Insert Picture: {image_path}")
self._insert_image_at_run(insert_run, image_path, alt_text)
logger.info(f"[Image Replacement] Placeholder replacement complete: {placeholder} -> {image_path}")
else:
# If a suitable location is not found, use the original method
logger.warning(f"[Image Replacement] No Fit Foundrunlocation, using alternate methods: {image_path}")
self._insert_image(paragraph, image_path, alt_text)
def _insert_image_at_run(self, run, image_path, alt_text):
"""
In DesignationrunPosition Insert Picture
Args:
run: Word runObjects
image_path: Image Path
alt_text: Image Alt Text
"""
try:
from docx.shared import Inches
# Checks to see if file exists.
if not os.path.exists(image_path):
logger.warning(f"[Image Rendering] Image file does not exist, use original path: {image_path}")
run.text = image_path
return
# Check file size and format
file_size = os.path.getsize(image_path)
file_ext = os.path.splitext(image_path)[1].lower()
logger.debug(f"[Image Rendering] Ready to insert image: path={image_path}, size={file_size}byte, ext={file_ext}")
# Directly in the currentrunInsert image in
run.add_picture(image_path, width=Inches(4)) # Default Width4Inch
# Emptyruntext in to avoid displaying excess text
run.text = ""
logger.info(f"[Image Rendering] InsiderunLocation Insert Picture Successful: {image_path}, size={file_size}byte")
except Exception as e:
logger.error(f"[Image Rendering] runLocation Insert Picture Failed: {image_path}, Error-free: {e}")
logger.debug(f"[Image Rendering] Error Details of error: {type(e).__name__}: {str(e)}")
try:
# Use original insertion method as backup
logger.info(f"[Image Rendering] Try alternate insertion methods: {image_path}")
# DapatkanrunParent Paragraph Object of
paragraph = run._element.getparent()
# Convert Topython-docxParagraph object
from docx.text.paragraph import Paragraph
para_obj = Paragraph(paragraph, run.part)
self._insert_image(para_obj, image_path, alt_text)
logger.info(f"[Image Rendering] Alternate method inserted successfully: {image_path}")
except Exception as backup_e:
logger.error(f"[Image Rendering] Alternate insertion methods also failed: {image_path}, Error-free: {backup_e}")
logger.error(f"[Image Rendering] Alternate Method Detail Error: {type(backup_e).__name__}: {str(backup_e)}")
# If all methods fail, use the original path
run.text = image_path
logger.warning(f"[Image Rendering] All insert methods failed, use original path text: {image_path}")
def _replace_placeholder_in_structured_paragraph(self, paragraph, placeholder: str, table_data: List[List[str]]):
"""
Simplified table replacement: insert the table directly at the placeholder without adding any structural tags
"""
# Clear placeholder text
self._clear_placeholder_from_paragraph(paragraph, placeholder)
# Insert table directly at paragraph position
self._insert_table(paragraph, table_data)
logger.info(f"Form Replacement Complete of {len(table_data)} Row")
def _clear_placeholder_from_paragraph(self, paragraph, placeholder):
"""
Precisely clear placeholders from paragraphs to handle spanningrunCONDITION....
Args:
paragraph: WordParagraph object
placeholder: Placeholders to be cleared
"""
# Get the full text of the paragraph
paragraph_text = paragraph.text
# Find the location of the placeholder
placeholder_start = paragraph_text.find(placeholder)
if placeholder_start == -1:
return # Placeholder does not exist
placeholder_end = placeholder_start + len(placeholder)
# Anchor placeholder inrunsPosition in
current_pos = 0
start_run_index = -1
start_run_pos = 0
end_run_index = -1
end_run_pos = 0
for i, run in enumerate(paragraph.runs):
run_len = len(run.text)
# Find the placeholder start position
if start_run_index == -1 and current_pos + run_len > placeholder_start:
start_run_index = i
start_run_pos = placeholder_start - current_pos
# Find the placeholder end position
if current_pos + run_len >= placeholder_end:
end_run_index = i
end_run_pos = placeholder_end - current_pos
break
current_pos += run_len
# Clear placeholder text
if start_run_index == end_run_index:
# Placeholder is the samerunand within
run = paragraph.runs[start_run_index]
run.text = run.text[:start_run_pos] + run.text[end_run_pos:]
else:
# Placeholders span multipleruns
# Clear startrunSections in
start_run = paragraph.runs[start_run_index]
start_run.text = start_run.text[:start_run_pos]
# Clear EndrunSections in
end_run = paragraph.runs[end_run_index]
end_run.text = end_run.text[end_run_pos:]
# Clear middleruns
for i in range(end_run_index - 1, start_run_index, -1):
paragraph.runs[i].text = ""
def _process_resource_placeholders(self, doc, placeholder_map):
"""
Work with hybrid placeholders in paragraphs in positional order
Args:
doc: WordDocument object
placeholder_map: Placeholder Mapping Dictionary
"""
# Work with placeholders in all paragraphs
paragraphs_to_process = list(doc.paragraphs) # Create a copy as we may modify the paragraph structure
for i, p in enumerate(paragraphs_to_process):
paragraph_text = p.text
if not paragraph_text:
continue
# Find all placeholders and their positions in the paragraph
placeholders_with_positions = []
for placeholder, resource_info in placeholder_map.items():
pos = paragraph_text.find(placeholder)
if pos != -1:
placeholders_with_positions.append({
'placeholder': placeholder,
'resource_info': resource_info,
'position': pos,
'end_position': pos + len(placeholder)
})
if not placeholders_with_positions:
continue
# Sort by location from to post-processing
placeholders_with_positions.sort(key=lambda x: x['position'])
# Split paragraph into text segments and placeholder segments
self._process_mixed_content_paragraph(doc, p, placeholders_with_positions, paragraph_text)
# Working with Placeholders in Table Cells
for table in doc.tables:
for i, row in enumerate(table.rows):
for j, cell in enumerate(row.cells):
for one in cell.paragraphs:
cell_text = one.text
if not cell_text:
continue
# Check for placeholders in cells
for placeholder, resource_info in placeholder_map.items():
if placeholder in cell_text:
if resource_info["type"] == "image":
# Insert Actual Picture in Table Cell
image_path = resource_info.get("local_path") or resource_info.get("path", "")
if image_path and os.path.exists(image_path):
try:
# Insert a picture in a cell (this will empty the cell and insert the picture)
self._insert_image_in_table_cell(cell, image_path)
logger.info(f"✅ Picture successfully inserted in table cell: {image_path}")
# Marker placeholder processed, no need to update text
cell_text = ""
except Exception as e:
logger.error(f"❌ Table Cell Insert Picture Failed: {str(e)}")
# Show file name on failure
cell_text = cell_text.replace(placeholder, os.path.basename(image_path))
else:
# Image file does not exist, display path
cell_text = cell_text.replace(placeholder,
resource_info.get("path", placeholder))
elif resource_info["type"] == "excel":
cell_text = cell_text.replace(placeholder, "[ExcelTable Filter]")
elif resource_info["type"] == "csv":
cell_text = cell_text.replace(placeholder, "[CSVTable Filter]")
elif resource_info["type"] == "markdown_table":
cell_text = cell_text.replace(placeholder, "[MarkdownTable Filter]")
logger.info(f"Process table cell placeholders: {placeholder}")
# Update cell text
if cell_text != one.text:
if one.runs:
one.runs[0].text = cell_text
for r_index in range(1, len(one.runs)):
one.runs[r_index].text = ""
else:
one.add_run(cell_text)
def _process_mixed_content_paragraph(self, doc, paragraph, placeholders_with_positions, original_text):
"""
Working with paragraphs with mixed content - Insert images and tables inline
Args:
doc: WordDocument object
paragraph: Original paragraph
placeholders_with_positions: List of placeholder information by location
original_text: Original paragraph text
"""
# Extract style information from the original paragraph for subsequent paragraph creation
original_style_info = self._extract_paragraph_style_info(paragraph)
# Split the text into fragments, keeping the original order
segments = []
last_end = 0
for item in placeholders_with_positions:
# Add text before placeholder
if item['position'] > last_end:
text_before = original_text[last_end:item['position']]
if text_before: # Keep all text, including whitespaces
segments.append({
'type': 'text',
'content': text_before
})
# Add the resource corresponding to the placeholder
segments.append({
'type': 'resource',
'placeholder': item['placeholder'],
'resource_info': item['resource_info']
})
last_end = item['end_position']
# Add last remaining text
if last_end < len(original_text):
text_after = original_text[last_end:]
if text_after:
segments.append({
'type': 'text',
'content': text_after
})
# Clear original paragraph
paragraph.clear()
# Redesign: segmentation to keep document structure consistent
current_paragraph = paragraph
i = 0
while i < len(segments):
segment = segments[i]
if segment['type'] == 'text':
# Add text content directly without any cleanup
current_paragraph.add_run(segment['content'])
elif segment['type'] == 'resource':
resource_info = segment['resource_info']
if resource_info['type'] == 'image':
# Images can be really inlined in paragraphs
self._insert_inline_image(current_paragraph, resource_info['path'],
resource_info.get('alt_text', ''))
elif resource_info['type'] in ['excel', 'csv', 'markdown_table']:
# Table Inline Processing: Insert the table at its current position, then create a new paragraph for subsequent content
if resource_info['type'] == 'markdown_table':
table_data, alignments = self._markdown_table_to_data(resource_info["content"])
else:
table_data = resource_info.get('table_data', [["Table data parsing failed"]])
alignments = None
# Insert table immediately after current paragraph
table_element = self._create_table_element(table_data)
# Get the position of the current paragraph in the document
paragraph_element = current_paragraph._element
paragraph_parent = paragraph_element.getparent()
paragraph_index = list(paragraph_parent).index(paragraph_element)
# Insert table after current paragraph
paragraph_parent.insert(paragraph_index + 1, table_element)
# Create a new paragraph for the subsequent text and update itcurrent_paragraph
if i + 1 < len(segments):
next_paragraph = self._create_new_paragraph_after_table(paragraph_parent, paragraph_index + 1,
original_style_info)
current_paragraph = next_paragraph
i += 1
def _insert_table_inline(self, current_paragraph, table_data, segments, current_index):
"""
Insert tables inline for text continuity
Args:
current_paragraph: Current paragraph
table_data: Form Data
segments: All phrases
current_index: Current Snippet Index
Returns:
int: Number of subsequent text fragments skipped
"""
# 1. Insert table paragraph after current paragraph
table_paragraph = self._create_new_paragraph_after(current_paragraph)
self._insert_table(table_paragraph, table_data)
# 2. Check if there is a subsequent text fragment and create a new one if there is one
remaining_segments = segments[current_index + 1:]
text_segments = [seg for seg in remaining_segments if seg['type'] == 'text']
if text_segments:
# Create a new paragraph for the subsequent text
next_text_paragraph = self._create_new_paragraph_after(table_paragraph)
# Add all subsequent text fragments to the new paragraph
for j, seg in enumerate(text_segments):
next_text_paragraph.add_run(seg['content'])
# Returns the number of text fragments skipped (so that the main loop no longer processes these fragments)
skipped_count = len(text_segments)
return skipped_count
else:
return 0
def _create_new_paragraph_after(self, paragraph):
"""
Create a new paragraph after the specified paragraph
Args:
paragraph: Topic Paragraph reference
Returns:
Newly created paragraph object
"""
from docx.oxml import parse_xml
from docx.text.paragraph import Paragraph
# Create a new paragraph element
new_p_xml = ''
new_p_element = parse_xml(new_p_xml)
# Insert a new paragraph after the current paragraph
paragraph._element.getparent().insert(
list(paragraph._element.getparent()).index(paragraph._element) + 1,
new_p_element
)
# Returns the wrapped paragraph object
return Paragraph(new_p_element, paragraph._parent)
def _insert_inline_image(self, paragraph, image_path: str, alt_text: str = ""):
"""
Insert an image inline in the paragraph, and the image will be in the correct position in the text stream
"""
try:
# Use an existing image insertion method, but make sure it's inline
run = paragraph.add_run()
run.add_picture(image_path, width=Inches(4.0))
logger.info(f"[Inline Image] ✅ Successfully inserted image inline: {image_path}")
except Exception as e:
logger.error(f"[Inline Image] ❌ Failed to insert image: {str(e)}")
# Insert original image path
paragraph.add_run(image_path)
def _insert_image_in_table_cell(self, cell, image_path: str):
"""
Insert a picture in a table cell
Args:
cell: Table Cell Object
image_path: Image file path
"""
try:
from docx.shared import Inches
import os
if not os.path.exists(image_path):
logger.warning(f"[Table Cell Picture] Image file does not exist: {image_path}")
return
# Empty cell contents
for paragraph in cell.paragraphs:
paragraph.clear()
# Insert image in first paragraph
first_paragraph = cell.paragraphs[0] if cell.paragraphs else cell.add_paragraph()
run = first_paragraph.add_run()
# Insert an image and set the appropriate size
run.add_picture(image_path, width=Inches(2.0)) # Use smaller dimensions in table cells
logger.info(f"[Table Cell Picture] ✅ Successfully inserted image: {image_path}")
except Exception as e:
logger.error(f"[Table Cell Picture] ❌ Failed to insert: {str(e)}")
raise e
def _csv_to_table(self, csv_path: str) -> List[List[str]]:
"""
will beCSVConvert file to table data
Args:
csv_path: CSVFilePath
Returns:
List[List[str]]: Form Data
"""
try:
# read outCSVfiles, automatically detecting encodings and delimiters
import csv
import chardet
# Test file code
with open(csv_path, "rb") as f:
raw_data = f.read()
encoding_result = chardet.detect(raw_data)
encoding = encoding_result["encoding"] if encoding_result["encoding"] else "utf-8"
table_data = []
# Try a different delimiter
delimiters = [",", ";", "\t", "|"]
for delimiter in delimiters:
try:
with open(csv_path, "r", encoding=encoding, newline="") as f:
# Read a small part first to detect the delimiter
sample = f.read(1024)
f.seek(0)
sniffer = csv.Sniffer()
try:
detected_delimiter = sniffer.sniff(sample).delimiter
except:
detected_delimiter = delimiter
reader = csv.reader(f, delimiter=detected_delimiter)
table_data = [row for row in reader]
# Delimiter is considered correct if data is read successfully and there are multiple columns
if table_data and len(table_data[0]) > 1:
break
except Exception as e:
logger.debug(f"Try Delimiter '{delimiter}' Kalah: {str(e)}")
continue
if not table_data:
# If all delimiters fail, try a simple line-by-line read
with open(csv_path, "r", encoding=encoding) as f:
lines = f.readlines()
table_data = [[line.strip()] for line in lines if line.strip()]
# Cleanup data - Fully preserve the original table structure, including empty rows
cleaned_data = []
for row in table_data:
cleaned_row = [str(cell).strip() if cell is not None else "" for cell in row]
# Removes blank row filtering, leaving the original table structure intact
cleaned_data.append(cleaned_row)
logger.info(f"Successfully parsedCSVDoc.: {csv_path}, # of Lines: {len(cleaned_data)}")
return cleaned_data
except Exception as e:
logger.error(f"analyzingCSVFile failed: {csv_path}, Error-free: {str(e)}")
return [["CSVFile parsing failed", str(e)]]
def _excel_to_table(self, excel_path: str) -> List[List[str]]:
"""
will beExcelConvert files to tabular data, preserving data formats and types
Args:
excel_path: ExcelFilePath
Returns:
List[List[str]]: Form Data
"""
try:
# read outExcelfiles, keeping more of the original formatting
df = pd.read_excel(excel_path, sheet_name=0, dtype=str, keep_default_na=False)
# Limit table size to avoid overly large table effectsWordDocumentation
max_rows = 500 # Max row count
max_cols = 20 # Maximum Columns
if len(df) > max_rows:
logger.warning(f"ExcelToo many file lines({len(df)}), before interception{max_rows}Parade")
df = df.head(max_rows)
if len(df.columns) > max_cols:
logger.warning(f"ExcelToo many file columns({len(df.columns)}), before interception{max_cols}column")
df = df.iloc[:, :max_cols]
# Convert to list format
table_data = []
# Process header, clean up column names
headers = []
for col in df.columns:
col_str = str(col).strip()
# Medical Treatment:ExcelAutomatically generated column names (e.g.Unnamed: 0）
if col_str.startswith("Unnamed:"):
col_str = "" # Empty column name
headers.append(col_str)
table_data.append(headers)
# Add data rows, smart formatting
for _, row in df.iterrows():
row_data = []
for cell in row:
cell_str = str(cell).strip()
# Handle Empty Values
if cell_str.lower() in ["nan", "none", "null", ""]:
cell_str = ""
# Handle long numbers for readability
elif cell_str.replace(".", "").replace("-", "").isdigit():
try:
# Try formatting numbers
if "." in cell_str:
# Float, keep reasonable decimal places
num = float(cell_str)
if abs(num) >= 1000:
cell_str = f"{num:,.2f}" # Range Slider Thousand Separator
else:
cell_str = f"{num:.2f}".rstrip("0").rstrip(".")
else:
# integer, adding thousands separator
num = int(cell_str)
if abs(num) >= 1000:
cell_str = f"{num:,}"
except ValueError:
pass # Keep original string
# Limit cell content length to avoid too long content affecting layout
if len(cell_str) > 100:
cell_str = cell_str[:97] + "..."
row_data.append(cell_str)
table_data.append(row_data)
# Make sure all rows have the same number of columns
if table_data:
max_cols = max(len(row) for row in table_data)
for row in table_data:
while len(row) < max_cols:
row.append("")
logger.info(f"Successfully parsedExcelDoc.: {excel_path}, size: {len(table_data)}Parade x {max_cols}column")
return table_data
except Exception as e:
logger.error(f"analyzingExcelFile failed: {excel_path}, Error-free: {str(e)}")
return [["ExcelFile parsing failed", str(e)]]
def _markdown_table_to_data(self, markdown_table: str) -> Tuple[List[List[str]], List[str]]:
"""
will beMarkdownTable to table data and parse alignment information
Args:
markdown_table: MarkdownTable Text
Returns:
tuple: (Form Data, Alignment Info List)
"""
try:
lines = [line.strip() for line in markdown_table.strip().split("\n") if line.strip()]
if len(lines) < 2:
logger.warning("MarkdownTable format is incomplete, at least table header and delimiter lines are required")
return [["Format salah.", "The form is incomplete."]], ["left"]
table_data = []
alignments = []
separator_found = False
for i, line in enumerate(lines):
# Check if it is a delimiter line
if self._is_separator_line(line):
alignments = self._parse_alignments(line)
separator_found = True
continue
# Parse Data Rows
cells = self._parse_table_row(line)
if cells:
# Clean cell contents
cleaned_cells = []
for cell in cells:
cleaned_cell = self._clean_cell_content(cell)
cleaned_cells.append(cleaned_cell)
table_data.append(cleaned_cells)
# Validate table structure
if not separator_found:
logger.warning("MarkdownTable is missing a delimiter row, use default left alignment")
alignments = ["left"] * (len(table_data[0]) if table_data else 1)
# Make sure all rows have the same number of columns
if table_data:
max_cols = max(len(row) for row in table_data)
for row in table_data:
while len(row) < max_cols:
row.append("")
# Make sure the number of aligned messages matches the number of columns
while len(alignments) < max_cols:
alignments.append("left")
alignments = alignments[:max_cols]
logger.info(f"Successfully parsedMarkdownTable, Size: {len(table_data)}Parade x {len(alignments)}column")
logger.info(f"Column alignment: {alignments}")
return table_data, alignments
except Exception as e:
logger.error(f"analyzingMarkdownTable failed: {str(e)}")
return [["MarkdownTable parsing failed", str(e)]], ["left"]
def _is_separator_line(self, line: str) -> bool:
"""
Determine if yesMarkdownTable Separator Row
Args:
line: Table Row Content
Returns:
bool: Is Separator Row
"""
# Remove leading and trailing|Symbols
content = line.strip().strip("|").strip()
if not content:
return False
# The delimiter line should mainly contain-And:characters.
cells = [cell.strip() for cell in content.split("|")]
for cell in cells:
if not cell:
continue
# Each cell should be predominantly composed of-And:Composition
clean_cell = cell.replace("-", "").replace(":", "").strip()
if clean_cell: # Not a delimiter line if there are other characters
return False
return True
def _parse_alignments(self, separator_line: str) -> List[str]:
"""
Parse column alignment from delimiter rows
Args:
separator_line: Separator Row
Returns:
List[str]: Alignment List ("left", "center", "right")
"""
alignments = []
content = separator_line.strip().strip("|").strip()
cells = [cell.strip() for cell in content.split("|")]
for cell in cells:
if not cell:
alignments.append("left")
continue
# Determine alignment
if cell.startswith(":") and cell.endswith(":"):
alignments.append("center") # :---:
elif cell.endswith(":"):
alignments.append("right") # ---:
else:
alignments.append("left") # --- OR :---
return alignments
def _parse_table_row(self, line: str) -> List[str]:
"""
Parse table rows, handle escapes and special characters
Args:
line: Table Row Content
Returns:
List[str]: Cell Contents List
"""
# Remove leading and trailing|Symbols
content = line.strip()
if content.startswith("|"):
content = content[1:]
if content.endswith("|"):
content = content[:-1]
# Split cells, but handle escapes|
cells = []
current_cell = ""
escaped = False
for char in content:
if escaped:
current_cell += char
escaped = False
elif char == "\\":
escaped = True
current_cell += char
elif char == "|":
cells.append(current_cell.strip())
current_cell = ""
else:
current_cell += char
# Add last cell
if current_cell or cells: # Process Empty Rows
cells.append(current_cell.strip())
return cells
def _clean_cell_content(self, cell: str) -> str:
"""
Clean up cell contents, processMarkdownFormat
Args:
cell: Raw Cell Contents
Returns:
str: Post-Cleanup Content
"""
if not cell:
return ""
# Remove extra spaces
cleaned = cell.strip()
# Medical Treatment:MarkdownFormatting tags (simplified processing)
# Remove bold mark
cleaned = cleaned.replace("**", "")
# Remove italic mark
cleaned = cleaned.replace("*", "")
# Remove code marker
cleaned = cleaned.replace("`", "")
# Working with link formats [text](url) -> text
import re
cleaned = re.sub(r"\[([^\]]+)\]\([^\)]+\)", r"\1", cleaned)
# Limit length
if len(cleaned) > 100:
cleaned = cleaned[:97] + "..."
return cleaned
def _insert_markdown_table(self, paragraph, table_data: List[List[str]], alignments: List[str]):
"""
Insert ExclusivelyMarkdownTable with support for alignment information
Args:
paragraph: WordParagraph object
table_data: Form Data
alignments: Alignment Info List
"""
try:
if not table_data:
paragraph.add_run("Table data is empty")
return
rows = len(table_data)
cols = len(alignments)
# Simplify processing: insert a table directly at the paragraph position
paragraph_element = paragraph._element
paragraph_parent = paragraph_element.getparent()
# Create Table
table = self.doc.add_table(rows=rows, cols=cols)
table_element = table._tbl
# Insert table directly after paragraph
paragraph_index = list(paragraph_parent).index(paragraph_element)
paragraph_parent.insert(paragraph_index + 1, table_element)
# PengaturanMarkdownTable-specific styles
try:
table.style = "Light List - Accent 1" # Refreshing list style
except Exception:
try:
table.style = "Table Grid" # Alternative style
except Exception:
pass
# Populate table data and set alignment
for i, row_data in enumerate(table_data):
for j, cell_data in enumerate(row_data):
if j < cols:
cell = table.cell(i, j)
cell.text = str(cell_data)
# accordingMarkdownAlignment Info Set Cell Alignment
cell_paragraphs = cell.paragraphs
if cell_paragraphs and j < len(alignments):
alignment = alignments[j]
if alignment == "center":
cell_paragraphs[0].alignment = 1 # Center
elif alignment == "right":
cell_paragraphs[0].alignment = 2 # Halign Right
else: # left
cell_paragraphs[0].alignment = 0 # Align left
# MarkdownTable style: The first row is usually the header
if rows > 0:
for cell in table.rows[0].cells:
# Header Style
for paragraph in cell.paragraphs:
for run in paragraph.runs:
run.bold = True
# Set header background color (light gray, better forMarkdownSTYLE #
try:
from docx.oxml import parse_xml
shading_elm = parse_xml(
r''.format(
'xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"'
)
)
cell._tc.get_or_add_tcPr().append(shading_elm)
except Exception:
pass
# Set Table Layout
try:
table.autofit = True
from docx.shared import Inches
table.width = Inches(6.5)
except Exception:
pass
# Set a more compact line height (MarkdownSTYLE #
try:
for row in table.rows:
row.height = Inches(0.25) # .ExcelForms are more compact
for cell in row.cells:
# Smaller margin
cell.margin_left = Inches(0.03)
cell.margin_right = Inches(0.03)
cell.margin_top = Inches(0.01)
cell.margin_bottom = Inches(0.01)
except Exception:
pass
logger.info(f"Succesfully insertedMarkdownTable, Size: {rows}x{cols}Align: {alignments}")
except Exception as e:
paragraph.add_run(f"Table insertion failed: {str(e)}")
logger.error(f"InsertMarkdownTable failed: {str(e)}")
def _insert_table_at_position(self, paragraph, table_data: List[List[str]]):
"""
Insert a table at the current position of the paragraph, and the table will appear before the paragraph
Args:
paragraph: WordParagraph object
table_data: Form Data
"""
try:
if not table_data:
paragraph.add_run("[Table data is empty]")
return
rows = len(table_data)
cols = max(len(row) for row in table_data) if table_data else 1
# Get the position of the paragraph in the document
paragraph_element = paragraph._element
paragraph_parent = paragraph_element.getparent()
paragraph_index = list(paragraph_parent).index(paragraph_element)
# Create Table
table = self.doc.add_table(rows=rows, cols=cols)
table_element = table._tbl
# Insert table before paragraph (so the table appears in the current content position)
paragraph_parent.insert(paragraph_index, table_element)
# Fill in the table data and style it
self._fill_and_style_table(table, table_data)
except Exception as e:
paragraph.add_run(f"[Table insertion failed: {str(e)}]")
logger.error(f"Failed to insert table: {str(e)}")
def _insert_table(self, paragraph, table_data: List[List[str]]):
"""
Insert high quality table after paragraph
Args:
paragraph: WordParagraph object
table_data: Form Data
"""
try:
if not table_data:
paragraph.add_run("[Table data is empty]")
return
# Add table after paragraph
rows = len(table_data)
cols = max(len(row) for row in table_data) if table_data else 1
# Simplify processing: insert a table directly at the paragraph position
paragraph_element = paragraph._element
paragraph_parent = paragraph_element.getparent()
# Create Table
table = self.doc.add_table(rows=rows, cols=cols)
table_element = table._tbl
# Insert table directly after paragraph
paragraph_index = list(paragraph_parent).index(paragraph_element)
paragraph_parent.insert(paragraph_index + 1, table_element)
# Fill in the table data and style it
self._fill_and_style_table(table, table_data)
except Exception as e:
paragraph.add_run(f"[Table insertion failed: {str(e)}]")
logger.error(f"Failed to insert table: {str(e)}")
def _fill_and_style_table(self, table, table_data: List[List[str]]):
"""
Fill in the table data and style it
Args:
table: WordTable Object
table_data: Form Data
"""
rows = len(table_data)
# Calculate the maximum number of columns, taking into account possible empty rows
cols = max((len(row) for row in table_data if row), default=1)
# Set a more professional table style
try:
# Try a better built-in style
table.style = "Light Shading - Accent 1" # Light Shadow Style
except Exception:
try:
table.style = "Table Grid" # Alternative style
except Exception:
pass # If the style does not exist, use the default style
# Populate table data
for i, row_data in enumerate(table_data):
# Ensure that empty lines can also be handled correctly
current_row = row_data if row_data else []
# Populate all columns of the row
for j in range(cols):
cell = table.cell(i, j)
if j < len(current_row):
cell_data = current_row[j]
else:
cell_data = "" # Empty Cells
cell.text = str(cell_data)
# Set cell alignment
cell_paragraphs = cell.paragraphs
if cell_paragraphs:
# Numbers aligned right, text aligned left
cell_text = str(cell_data).strip()
if self._is_number(cell_text):
cell_paragraphs[0].alignment = 2 # Halign Right
else:
cell_paragraphs[0].alignment = 0 # Align left
# Set header style (first line)
if rows > 0:
for cell in table.rows[0].cells:
# Align Header Centered
for paragraph in cell.paragraphs:
paragraph.alignment = 1 # Center Alignment
for run in paragraph.runs:
run.bold = True
# Try setting the header background color (if supported)
try:
from docx.oxml import parse_xml
shading_elm = parse_xml(
r''.format(
'xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"'
)
)
cell._tc.get_or_add_tcPr().append(shading_elm)
except Exception:
pass # If setting the background color fails, continue
# Auto Fit Column Width
try:
table.autofit = True
# Set table width to page width
from docx.shared import Inches
table.width = Inches(6.5) # ca.A4Width (%)
except Exception:
pass
# Set row height and cell margin
try:
for row in table.rows:
row.height = Inches(0.3) # Set Row Heights
for cell in row.cells:
# Set cell margin
cell.margin_left = Inches(0.05)
cell.margin_right = Inches(0.05)
cell.margin_top = Inches(0.02)
cell.margin_bottom = Inches(0.02)
except Exception:
pass
logger.info(f"Successfully set table style and data, size: {rows}x{cols}")
def _create_table_element(self, table_data: List[List[str]]):
"""
Create a table element
Args:
table_data: Form Data [[row1_col1, row1_col2], [row2_col1, row2_col2]]
Returns:
Form Elements
"""
try:
# Handling completely data-free scenarios: creating minimal tables
if not table_data:
logger.info("Table data isNoneor an empty list to create a minimal table:1x1")
table = self.doc.add_table(rows=1, cols=1)
table.cell(0, 0).text = "" # Empty Cells
return table._tbl
# Calculate table size: keep original structure even if all rows are empty
rows = len(table_data)
if rows == 0:
logger.info("The number of rows in the table is0, create a minimal table:1x1")
table = self.doc.add_table(rows=1, cols=1)
table.cell(0, 0).text = ""
return table._tbl
# Calculate the maximum number of columns, if all rows are empty, the default is1column
cols = max((len(row) for row in table_data if row), default=1)
logger.info(f"Create Table{rows}Parade x {cols}column")
table = self.doc.add_table(rows=rows, cols=cols)
# Fill in the table data and style it
self._fill_and_style_table(table, table_data)
return table._tbl
except Exception as e:
logger.error(f"Form element creation failed: {str(e)}", exc_info=True)
return None
def _create_new_paragraph_after_table(self, parent, table_index, style_info=None):
"""
Create new paragraph after table
Args:
parent: Parent Container
table_index: Index of the table in the parent container
style_info: Style information dictionary for applying to new paragraphs
Returns:
New Paragraph Object
"""
try:
# Create a new paragraph
new_paragraph = self.doc.add_paragraph()
# Apply style information
if style_info:
self._apply_style_info(new_paragraph, style_info)
alignment_info = f"inherit styles, aligning={style_info.get('alignment', 'None')}"
else:
# Back of pocket: set to left alignment
from docx.enum.text import WD_ALIGN_PARAGRAPH
new_paragraph.alignment = WD_ALIGN_PARAGRAPH.LEFT
alignment_info = "Default left alignment"
paragraph_element = new_paragraph._element
# Insert new paragraph after table
parent.insert(table_index + 1, paragraph_element)
return new_paragraph
except Exception as e:
logger.error(f"Failed to create new paragraph: {str(e)}", exc_info=True)
return None
def _extract_paragraph_style_info(self, paragraph):
"""
Extract full style information for paragraphs
Args:
paragraph: Paragraph object
Returns:
dict: Dictionary containing paragraph style information
"""
try:
style_info = {
'alignment': paragraph.alignment,
'paragraph_format': {},
'style_name': None
}
# Extract paragraph formatting information
if hasattr(paragraph, 'paragraph_format'):
pf = paragraph.paragraph_format
style_info['paragraph_format'] = {
'space_before': pf.space_before,
'space_after': pf.space_after,
'line_spacing': pf.line_spacing,
'left_indent': pf.left_indent,
'right_indent': pf.right_indent,
'first_line_indent': pf.first_line_indent,
}
# Extract Style Name
if hasattr(paragraph, 'style') and paragraph.style:
style_info['style_name'] = paragraph.style.name
return style_info
except Exception as e:
logger.warning(f"Failed to extract paragraph style: {str(e)}")
return {'alignment': None, 'paragraph_format': {}, 'style_name': None}
def _apply_style_info(self, paragraph, style_info):
"""
Apply style information to paragraphs
Args:
paragraph: Target Paragraph Object
style_info: Style Information Dictionary
"""
try:
# Apply alignment
if style_info.get('alignment') is not None:
paragraph.alignment = style_info['alignment']
# Apply Paragraph Formatting
paragraph_format = style_info.get('paragraph_format', {})
if paragraph_format:
pf = paragraph.paragraph_format
for attr_name, value in paragraph_format.items():
if value is not None and hasattr(pf, attr_name):
try:
setattr(pf, attr_name, value)
except Exception:
pass # Ignore formatting application failed
# Apply style name
style_name = style_info.get('style_name')
if style_name:
try:
paragraph.style = style_name
except Exception:
pass # Ignore style application failed
except Exception as e:
logger.warning(f"Failed to apply style information: {str(e)}")
def _is_number(self, text: str) -> bool:
"""
Determine if the text is a number
Args:
text: Text to judge
Returns:
bool: Is numeric
"""
if not text:
return False
# Remove thousands separator and spaces
clean_text = text.replace(",", "").replace(" ", "").replace("%", "")
try:
float(clean_text)
return True
except ValueError:
return False
def render(self, template_def, resources: Dict[str, List[Dict[str, Any]]] = None):
"""
Render templates for image and table insertion
Args:
template_def: Template Definition List
resources: A dictionary of resource information, including pictures,Exceldocumentation andMarkdownTable Filter
"""
doc = self.doc
# If no incomingresources, using an empty dictionary
if resources is None:
resources = {"images": [], "excel_files": [], "csv_files": [], "markdown_tables": []}
# Create placeholder-to-resource mapping
placeholder_map = {}
# Image placeholder mapping
for img_info in resources.get("images", []):
placeholder_map[img_info["placeholder"]] = {
"type": "image",
"path": img_info["local_path"],
"alt_text": img_info["alt_text"],
"resource_type": img_info["type"],
}
# ExcelFile placeholder mapping
for excel_info in resources.get("excel_files", []):
placeholder_map[excel_info["placeholder"]] = {
"type": "excel",
"path": excel_info["local_path"],
"resource_type": excel_info["type"],
"table_data": excel_info.get("table_data", []), # Tambahtable_dataData field
"alignments": excel_info.get("alignments", None), # TambahalignmentsData field
}
# CSVFile placeholder mapping
for csv_info in resources.get("csv_files", []):
placeholder_map[csv_info["placeholder"]] = {
"type": "csv",
"file_name": csv_info.get("file_name", "UnknownCSVDoc."),
"table_data": csv_info.get("table_data", []),
"resource_type": csv_info.get("type", "content"),
}
# MarkdownTable placeholder mapping
for table_info in resources.get("markdown_tables", []):
placeholder_map[table_info["placeholder"]] = {"type": "markdown_table", "content": table_info["content"]}
# Original Text Replacement Logic
for replace_info in template_def:
k1 = replace_info[0]
v1 = replace_info[1]
# Work with placeholders in tables
for table in doc.tables:
for i, row in enumerate(table.rows):
for j, cell in enumerate(row.cells):
if k1 in cell.text:
for one in cell.paragraphs:
if k1 in one.text:
# Checks for placeholders requiring special handling
cell_text = one.text.replace(k1, v1)
# Processing Placeholders
for placeholder, resource_info in placeholder_map.items():
if placeholder in cell_text:
if resource_info["type"] == "image":
# Insert Actual Picture in Table Cell
image_path = resource_info.get("local_path") or resource_info.get(
"path", "")
if image_path and os.path.exists(image_path):
try:
# Clear cell text
cell_text = cell_text.replace(placeholder, "")
# Insert Picture in Cell
self._insert_image_in_table_cell(cell, image_path)
logger.info(f"✅ Picture successfully inserted in table cell: {image_path}")
except Exception as e:
logger.error(f"❌ Table Cell Insert Picture Failed: {str(e)}")
# Show file name on failure
cell_text = cell_text.replace(placeholder,
os.path.basename(image_path))
else:
# Image file does not exist, display path
cell_text = cell_text.replace(placeholder, resource_info.get("path",
placeholder))
elif resource_info["type"] == "excel":
cell_text = cell_text.replace(placeholder, "[ExcelTable Filter]")
elif resource_info["type"] == "csv":
cell_text = cell_text.replace(placeholder, "[CSVTable Filter]")
elif resource_info["type"] == "markdown_table":
cell_text = cell_text.replace(placeholder, "[MarkdownTable Filter]")
one.runs[0].text = cell_text
for r_index, r in enumerate(one.runs):
if r_index == 0:
continue
r.text = ""
# Processing Placeholders in Paragraphs
for p in doc.paragraphs:
if k1 not in p.text:
continue
runs_cnt = len(p.runs)
s_e = []
i = 0
while i < runs_cnt:
new_i = i + 1
for j in range(i + 1, runs_cnt + 1):
part_text = "".join([r.text for r in p.runs[i:j]])
if k1 in part_text:
# Find the smallest range containingk1right of privacyruns
tmp_i, tmp_j = i, j
while tmp_i <= tmp_j:
tmp_part_text = "".join([r.text for r in p.runs[tmp_i:tmp_j]])
if k1 in tmp_part_text:
tmp_i += 1
continue
else:
tmp_i -= 1
break
s_e.append((tmp_i, j))
new_i = j
break
i = new_i
for one in s_e:
s, e = one
assert e > 0, [r.text for r in p.runs]
if e - s == 1:
replace_mapping = [(k1, v1)]
elif e - s == 2:
s_tgt_text = p.runs[s].text
comm_str, max_num = find_lcs(k1, s_tgt_text)
assert k1.startswith(comm_str)
p1 = comm_str
p2 = k1[max_num:]
n = len(v1)
sub_n1 = int(1.0 * len(p1) / (len(p1) + len(p2)) * n)
replace_mapping = [(p1, v1[:sub_n1]), (p2, v1[sub_n1:])]
elif e - s == 3:
m_text = p.runs[s + 1].text
head_tail = k1.split(m_text, 1)
assert len(head_tail) == 2
h_text = head_tail[0]
t_text = head_tail[1]
replace_mapping = [(h_text, ""), (m_text, v1), (t_text, "")]
else:
m_texts = [p.runs[i].text for i in range(s + 1, e - 1)]
m_text = "".join(m_texts)
head_tail = k1.split(m_text, 1)
assert len(head_tail) == 2
h_text = head_tail[0]
t_text = head_tail[1]
replace_mapping = [(h_text, "")]
replace_mapping.append((m_texts[0], v1))
for text in m_texts[1:]:
replace_mapping.append((text, ""))
replace_mapping.append((t_text, ""))
for i in range(s, e):
_k, _v = replace_mapping[i - s]
p.runs[i].text = p.runs[i].text.replace(_k, _v)
# Unify resource placeholders after all variable replacements are complete
self._process_resource_placeholders(doc, placeholder_map)
# Add Final Document Content Check
self._log_final_document_content(doc)
return doc
def _log_final_document_content(self, doc):
"""Review final document content"""
try:
# Check for unprocessedfile contentlabel
for i, paragraph in enumerate(doc.paragraphs):
paragraph_text = paragraph.text.strip()
if paragraph_text and 'file content' in paragraph_text.lower():
logger.warning(f"Unprocessed foundfile contentLabel, paragraph{i}: {paragraph_text[:200]}...")
except Exception as e:
logger.error(f"Failed to check document content: {str(e)}")
def test_replace_string(template_file, kv_dict: dict, file_name: str):
# If the file is a web path, download it to a temporary file, and use that
if not os.path.isfile(template_file) and _is_valid_url(template_file):
r = requests.get(template_file)
if r.status_code != 200:
raise ValueError("Check the url of your file; returned status code %s" % r.status_code)
temp_dir = tempfile.TemporaryDirectory()
temp_file = Path(temp_dir.name) / unquote(urlparse(template_file).path.split("/")[-1])
with open(temp_file, mode="wb") as f:
f.write(r.content)
template_file = temp_file
elif not os.path.isfile(template_file):
raise ValueError("File path %s is not a valid file or url" % template_file)
template_dict = []
for k, v in kv_dict.items():
template_dict.append(["{{" + k + "}}", v])
doc = DocxTemplateRender(str(template_file))
output = doc.render(template_dict)
temp_dir = tempfile.TemporaryDirectory()
temp_file = Path(temp_dir.name) / file_name
output.save(temp_file)
minio_client = get_minio_storage_sync()
minio_client.put_object_sync(bucket_name=minio_client.bucket, object_name=file_name, file=temp_file)
return file_name
================================================
FILE: src/backend/bisheng/utils/exceptions.py
================================================
class IgnoreException(Exception):
def __init__(self, message):
self.message = message
super().__init__(self.message)
def __str__(self):
return self.message
# For business exception throwing error message
class MessageException(Exception):
def __init__(self, message):
self.message = message
super().__init__(self.message)
def __str__(self):
return self.message
class FileParseException(Exception):
def __init__(self, message):
self.message = message
super().__init__(self.message)
def __str__(self):
return self.message
class EtlException(Exception):
def __init__(self, message):
self.message = message
super().__init__(self.message)
def __str__(self):
return self.message
================================================
FILE: src/backend/bisheng/utils/http_middleware.py
================================================
# Define a custom middleware class
from time import time
from fastapi import Request
from loguru import logger
from starlette.middleware.base import BaseHTTPMiddleware, RequestResponseEndpoint
from bisheng.core.logger import trace_id_generator, trace_id_var
from bisheng.utils import get_request_ip
class CustomMiddleware(BaseHTTPMiddleware):
"""Cutting Procedure"""
async def dispatch(self, request: Request, call_next: RequestResponseEndpoint):
# You can modify the request before passing it to the next middleware or endpoint
if request.headers.get('x-trace-id'):
trace_id = request.headers.get('x-trace-id')
else:
trace_id = trace_id_generator()
# WithNginx choose one of two Gotta see.NGINX Configuration of
ip = get_request_ip(request)
path = request.url
trace_id_var.set(trace_id)
logger.info(f"| {ip} | {request.method} {path}")
start_time = time()
response = await call_next(request)
process_time = round(time() - start_time, 4)
response.headers["X-Process-Time"] = str(process_time)
response.headers["X-Trace-ID"] = trace_id
logger.info(f"| {ip} | {request.method} {path} | process_time={process_time}s")
return response
class WebSocketLoggingMiddleware:
"""WebSocket Log middleware"""
def __init__(self, app):
self.app = app
async def __call__(self, scope, receive, send):
if scope["type"] == "websocket":
trace_id = trace_id_generator()
trace_id_var.set(trace_id)
await self.app(scope, receive, send)
================================================
FILE: src/backend/bisheng/utils/lazy_load.py
================================================
class LazyLoadDictBase:
def __init__(self):
self._all_types_dict = None
@property
def all_types_dict(self):
if self._all_types_dict is None:
self._all_types_dict = self._build_dict()
return self._all_types_dict
def _build_dict(self):
raise NotImplementedError
def get_type_dict(self):
raise NotImplementedError
================================================
FILE: src/backend/bisheng/utils/linked_list.py
================================================
# -*- coding: utf-8 -*-
from __future__ import print_function
class DoubleNode(object):
"""node_amb"""
def __init__(self, data):
# Identity Data Domain
self.data = data
# Identify the previous linked domain
self.prev = None
# Identify the next linked domain
self.next = None
class DoubleLinkList(object):
"""Double linked watch"""
def head(self):
return self.__head
def __init__(self, node=None):
# Private Attribute Header Node
self.__head = node
# is_empty() Is the linked list empty
def is_empty(self):
return self.__head is None
# length() Link List Length
def length(self):
count = 0 # Number
# Current Node
current = self.__head
while current is not None:
count += 1
# Move the current node backward
current = current.next
return count
# travel() Traverse the entire linked list
def travel(self):
# Current node accessed
current = self.__head
print('[ ', end='')
while current is not None:
print(current.data, end=' ')
current = current.next
print(']')
# add(item) Adding elements to the head of a linked list
def add(self, item):
node = DoubleNode(item)
# The next node of the new node is the head node of the old linked list
node.next = self.__head
# The header node of the new linked list is the new node
self.__head = node
# The previous node of the next node points to the newly added node
node.next.prev = node
# append(item) Add an element at the end of the linked list
def append(self, item):
node = DoubleNode(item)
if self.is_empty():
# When empty node
self.__head = node
else:
# Let the pointer point to the last node
current = self.__head
while current.next is not None:
current = current.next
# The next of the last node is the newly addednode
current.next = node
# Newly added node The previous node is the current node
node.prev = current
# search(item) Find out if the node exists
def search(self, item):
# Current Node
current = self.__head
while current is not None:
if current.data == item:
# Found it.
return True
else:
current = current.next
return False
def find(self, item):
# Current Node
current = self.__head
while current is not None:
if current.data == item:
# Found it.
return current
else:
current = current.next
return None
# insert(index, item) Specify location (from0Start) Add Element
def insert(self, index, item):
if index <= 0:
# Insert Before
self.add(item)
elif index > (self.length() - 1):
# Add at the end
self.append(item)
else:
# Create a new section
node = DoubleNode(item)
current = self.__head
# Number of traversals
count = 0
# Find the previous node of the inserted node
while count < index:
count += 1
current = current.next
# The next node of the new node points to the current node
node.next = current
# The previous node of the new node points to the previous node of the current node
node.prev = current.prev
# The next node of the previous node of the current node points to the new node
current.prev.next = node
# The previous node of the current node points to the new node
current.prev = node
# remove(item) Delete node
def remove(self, item):
current = self.__head
while current is not None:
if current.data == item:
# Locate the node element you want to delete
if current == self.__head:
# Head Node
self.__head = current.next
if current.next:
# If there is not only one node left,
current.next.prev = None
else:
# The next node of the previous node of the current node points to the next node of the current node
current.prev.next = current.next
if current.next:
# If the last element is not deleted, the previous node of the next node of the current node points to the previous node of the current node
current.next.prev = current.prev
return # Return to current node
else:
# Nothing found, move backward
current = current.next
if __name__ == '__main__':
print('test:')
double_link_list = DoubleLinkList()
for i in range(10):
double_link_list.append(str(i)+":hello")
print('--------Determine if it is empty-------')
print(double_link_list.is_empty())
node = double_link_list.find('0:hello')
t = {}
t["abc"]="abc"
t["abc1"]="abc1"
print("{}",node.next.data)
print('-----------Longitudinal------------')
print(double_link_list.length())
double_link_list.append(2)
double_link_list.append(3)
double_link_list.append(5)
print('-----------Step Through------------')
double_link_list.travel()
double_link_list.add(1)
print('-----------Step Through------------')
double_link_list.travel()
double_link_list.add(0)
print('-----------Step Through------------')
double_link_list.travel()
double_link_list.insert(4, 4)
print('-----------Step Through------------')
double_link_list.travel()
double_link_list.insert(-1, -1)
print('-----------Step Through------------')
double_link_list.travel()
print('-----------Cari------------')
print(double_link_list.search(4))
print('-----------Delete------------')
double_link_list.remove(5)
double_link_list.remove(-1)
print('-----------Step Through------------')
double_link_list.travel()
print('-----------Longitudinal------------')
print(double_link_list.length())
================================================
FILE: src/backend/bisheng/utils/mask_data.py
================================================
import json
import re
from copy import deepcopy
from typing import Dict
class JsonFieldMasker:
def __init__(self):
# Define sensitive fields and corresponding desensitization rules
self.sensitive_fields = {
'phone': self.mask_phone,
'mobile': self.mask_phone,
'phone_number': self.mask_phone,
'tel': self.mask_phone,
'id_card': self.mask_id_card,
'identity_card': self.mask_id_card,
'email': self.mask_email,
'password': self.mask_password,
'credit_card': self.mask_credit_card,
'bank_card': self.mask_bank_card,
'address': self.mask_address,
'api_key': self.mask_api_key,
'api_secret': self.mask_api_key,
'openai_api_key': self.mask_api_key,
'siliconflow_api_key': self.mask_api_key,
'email_password': self.mask_api_key,
'jina_api_key': self.mask_api_key,
'app_secret': self.mask_api_key,
}
# Record regular expressions for desensitization mode
self.mask_patterns = {
'phone': r'^\d{3}\*{4}\d{4}$',
'id_card': r'^\d{6}\*{8}\d{4}$|^\d{3}\*{4}\d{3}\*{3}$',
'email': r'^[^*@]+\*[^*@]*@',
'password': r'^\*+$',
'credit_card': r'^\d{4}\*{8,12}\d{4}$',
'bank_card': r'^\d{4}\*{8,12}\d{4}$',
'name': r'^[^*]+\*+[^*]*$',
'address': r'^[^*]+\*+[^*]*$',
'api_key': r'^\*+$',
'api_secret': r'^\*+$',
'openai_api_key': r'^\*+$',
'siliconflow_api_key': r'^\*+$',
'email_password': r'^\*+$',
'jina_api_key': r'^\*+$',
'app_secret': r'^\*+$',
}
def mask_api_key(self, api_key: str) -> str:
"""ImmunosuppressionAPIKey"""
if not api_key:
return api_key
return '********'
def mask_phone(self, phone: str) -> str:
"""Desensitized phone number"""
if not phone or len(phone) < 7:
return phone
return phone[:3] + '****' + phone[-4:]
def mask_id_card(self, id_card: str) -> str:
"""Desensitization ID number"""
if not id_card:
return id_card
if len(id_card) == 18: # 18bits ID card
return id_card[:6] + '********' + id_card[-4:]
elif len(id_card) == 15: # 15bits ID card
return id_card[:3] + '****' + id_card[7:10] + '***'
return '*' * 8
def mask_email(self, email: str) -> str:
"""Desensitization email"""
if '@' not in email:
return email
username, domain = email.split('@', 1)
if len(username) > 2:
masked_username = username[0] + '*' * 3 + username[-1]
else:
masked_username = '*' * len(username)
return f'{masked_username}@{domain}'
def mask_password(self, password: str) -> str:
"""Desensitization password"""
return '********'
def mask_credit_card(self, card: str) -> str:
"""Desensitized credit card number"""
if not card or len(card) < 8:
return card
return card[:4] + '*' * (len(card) - 8) + card[-4:]
def mask_bank_card(self, card: str) -> str:
"""Desensitized bank card number"""
if not card or len(card) < 8:
return card
return card[:4] + '*' * (len(card) - 8) + card[-4:]
def mask_name(self, name: str) -> str:
"""Desensitization Name"""
if not name:
return name
if len(name) == 2:
return name[0] + '*'
elif len(name) > 2:
return name[0] + '*' * (len(name) - 2) + name[-1]
return '*'
def mask_address(self, address: str) -> str:
"""Desensitization Address"""
if not address or len(address) <= 4:
return address
visible_length = min(4, len(address) // 3)
return address[:visible_length] + '****' + address[-visible_length:]
def is_masked_value(self, value: str, field_type: str) -> bool:
"""Determine if the value is already desensitized"""
if not isinstance(value, str):
return False
if field_type in self.mask_patterns:
pattern = self.mask_patterns[field_type]
return bool(re.match(pattern, value))
return False
def mask_json(self, data: Dict) -> Dict:
"""
ImmunosuppressionJSONDATA
Args:
data: OriginalJSONDATA
Returns:
Data after desensitization
"""
if not isinstance(data, dict):
return data
result = {}
for key, value in data.items():
if key in self.sensitive_fields:
mask_func = self.sensitive_fields[key]
if isinstance(value, str):
result[key] = mask_func(value)
elif isinstance(value, dict):
# If the value is a dictionary, recursive processing
result[key] = self.mask_json(value)
else:
# Desensitization after conversion of other types to strings
result[key] = mask_func(str(value))
elif isinstance(value, dict):
# Insensitive fields, recursive processing of nested dictionaries
result[key] = self.mask_json(value)
else:
# Non-sensitive fields, leave as is
result[key] = value
return result
def update_json_with_masked(self, original: Dict, masked: Dict) -> Dict:
"""
After desensitizationJSONUpdate originalJSON
Rule: If the field value after desensitization is still desensitized, it will not be updated;
Otherwise, update the original value with the desensitized value
Args:
original: OriginalJSONDATA
masked: After desensitizationJSONDATA
Returns:
Updated data
"""
if not isinstance(original, dict) or not isinstance(masked, dict):
return masked if isinstance(masked, dict) else original
result = deepcopy(original)
for key, masked_value in masked.items():
if key in result:
original_value = result[key]
if key in self.sensitive_fields:
# Sensitive Fields: Check for desensitization
if isinstance(masked_value, str) and isinstance(original_value, str):
if self.is_masked_value(masked_value, key):
# Still desensitized, don't update
result[key] = original_value
else:
# Restored to desensitized, updated
result[key] = masked_value
elif isinstance(masked_value, dict) and isinstance(original_value, dict):
# Nested dictionaries, recursive processing
result[key] = self.update_json_with_masked(original_value, masked_value)
else:
# Type mismatch, keep original value
result[key] = original_value
else:
# Non-Sensitive Fields
if isinstance(masked_value, dict) and isinstance(original_value, dict):
# Nested dictionaries, recursive processing
result[key] = self.update_json_with_masked(original_value, masked_value)
else:
# Non-nested fields, update directly
result[key] = masked_value
else:
# Keys that are not in the original data, add them directly
result[key] = masked_value
return result
def safe_update_json(self, original_json: str, masked_json: str) -> str:
"""
SafeJSONUpdate: InsightsJSONString, update, re-serialize
Args:
original_json: OriginalJSONString
masked_json: After desensitizationJSONString
Returns:
Post UpdateJSONString
"""
try:
original_data = json.loads(original_json)
masked_data = json.loads(masked_json)
if not isinstance(original_data, dict) or not isinstance(masked_data, dict):
raise ValueError("JSONData must be of object type")
updated_data = self.update_json_with_masked(original_data, masked_data)
return json.dumps(updated_data, ensure_ascii=False, indent=2)
except json.JSONDecodeError as e:
raise ValueError(f"InvalidJSONDATA: {e}")
# Example Use
def main():
masker = JsonFieldMasker()
# OriginalJSONDATA
original_data = {
"user": {
"id": 1,
"name": "Zhang San",
"phone": "13800138000",
"email": "zhangsan@example.com",
"id_card": "110101199001011234",
"bank_card": "6228480402564890018",
"address": "Jianguomenwai Street, Chaoyang District, Beijing1.",
"details": {
"emergency_contact": "Lisi",
"emergency_phone": "13987654321"
}
},
"password": "mysecret123",
"timestamp": "2023-10-01T12:00:00Z"
}
print("OriginalJSONDATA:")
print(json.dumps(original_data, ensure_ascii=False, indent=2))
print("\n" + "=" * 50 + "\n")
# 1. Perform desensitization
masked_data = masker.mask_json(original_data)
print("After desensitizationJSONDATA:")
print(json.dumps(masked_data, ensure_ascii=False, indent=2))
print("\n" + "=" * 50 + "\n")
# 2. Simulated modified desensitization data
modified_masked_data = deepcopy(masked_data)
# Modify some fields
modified_masked_data["user"]["name"] = "Pcs*" # Keep desensitization
modified_masked_data["user"]["phone"] = "13812345678" # Change to a new phone number
modified_masked_data["user"]["email"] = "zhang@newmail.com" # Modify email
modified_masked_data["password"] = "newpassword123" # Change the password
modified_masked_data["timestamp"] = "2023-10-02T12:00:00Z" # Revision timestamp
print("Modified desensitizationJSONDATA:")
print(json.dumps(modified_masked_data, ensure_ascii=False, indent=2))
print("\n" + "=" * 50 + "\n")
# 3. Update raw data with desensitized data
updated_data = masker.update_json_with_masked(original_data, modified_masked_data)
print("Post UpdateJSONDATA:")
print(json.dumps(updated_data, ensure_ascii=False, indent=2))
# Show updated results analysis
print("\n" + "=" * 50)
print("Field Update Situation Analysis:")
print("-" * 30)
def analyze_updates(orig, updated, path=""):
for key in sorted(set(orig.keys()) | set(updated.keys())):
current_path = f"{path}.{key}" if path else key
if key in orig and key in updated:
orig_val = orig[key]
updated_val = updated[key]
if orig_val != updated_val:
is_sensitive = key in masker.sensitive_fields
if is_sensitive:
is_masked = masker.is_masked_value(str(updated_val), key) if isinstance(updated_val,
str) else False
status = "✓ Updated" if not is_masked else "✗ Not updated (still desensitized)"
else:
status = "✓ Updated (non-sensitive field)"
print(f"{current_path}: {status}")
print(f" Original Value: {orig_val}")
print(f" New Value: {updated_val}")
elif isinstance(orig_val, dict) and isinstance(updated_val, dict):
analyze_updates(orig_val, updated_val, current_path)
elif key in updated and key not in orig:
print(f"{current_path}: ✓ Added (new field)")
print(f" New Value: {updated[key]}")
analyze_updates(original_data, updated_data)
# Test StringJSONMedical Treatment:
print("\n" + "=" * 50)
print("Test StringJSONMedical Treatment::")
print("-" * 30)
original_json_str = json.dumps(original_data, ensure_ascii=False)
masked_json_str = json.dumps(modified_masked_data, ensure_ascii=False)
try:
updated_json_str = masker.safe_update_json(original_json_str, masked_json_str)
print("Post UpdateJSONString:")
print(updated_json_str)
except ValueError as e:
print(f"Error-free: {e}")
if __name__ == "__main__":
main()
================================================
FILE: src/backend/bisheng/utils/payload.py
================================================
import contextlib
import re
from typing import Dict
def extract_input_variables(nodes):
"""
Extracts input variables from the template
and adds them to the input_variables field.
"""
for node in nodes:
with contextlib.suppress(Exception):
if 'input_variables' in node['data']['node']['template']:
if node['data']['node']['template']['_type'] == 'prompt':
variables = re.findall(
r'\{(.*?)\}',
node['data']['node']['template']['template']['value'],
)
elif node['data']['node']['template']['_type'] == 'few_shot':
variables = re.findall(
r'\{(.*?)\}',
node['data']['node']['template']['prefix']['value'] +
node['data']['node']['template']['suffix']['value'],
)
else:
variables = []
node['data']['node']['template']['input_variables']['value'] = variables
return nodes
def get_root_node(graph):
"""
Returns the root node of the template.
"""
incoming_edges = {edge.source_id for edge in graph.edges}
input_node = {
edge.source_id
for edge in graph.edges if graph.get_vertex(edge.source_id).base_type == 'inputOutput'
}
if not incoming_edges and len(graph.vertices) == 1:
return graph.vertices[0]
node = {node for node in graph.vertices if node.id not in incoming_edges}
if input_node:
if node:
input_node = input_node.union(node)
return input_node
return node
def get_root_vertex(graph):
"""
Returns the root node of the template.
"""
incoming_edges = {edge.source_id for edge in graph.edges}
if not incoming_edges and len(graph.vertices) == 1:
return graph.vertices[0]
return next((node for node in graph.vertices if node.id not in incoming_edges), None)
def build_json(root, graph) -> Dict:
if 'node' not in root.data:
# If the root node has no "node" key, then it has only one child,
# which is the target of the single outgoing edge
edge = root.edges[0]
local_nodes = [edge.target]
else:
# Otherwise, find all children whose type matches the type
# specified in the template
node_type = root.node_type
local_nodes = graph.get_nodes_with_target(root)
if len(local_nodes) == 1:
return build_json(local_nodes[0], graph)
# Build a dictionary from the template
template = root.data['node']['template']
final_dict = template.copy()
for key, value in final_dict.items():
if key == '_type':
continue
node_type = value['type']
if 'value' in value and value['value'] is not None:
# If the value is specified, use it
value = value['value']
elif 'dict' in node_type:
# If the value is a dictionary, create an empty dictionary
value = {}
else:
# Otherwise, recursively build the child nodes
children = []
for local_node in local_nodes:
node_children = graph.get_children_by_node_type(local_node, node_type)
children.extend(node_children)
if value['required'] and not children:
raise ValueError(f'No child with type {node_type} found')
values = [build_json(child, graph) for child in children]
value = (
list(values) if value['list'] else next(iter(values), None) # type: ignore
)
final_dict[key] = value
return final_dict
================================================
FILE: src/backend/bisheng/utils/threadpool.py
================================================
import asyncio
import concurrent.futures
import threading
import time
from typing import Dict, List, Set, Tuple
from loguru import logger
from bisheng.core.logger import trace_id_var
class ThreadPoolManager:
def __init__(self, max_workers, thread_name_prefix='pool'):
self.thread_group = thread_name_prefix
self.executor = concurrent.futures.ThreadPoolExecutor(
max_workers=max_workers, thread_name_prefix=thread_name_prefix)
# Design one syndication per sync thread
self.future_dict: Dict[str, List[concurrent.futures.Future]] = {}
self.async_task: Dict[str, List[concurrent.futures.Future]] = {}
self.lock = threading.Lock()
def submit(self, key: str, fn, *args, **kwargs):
with self.lock:
if key not in self.future_dict:
self.future_dict[key] = []
if key not in self.async_task:
self.async_task[key] = []
if asyncio.coroutines.iscoroutinefunction(fn):
future = asyncio.create_task(self.acontext_wrapper(fn, *args, **kwargs))
self.async_task[key].append(future)
else:
future = self.executor.submit(self.context_wrapper, fn, *args, **kwargs)
self.future_dict[key].append(future)
return future
async def acontext_wrapper(self, func, *args, **kwargs):
start_wait = time.time()
kwargs.pop('trace_id', '2')
result = await func(*args, **kwargs)
end_wait = time.time() # Time when the task actually started
logger.info(
f'aTask_waited={end_wait - start_wait:.2f} seconds and {func.__name__} executed={time.time() - end_wait:.2f} seconds',
)
return result
def context_wrapper(self, func, *args, **kwargs):
trace_id = kwargs.pop('trace_id', '2')
start_wait = time.time()
trace_id_var.set(trace_id)
result = func(*args, **kwargs)
end_wait = time.time() # Time when the task actually started
logger.info(
f'Task_waited={end_wait - start_wait:.2f} seconds and executed={time.time() - end_wait:.2f} seconds',
)
return result
async def as_completed(self,
key_list: Set[str]) -> List[Tuple[str, concurrent.futures.Future]]:
with self.lock:
completed_futures = []
for k, lf in list(self.future_dict.items()):
for f in lf:
if f.done():
if k in key_list:
completed_futures.append((k, f))
self.future_dict[k].remove(f)
if len(lf) == 0:
self.future_dict.pop(k)
pending_count = 0
for k, lf in list(self.async_task.items()):
for f in lf:
if f.done():
# Dapatkantask
if k in key_list:
completed_futures.append((k, f))
self.async_task[k].remove(f)
else:
pending_count += 1
if len(lf) == 0:
self.async_task.pop(k)
if pending_count > 0:
logger.info('async_wait_count={}', pending_count)
return completed_futures
# async def async_done_callback(self, future):
# self.async_task_result.append(future)
def cancel_task(self, key_list: List[str]):
res = [False] * len(key_list)
with self.lock:
for index, key in enumerate(key_list):
if self.async_task.get(key):
for task in self.async_task.get(key):
cancel_res = task.cancel()
logger.info('clean_pending_task key={} task={} res={}', key, task,
cancel_res)
res[index] = cancel_res
if self.future_dict.get(key):
for task in self.future_dict.get(key):
res.append(task.cancel())
return res
def tear_down(self):
key_list = list(self.async_task.keys())
self.cancel_task(key_list)
self.executor.shutdown(cancel_futures=True)
# Create a thread pool manager
thread_pool = ThreadPoolManager(5)
if __name__ == '__main__':
def wait_(name: str):
logger.info('{} enter wait {}', threading.current_thread(), name)
time.sleep(10)
logger.info('{} done {}', threading.current_thread(), name)
async def await_(name: str = 1):
logger.info('{} enter wait {}', threading.current_thread(), name)
await asyncio.sleep(10)
logger.info('{} done {}', threading.current_thread(), name)
thread_pool.submit('ABB', wait_, 'NO.1')
thread_pool.submit('AAA', await_)
# time.sleep(3)
# thread_pool.submit("AA", wait_, "~~~~~~~~~~~~")
# thread_pool.submit("AA", await_, "NO.3")
# time.sleep(2)
# thread_pool.tear_down(["AAA"])
================================================
FILE: src/backend/bisheng/utils/util.py
================================================
import asyncio
import functools
import hashlib
import importlib
import inspect
import io
import logging
import re
import time
import zipfile
from functools import wraps
from typing import Dict, Optional, Union, List, Tuple
from urllib.parse import urlparse
from bisheng.template.frontend_node.constants import FORCE_SHOW_FIELDS
from bisheng.utils import constants
from docstring_parser import parse # type: ignore
logger = logging.getLogger(__name__)
def build_template_from_function(name: str, type_to_loader_dict: Dict, add_function: bool = False):
classes = [item.__annotations__['return'].__name__ for item in type_to_loader_dict.values()]
# Raise error if name is not in chains
if name not in classes:
raise ValueError(f'{name} not found')
for _type, v in type_to_loader_dict.items():
if v.__annotations__['return'].__name__ == name:
_class = v.__annotations__['return']
# Get the docstring
docs = parse(_class.__doc__)
variables = {'_type': _type}
for class_field_items, value in _class.__fields__.items():
if class_field_items in ['callback_manager']:
continue
variables[class_field_items] = {}
for name_, value_ in value.__repr_args__():
if name_ == 'default_factory':
try:
variables[class_field_items]['default'] = get_default_factory(
module=_class.__base__.__module__, function=value_)
except Exception:
variables[class_field_items]['default'] = None
elif name_ not in ['name']:
variables[class_field_items][name_] = value_
variables[class_field_items]['placeholder'] = (
docs.params[class_field_items] if class_field_items in docs.params else '')
# Adding function to base classes to allow
# the output to be a function
base_classes = get_base_classes(_class)
if add_function:
base_classes.append('function')
return {
'template': format_dict(variables, name),
'description': docs.short_description or '',
'base_classes': base_classes,
}
def build_template_from_class(name: str, type_to_cls_dict: Dict, add_function: bool = False):
classes = [item.__name__ for item in type_to_cls_dict.values()]
# Raise error if name is not in chains
if name not in classes:
raise ValueError(f'{name} not found.')
for _type, v in type_to_cls_dict.items():
if v.__name__ == name:
_class = v
# Get the docstring
docs = parse(_class.__doc__)
variables = {'_type': _type}
if getattr(_class, 'model_fields', None):
for class_field_items, value in _class.model_fields.items():
if class_field_items in ['callback_manager']:
continue
variables[class_field_items] = {}
for name_, value_ in value.__repr_args__():
if name_ == 'default_factory':
try:
variables[class_field_items]['default'] = get_default_factory(
module=_class.__base__.__module__, function=value_)
except Exception:
variables[class_field_items]['default'] = None
elif name_ not in ['name']:
variables[class_field_items][name_] = value_
variables[class_field_items]['placeholder'] = (
docs.params[class_field_items] if class_field_items in docs.params else '')
else:
for name, param in inspect.signature(_class.__init__).parameters.items():
if name == 'self':
continue
variables[name] = {}
variables[name]['default'] = get_default_factory(module=_class.__base__.__module__,
function=str(param.annotation))
variables[name]['annotation'] = str(param.annotation)
variables[name]['required'] = False
base_classes = get_base_classes(_class)
# Adding function to base classes to allow
# the output to be a function
if add_function:
base_classes.append('function')
return {
'template': format_dict(variables, name),
'description': docs.short_description or '',
'base_classes': base_classes,
}
return None
def build_template_from_method(
class_name: str,
method_name: str,
type_to_cls_dict: Dict,
add_function: bool = False,
):
classes = [item.__name__ for item in type_to_cls_dict.values()]
# Raise error if class_name is not in classes
if class_name not in classes:
raise ValueError(f'{class_name} not found.')
for _type, v in type_to_cls_dict.items():
if v.__name__ == class_name:
_class = v
# Check if the method exists in this class
if not hasattr(_class, method_name):
raise ValueError(f'Method {method_name} not found in class {class_name}')
# Get the method
method = getattr(_class, method_name)
# Get the docstring
docs = parse(method.__doc__)
# Get the signature of the method
sig = inspect.signature(method)
# Get the parameters of the method
params = sig.parameters
# Initialize the variables dictionary with method parameters
variables = {
'_type': _type,
**{
name: {
'default': param.default if param.default != param.empty else None,
'type': param.annotation if param.annotation != param.empty else None,
'required': param.default == param.empty,
}
for name, param in params.items() if name not in ['self', 'kwargs', 'args']
},
}
base_classes = get_base_classes(_class)
# Adding function to base classes to allow the output to be a function
if add_function:
base_classes.append('function')
return {
'template': format_dict(variables, class_name),
'description': docs.short_description or '',
'base_classes': base_classes,
}
def get_base_classes(cls):
"""Get the base classes of a class.
These are used to determine the output of the nodes.
"""
if hasattr(cls, '__bases__') and cls.__bases__:
bases = cls.__bases__
result = []
for base in bases:
if any(type in base.__module__ for type in ['pydantic', 'abc']):
continue
result.append(base.__name__)
base_classes = get_base_classes(base)
# check if the base_classes are in the result
# if not, add them
for base_class in base_classes:
if base_class not in result:
result.append(base_class)
else:
result = [cls.__name__]
if not result:
result = [cls.__name__]
return list(set(result + [cls.__name__]))
def get_default_factory(module: str, function: str):
pattern = r''
if match := re.search(pattern, function):
imported_module = importlib.import_module(module)
return getattr(imported_module, match[1])()
return None
def type_to_string(tp):
if getattr(tp, '__args__', None):
args_str = ','.join(type_to_string(arg) for arg in tp.__args__
if arg is not type(None)) # noqa
return f'{tp.__name__}[{args_str}]'
else:
return tp.__name__
def format_dict(d, name: Optional[str] = None):
"""
Formats a dictionary by removing certain keys and modifying the
values of other keys.
Args:
d: the dictionary to format
name: the name of the class to format
Returns:
A new dictionary with the desired modifications applied.
"""
need_remove_key = []
# Process remaining keys
for key, value in d.items():
if key == '_type':
continue
_type = value['type'] if 'type' in value else value['annotation']
if not isinstance(_type, str):
_type = type_to_string(_type)
# Remove 'Optional' wrapper
if 'Optional' in _type:
_type = _type.replace('Optional[', '')[:-1]
value['required'] = False
# Check for list type
if 'List' in _type or 'Sequence' in _type or 'Set' in _type:
_type = (_type.replace('List[', '').replace('Sequence[', '').replace('Set[', '')[:-1])
value['list'] = True
else:
value['list'] = False
# Replace 'Mapping' with 'dict'
if 'Mapping' in _type:
_type = _type.replace('Mapping', 'dict')
# Change type from str to Tool
value['type'] = 'Tool' if key in ['allowed_tools'] else _type
value['type'] = 'int' if key in ['max_value_length'] else value['type']
# Show or not field
value['show'] = bool(
(value['required'] and (key not in ['input_variables'] or name == 'SequentialChain'))
or key in FORCE_SHOW_FIELDS or 'api_key' in key)
# Add password field
value['password'] = any(text in key.lower()
for text in ['password', 'token', 'api', 'key'])
# Add multline
value['multiline'] = key in [
'suffix',
'prefix',
'template',
'examples',
'code',
'headers',
'format_instructions',
]
# Replace dict type with str
if 'dict' in value['type'].lower():
value['type'] = 'code'
if key == 'dict_':
value['type'] = 'file'
value['suffixes'] = ['.json', '.yaml', '.yml']
value['fileTypes'] = ['json', 'yaml', 'yml']
# Replace default value with actual value
if 'default' in value:
value['value'] = value['default']
value.pop('default')
if key == 'headers':
value['value'] = """{'Authorization':
'Bearer '}"""
# Add options to openai
if name == 'OpenAI' and key == 'model_name':
value['options'] = constants.OPENAI_MODELS
value['list'] = True
value['value'] = constants.OPENAI_MODELS[0]
elif name == 'ChatOpenAI' and key == 'model_name':
value['options'] = constants.CHAT_OPENAI_MODELS
value['list'] = True
value['value'] = constants.CHAT_OPENAI_MODELS[0]
elif (name == 'Anthropic' or name == 'ChatAnthropic') and key == 'model_name':
value['options'] = constants.ANTHROPIC_MODELS
value['list'] = True
value['value'] = constants.ANTHROPIC_MODELS[0]
if 'value' in value and type(value['value']) == set:
value['value'] = list(value['value'])
if 'value' in value and inspect.isfunction(value['value']):
need_remove_key.append(key)
for one in need_remove_key:
del d[one]
return d
def update_verbose(d: dict, new_value: bool) -> dict:
"""
Recursively updates the value of the 'verbose' key in a dictionary.
Args:
d: the dictionary to update
new_value: the new value to set
Returns:
The updated dictionary.
"""
for k, v in d.items():
if isinstance(v, dict):
update_verbose(v, new_value)
elif k == 'verbose':
d[k] = new_value
return d
def sync_to_async(func):
"""
Decorator to convert a sync function to an async function.
"""
@wraps(func)
async def async_wrapper(*args, **kwargs):
return func(*args, **kwargs)
return async_wrapper
def run_async(coro, loop=None):
"""
Run asynchronous functions
:param coro:
:param loop:
:return:
"""
if loop is None:
try:
loop = asyncio.get_running_loop()
return loop.run_until_complete(coro)
except RuntimeError:
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
return loop.run_until_complete(coro)
return loop.run_until_complete(coro)
def get_cache_key(flow_id: str, chat_id: str, vertex_id: str = None):
return f'{flow_id}_{chat_id}_{vertex_id}'
def _is_valid_url(url: str) -> bool:
"""Check if the url is valid."""
parsed = urlparse(url)
return bool(parsed.netloc) and bool(parsed.scheme)
# Retry decorator Asynchronous
def retry_async(num_retries=3, delay=0.5, return_exceptions=False):
def wrapper(func):
async def wrapped(*args, **kwargs):
for i in range(num_retries):
try:
return await func(*args, **kwargs)
except Exception as e:
logger.info(
f"Retrying {func.__name__} in {delay} seconds... Attempt {i + 1} of {num_retries}... error: {e}")
if i == num_retries - 1:
if return_exceptions:
# Return Exception Parameters will bee.argsSplit into tuples
return e.args if len(e.args) > 1 else e.args[0]
logger.error(f"Failed to execute {func.__name__} after {num_retries} retries")
raise e
await asyncio.sleep(delay)
return None
return wrapped
return wrapper
# Retry decorator
def retry_sync(num_retries=3, delay=0.5, return_exceptions=False):
def wrapper(func):
def wrapped(*args, **kwargs):
for i in range(num_retries):
try:
return func(*args, **kwargs)
except Exception as e:
logger.info(
f"Retrying {func.__name__} in {delay} seconds... Attempt {i + 1} of {num_retries}... error: {e}")
if i == num_retries - 1:
if return_exceptions:
# Return Exception Parameters will bee.argsSplit into tuples
return e.args if len(e.args) > 1 else e.args[0]
logger.error(f"Failed to execute {func.__name__} after {num_retries} retries")
raise e
time.sleep(delay)
return None
return wrapped
return wrapper
def calculate_md5(file: Union[str, bytes]):
"""Calculating the Document's MD5 .
Returns:
str: of the document MD5 .
"""
md5_hash = hashlib.md5()
if isinstance(file, bytes):
md5_hash.update(file)
return md5_hash.hexdigest()
else:
# Reading Files in Binary Form
with open(file, "rb") as f:
# Read files by block to avoid large files taking up too much memory
for chunk in iter(lambda: f.read(4096), b""):
md5_hash.update(chunk)
return md5_hash.hexdigest()
async def async_calculate_md5(file: Union[str, bytes]):
"""of the asynchronous computation document MD5 .
Returns:
str: of the document MD5 .
"""
import aiofiles
md5_hash = hashlib.md5()
if isinstance(file, bytes):
md5_hash.update(file)
return md5_hash.hexdigest()
else:
# Read files asynchronously in binary form
async with aiofiles.open(file, "rb") as f:
# Read files asynchronously by block to avoid large files taking up too much memory
while True:
chunk = await f.read(4096)
if not chunk:
break
md5_hash.update(chunk)
return md5_hash.hexdigest()
# Read all files in the directory
def read_files_in_directory(path: str):
"""
Reads all files in the directory and returns a list of filenames.
Args:
path (str): Directory Path
Returns:
list: List of filenames.
"""
import os
if not os.path.exists(path):
logger.error(f"Path {path} does not exist.")
return []
files = []
for root, _, filenames in os.walk(path):
for filename in filenames:
files.append(os.path.join(root, filename))
return files
def sync_func_to_async(func, executor=None):
@functools.wraps(func)
async def wrapper(*args, **kwargs):
loop = asyncio.get_running_loop()
bound_func = functools.partial(func, *args, **kwargs)
return await loop.run_in_executor(executor, bound_func)
return wrapper
def bytes_to_zip(
files: List[Tuple[str, bytes]],
compress_level: int = 6
) -> bytes:
"""
Package byte stream data intoZIPfiles, back toZIPByte Stream for Files
Parameters:
files: Contains(The file name, Byte Stream)List of tuples
compress_level: compression level(0-9)，0Indicates no compression,9Represents the highest compression rate
Return:
Date GeneratedZIPFile Byte Stream
"""
try:
# Verify compression level
if not 0 <= compress_level <= 9:
raise ValueError("The compression level must be0to9Between")
# Create in-memory byte streams for storageZIPDATA
zip_buffer = io.BytesIO()
# BuatZIPFile and add byte stream data
with zipfile.ZipFile(
zip_buffer,
'w',
zipfile.ZIP_DEFLATED,
compresslevel=compress_level
) as zipf:
for filename, data in files:
# Enter your messageZIPAdd byte stream data to the file
zipf.writestr(filename, data)
print(f"Was added: {filename} (size: {len(data) / 1024:.2f} KB)")
# will beZIPData is positioned to the starting position and returns a byte stream
zip_buffer.seek(0)
zip_data = zip_buffer.getvalue()
logger.debug(f"\nZIPFile created successfully, total size: {len(zip_data) / 1024:.2f} KB")
return zip_data
except Exception as e:
logger.error(f"Packaging process error: {str(e)}")
raise e
================================================
FILE: src/backend/bisheng/utils/validate.py
================================================
import ast
import contextlib
import importlib
from types import FunctionType
from typing import Dict, List, Optional, Union
from bisheng.field_typing.constants import CUSTOM_COMPONENT_SUPPORTED_TYPES
def add_type_ignores():
if not hasattr(ast, 'TypeIgnore'):
class TypeIgnore(ast.AST):
_fields = ()
ast.TypeIgnore = TypeIgnore
def validate_code(code):
# Initialize the errors dictionary
errors = {'imports': {'errors': []}, 'function': {'errors': []}}
# Parse the code string into an abstract syntax tree (AST)
try:
tree = ast.parse(code)
except Exception as e:
errors['function']['errors'].append(str(e))
return errors
# Add a dummy type_ignores field to the AST
add_type_ignores()
tree.type_ignores = []
# Evaluate the import statements
for node in tree.body:
if isinstance(node, ast.Import):
for alias in node.names:
try:
importlib.import_module(alias.name)
except ModuleNotFoundError as e:
errors['imports']['errors'].append(str(e))
# Evaluate the function definition
for node in tree.body:
if isinstance(node, ast.FunctionDef):
code_obj = compile(ast.Module(body=[node], type_ignores=[]), '', 'exec')
try:
exec(code_obj)
except Exception as e:
errors['function']['errors'].append(str(e))
# Return the errors dictionary
return errors
def eval_function(function_string: str):
# Create an empty dictionary to serve as a separate namespace
namespace: Dict = {}
# Execute the code string in the new namespace
exec(function_string, namespace)
function_object = next(
(
obj
for name, obj in namespace.items()
if isinstance(obj, FunctionType) and obj.__code__.co_filename == ''
),
None,
)
if function_object is None:
raise ValueError('Function string does not contain a function')
return function_object
def execute_function(code, function_name, *args, **kwargs):
add_type_ignores()
module = ast.parse(code)
exec_globals = globals().copy()
for node in module.body:
if isinstance(node, ast.Import):
for alias in node.names:
try:
exec(
f"{alias.asname or alias.name} = importlib.import_module('{alias.name}')",
exec_globals,
locals(),
)
exec_globals[alias.asname or alias.name] = importlib.import_module(alias.name)
except ModuleNotFoundError as e:
raise ModuleNotFoundError(f'Module {alias.name} not found. Please install it and try again.') from e
function_code = next(
node for node in module.body if isinstance(node, ast.FunctionDef) and node.name == function_name
)
function_code.parent = None
code_obj = compile(ast.Module(body=[function_code], type_ignores=[]), '', 'exec')
try:
exec(code_obj, exec_globals, locals())
except Exception as exc:
raise ValueError('Function string does not contain a function') from exc
# Add the function to the exec_globals dictionary
exec_globals[function_name] = locals()[function_name]
return exec_globals[function_name](*args, **kwargs)
def create_function(code, function_name):
if not hasattr(ast, 'TypeIgnore'):
class TypeIgnore(ast.AST):
_fields = ()
ast.TypeIgnore = TypeIgnore
module = ast.parse(code)
exec_globals = globals().copy()
for node in module.body:
if isinstance(node, ast.Import):
for alias in node.names:
try:
exec_globals[alias.asname or alias.name] = importlib.import_module(alias.name)
except ModuleNotFoundError as e:
raise ModuleNotFoundError(f'Module {alias.name} not found. Please install it and try again.') from e
function_code = next(
node for node in module.body if isinstance(node, ast.FunctionDef) and node.name == function_name
)
function_code.parent = None
code_obj = compile(ast.Module(body=[function_code], type_ignores=[]), '', 'exec')
with contextlib.suppress(Exception):
exec(code_obj, exec_globals, locals())
exec_globals[function_name] = locals()[function_name]
# Return a function that imports necessary modules and calls the target function
def wrapped_function(*args, **kwargs):
for module_name, module in exec_globals.items():
if isinstance(module, type(importlib)):
globals()[module_name] = module
return exec_globals[function_name](*args, **kwargs)
return wrapped_function
def create_class(code, class_name):
"""
Dynamically create a class from a string of code and a specified class name.
:param code: String containing the Python code defining the class
:param class_name: Name of the class to be created
:return: A function that, when called, returns an instance of the created class
"""
if not hasattr(ast, 'TypeIgnore'):
ast.TypeIgnore = create_type_ignore_class()
module = ast.parse(code)
exec_globals = prepare_global_scope(code, module)
class_code = extract_class_code(module, class_name)
compiled_class = compile_class_code(class_code)
return build_class_constructor(compiled_class, exec_globals, class_name)
def create_type_ignore_class():
"""
Create a TypeIgnore class for AST module if it doesn't exist.
:return: TypeIgnore class
"""
class TypeIgnore(ast.AST):
_fields = ()
return TypeIgnore
def prepare_global_scope(code, module):
"""
Prepares the global scope with necessary imports from the provided code module.
:param module: AST parsed module
:return: Dictionary representing the global scope with imported modules
"""
exec_globals = globals().copy()
exec_globals.update(get_default_imports(code))
for node in module.body:
if isinstance(node, ast.Import):
for alias in node.names:
try:
exec_globals[alias.asname or alias.name] = importlib.import_module(alias.name)
except ModuleNotFoundError as e:
raise ModuleNotFoundError(f'Module {alias.name} not found. Please install it and try again.') from e
elif isinstance(node, ast.ImportFrom) and node.module is not None:
try:
imported_module = importlib.import_module(node.module)
for alias in node.names:
exec_globals[alias.name] = getattr(imported_module, alias.name)
except ModuleNotFoundError as e:
raise ModuleNotFoundError(f'Module {node.module} not found. Please install it and try again.') from e
return exec_globals
def extract_class_code(module, class_name):
"""
Extracts the AST node for the specified class from the module.
:param module: AST parsed module
:param class_name: Name of the class to extract
:return: AST node of the specified class
"""
class_code = next(node for node in module.body if isinstance(node, ast.ClassDef) and node.name == class_name)
class_code.parent = None
return class_code
def compile_class_code(class_code):
"""
Compiles the AST node of a class into a code object.
:param class_code: AST node of the class
:return: Compiled code object of the class
"""
code_obj = compile(ast.Module(body=[class_code], type_ignores=[]), '', 'exec')
return code_obj
def build_class_constructor(compiled_class, exec_globals, class_name):
"""
Builds a constructor function for the dynamically created class.
:param compiled_class: Compiled code object of the class
:param exec_globals: Global scope with necessary imports
:param class_name: Name of the class
:return: Constructor function for the class
"""
exec(compiled_class, exec_globals, locals())
exec_globals[class_name] = locals()[class_name]
# Return a function that imports necessary modules and creates an instance of the target class
def build_custom_class(*args, **kwargs):
for module_name, module in exec_globals.items():
if isinstance(module, type(importlib)):
globals()[module_name] = module
instance = exec_globals[class_name](*args, **kwargs)
return instance
build_custom_class.__globals__.update(exec_globals)
return build_custom_class
def get_default_imports(code_string):
"""
Returns a dictionary of default imports for the dynamic class constructor.
"""
default_imports = {
'Optional': Optional,
'List': List,
'Dict': Dict,
'Union': Union,
}
langflow_imports = list(CUSTOM_COMPONENT_SUPPORTED_TYPES.keys())
necessary_imports = find_names_in_code(code_string, langflow_imports)
langflow_module = importlib.import_module('bisheng.field_typing')
default_imports.update({name: getattr(langflow_module, name) for name in necessary_imports})
return default_imports
def find_names_in_code(code, names):
"""
Finds if any of the specified names are present in the given code string.
:param code: The source code as a string.
:param names: A list of names to check for in the code.
:return: A set of names that are found in the code.
"""
found_names = {name for name in names if name in code}
return found_names
def extract_function_name(code):
module = ast.parse(code)
for node in module.body:
if isinstance(node, ast.FunctionDef):
return node.name
raise ValueError('No function definition found in the code string')
def extract_class_name(code):
module = ast.parse(code)
for node in module.body:
if isinstance(node, ast.ClassDef):
return node.name
raise ValueError('No class definition found in the code string')
================================================
FILE: src/backend/bisheng/worker/README.md
================================================
# 毕昇异步任务模块
================================================
FILE: src/backend/bisheng/worker/__init__.py
================================================
# register tasks
from bisheng.worker.knowledge.file_worker import file_copy_celery, parse_knowledge_file_celery, \
retry_knowledge_file_celery
from bisheng.worker.knowledge.rebuild_knowledge_worker import rebuild_knowledge_celery
from bisheng.worker.telemetry.mid_table import sync_mid_user_increment, sync_mid_knowledge_increment, \
sync_mid_app_increment, sync_mid_user_interact_dtl
from bisheng.worker.test.test import add
from bisheng.worker.workflow.tasks import execute_workflow, continue_workflow, stop_workflow
================================================
FILE: src/backend/bisheng/worker/config.py
================================================
from bisheng.common.services.config_service import settings
broker_url = settings.celery_redis_url
task_serializer = 'json'
result_serializer = 'json'
accept_content = ['json']
timezone = 'Asia/Shanghai'
enable_utc = False
task_routes = settings.celery_task.task_routers
# redisHealth check interval, unit sec
redis_backend_health_check_interval = 5
beat_schedule = settings.celery_task.beat_schedule
================================================
FILE: src/backend/bisheng/worker/knowledge/__init__.py
================================================
================================================
FILE: src/backend/bisheng/worker/knowledge/file_worker.py
================================================
import json
from typing import List
from loguru import logger
from pymilvus import Collection, MilvusException
from bisheng.api.services.knowledge_imp import decide_vectorstores, process_file_task, delete_knowledge_file_vectors, \
KnowledgeUtils, delete_vector_files
from bisheng.api.v1.schemas import FileProcessBase
from bisheng.common.errcode.knowledge import KnowledgeFileFailedError
from bisheng.core.logger import trace_id_var
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync
from bisheng.interface.embeddings.custom import FakeEmbedding
from bisheng.knowledge.domain.models.knowledge import Knowledge, KnowledgeDao, KnowledgeTypeEnum, KnowledgeState
from bisheng.knowledge.domain.models.knowledge_file import (
KnowledgeFile,
KnowledgeFileDao,
KnowledgeFileStatus,
)
from bisheng.utils import generate_uuid
from bisheng.worker.main import bisheng_celery
from bisheng_langchain.vectorstores import ElasticKeywordsSearch, Milvus
@bisheng_celery.task(acks_late=True)
def file_copy_celery(param: json) -> str:
"""Copy a Knowledge Base file to another Knowledge Base
1. mysqlCopy of
2. Copying Files
3. Replication of vectors
"""
source_knowledge_id = param.get("source_knowledge_id")
target_id = param.get("target_id")
login_user_id = param.get("login_user_id")
logger.info(
"file_copy_celery start source_id={} target_id={}",
source_knowledge_id,
target_id,
)
page_size = 20
page_num = 1
source_knowledge = KnowledgeDao.query_by_id(source_knowledge_id)
target_knowledge = KnowledgeDao.query_by_id(target_id)
target_list: List[KnowledgeFile] = KnowledgeFileDao.get_file_by_condition(
target_id
) # All files
if target_list:
target_list = [t.md5 for t in target_list]
while True:
if source_knowledge.type == KnowledgeTypeEnum.NORMAL.value:
files = KnowledgeFileDao.get_file_by_filters(
source_knowledge_id, page=page_num, page_size=page_size
)
if not files:
# No more documents Buatmilvus collection And es index
create_milvus_col_and_es_index(source_knowledge, target_knowledge)
break
for one in files:
if target_list and one.md5 in target_list:
# Duplicate Tasks Prevent Duplicate Writes
continue
try:
copy_normal(
one,
source_knowledge,
target_knowledge,
login_user_id,
)
except Exception as e:
logger.error(f"copy file error: {one.file_name} {e}")
page_num += 1
if not files or len(files) < page_size:
break
# Recovery status
logger.info("file_copy_celery end")
target_knowledge.state = 1
KnowledgeDao.update_state(knowledge_id=source_knowledge.id, state=KnowledgeState.PUBLISHED,
update_time=source_knowledge.update_time)
KnowledgeDao.update_one(target_knowledge)
return "copy task done"
def copy_normal(
one: KnowledgeFile,
source_knowledge: Knowledge,
target_knowledge: Knowledge,
op_user_id: int,
):
one_dict = one.model_dump()
one_dict.pop("id")
one_dict.pop("update_time")
one_dict["user_id"] = op_user_id
one_dict["knowledge_id"] = target_knowledge.id
one_dict["status"] = KnowledgeFileStatus.PROCESSING.value
source_file_pdf = one.id
source_file = one.object_name
source_file_ext = one.object_name.split('.')[-1]
bbox_file = one.bbox_object_name
knowledge_new = KnowledgeFile(**one_dict)
knowledge_new = KnowledgeFileDao.add_file(knowledge_new)
# migrate file
try:
target_source_file = KnowledgeUtils.get_knowledge_file_object_name(knowledge_new.id, knowledge_new.file_name)
minio_client = get_minio_storage_sync()
# Copy source file
if minio_client.object_exists_sync(minio_client.bucket, source_file):
minio_client.copy_object_sync(source_bucket=minio_client.bucket, source_object=source_file,
dest_object=target_source_file, dest_bucket=minio_client.bucket)
knowledge_new.object_name = target_source_file
# Copy GeneratedpdfDoc.
if minio_client.object_exists_sync(minio_client.bucket, f"{source_file_pdf}"):
minio_client.copy_object_sync(source_bucket=minio_client.bucket, source_object=f"{source_file_pdf}",
dest_object=f"{knowledge_new.id}", dest_bucket=minio_client.bucket)
# Copies:bboxDoc.
if minio_client.object_exists_sync(object_name=bbox_file):
target_bbox_file = KnowledgeUtils.get_knowledge_bbox_file_object_name(knowledge_new.id)
minio_client.copy_object_sync(source_bucket=minio_client.bucket, source_object=bbox_file,
dest_object=target_bbox_file, dest_bucket=minio_client.bucket)
knowledge_new.bbox_object_name = target_bbox_file
preview_file = None
target_preview_file = None
# Copy preview file
if source_file_ext in ['doc', 'ppt', 'pptx']:
preview_file = KnowledgeUtils.get_knowledge_preview_file_object_name(one.id, one.file_name)
target_preview_file = KnowledgeUtils.get_knowledge_preview_file_object_name(knowledge_new.id,
knowledge_new.file_name)
if preview_file and target_preview_file:
if minio_client.object_exists_sync(minio_client.bucket, preview_file):
minio_client.copy_object_sync(source_bucket=minio_client.bucket, source_object=preview_file,
dest_object=target_preview_file, dest_bucket=minio_client.bucket)
except Exception as e:
logger.exception(f"copy_file_error file_id={knowledge_new.id}")
knowledge_new.remark = KnowledgeFileFailedError(exception=e).to_json_str()
knowledge_new.status = KnowledgeFileStatus.FAILED.value
KnowledgeFileDao.update(knowledge_new)
return
# copy vector
try:
if one.status == KnowledgeFileStatus.SUCCESS.value:
copy_vector(source_knowledge, target_knowledge, one.id, knowledge_new.id)
knowledge_new.status = KnowledgeFileStatus.SUCCESS.value
else:
knowledge_new.status = one.status
KnowledgeFileDao.update(knowledge_new)
except Exception as e:
logger.exception(e)
logger.error("source={} new={} e={}", one.id, knowledge_new.id, e)
knowledge_new.remark = KnowledgeFileFailedError(exception=e).to_json_str()
knowledge_new.status = KnowledgeFileStatus.FAILED.value
KnowledgeFileDao.update(knowledge_new)
def copy_vector(
source_konwledge: Knowledge,
target_knowledge: Knowledge,
source_file_id: int,
target_file_id: int,
):
# migrate vectordb
embedding = FakeEmbedding()
source_col = source_konwledge.collection_name
source_milvus: Milvus = decide_vectorstores(source_col, "Milvus", embedding)
# Saat Inies Exclusion:vector
fields = [s.name for s in source_milvus.col.schema.fields if s.name != "pk"]
source_data = source_milvus.col.query(
expr=f"document_id=={source_file_id} && knowledge_id=={source_konwledge.id}",
output_fields=fields,
)
for data in source_data:
data["knowledge_id"] = target_knowledge.id
data["document_id"] = target_file_id
milvus_db: Milvus = decide_vectorstores(
target_knowledge.collection_name, "Milvus", embedding
)
# Create a new one for the first time collection
if milvus_db.col is None:
new_col = Collection(name=target_knowledge.collection_name, schema=source_milvus.col.schema,
using=source_milvus.alias,
consistency_level=source_milvus.consistency_level)
milvus_db: Milvus = decide_vectorstores(
target_knowledge.collection_name, "Milvus", embedding
)
if milvus_db:
insert_milvus(source_data, fields, milvus_db)
es_db = decide_vectorstores(
target_knowledge.index_name, "ElasticKeywordsSearch", embedding
)
if es_db:
insert_es(source_data, es_db)
def create_milvus_col_and_es_index(source_konwledge: Knowledge, target_knowledge: Knowledge):
embedding = FakeEmbedding()
source_col = source_konwledge.collection_name
source_milvus: Milvus = decide_vectorstores(source_col, "Milvus", embedding)
milvus_db: Milvus = decide_vectorstores(
target_knowledge.collection_name, "Milvus", embedding
)
if milvus_db.col is None and source_milvus.col is not None:
new_col = Collection(name=target_knowledge.collection_name, schema=source_milvus.col.schema,
using=source_milvus.alias,
consistency_level=source_milvus.consistency_level)
new_col.load()
# Buates index
es_db = decide_vectorstores(
target_knowledge.index_name, "ElasticKeywordsSearch", embedding
)
es_db.client.indices.create(index=target_knowledge.index_name, ignore=400)
def insert_milvus(li: List, fields: list, target: Milvus):
total_count = len(li)
batch_size = 1000
res_list = []
for i in range(0, total_count, batch_size):
# Grab end index
end = min(i + batch_size, total_count)
# Convert dict to list of lists batch for insertion
insert_list = [[data[x] for data in li[i:end]] for x in fields]
# Insert into the collection.
try:
res: Collection
res = target.col.insert(insert_list, timeout=100)
res_list.extend(res.primary_keys)
except MilvusException as e:
logger.error(
"Failed to insert batch starting at entity: %s/%s", i, total_count
)
raise e
logger.info("copy_done pk_size={}", len(res_list))
def insert_es(li: List, target: ElasticKeywordsSearch):
from elasticsearch.helpers import bulk
res_list = []
ids = [generate_uuid() for _ in li]
requests = []
for i, data in enumerate(li):
text = data.pop("text")
data.pop("vector", "") # es Exclusion:vector
metadata = data
request = {
"_op_type": "index",
"_index": target.index_name,
"text": text,
"metadata": metadata,
"_id": ids[i],
}
requests.append(request)
bulk(target.client, requests)
target.client.indices.refresh(index=target.index_name)
logger.info("copy_es_done pk_size={}", len(res_list))
@bisheng_celery.task(acks_late=True)
def parse_knowledge_file_celery(file_id: int, preview_cache_key: str = None, callback_url: str = None):
""" Asynchronously parse one incoming successful file """
trace_id_var.set(f'parse_file_{file_id}')
logger.info(
f"parse_knowledge_file_celery start preview_cache_key={preview_cache_key}, callback_url={callback_url}")
try:
# After the warehousing is successful, it is judged whether the file information still exists, and if not, it is deleted.
_, knowledge = _parse_knowledge_file(file_id, preview_cache_key, callback_url)
except Exception as e:
logger.error("parse_knowledge_file_celery error: {}", str(e))
finally:
db_file = KnowledgeFileDao.get_file_by_ids([file_id])
if not db_file:
logger.debug(f"delete_knowledge_file_celery file_id={file_id}")
# If it does not exist, it may have been deleted during the parsing process,
# and the data of the vector database needs to be deleted.
delete_vector_files([db_file[0].id], knowledge)
def _parse_knowledge_file(file_id: int, preview_cache_key: str = None, callback_url: str = None):
db_file = KnowledgeFileDao.get_file_by_ids([file_id])
if not db_file:
logger.error("file_id={} not found in db", file_id)
return
db_file = db_file[0]
if db_file.status not in [KnowledgeFileStatus.PROCESSING.value, KnowledgeFileStatus.WAITING.value]:
logger.error(
"file_id={} status={} not processing, skip parse",
file_id,
db_file.status,
)
return
db_knowledge = KnowledgeDao.query_by_id(db_file.knowledge_id)
if not db_knowledge:
logger.error("knowledge_id={} not found", db_file.knowledge_id)
return
if db_file.status == KnowledgeFileStatus.WAITING.value:
db_file.status = KnowledgeFileStatus.PROCESSING.value
KnowledgeFileDao.update_file_status([db_file.id], KnowledgeFileStatus.PROCESSING)
# Get Splitting Rules
file_rule = FileProcessBase(**json.loads(db_file.split_rule))
logger.debug("parse_knowledge_file_celery_start", file_id)
process_file_task(db_knowledge,
db_files=[db_file],
separator=file_rule.separator,
separator_rule=file_rule.separator_rule,
chunk_size=file_rule.chunk_size,
chunk_overlap=file_rule.chunk_overlap,
callback_url=callback_url,
extra_metadata=db_file.user_metadata,
preview_cache_keys=[preview_cache_key],
retain_images=file_rule.retain_images,
enable_formula=file_rule.enable_formula,
force_ocr=file_rule.force_ocr,
filter_page_header_footer=file_rule.filter_page_header_footer)
logger.debug("parse_knowledge_file_celery_over", file_id)
return db_file, db_knowledge
@bisheng_celery.task(acks_late=True)
def retry_knowledge_file_celery(file_id: int, preview_cache_key: str = None, callback_url: str = None):
""" Retry parsing a file that failed to enter the repository or has a different name """
trace_id_var.set(f'retry_knowledge_file_{file_id}')
logger.info("retry_knowledge_file_celery start file_id={}", file_id)
try:
delete_knowledge_file_vectors(
file_ids=[file_id], clear_minio=False
)
except Exception as e:
logger.exception("retry_knowledge_file_celery delete vectors error: {}", str(e))
KnowledgeFileDao.update_file_status([file_id], KnowledgeFileStatus.FAILED,
KnowledgeFileFailedError(exception=e).to_json_str())
return
try:
_parse_knowledge_file(file_id, preview_cache_key, callback_url)
except Exception as e:
logger.error("retry_knowledge_file_celery error: {}", str(e))
finally:
db_file = KnowledgeFileDao.get_file_by_ids([file_id])
if not db_file:
logger.debug(f"delete_knowledge_file_celery file_id={file_id}")
# If it does not exist, it may have been deleted during the parsing process, and the data of the vector database needs to be deleted.
delete_vector_files([db_file[0].id], knowledge)
@bisheng_celery.task()
def delete_knowledge_file_celery(file_ids: List[int], knowledge_id: int, clear_minio: bool = True):
""" Asynchronous deletion of knowledge files and their vectors """
trace_id_var.set(f'delete_knowledge_file_{file_ids}')
logger.info("delete_knowledge_file_celery start file_ids={}", file_ids)
try:
knowledge = KnowledgeDao.query_by_id(knowledge_id)
if not knowledge:
logger.warning(f"knowledge_id={knowledge_id} is deleted, skip delete file")
return
delete_vector_files(file_ids, knowledge)
except Exception as e:
logger.error("delete_knowledge_file_celery error: {}", str(e))
================================================
FILE: src/backend/bisheng/worker/knowledge/qa.py
================================================
from typing import List
from loguru import logger
from pymilvus import Collection
from bisheng.api.services.knowledge import KnowledgeService
from bisheng.api.services.knowledge_imp import QA_save_knowledge, decide_vectorstores
from bisheng.common.errcode.knowledge import KnowledgeFileFailedError
from bisheng.core.logger import trace_id_var
from bisheng.interface.embeddings.custom import FakeEmbedding
from bisheng.knowledge.domain.models.knowledge import KnowledgeDao, KnowledgeState
from bisheng.knowledge.domain.models.knowledge_file import (
QAKnoweldgeDao, QAKnowledge, QAKnowledgeUpsert, QAStatus,
)
from bisheng.llm.domain import LLMService
from bisheng.worker.main import bisheng_celery
from bisheng_langchain.vectorstores import Milvus, ElasticKeywordsSearch
@bisheng_celery.task
def insert_qa_celery(qa_id: int):
"""
Insert a QA pair into the milvus and es.
"""
trace_id_var.set(f'insert_qa_{qa_id}')
qa_info = QAKnoweldgeDao.get_qa_knowledge_by_primary_id(qa_id)
if not qa_info:
logger.error(f"QA with id {qa_id} not found.")
return
knowledge_info = KnowledgeDao.query_by_id(qa_info.knowledge_id)
if not knowledge_info:
logger.error(f"Knowledge with id {qa_info.knowledge_id} not found.")
return
QA_save_knowledge(knowledge_info, qa_info)
@bisheng_celery.task
def copy_qa_knowledge_celery(source_knowledge_id: int, target_knowledge_id: int, login_user_id: int):
"""
SalinQAkey learning points
:param login_user_id: Login userID
:param source_knowledge_id: Source Knowledge PointsID
:param target_knowledge_id: Target Knowledge PointsID
:return:
"""
trace_id_var.set(f'copy_qa_knowledge_{source_knowledge_id}_{target_knowledge_id}')
try:
source_knowledge = KnowledgeDao.query_by_id(source_knowledge_id)
target_knowledge = KnowledgeDao.query_by_id(target_knowledge_id)
qa_count = QAKnoweldgeDao.count_by_id(source_knowledge_id)
if qa_count == 0:
logger.info(f"No QA knowledge to copy from knowledge id {source_knowledge_id}.")
return
source_milvus: Milvus = decide_vectorstores(source_knowledge.collection_name, "Milvus", FakeEmbedding())
# create new collection name for target knowledge
new_col = Collection(name=target_knowledge.collection_name, schema=source_milvus.col.schema,
using=source_milvus.alias,
consistency_level=source_milvus.consistency_level)
target_milvus: Milvus = decide_vectorstores(target_knowledge.collection_name, "Milvus", FakeEmbedding())
# Batched SalinQAkey learning points Start from the first page
batch_size = 100
for page in range((qa_count + batch_size - 1) // batch_size):
page += 1
qa_list: List[QAKnowledge] = QAKnoweldgeDao.get_qa_knowledge_by_knowledge_id(
knowledge_id=source_knowledge_id,
page=page,
page_size=batch_size
)
try:
# SalinQAkey learning points Batch Insert
new_qa_list = []
for qa in qa_list:
qa_dict = qa.model_dump()
qa_dict.pop("id")
qa_dict.pop("create_time")
qa_dict.pop("update_time")
qa_dict["user_id"] = login_user_id
qa_dict["knowledge_id"] = target_knowledge_id
new_qa = QAKnowledgeUpsert(**qa_dict)
new_qa_list.append(new_qa)
result = QAKnoweldgeDao.batch_insert_qa(new_qa_list)
id_mapping = {qa_list[i].id: result[i].id for i in range(len(qa_list))}
# Copy Vector
source_ids = [int(qa.id) for qa in qa_list if qa.status == QAStatus.ENABLED.value]
fields = [s.name for s in source_milvus.col.schema.fields if s.name != "pk"]
vectors = source_milvus.col.query(
expr=f"file_id in {source_ids} && knowledge_id == '{source_knowledge_id}'",
output_fields=fields)
for vector in vectors:
vector["file_id"] = id_mapping[vector["file_id"]]
vector["knowledge_id"] = str(target_knowledge_id)
vector.pop("pk")
if vectors.__len__() != 0:
target_milvus.col.insert(vectors)
logger.info(f"Copied {len(qa_list)} QA knowledge from knowledge id {source_knowledge_id} "
f"to knowledge id {target_knowledge_id}.")
# es Salin
es_db = decide_vectorstores(
target_knowledge.index_name, "ElasticKeywordsSearch", FakeEmbedding()
)
es_texts = []
es_metadatas = []
for vector in vectors:
text = vector.pop("text")
vector.pop("vector")
es_texts.append(text)
es_metadatas.append(vector)
if es_texts.__len__() != 0:
es_db.add_texts(es_texts, es_metadatas)
logger.info(
f"Updated status to SUCCESS for copied QA knowledge in knowledge id {target_knowledge_id}.")
logger.info(f"Finished copying batch {page + 1} of QA knowledge.")
except Exception as e:
logger.error(f"Error copying batch {page + 1} of QA knowledge: {e}")
QAKnoweldgeDao.batch_update_status_by_ids(
qa_ids=[new_qa.id for new_qa in result],
status=QAStatus.FAILED,
remark=KnowledgeFileFailedError(exception=e).to_json_str()
)
# All copied Update statuses
target_knowledge.state = KnowledgeState.PUBLISHED.value
KnowledgeDao.update_state(knowledge_id=source_knowledge.id, state=KnowledgeState.PUBLISHED,
update_time=source_knowledge.update_time)
KnowledgeDao.update_one(target_knowledge)
logger.info(f"Finished copying all QA knowledge from knowledge id {source_knowledge_id} "
f"to knowledge id {target_knowledge_id}.")
except Exception as e:
logger.error(f"Error copying QA knowledge from knowledge id {source_knowledge_id} "
f"to knowledge id {target_knowledge_id}: {e}")
KnowledgeDao.update_state(knowledge_id=target_knowledge_id, state=KnowledgeState.FAILED)
KnowledgeDao.update_state(knowledge_id=source_knowledge.id, state=KnowledgeState.PUBLISHED)
@bisheng_celery.task
def rebuild_qa_knowledge_celery(knowledge_id: int, embedding_model_id: int, invoke_user_id: int):
"""
RebuildQAThe knowledge base upon,Vector Storage
:param knowledge_id:
:param embedding_model_id:
:param invoke_user_id:
:return:
"""
trace_id_var.set(f'rebuild_qa_knowledge_{knowledge_id}')
knowledge_info = KnowledgeDao.query_by_id(knowledge_id)
try:
if not knowledge_info:
logger.error(f"Knowledge with id {knowledge_id} not found.")
return
# DeletemilvusCorresponding data in
KnowledgeService.delete_knowledge_file_in_vector(knowledge=knowledge_info, del_es=False)
es_db: ElasticKeywordsSearch = decide_vectorstores(
knowledge_info.index_name, "ElasticKeywordsSearch", FakeEmbedding()
)
# InquiryesAll data in Delete
es_result = es_db.client.search(body={
"query": {
"term": {
"metadata.knowledge_id": str(knowledge_id)
}
}
}, filter_path=["hits.total.value"])
total = es_result.get("hits", {}).get("total", {}).get("value", 0)
logger.info(f"Found {total} documents in Elasticsearch for knowledge id {knowledge_id}.")
if total <= 0:
logger.info(f"No documents to delete in Elasticsearch for knowledge id {knowledge_id}.")
knowledge_info.state = KnowledgeState.PUBLISHED.value
return
embeddings = LLMService.get_bisheng_knowledge_embedding_sync(model_id=embedding_model_id,
invoke_user_id=invoke_user_id)
milvus_db: Milvus = decide_vectorstores(
knowledge_info.collection_name, "Milvus", embeddings
)
knowledge_info.state = KnowledgeState.PUBLISHED.value
# Batched RebuildQAThe knowledge base upon Start from the first page
batch_size = 100
for page in range((total + batch_size - 1) // batch_size):
page += 1
texts = []
metadatas = []
es_result = es_db.client.search(body={
"query": {
"term": {
"metadata.knowledge_id": str(knowledge_id)
}
},
"from": (page - 1) * batch_size,
"size": batch_size
}, filter_path=["hits.hits._source"])
hits = es_result.get("hits", {}).get("hits", [])
file_ids = [hit.get("_source", {}).get("metadata", {}).get("file_id") for hit in hits]
try:
QAKnoweldgeDao.batch_update_status_by_ids(
qa_ids=file_ids,
status=QAStatus.PROCESSING
)
for hit in hits:
source = hit.get("_source", {})
text = source.get("text", "")
metadata = source.get("metadata", {})
texts.append(text)
metadata.pop("vector", None)
metadatas.append(metadata)
# Insertmilvus
milvus_db.add_texts(texts, metadatas)
QAKnoweldgeDao.batch_update_status_by_ids(
qa_ids=file_ids,
status=QAStatus.ENABLED
)
logger.info(f"Rebuilt batch {page} of QA knowledge into Milvus for knowledge id {knowledge_id}.")
except Exception as e:
logger.error(f"Error rebuilding batch {page} of QA knowledge: {e}")
QAKnoweldgeDao.batch_update_status_by_ids(
qa_ids=file_ids,
status=QAStatus.FAILED,
remark=KnowledgeFileFailedError(exception=e).to_json_str()
)
knowledge_info.state = KnowledgeState.FAILED.value
logger.info(f"Finished rebuilding QA knowledge for knowledge id {knowledge_id}.")
except Exception as e:
logger.error(f"Error rebuilding QA knowledge for knowledge id {knowledge_id}: {e}")
# DeletemilvusCorresponding data in
KnowledgeService.delete_knowledge_file_in_vector(knowledge=knowledge_info, del_es=False)
knowledge_info.state = KnowledgeState.FAILED.value
QAKnoweldgeDao.update_status_by_knowledge_id(knowledge_id=knowledge_id, status=QAStatus.FAILED,
remark=KnowledgeFileFailedError(exception=e).to_json_str())
finally:
if knowledge_info:
KnowledgeDao.update_one(knowledge_info)
================================================
FILE: src/backend/bisheng/worker/knowledge/rebuild_knowledge_worker.py
================================================
from typing import List
from loguru import logger
from bisheng.api.services.knowledge import KnowledgeService
from bisheng.api.services.knowledge_imp import (
decide_vectorstores
)
from bisheng.common.errcode.knowledge import KnowledgeFileFailedError
from bisheng.core.logger import trace_id_var
from bisheng.interface.embeddings.custom import FakeEmbedding
from bisheng.knowledge.domain.models.knowledge import Knowledge, KnowledgeDao, KnowledgeState
from bisheng.knowledge.domain.models.knowledge_file import (
KnowledgeFile,
KnowledgeFileDao,
KnowledgeFileStatus
)
from bisheng.llm.domain import LLMService
from bisheng.worker.main import bisheng_celery
@bisheng_celery.task(acks_late=True)
def rebuild_knowledge_celery(knowledge_id: int, new_model_id: int, invoke_user_id: int) -> str:
"""
Asynchronous task to rebuild knowledge base
Args:
knowledge_id: The knowledge base uponID
new_model_id: New.. embeddingModelsID
invoke_user_id: Call UserID
Returns:
str: Task Execution Results
"""
trace_id_var.set(f'rebuild_knowledge_{knowledge_id}')
logger.info(f"rebuild_knowledge_celery start knowledge_id={knowledge_id} new_model_id={new_model_id}")
try:
# Get Knowledge Base Information
knowledge = KnowledgeDao.query_by_id(knowledge_id)
if not knowledge:
logger.error(f"knowledge_id={knowledge_id} not found")
return f"knowledge {knowledge_id} not found"
# 1. according knowledge_id Found knowledgefile All in the tablestatus=2Andstatus=4File, put thestatusto4
files = KnowledgeFileDao.get_files_by_multiple_status(
knowledge_id,
[KnowledgeFileStatus.SUCCESS.value, KnowledgeFileStatus.REBUILDING.value]
)
# 2. According to thecollection_namewentmilvusDelete Vector Store in
KnowledgeService.delete_knowledge_file_in_vector(knowledge=knowledge, del_es=False)
if not files:
logger.info(f"knowledge_id={knowledge_id} has no success files")
# Directly update knowledge base status to success
knowledge.state = KnowledgeState.PUBLISHED.value
KnowledgeDao.update_one(knowledge)
return f"knowledge {knowledge_id} rebuild completed (no files)"
# Updating file status to rebuild in progress
file_ids = [f.id for f in files]
KnowledgeFileDao.update_status_bulk(file_ids, KnowledgeFileStatus.REBUILDING)
logger.info(f"Updated {len(files)} files to rebuilding status")
# 3. accordingindex_nameFROMesGot it inchunkinformation, reembeddingInsertmilvus
success_files, failed_files = _rebuild_embeddings(knowledge, files, new_model_id, invoke_user_id)
# 4. Update file status
KnowledgeFileDao.update_status_bulk(success_files, KnowledgeFileStatus.SUCCESS)
for file_id in failed_files:
file = next((f for f in files if f.id == file_id), None)
if file:
file.status = KnowledgeFileStatus.FAILED.value
file.remark = KnowledgeFileFailedError(data={"exception": "rebuild error"}).to_json_str()
KnowledgeFileDao.update(file)
# 5. Update knowledge base status
if failed_files:
# DeleteesIndex andmilvusCollections to avoid data inconsistencies
_delete_es_files(knowledge, failed_files)
knowledge.state = KnowledgeState.FAILED.value
logger.error(f"knowledge_id={knowledge_id} rebuild failed, failed_files={failed_files}")
else:
knowledge.state = KnowledgeState.PUBLISHED.value
logger.info(f"knowledge_id={knowledge_id} rebuild completed successfully")
KnowledgeDao.update_one(knowledge)
return f"knowledge {knowledge_id} rebuild completed"
except Exception as e:
logger.exception(f"rebuild_knowledge_celery error: {str(e)}")
# Unexpected handles during asynchronous tasksknowledgeSet to4
try:
knowledge = KnowledgeDao.query_by_id(knowledge_id)
if knowledge:
knowledge.state = KnowledgeState.FAILED.value
KnowledgeDao.update_one(knowledge)
except Exception as e2:
logger.exception(f"Failed to update knowledge state after error: {str(e2)}")
raise e
def _delete_es_files(knowledge: Knowledge, file_ids: List[int]):
"""DeleteESFile data in"""
try:
index_name = knowledge.index_name or knowledge.collection_name
embeddings = FakeEmbedding()
es_client = decide_vectorstores(index_name, "ElasticKeywordsSearch", embeddings)
if not es_client.client.indices.exists(index=index_name):
logger.warning(f"ES index {index_name} does not exist, skipping deletion")
return
for file_id in file_ids:
delete_query = {
"query": {
"match": {
"metadata.document_id": file_id
}
}
}
response = es_client.client.delete_by_query(index=index_name, body=delete_query)
deleted = response.get("deleted", 0)
logger.info(f"Deleted {deleted} documents from ES for file_id={file_id}")
except Exception as e:
logger.exception(f"Failed to delete ES files for knowledge_id={knowledge.id}: {str(e)}")
def _rebuild_embeddings(knowledge: Knowledge, files: List[KnowledgeFile], new_model_id: int, invoke_user_id: int) -> \
tuple[List[int], List[int]]:
"""
Rebuildembeddings
Returns:
tuple: (Success FilesIDVertical, Failed FilesIDVertical)
"""
success_files = []
failed_files = []
vector_client = None
try:
# DapatkanEShitting the nail on the headchunkMessage
index_name = knowledge.index_name or knowledge.collection_name
embeddings = FakeEmbedding()
es_client = decide_vectorstores(index_name, "ElasticKeywordsSearch", embeddings)
# Get newembeddingModel and createMilvusClient
logger.info(f"[DEBUG] Begin initializing newembeddingModelsmodel_id={new_model_id}")
new_embeddings = LLMService.get_bisheng_knowledge_embedding_sync(model_id=new_model_id,
invoke_user_id=invoke_user_id)
logger.info(
f"[DEBUG] Slider Created Successfully.embeddingModel Instance: {type(new_embeddings).__name__}, model_id={getattr(new_embeddings, 'model_id', 'unknown')}")
# TestembeddingIs the model available
try:
test_result = new_embeddings.embed_query("Test text")
logger.info(
f"[DEBUG] EmbeddingModel tested successfully, dimension returned: {len(test_result) if test_result else 'None'}")
except Exception as e:
logger.error(f"[DEBUG] EmbeddingModel Test Failed: {str(e)}")
# Model test failure should terminate the entire process, not continue
raise Exception(f"EmbeddingModel not available: {str(e)}")
vector_client = decide_vectorstores(knowledge.collection_name, "Milvus", new_embeddings)
logger.info(f"[DEBUG] Slider Created Successfully.MilvusClientcollection_name={knowledge.collection_name}")
# OthersESWhether the index is present (check in advance, avoid double-checking in the loop)
if not es_client.client.indices.exists(index=index_name):
logger.error(f"ES index {index_name} does not exist")
# Index does not exist, all files failed
failed_files = [f.id for f in files]
return success_files, failed_files
# Regenerate for each fileembeddings
for file in files:
try:
success = _process_single_file(file, es_client, index_name, vector_client)
if success:
success_files.append(file.id)
logger.info(f"Successfully rebuilt embeddings for file_id={file.id}")
else:
failed_files.append(file.id)
except Exception as e:
logger.exception(f"Failed to rebuild embeddings for file_id={file.id}: {str(e)}")
failed_files.append(file.id)
except Exception as e:
logger.exception(f"Failed to rebuild embeddings: {str(e)}")
# If the entire process fails, all unsuccessful files are marked as failed
failed_files.extend([f.id for f in files if f.id not in success_files])
return success_files, failed_files
def _process_single_file(file, es_client, index_name, vector_client):
"""Processing of individual filesembeddingRebuild"""
logger.info(f"Rebuilding embeddings for file_id={file.id}")
# FROMESGet all of this file inchunks
search_query = {
"query": {
"match": {
"metadata.document_id": file.id
}
},
"size": 10000
}
logger.debug(f"ES search query: {search_query}")
response = es_client.client.search(index=index_name, body=search_query)
chunks = response.get("hits", {}).get("hits", [])
logger.info(f"Found {len(chunks)} chunks in ES for file_id={file.id}")
if not chunks:
logger.warning(f"No chunks found for file_id={file.id}")
return True # No data to process, considered a success
# Extract text and metadata
texts = []
metadatas = []
for chunk in chunks:
source = chunk["_source"]
texts.append(source["text"])
# Removepkfields, avoid insertingMilvusTime Conflict
if "pk" in source["metadata"]:
del source["metadata"]["pk"]
metadatas.append(source["metadata"])
logger.info(f"Found {len(texts)} chunks for file_id={file.id}")
# Insert data intoMilvus
logger.info(f"[DEBUG] Upcoming Callsvector_client.add_texts，textsQuantity={len(texts)}")
logger.info(f"[DEBUG] First text example: {texts[0][:100] if texts else 'No texts'}...")
try:
vector_client.add_texts(texts=texts, metadatas=metadatas)
logger.info(f"[DEBUG] vector_client.add_textsCall successful")
return True
except Exception as add_error:
logger.error(f"[DEBUG] vector_client.add_textsCall failed: {str(add_error)}")
raise add_error
================================================
FILE: src/backend/bisheng/worker/main.py
================================================
import threading
import time
from typing import List
from celery import Celery
from celery.signals import celeryd_after_setup, worker_shutting_down
from loguru import logger
from bisheng.common.services.config_service import settings
from bisheng.core.cache.redis_manager import get_redis_client_sync
from bisheng.core.logger import set_logger_config
def create_celery_app():
"""
Celery Asynchronous Tasks
:return:
"""
set_logger_config(settings.logger_conf)
# loop = app_ctx.get_event_loop()
bisheng_celery = Celery('bisheng', include=['bisheng.worker'])
bisheng_celery.config_from_object('bisheng.worker.config')
return bisheng_celery
_WORKER_START = False
_WORKER_BEAT_SLEEP = 5 # seconds
WORKER_ALIVE_KEY = "celery_worker_alive_queues"
bisheng_celery = create_celery_app()
def worker_alive_beat(all_queues: List[str]):
"""Worker heartbeat function."""
logger.debug(f"Worker heartbeat function: {all_queues}")
while _WORKER_START:
try:
# upload worker alive timestamp to redis
current_timestamp = str(int(time.time()))
redis_client = get_redis_client_sync()
redis_client.hset(WORKER_ALIVE_KEY, mapping={one: current_timestamp for one in all_queues})
time.sleep(_WORKER_BEAT_SLEEP)
except Exception as e:
logger.error(f"Error in worker alive beat: {e}")
time.sleep(_WORKER_BEAT_SLEEP * 2)
continue
logger.debug('Worker alive beat stopped.')
@celeryd_after_setup.connect
def on_worker_init(*args, **kwargs):
global _WORKER_START
"""Worker initialization signal handler."""
queues = bisheng_celery.amqp.queues
all_queues = []
for queue_name, _ in queues.items():
all_queues.append(queue_name)
_WORKER_START = True
t = threading.Thread(target=worker_alive_beat, args=(all_queues,), daemon=True)
t.start()
logger.debug("Celery worker alive beat started.")
@worker_shutting_down.connect
def on_worker_shutdown(*args, **kwargs):
logger.debug("Celery worker shutting down.")
global _WORKER_START
_WORKER_START = False
================================================
FILE: src/backend/bisheng/worker/telemetry/__init__.py
================================================
================================================
FILE: src/backend/bisheng/worker/telemetry/mid_table.py
================================================
from datetime import datetime, timedelta
from typing import List
from loguru import logger
from bisheng.api.services.workflow import WorkFlowService
from bisheng.common.constants.enums.telemetry import ApplicationTypeEnum
from bisheng.common.schemas.telemetry.base_telemetry_schema import UserGroupInfo, UserRoleInfo
from bisheng.core.logger import trace_id_var
from bisheng.database.models.flow import FlowType
from bisheng.knowledge.domain.services.knowledge_service import KnowledgeService
from bisheng.telemetry.domain.mid_table.app_increment import AppIncrement, AppIncrementRecord
from bisheng.telemetry.domain.mid_table.base import BaseMidTable
from bisheng.telemetry.domain.mid_table.knowledge_increment import KnowledgeIncrement, KnowledgeIncrementRecord
from bisheng.telemetry.domain.mid_table.user_increment import UserIncrement, UserIncrementRecord
from bisheng.telemetry.domain.mid_table.user_interact import UserInteract, UserInteractRecord
from bisheng.user.domain.services.user import UserService
from bisheng.utils import generate_uuid
from bisheng.worker.main import bisheng_celery
def get_yesterday_date_range(mid_table: BaseMidTable, start_date: str = None, end_date: str = None) -> (datetime,
datetime):
if start_date is None or end_date is None:
# default to yesterday's date
now = datetime.now()
yesterday = now - timedelta(days=1)
start_date = datetime(year=yesterday.year, month=yesterday.month, day=yesterday.day, hour=0, minute=0, second=0)
end_date = datetime(year=now.year, month=now.month, day=now.day, hour=0, minute=0, second=0)
else:
start_date = datetime.fromisoformat(start_date)
end_date = datetime.fromisoformat(end_date)
lastest_time = mid_table.get_latest_record_time_sync()
if lastest_time:
start_date = datetime.fromtimestamp(lastest_time) + timedelta(seconds=1)
if end_date < start_date:
logger.error(f"end_date {end_date} is before start_date {start_date}")
return None, None
return start_date, end_date
def convert_flow_type(flow_type: int) -> ApplicationTypeEnum:
flow_type_mapping = {
FlowType.FLOW.value: ApplicationTypeEnum.SKILL,
FlowType.ASSISTANT.value: ApplicationTypeEnum.ASSISTANT,
FlowType.WORKFLOW.value: ApplicationTypeEnum.WORKFLOW,
}
return flow_type_mapping.get(flow_type, ApplicationTypeEnum.UNKNOWN)
@bisheng_celery.task()
def sync_mid_user_increment(start_date: str = None, end_date: str = None):
trace_id_var.set(f"sync_mid_user_increment_task_{generate_uuid()}")
mid_table = UserIncrement()
start_date, end_date = get_yesterday_date_range(mid_table, start_date, end_date)
if start_date is None or end_date is None:
return
logger.info(f"Syncing mid_user_increment from {start_date} to {end_date}")
# Here would be the logic to fetch data from the source and insert into mid_user_increment
page, page_size = 1, 1000
while True:
user_list = UserService.get_user_all_info(start_time=start_date, end_time=end_date,
page=page, page_size=page_size)
page += 1
if not user_list:
break
records = []
for user in user_list:
records.append(UserIncrementRecord(
es_id=f"user_{user.user_id}",
user_id=user.user_id,
user_name=user.user_name,
user_group_infos=[UserGroupInfo(user_group_id=group.id, user_group_name=group.group_name)
for group in user.groups],
user_role_infos=[UserRoleInfo(role_id=role.id, role_name=role.role_name, group_id=role.group_id)
for role in user.roles],
timestamp=int(user.create_time.timestamp())
))
mid_table.insert_records_sync(records)
# This is a placeholder for the actual data synchronization logic
logger.info(f"Successfully synced mid_user_increment from {start_date} to {end_date}")
def get_user_from_ids_with_cache(user_ids: List[int], user_map: dict):
if user_ids:
user_list = UserService.get_user_all_info(user_ids=user_ids, page=0, page_size=0)
user_map.update({user.user_id: user for user in user_list})
return user_map
@bisheng_celery.task()
def sync_mid_app_increment(start_date: str = None, end_date: str = None):
# Placeholder for syncing mid_app_increment table
trace_id_var.set(f"sync_mid_app_increment_task_{generate_uuid()}")
logger.info("Syncing mid_app_increment table...")
mid_table = AppIncrement()
start_date, end_date = get_yesterday_date_range(mid_table, start_date, end_date)
if start_date is None or end_date is None:
return
logger.info(f"Syncing mid_app_increment from {start_date} to {end_date}")
page, page_size = 1, 1000
user_map = {}
while True:
app_list = WorkFlowService.get_all_apps_by_time_range_sync(start_time=start_date, end_time=end_date, page=page,
page_size=page_size)
page += 1
if not app_list:
break
records = []
user_ids = set()
for app in app_list:
if app['user_id'] not in user_map:
user_ids.add(app['user_id'])
user_map = get_user_from_ids_with_cache(list(user_ids), user_map)
for app in app_list:
user = user_map.get(app['user_id'], None)
records.append(AppIncrementRecord(
es_id=f"app_{app['id']}",
user_id=app['user_id'],
user_name=user.user_name if user else "",
user_group_infos=[UserGroupInfo(user_group_id=group.id, user_group_name=group.group_name)
for group in user.groups] if user else [],
user_role_infos=[UserRoleInfo(role_id=role.id, role_name=role.role_name, group_id=role.group_id)
for role in user.roles] if user else [],
app_id=app['id'],
app_name=app['name'],
app_type=convert_flow_type(app['flow_type']),
timestamp=int(app['create_time'].timestamp())
))
mid_table.insert_records_sync(records)
# Implement the actual logic here
logger.info("Successfully synced mid_app_increment table.")
@bisheng_celery.task()
def sync_mid_knowledge_increment(start_date: str = None, end_date: str = None):
# Placeholder for syncing mid_knowledge_increment table
trace_id_var.set(f"sync_mid_knowledge_increment_task_{generate_uuid()}")
logger.info("Syncing mid_knowledge_increment table...")
mid_table = KnowledgeIncrement()
start_date, end_date = get_yesterday_date_range(mid_table, start_date, end_date)
if start_date is None or end_date is None:
return
logger.info(f"Syncing mid_knowledge_increment from {start_date} to {end_date}")
page, page_size = 1, 1000
user_map = {}
while True:
knowledge_list = KnowledgeService.get_all_knowledge_by_time_range(start_date, end_date, page=page,
page_size=page_size)
page += 1
if not knowledge_list:
break
user_ids = set()
for knowledge in knowledge_list:
if knowledge.user_id not in user_map:
user_ids.add(knowledge.user_id)
user_map = get_user_from_ids_with_cache(list(user_ids), user_map)
records = []
for knowledge in knowledge_list:
user = user_map.get(knowledge.user_id, None)
records.append(KnowledgeIncrementRecord(
es_id=f"knowledge_{knowledge.id}",
user_id=knowledge.user_id,
user_name=user.user_name if user else "",
user_group_infos=[UserGroupInfo(user_group_id=group.id, user_group_name=group.group_name)
for group in user.groups] if user else [],
user_role_infos=[UserRoleInfo(role_id=role.id, role_name=role.role_name, group_id=role.group_id)
for role in user.roles] if user else [],
knowledge_id=knowledge.id,
knowledge_name=knowledge.name,
knowledge_type=knowledge.type,
timestamp=int(knowledge.create_time.timestamp())
))
mid_table.insert_records_sync(records)
# Implement the actual logic here
logger.info("Successfully synced mid_knowledge_increment table.")
@bisheng_celery.task()
def sync_mid_user_interact_dtl(start_date: str = None, end_date: str = None):
# Placeholder for syncing mid_user_interact_dtl table
trace_id_var.set(f"sync_mid_user_interact_dtl_task_{generate_uuid()}")
logger.info("Syncing mid_user_interact_dtl table...")
mid_table = UserInteract()
start_date, end_date = get_yesterday_date_range(mid_table, start_date, end_date)
if start_date is None or end_date is None:
return
page, page_size = 1, 1000
while True:
result = mid_table.get_records_by_time_range_sync(start_time=int(start_date.timestamp()),
end_time=int(end_date.timestamp()),
page=page,
page_size=page_size)
page += 1
if not result:
break
records = []
for record in result:
es_id = record['_id']
record = record['_source']
records.append(UserInteractRecord(
es_id=es_id,
user_id=record['user_context']['user_id'],
user_name=record['user_context']['user_name'],
user_group_infos=[UserGroupInfo(user_group_id=group['user_group_id'],
user_group_name=group['user_group_name'])
for group in record['user_context'].get('user_group_infos', [])],
user_role_infos=[UserRoleInfo(role_id=role['role_id'],
role_name=role['role_name'],
group_id=role.get('group_id', 0))
for role in record['user_context'].get('user_role_infos', [])],
event_id=record['event_id'],
timestamp=record['timestamp'],
message_id=record['event_data']['message_feedback_message_id'],
interact_type=record['event_data']['message_feedback_operation_type'],
app_id=record['event_data']['message_feedback_app_id'],
app_name=record['event_data']['message_feedback_app_name'],
))
mid_table.insert_records_sync(records)
# Implement the actual logic here
logger.info("Successfully synced mid_user_interact_dtl table.")
================================================
FILE: src/backend/bisheng/worker/test/__init__.py
================================================
================================================
FILE: src/backend/bisheng/worker/test/test.py
================================================
from loguru import logger
from bisheng.worker.main import bisheng_celery
@bisheng_celery.task
def add(x, y):
logger.info(f"add {x} + {y}")
return x + y
================================================
FILE: src/backend/bisheng/worker/utils/__init__.py
================================================
================================================
FILE: src/backend/bisheng/worker/utils/stateful_worker.py
================================================
import asyncio
import time
from typing import List, Dict
from httpcore._synchronization import ThreadLock
from loguru import logger
from bisheng.core.cache.redis_manager import get_redis_client, get_redis_client_sync
from bisheng.utils.consisten_hash import ConsistentHash
from bisheng.worker.main import WORKER_ALIVE_KEY
class StatefulWorker:
def __init__(self,
queue_prefix: str = "workflow_celery",
bound_nodes_prefix: str = "workflow_bound_nodes:",
**kwargs):
self.queue_prefix = queue_prefix
self.bound_nodes_prefix = bound_nodes_prefix
self.alive_times = kwargs.get("alive_time", 20) # how much seconds to consider a worker alive
self.consistent_hash = ConsistentHash(virtual_replicas=kwargs.get("virtual_replicas", 10))
self.sync_timestamp = 0
self.sync_timestamp_sleep = 5 # seconds
self.thread_lock = ThreadLock()
self.async_lock = asyncio.Lock()
def _base_update_alive_nodes(self, alive_nodes: List[str]):
consistent_nodes = self.consistent_hash.get_all_nodes()
add_nodes = set(alive_nodes) - set(consistent_nodes)
remove_nodes = set(consistent_nodes) - set(alive_nodes)
for node in add_nodes:
self.consistent_hash.add_node(node)
for node in remove_nodes:
self.consistent_hash.remove_node(node)
logger.debug(f'Worker alive nodes updated: {self.consistent_hash.get_all_nodes()}')
async def update_alive_nodes(self):
async with self.async_lock:
current_time = time.time()
if current_time - self.sync_timestamp < self.alive_times:
return
alive_nodes = await self.get_all_alive_queues()
self._base_update_alive_nodes(alive_nodes)
def update_alive_nodes_sync(self):
with self.thread_lock:
current_time = time.time()
if current_time - self.sync_timestamp < self.alive_times:
return
alive_nodes = self.get_all_alive_queues_sync()
self._base_update_alive_nodes(alive_nodes)
def _base_get_all_alive_queues(self, all_queues: Dict[bytes, bytes]) -> (List[str], List[str]):
if not all_queues:
return [], []
current_timestamp = int(time.time())
alive_queues = []
need_remove_queues = []
for queue_name, queue_timestamp in all_queues.items():
queue_name = queue_name.decode() if isinstance(queue_name, bytes) else queue_name
queue_timestamp = queue_timestamp.decode() if isinstance(queue_timestamp, bytes) else queue_timestamp
if not queue_name.startswith(self.queue_prefix):
continue
if current_timestamp - int(queue_timestamp) < self.alive_times:
alive_queues.append(queue_name)
else:
need_remove_queues.append(queue_name)
return alive_queues, need_remove_queues
async def get_all_alive_queues(self) -> List[str]:
redis_client = await get_redis_client()
all_queues = await redis_client.ahgetall(WORKER_ALIVE_KEY)
alive_nodes, remove_nodes = self._base_get_all_alive_queues(all_queues)
if remove_nodes:
await redis_client.ahdel(WORKER_ALIVE_KEY, *remove_nodes)
return alive_nodes
def get_all_alive_queues_sync(self):
redis_client = get_redis_client_sync()
all_queues = redis_client.hgetall(WORKER_ALIVE_KEY)
alive_nodes, remove_nodes = self._base_get_all_alive_queues(all_queues)
if remove_nodes:
redis_client.hdel(WORKER_ALIVE_KEY, *remove_nodes)
return alive_nodes
def is_node_alive(self, node: str) -> bool:
all_nodes = self.consistent_hash.get_all_nodes()
return node in all_nodes
async def _find_bound_node(self, hash_key: str) -> str | None:
redis_client = await get_redis_client()
return await redis_client.aget(f"{self.bound_nodes_prefix}{hash_key}")
async def _save_bound_node(self, hash_key: str, node: str) -> None:
redis_client = await get_redis_client()
return await redis_client.aset(f"{self.bound_nodes_prefix}{hash_key}", node)
def _find_bound_node_sync(self, hash_key: str) -> str | None:
redis_client = get_redis_client_sync()
return redis_client.get(f"{self.bound_nodes_prefix}{hash_key}")
def _save_bound_node_sync(self, hash_key: str, node: str) -> None:
redis_client = get_redis_client_sync()
return redis_client.set(f"{self.bound_nodes_prefix}{hash_key}", node)
async def find_task_node(self, hash_key: str) -> str | None:
await self.update_alive_nodes()
# Get the node bound to the key
bound_node = await self._find_bound_node(hash_key)
if bound_node and self.is_node_alive(bound_node):
# judge if the bound node is alive
logger.debug(f"Assigned node {bound_node} for key {hash_key}")
return bound_node
# reassign the node
assigned_node = self.consistent_hash.find_node(hash_key)
if assigned_node:
await self._save_bound_node(hash_key, assigned_node)
logger.debug(f"Assigned node {assigned_node} for key {hash_key}")
return assigned_node
def find_task_node_sync(self, hash_key: str) -> str | None:
self.update_alive_nodes()
# Get the node bound to the key
bound_node = self._find_bound_node_sync(hash_key)
if bound_node and self.is_node_alive(bound_node):
# judge if the bound node is alive
logger.debug(f"Assigned node {bound_node} for key {hash_key}")
return bound_node
# reassign the node
assigned_node = self.consistent_hash.find_node(hash_key)
if assigned_node:
self._save_bound_node_sync(hash_key, assigned_node)
logger.debug(f"Assigned node {assigned_node} for key {hash_key}")
return assigned_node
================================================
FILE: src/backend/bisheng/worker/workflow/__init__.py
================================================
================================================
FILE: src/backend/bisheng/worker/workflow/redis_callback.py
================================================
import asyncio
import json
import os
import time
import uuid
from typing import AsyncIterator, Iterator, Dict, List
from langchain_core.documents import Document
from loguru import logger
from bisheng.api.v1.schema.workflow import WorkflowEventType
from bisheng.api.v1.schemas import ChatResponse
from bisheng.chat.utils import sync_judge_source, sync_process_source_document
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum, ApplicationTypeEnum
from bisheng.common.errcode.flow import WorkFlowNodeRunMaxTimesError, WorkFlowWaitUserTimeoutError, \
WorkFlowNodeUpdateError, WorkFlowVersionUpdateError, WorkFlowTaskBusyError, WorkFlowTaskOtherError
from bisheng.common.errcode.http_error import ServerError
from bisheng.common.schemas.telemetry.event_data_schema import NewMessageSessionEventData
from bisheng.common.services import telemetry_service
from bisheng.common.services.config_service import settings
from bisheng.core.cache.redis_manager import get_redis_client_sync
from bisheng.core.logger import trace_id_var
from bisheng.database.models.flow import FlowDao, FlowType
from bisheng.database.models.message import ChatMessageDao, ChatMessage
from bisheng.database.models.session import MessageSessionDao, MessageSession
from bisheng.utils.threadpool import thread_pool
from bisheng.workflow.callback.base_callback import BaseCallback
from bisheng.workflow.callback.event import NodeStartData, NodeEndData, UserInputData, GuideWordData, GuideQuestionData, \
OutputMsgData, StreamMsgData, StreamMsgOverData, OutputMsgChooseData, OutputMsgInputData
from bisheng.workflow.common.workflow import WorkflowStatus
class RedisCallback(BaseCallback):
def __init__(self, unique_id: str, workflow_id: str, chat_id: str, user_id: int, **kwargs):
super(RedisCallback, self).__init__()
# Unique for asynchronous tasksID
self.unique_id = unique_id
self.workflow_id = workflow_id
self.chat_id = chat_id
self.user_id = user_id
self.workflow = None
self.create_session = False
self.source = kwargs.get('source', 'platform') # only platform or api
self.redis_client = get_redis_client_sync()
self.workflow_data_key = f'workflow:{unique_id}:data'
self.workflow_status_key = f'workflow:{unique_id}:status'
self.workflow_event_key = f'workflow:{unique_id}:event'
self.workflow_input_key = f'workflow:{unique_id}:input'
self.workflow_stop_key = f'workflow:{unique_id}:stop'
self.workflow_expire_time = settings.get_workflow_conf().timeout * 60 + 60
def set_workflow_data(self, data: Dict, override: Dict = None):
data = self.override_nodes_params(data, override)
self.redis_client.set(self.workflow_data_key, data, expiration=self.workflow_expire_time)
@staticmethod
def override_nodes_params(data: Dict, override: Dict = None) -> Dict:
if not override:
return data
def replace_param(one_params: List[Dict], one_node_id: str):
for param in one_params:
param_key = param.get('key')
if param_key not in override[node_id]:
continue
param['value'] = override[one_node_id][param_key]
nodes = data.get('nodes', [])
for node in nodes:
node_data = node.get('data', {})
node_id = node_data.get('id')
if node_id not in override:
continue
group_params = node_data.get('group_params', [])
for group_param in group_params:
replace_param(group_param.get('params', []), node_id)
return data
async def async_set_workflow_data(self, data: dict):
await self.redis_client.aset(self.workflow_data_key, data, expiration=self.workflow_expire_time)
def get_workflow_data(self) -> dict:
return self.redis_client.get(self.workflow_data_key)
def set_workflow_status(self, status: str, reason: str = None):
self.redis_client.set(self.workflow_status_key,
{'status': status, 'reason': reason, 'time': time.time()},
expiration=3600 * 24 * 7)
if status in [WorkflowStatus.FAILED.value, WorkflowStatus.SUCCESS.value]:
# Message Events and StatuskeyConsumption may also be required
self.redis_client.delete(self.workflow_data_key)
self.redis_client.delete(self.workflow_input_key)
async def async_set_workflow_status(self, status: str, reason: str = None):
await self.redis_client.aset(self.workflow_status_key,
{'status': status, 'reason': reason, 'time': time.time()},
expiration=3600 * 24 * 7)
if status in [WorkflowStatus.FAILED.value, WorkflowStatus.SUCCESS.value]:
# Message Events and StatuskeyConsumption may also be required
await self.redis_client.adelete(self.workflow_data_key)
await self.redis_client.adelete(self.workflow_input_key)
def get_workflow_status(self) -> dict | None:
workflow_status = self.redis_client.get(self.workflow_status_key)
return workflow_status
async def async_get_workflow_status(self) -> dict | None:
workflow_status = await self.redis_client.aget(self.workflow_status_key)
return workflow_status
def clear_workflow_status(self):
self.redis_client.delete(self.workflow_status_key)
self.redis_client.delete(self.workflow_stop_key)
self.redis_client.delete(self.workflow_data_key)
async def async_clear_workflow_status(self):
await self.redis_client.adelete(self.workflow_status_key)
await self.redis_client.adelete(self.workflow_stop_key)
await self.redis_client.adelete(self.workflow_data_key)
def insert_workflow_response(self, event: dict):
self.redis_client.rpush(self.workflow_event_key, json.dumps(event), expiration=self.workflow_expire_time)
def get_workflow_response(self) -> ChatResponse | None:
response = self.redis_client.lpop(self.workflow_event_key)
if self.get_workflow_stop():
self.redis_client.delete(self.workflow_event_key)
return None
if response:
response = ChatResponse(**json.loads(response))
return response
async def async_get_workflow_response(self) -> ChatResponse | None:
response = await self.redis_client.alpop(self.workflow_event_key)
if await self.async_get_workflow_stop():
await self.redis_client.adelete(self.workflow_event_key)
return None
if response:
response = ChatResponse(**json.loads(response))
return response
def build_chat_response(self, category, category_type, message, extra=None, files=None):
return ChatResponse(
user_id=self.user_id,
chat_id=self.chat_id,
flow_id=self.workflow_id,
type=category_type,
message=message,
category=category,
extra=extra,
files=files,
)
def parse_workflow_failed(self, status_info: dict) -> ChatResponse | None:
if status_info['reason'].find('-- has run more than the maximum number of times') != -1:
return self.build_chat_response(WorkflowEventType.Error.value, 'over',
message=WorkFlowNodeRunMaxTimesError(
exception=status_info['reason'].split('--')[0]).to_dict())
elif status_info['reason'].find('workflow wait user input timeout') != -1:
return self.build_chat_response(WorkflowEventType.Error.value, 'over',
message=WorkFlowWaitUserTimeoutError().to_dict())
elif status_info['reason'].find('-- node params is error') != -1:
return self.build_chat_response(WorkflowEventType.Error.value, 'over',
message=WorkFlowNodeUpdateError(
exception=status_info['reason'].split('--')[0]).to_dict())
elif status_info['reason'].find('-- workflow node is update') != -1:
return self.build_chat_response(WorkflowEventType.Error.value, 'over',
message=WorkFlowVersionUpdateError(
exception=status_info['reason'].split('--')[0]).to_dict())
elif status_info['reason'].find('stop by user') != -1:
return None
else:
return self.build_chat_response(WorkflowEventType.Error.value, 'over',
WorkFlowTaskOtherError(exception=status_info['reason']).to_dict())
def sync_get_response_until_break(self) -> Iterator[ChatResponse]:
while True:
# get workflow status
status_info = self.get_workflow_status()
if not status_info:
yield self.build_chat_response(WorkflowEventType.Error.value, 'over',
message=WorkFlowTaskOtherError(
exception=Exception("workflow status not found")).to_dict())
break
elif status_info['status'] in [WorkflowStatus.FAILED.value, WorkflowStatus.SUCCESS.value]:
while True:
chat_response = self.get_workflow_response()
if not chat_response:
break
yield chat_response
if status_info['status'] == WorkflowStatus.FAILED.value:
error_resp = self.parse_workflow_failed(status_info)
if error_resp:
yield error_resp
break
elif status_info['status'] == WorkflowStatus.INPUT.value:
while True:
chat_response = self.get_workflow_response()
if not chat_response:
break
yield chat_response
break
elif status_info['status'] in [WorkflowStatus.WAITING.value,
WorkflowStatus.INPUT_OVER.value] and time.time() - status_info['time'] > 10:
# 10No status update received in seconds, descriptionworkflowNot started, could becelery workerThreads full
self.set_workflow_status(WorkflowStatus.FAILED.value, 'workflow task execute busy')
yield self.build_chat_response(WorkflowEventType.Error.value, 'over',
message=WorkFlowTaskBusyError().to_dict())
break
elif time.time() - status_info['time'] > 86400:
yield self.build_chat_response(WorkflowEventType.Error.value, 'over',
WorkFlowTaskOtherError(
exception=Exception(
"workflow status not update over 1 day")).to_dict())
self.set_workflow_status(WorkflowStatus.FAILED.value, 'workflow status not update over 1 day')
self.set_workflow_stop()
break
else:
chat_response = self.get_workflow_response()
if not chat_response:
time.sleep(1)
continue
yield chat_response
async def get_response_until_break(self) -> AsyncIterator[ChatResponse]:
""" Continuous accessworkflowright of privacyresponseuntil the end of the run is encountered or pending entry """
while True:
# get workflow status
status_info = await self.async_get_workflow_status()
if not status_info:
yield self.build_chat_response(WorkflowEventType.Error.value, 'over',
message=WorkFlowTaskOtherError(
exception=Exception("workflow status not found")).to_dict())
break
elif status_info['status'] in [WorkflowStatus.FAILED.value, WorkflowStatus.SUCCESS.value]:
while True:
chat_response = await self.async_get_workflow_response()
if not chat_response:
break
yield chat_response
if status_info['status'] == WorkflowStatus.FAILED.value:
error_resp = self.parse_workflow_failed(status_info)
if error_resp:
yield error_resp
break
elif status_info['status'] == WorkflowStatus.INPUT.value:
while True:
chat_response = await self.async_get_workflow_response()
if not chat_response:
break
yield chat_response
break
elif status_info['status'] in [WorkflowStatus.WAITING.value,
WorkflowStatus.INPUT_OVER.value] and time.time() - status_info['time'] > 10:
# 10No status update received in seconds, descriptionworkflowNot started, could becelery workerThreads full
await self.async_set_workflow_status(WorkflowStatus.FAILED.value, 'workflow task execute busy')
yield self.build_chat_response(WorkflowEventType.Error.value, 'over',
message=WorkFlowTaskBusyError().to_dict())
break
elif time.time() - status_info['time'] > 86400:
yield self.build_chat_response(WorkflowEventType.Error.value, 'over',
message=WorkFlowTaskOtherError(exception=Exception(
"workflow status not update over 1 day")).to_dict())
await self.async_set_workflow_status(WorkflowStatus.FAILED.value,
'workflow status not update over 1 day')
await self.async_set_workflow_stop()
break
else:
chat_response = await self.async_get_workflow_response()
if not chat_response:
await asyncio.sleep(0.01)
continue
yield chat_response
def set_user_input(self, data: dict, message_id: int = None, message_content: str = None,
verify_input: bool = False):
if self.chat_id and message_id:
message_db = ChatMessageDao.get_message_by_id(message_id)
self.update_old_message(data, message_db, message_content, verify_input)
# Notify Asynchronous Task User Input
self.redis_client.set(self.workflow_input_key, data, expiration=self.workflow_expire_time)
return
async def async_set_user_input(self, data: dict, message_id: int = None, message_content: str = None,
verify_input: bool = False):
if self.chat_id and message_id:
message_db = await ChatMessageDao.aget_message_by_id(message_id)
await self.async_update_old_message(data, message_db, message_content, verify_input)
# Notify Asynchronous Task User Input
await self.redis_client.aset(self.workflow_input_key, data, expiration=self.workflow_expire_time)
return
@staticmethod
def _verify_input_schema(input_schema_message: Dict, user_input: Dict):
""" Verify that the user input matches the input schema """
node_id = input_schema_message['node_id']
if node_id not in user_input:
raise ServerError(msg="node_id not found in user input")
user_input = user_input[node_id]
input_schema = input_schema_message['input_schema']
if input_schema["tab"] == "form_input":
user_input_keys = {one: None for one in user_input.keys()}
for key_info in input_schema['value']:
key = key_info['key']
if key not in user_input:
raise ServerError(msg=f"key {key} not found in user input")
user_input_keys.pop(key)
if user_input_keys:
raise ServerError(msg=f"extra key {list(user_input_keys.keys())} found in user input")
else:
if input_schema["key"] not in user_input:
raise ServerError(msg=f"key {input_schema['key']} not found in user input")
@classmethod
def _update_old_message(cls, user_input: dict, message_db: ChatMessage, message_content: str,
verify_input: bool = False):
"""
if ChatResponse is not None: add new message
if ChatMessage is not None: update old message
return ChatResponse | None, ChatMessage | None
"""
if not message_db:
if verify_input:
raise ServerError(msg="message info not found by message id")
return None
# Update the input and selection of the user in the output to be entered message
old_message = json.loads(message_db.message)
if message_db.category == WorkflowEventType.OutputWithInput.value:
old_message['hisValue'] = user_input[old_message['node_id']][old_message['key']]
elif message_db.category == WorkflowEventType.OutputWithChoose.value:
old_message['hisValue'] = user_input[old_message['node_id']][old_message['key']]
elif message_db.category == WorkflowEventType.UserInput.value:
if verify_input:
cls._verify_input_schema(old_message, user_input)
user_input = user_input[old_message['node_id']]
# If the front-end passes user input, the front-end content is used.
if message_content:
user_input_message = message_content
# Instructions are form inputs
elif old_message['input_schema']['tab'] == 'form_input':
user_input_message = ''
for key_info in old_message['input_schema']['value']:
user_input_message += f"{key_info['value']}:{user_input.get(key_info['key'], '')}\n"
else:
# Description Dialog Input, Uploaded file information needs to be added, It is related to the data structure of the input node.
user_input_message = user_input[old_message['input_schema']['key']]
dialog_files_content = user_input.get('dialog_files_content', [])
for one in dialog_files_content:
user_input_message += f"\n{os.path.basename(one).split('?')[0]}"
return ChatResponse(
message=user_input_message,
category='question',
), None
message_db.message = json.dumps(old_message, ensure_ascii=False)
return None, message_db
def update_old_message(self, user_input: dict, message_db: ChatMessage, message_content: str,
verify_input: bool = False):
chat_response, message = self._update_old_message(user_input, message_db, message_content, verify_input)
if chat_response:
self.save_chat_message(chat_response)
return
if message:
ChatMessageDao.update_message_model(message)
async def async_update_old_message(self, user_input: dict, message_db: ChatMessage, message_content: str,
verify_input: bool = False):
chat_response, message = self._update_old_message(user_input, message_db, message_content, verify_input)
if chat_response:
self.save_chat_message(chat_response)
return
if message:
await ChatMessageDao.aupdate_message_model(message)
def get_user_input(self) -> dict | None:
ret = self.redis_client.get(self.workflow_input_key)
if ret:
self.redis_client.delete(self.workflow_input_key)
return ret
def set_workflow_stop(self):
from bisheng.worker.workflow.tasks import stop_workflow
self.redis_client.set(self.workflow_stop_key, 1, expiration=3600 * 24)
stop_workflow.delay(self.unique_id, self.workflow_id, self.chat_id, self.user_id)
async def async_set_workflow_stop(self):
from bisheng.worker.workflow.tasks import stop_workflow
await self.redis_client.aset(self.workflow_stop_key, 1, expiration=3600 * 24)
stop_workflow.delay(self.unique_id, self.workflow_id, self.chat_id, self.user_id)
def get_workflow_stop(self) -> bool | None:
""" In order to stop in timeworkflow, Do not cache memory """
return self.redis_client.get(self.workflow_stop_key) == 1
async def async_get_workflow_stop(self) -> bool | None:
""" In order to stop in timeworkflow, Do not cache memory """
return await self.redis_client.aget(self.workflow_stop_key) == 1
def send_chat_response(self, chat_response: ChatResponse):
""" Send a chat message """
self.insert_workflow_response(chat_response.dict())
# Determine if it needs to be stoppedworkflow, Don't judge when streaming, queries are too frequent and can't be stoppedworkflow
if chat_response.category == WorkflowEventType.StreamMsg.value:
return
if self.workflow and self.get_workflow_stop():
self.workflow.stop()
def save_chat_message(self, chat_response: ChatResponse, source_documents=None) -> int | str | None:
""" save chat message to database
return message id
"""
if not self.chat_id:
# Generate a fake messageidPrevent duplicate front-end message rendering
return uuid.uuid4().hex
# Judgment traceability
if source_documents:
result = {}
extra = {}
if isinstance(source_documents, Document):
result = source_documents
source, result = sync_judge_source(result, source_documents, self.chat_id, extra)
chat_response.source = source
chat_response.extra = json.dumps(extra, ensure_ascii=False)
message = ChatMessageDao.insert_one(ChatMessage(
user_id=self.user_id,
chat_id=self.chat_id,
flow_id=self.workflow_id,
type=chat_response.type,
is_bot=chat_response.is_bot,
source=chat_response.source,
message=chat_response.message if isinstance(chat_response.message, str) else json.dumps(
chat_response.message, ensure_ascii=False),
extra=chat_response.extra,
category=chat_response.category,
files=json.dumps(chat_response.files, ensure_ascii=False)
))
# If the document is traceable, handle the recallchunk
if chat_response.source not in [0, 4]:
thread_pool.submit(f"workflow_source_document_{self.chat_id}",
sync_process_source_document,
source_documents, self.chat_id, message.id, chat_response.message.get('msg'))
# Determine if a new session is needed
if not self.create_session and chat_response.category != WorkflowEventType.UserInput.value:
# Insert a new session without session data
if not MessageSessionDao.get_one(self.chat_id):
db_workflow = FlowDao.get_flow_by_id(self.workflow_id)
MessageSessionDao.insert_one(MessageSession(
chat_id=self.chat_id,
flow_id=self.workflow_id,
flow_name=db_workflow.name,
flow_type=FlowType.WORKFLOW.value,
user_id=self.user_id,
))
# RecordTelemetryJournal
telemetry_service.log_event_sync(user_id=self.user_id,
event_type=BaseTelemetryTypeEnum.NEW_MESSAGE_SESSION,
trace_id=trace_id_var.get(),
event_data=NewMessageSessionEventData(
session_id=self.chat_id,
app_id=self.workflow_id,
source=self.source,
app_name=db_workflow.name,
app_type=ApplicationTypeEnum.WORKFLOW
)
)
self.create_session = True
return message.id
def on_node_start(self, data: NodeStartData):
""" node start event """
logger.debug(f'node start: {data}')
self.send_chat_response(
ChatResponse(message=data.dict(),
category=WorkflowEventType.NodeRun.value,
type='start',
flow_id=self.workflow_id,
chat_id=self.chat_id))
def on_node_end(self, data: NodeEndData):
""" node end event """
logger.debug(f'node end: {data}')
self.send_chat_response(
ChatResponse(message=data.dict(),
category=WorkflowEventType.NodeRun.value,
type='end',
flow_id=self.workflow_id,
chat_id=self.chat_id))
def on_user_input(self, data: UserInputData):
""" user input event """
logger.debug(f'user input: {data}')
chat_response = ChatResponse(message=data.dict(),
category=WorkflowEventType.UserInput.value,
type='over',
flow_id=self.workflow_id,
chat_id=self.chat_id)
msg_id = self.save_chat_message(chat_response)
if msg_id:
chat_response.message_id = msg_id
self.send_chat_response(chat_response)
def on_guide_word(self, data: GuideWordData):
""" guide word event """
logger.debug(f'guide word: {data}')
self.send_chat_response(
ChatResponse(message=data.dict(),
category=WorkflowEventType.GuideWord.value,
type='over',
flow_id=self.workflow_id,
chat_id=self.chat_id))
def on_guide_question(self, data: GuideQuestionData):
""" guide question event """
logger.debug(f'guide question: {data}')
self.send_chat_response(
ChatResponse(message=data.dict(),
category=WorkflowEventType.GuideQuestion.value,
type='over',
flow_id=self.workflow_id,
chat_id=self.chat_id))
def on_output_msg(self, data: OutputMsgData):
logger.debug(f'output msg: {data}')
chat_response = ChatResponse(message=data.dict(exclude={'source_documents'}),
category=WorkflowEventType.OutputMsg.value,
extra='',
type='over',
flow_id=self.workflow_id,
chat_id=self.chat_id,
files=data.files)
msg_id = self.save_chat_message(chat_response, source_documents=data.source_documents)
if msg_id:
chat_response.message_id = msg_id
self.send_chat_response(chat_response)
def on_stream_msg(self, data: StreamMsgData):
logger.debug(f'stream msg: {data}')
self.send_chat_response(
ChatResponse(message=data.dict(),
category=WorkflowEventType.StreamMsg.value,
extra='',
type='stream',
flow_id=self.workflow_id,
chat_id=self.chat_id))
def on_stream_over(self, data: StreamMsgOverData):
logger.debug(f'stream over: {data}')
# Replaceminioright of privacysharePrefix bynginxShare ugly solve
minio_share = settings.get_minio_conf().sharepoint
data.msg = data.msg.replace(f"http://{minio_share}", "")
chat_response = ChatResponse(message=data.dict(exclude={'source_documents'}),
category=WorkflowEventType.StreamMsg.value,
extra='',
type='end',
flow_id=self.workflow_id,
chat_id=self.chat_id)
msg_id = self.save_chat_message(chat_response, source_documents=data.source_documents)
if msg_id:
chat_response.message_id = msg_id
self.send_chat_response(chat_response)
def on_output_choose(self, data: OutputMsgChooseData):
logger.debug(f'output choose: {data}')
chat_response = ChatResponse(message=data.dict(exclude={'source_documents'}),
category=WorkflowEventType.OutputWithChoose.value,
extra='',
type='over',
flow_id=self.workflow_id,
chat_id=self.chat_id,
files=data.files)
msg_id = self.save_chat_message(chat_response, source_documents=data.source_documents)
if msg_id:
chat_response.message_id = msg_id
self.send_chat_response(chat_response)
def on_output_input(self, data: OutputMsgInputData):
logger.debug(f'output input: {data}')
chat_response = ChatResponse(message=data.dict(exclude={'source_documents'}),
category=WorkflowEventType.OutputWithInput.value,
extra='',
type='over',
flow_id=self.workflow_id,
chat_id=self.chat_id,
files=data.files)
msg_id = self.save_chat_message(chat_response, source_documents=data.source_documents)
if msg_id:
chat_response.message_id = msg_id
self.send_chat_response(chat_response)
================================================
FILE: src/backend/bisheng/worker/workflow/tasks.py
================================================
import time
from loguru import logger
from bisheng.api.services.workflow import WorkFlowService
from bisheng.common.constants.enums.telemetry import BaseTelemetryTypeEnum, ApplicationTypeEnum
from bisheng.common.schemas.telemetry.event_data_schema import ApplicationProcessEventData
from bisheng.common.services import telemetry_service
from bisheng.common.services.config_service import settings
from bisheng.core.logger import trace_id_var
from bisheng.database.models.flow import FlowDao
from bisheng.utils.exceptions import IgnoreException
from bisheng.worker.main import bisheng_celery
from bisheng.worker.utils.stateful_worker import StatefulWorker
from bisheng.worker.workflow.redis_callback import RedisCallback
from bisheng.workflow.common.workflow import WorkflowStatus
from bisheng.workflow.graph.workflow import Workflow
# workflow execute stateful worker for assigning tasks to bound workers
workflow_stateful_worker = StatefulWorker(queue_prefix="workflow_celery", bound_nodes_prefix="workflow_bound_nodes:")
# Stores global workflow objects
_global_workflow: dict[str, Workflow] = {}
def _clear_workflow_obj(unique_id: str):
""" Clear Global Workflow Objects """
if unique_id in _global_workflow:
del _global_workflow[unique_id]
logger.debug(f'clear workflow object for unique_id: {unique_id}')
else:
logger.warning(f'workflow object not found for unique_id: {unique_id}')
def _judge_workflow_status(redis_callback: RedisCallback, workflow: Workflow):
status = workflow.status()
reason = workflow.reason()
if workflow.status() in [WorkflowStatus.SUCCESS.value, WorkflowStatus.FAILED.value]:
redis_callback.set_workflow_status(status, reason)
_clear_workflow_obj(redis_callback.unique_id)
return
if workflow.status() == WorkflowStatus.INPUT.value:
# If it is an input state, place the object in memory
_global_workflow[redis_callback.unique_id] = workflow
# redis_callback.save_workflow_object(workflow)
redis_callback.set_workflow_status(status, reason)
return
logger.error(f'unexpected workflow status error: {status}')
redis_callback.set_workflow_status(WorkflowStatus.FAILED.value,
f'workflow run failed, unexpected status: {status}')
_clear_workflow_obj(redis_callback.unique_id)
def _execute_workflow(unique_id: str, workflow_id: str, chat_id: str, user_id: int, source: str = "platform"):
redis_callback = RedisCallback(unique_id, workflow_id, chat_id, user_id, source=source)
try:
# update workflow status
redis_callback.set_workflow_status(WorkflowStatus.RUNNING.value)
# get workflow data
workflow_data = redis_callback.get_workflow_data()
if not workflow_data:
raise Exception('workflow data not found maybe data is expired')
# init workflow
workflow_conf = settings.get_workflow_conf()
workflow_info = FlowDao.get_flow_by_id(workflow_id)
workflow_name = workflow_info.name if workflow_info else workflow_id
workflow = Workflow(workflow_id, workflow_name,
user_id, workflow_data, False,
workflow_conf.max_steps,
workflow_conf.timeout,
redis_callback)
redis_callback.workflow = workflow
status, reason = workflow.run()
_judge_workflow_status(redis_callback, workflow)
except IgnoreException as e:
logger.warning(f'execute_workflow ignore error: {e}')
redis_callback.set_workflow_status(WorkflowStatus.FAILED.value, str(e))
_clear_workflow_obj(redis_callback.unique_id)
except Exception as e:
logger.exception('execute_workflow error')
redis_callback.set_workflow_status(WorkflowStatus.FAILED.value, str(e)[:100])
_clear_workflow_obj(redis_callback.unique_id)
@bisheng_celery.task
def execute_workflow(unique_id: str, workflow_id: str, chat_id: str, user_id: int, source: str = "platform"):
""" Implementationworkflow """
trace_id_var.set(unique_id)
start_time = time.time()
try:
_execute_workflow(unique_id, workflow_id, chat_id, user_id, source)
finally:
end_time = time.time()
workflow_info = WorkFlowService.get_one_workflow_simple_info_sync(workflow_id)
telemetry_service.log_event_sync(user_id=user_id,
event_type=BaseTelemetryTypeEnum.APPLICATION_PROCESS,
trace_id=trace_id_var.get(),
event_data=ApplicationProcessEventData(
app_id=workflow_id,
app_name=workflow_info.name if workflow_info else workflow_id,
app_type=ApplicationTypeEnum.WORKFLOW,
chat_id=chat_id,
start_time=int(start_time),
end_time=int(end_time),
process_time=int((end_time - start_time) * 1000)
))
def _continue_workflow(unique_id: str, workflow_id: str, chat_id: str, user_id: int, source: str = "platform"):
""" Resumeworkflow """
redis_callback = RedisCallback(unique_id, workflow_id, chat_id, user_id, source=source)
try:
workflow = _global_workflow.get(redis_callback.unique_id, None)
if not workflow:
raise Exception('workflow object not found maybe data is expired')
if workflow.status() not in [WorkflowStatus.INPUT.value, WorkflowStatus.INPUT_OVER.value]:
raise Exception(f'workflow status is {workflow.status()} not INPUT')
user_input = redis_callback.get_user_input()
if not user_input:
raise IgnoreException('workflow continue not found user input')
redis_callback.set_workflow_status(WorkflowStatus.RUNNING.value)
status, reason = workflow.run(user_input)
_judge_workflow_status(redis_callback, workflow)
except IgnoreException as e:
logger.warning(f'continue_workflow ignore error: {e}')
redis_callback.set_workflow_status(WorkflowStatus.FAILED.value, str(e))
_clear_workflow_obj(redis_callback.unique_id)
except Exception as e:
logger.exception('continue_workflow error')
redis_callback.set_workflow_status(WorkflowStatus.FAILED.value, str(e)[:100])
_clear_workflow_obj(redis_callback.unique_id)
@bisheng_celery.task
def continue_workflow(unique_id: str, workflow_id: str, chat_id: str, user_id: int, source: str = "platform"):
""" Resumeworkflow """
trace_id_var.set(unique_id)
start_time = time.time()
try:
_continue_workflow(unique_id, workflow_id, chat_id, user_id, source)
finally:
end_time = time.time()
workflow_info = WorkFlowService.get_one_workflow_simple_info_sync(workflow_id)
telemetry_service.log_event_sync(user_id=user_id,
event_type=BaseTelemetryTypeEnum.APPLICATION_PROCESS,
trace_id=trace_id_var.get(),
event_data=ApplicationProcessEventData(
app_id=workflow_id,
app_name=workflow_info.name if workflow_info else workflow_id,
app_type=ApplicationTypeEnum.WORKFLOW,
chat_id=chat_id,
start_time=int(start_time),
end_time=int(end_time),
process_time=int((end_time - start_time) * 1000)
))
@bisheng_celery.task
def stop_workflow(unique_id: str, workflow_id: str, chat_id: str, user_id: int):
""" Stopworkflow """
trace_id_var.set(unique_id)
redis_callback = RedisCallback(unique_id, workflow_id, chat_id, user_id)
if unique_id not in _global_workflow:
redis_callback.set_workflow_status(WorkflowStatus.FAILED.value, 'workflow stop by user')
logger.warning("stop_workflow called but workflow not found in global cache")
return
workflow = _global_workflow[unique_id]
workflow.stop()
while workflow.status() == WorkflowStatus.RUNNING.value:
time.sleep(0.3)
status, reason = workflow.status(), workflow.reason()
if status != WorkflowStatus.FAILED.value:
status = WorkflowStatus.FAILED.value
reason = 'workflow stop by user'
redis_callback.set_workflow_status(status, reason)
_clear_workflow_obj(unique_id)
logger.info(f'workflow stop by user {user_id}')
================================================
FILE: src/backend/bisheng/workflow/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/callback/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/callback/base_callback.py
================================================
from abc import ABC
from bisheng.workflow.callback.event import NodeStartData, NodeEndData, OutputMsgChooseData, OutputMsgInputData, \
UserInputData, GuideWordData, GuideQuestionData, \
OutputMsgData, StreamMsgData, StreamMsgOverData
class BaseCallback(ABC):
def __init__(self, *args, **kwargs):
pass
def on_node_start(self, data: NodeStartData):
""" node start event """
pass
def on_node_end(self, data: NodeEndData):
""" node end event """
pass
def on_user_input(self, data: UserInputData):
""" user input event """
pass
def on_guide_word(self, data: GuideWordData):
""" guide word event """
pass
def on_guide_question(self, data: GuideQuestionData):
""" guide question event """
pass
def on_stream_msg(self, data: StreamMsgData):
pass
def on_stream_over(self, data: StreamMsgOverData):
pass
def on_output_msg(self, data: OutputMsgData):
pass
def on_output_choose(self, data: OutputMsgChooseData):
pass
def on_output_input(self, data: OutputMsgInputData):
pass
================================================
FILE: src/backend/bisheng/workflow/callback/event.py
================================================
from typing import Any, List, Optional
from pydantic import BaseModel, Field
# Node start event data
class NodeStartData(BaseModel):
unique_id: str = Field(..., description='Unique execution id')
node_id: str = Field(..., description='Node unique id')
name: str = Field(..., description='Node name')
class NodeEndData(NodeStartData):
reason: Optional[str] = Field(None, description='Reason for node exec error')
log_data: Any = Field(None, description='Log data on node exec success')
input_data: Any = Field(None, description='Input variable data for node exec')
class UserInputData(BaseModel):
node_id: str = Field(..., description='Node unique id')
name: str = Field(..., description='Node name')
input_schema: Any = Field(..., description='Input schema')
class GuideWordData(BaseModel):
unique_id: Optional[str] = Field(..., description='Unique execution id')
node_id: str = Field(..., description='Node unique id')
name: str = Field(..., description='Node name')
guide_word: str = Field(..., description='Guide word')
class GuideQuestionData(BaseModel):
unique_id: Optional[str] = Field(..., description='Unique execution id')
node_id: str = Field(..., description='Node unique id')
name: str = Field(..., description='Node name')
guide_question: List[str] = Field(..., description='Guide question')
class OutputMsgData(BaseModel):
unique_id: str = Field(..., description='Unique execution id')
node_id: str = Field(..., description='Node unique id')
name: str = Field(..., description='Node name')
msg: str = Field(default='', description='Output msg')
files: List[dict] = Field(default_factory=list, description='Output files', exclude=True)
output_key: str = Field(default="", description='Whether the message is stream')
source_documents: Optional[Any] = Field(default=None, description='Source documents')
class OutputMsgInputData(OutputMsgData):
key: str = Field('', description='variable key')
unique_id: Optional[str] = Field('', description='Unique execution id')
input_msg: str = Field('', description='default input msg')
class OutputMsgChooseData(OutputMsgData):
key: str = Field('', description='variable key')
unique_id: Optional[str] = Field('', description='Unique execution id')
options: Any = Field(None, description='default msg')
class StreamMsgData(BaseModel):
unique_id: str = Field(..., description='Unique execution id')
node_id: str = Field(..., description='Node unique id')
name: str = Field(..., description='Node name')
msg: Optional[str] = Field('', description='Stream msg')
reasoning_content: Optional[str] = Field(None, description='Reasoning content')
output_key: str = Field(..., description='Whether the message is stream')
class StreamMsgOverData(StreamMsgData):
source_documents: Optional[List[Any]] = Field(default=[], description='Source documents')
================================================
FILE: src/backend/bisheng/workflow/callback/llm_callback.py
================================================
from typing import Any, Dict, Optional, List, Union
from bisheng.workflow.callback.base_callback import BaseCallback
from bisheng.workflow.callback.event import OutputMsgData, StreamMsgData, StreamMsgOverData
from langchain_core.callbacks.base import BaseCallbackHandler
from langchain_core.outputs import LLMResult
from loguru import logger
class LLMNodeCallbackHandler(BaseCallbackHandler):
"""Callback handler for streaming LLM responses."""
def __init__(
self,
callback: BaseCallback,
unique_id: str,
node_id: str,
node_name: str,
output: bool,
output_key: str,
stream: bool = True,
tool_list: Optional[List[Any]] = None,
cancel_llm_end: bool = False,
):
self.callback_manager = callback
self.unique_id = unique_id
self.node_id = node_id
self.node_name = node_name
self.output = output
self.output_len = 0
self.output_key = output_key
self.stream = stream
self.tool_list = tool_list
self.cancel_llm_end = cancel_llm_end
self.reasoning_content = ''
logger.info('on_llm_new_token {} outkey={}', self.output, self.output_key)
async def on_tool_start(self, serialized: Dict[str, Any], input_str: str,
**kwargs: Any) -> Any:
"""Run when tool starts running."""
logger.debug(
f'on_tool_start serialized={serialized} input_str={input_str} kwargs={kwargs}')
if self.tool_list is not None:
self.tool_list.append({
'type': 'start',
'run_id': kwargs.get('run_id').hex,
'name': serialized['name'],
'input': input_str,
})
if serialized['name'] == 'sql_agent':
self.output = False
async def on_tool_end(self, output: str, **kwargs: Any) -> Any:
"""Run when tool ends running."""
logger.debug(f'on_tool_end output={output} kwargs={kwargs}')
result = output if isinstance(output, str) else getattr(output, 'content', output)
if self.tool_list is not None:
self.tool_list.append({
'type': 'end',
'run_id': kwargs.get('run_id').hex,
'name': kwargs['name'],
'output': result,
})
if kwargs['name'] == 'sql_agent':
self.output = True
async def on_tool_error(self, error: Union[Exception, KeyboardInterrupt],
**kwargs: Any) -> Any:
"""Run when tool errors."""
logger.debug(f'on_tool_error error={error} kwargs={kwargs}')
if self.tool_list is not None:
self.tool_list.append({
'type': 'error',
'run_id': kwargs.get('run_id').hex,
'error': str(error),
})
if kwargs['name'] == 'sql_agent':
self.output = True
def on_llm_new_token(self, token: str, **kwargs: Any) -> None:
chunk = kwargs.get('chunk', None)
# azureOccasionally returns aNone
if token is None and chunk is None:
return
if not self.output or not self.stream:
return
self.output_len += len(token) # Determine if the streaming output has been completed
self.callback_manager.on_stream_msg(
StreamMsgData(node_id=self.node_id,
name=self.node_name,
msg=token,
reasoning_content=getattr(chunk.message, 'additional_kwargs', {}).get('reasoning_content'),
unique_id=self.unique_id,
output_key=self.output_key))
def on_llm_end(self, response: LLMResult, **kwargs: Any) -> None:
self.reasoning_content = getattr(response.generations[0][0].message, 'additional_kwargs', {}).get('reasoning_content')
if self.cancel_llm_end:
return
if not self.output:
return
msg = response.generations[0][0].message
# ChatTongYi vl model special text
if isinstance(msg.content, list):
msg = ''.join([one.get('text', '') for one in msg.content])
else:
msg = msg.content
if not msg:
logger.warning('LLM output is empty')
return
if self.stream and self.output_len > 0:
# Streaming output end needs to return a streaming end event
self.callback_manager.on_stream_over(StreamMsgOverData(node_id=self.node_id,
name=self.node_name,
msg=msg,
reasoning_content=self.reasoning_content,
unique_id=self.unique_id,
output_key=self.output_key))
return
# If the output is required, but the stream output is not performed, a supplement is made. This happens when the cache is hit. This is also the case when output is required
self.callback_manager.on_output_msg(
OutputMsgData(node_id=self.node_id,
name=self.node_name,
msg=msg,
unique_id=self.unique_id,
output_key=self.output_key))
================================================
FILE: src/backend/bisheng/workflow/common/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/common/condition.py
================================================
class ComparisonType:
EQUAL = "equals"
NOT_EQUAL = "not_equals"
GREATER_THAN = "greater_than"
LESS_THAN = "less_than"
GREATER_THAN_OR_EQUAL = "greater_than_or_equal"
LESS_THAN_OR_EQUAL = "less_than_or_equal"
CONTAINS = "contains"
NOT_CONTAINS = "not_contains"
IS_EMPTY = "is_empty"
IS_NOT_EMPTY = "is_not_empty"
STARTS_WITH = "starts_with"
ENDS_WITH = "ends_with"
REGEX = "regex"
class LogicType:
AND = "and"
OR = "or"
================================================
FILE: src/backend/bisheng/workflow/common/const.py
================================================
================================================
FILE: src/backend/bisheng/workflow/common/knowledge.py
================================================
from datetime import datetime
from typing import List, Dict, Any, Literal
from langchain_core.documents import Document
from loguru import logger
from pydantic import BaseModel, Field
from bisheng.common.constants.enums.telemetry import ApplicationTypeEnum
from bisheng.common.constants.vectorstore_metadata import KNOWLEDGE_RAG_METADATA_SCHEMA
from bisheng.core.vectorstore.multi_retriever import MultiRetriever
from bisheng.knowledge.domain.knowledge_rag import KnowledgeRag
from bisheng.knowledge.domain.models.knowledge import Knowledge, MetadataFieldType
from bisheng.knowledge.domain.models.knowledge_file import KnowledgeFileDao
from bisheng.llm.domain import LLMService
from bisheng.tool.domain.langchain.knowledge import KnowledgeRetrieverTool
from bisheng.workflow.common.condition import ComparisonType
from bisheng.workflow.nodes.base import BaseNode
class ConditionOne(BaseModel):
id: str = Field(..., description='Unique id for condition')
knowledge_id: int = Field(..., description='metadata filed belong for knowledge`s id')
metadata_field: str = Field(..., description='knowledge metadata field')
comparison_operation: str = Field(..., description='Compare type')
right_value_type: str = Field(..., description='Right value type')
right_value: str = Field(..., description='Right value')
def convert_right_value(self, field_type: str, right_value: Any, is_preset: bool) -> Any:
# no need to convert right value for is_empty and is_not_empty
if self.comparison_operation in ['is_empty', 'is_not_empty'] or is_preset:
return right_value
# only for user metadata field, need to convert right value type
if field_type in [MetadataFieldType.STRING.value]:
right_value = str(right_value)
if not right_value:
raise ValueError("Right value cannot be empty for the selected comparison operation")
elif field_type in [MetadataFieldType.NUMBER.value]:
right_value = int(right_value)
elif field_type == MetadataFieldType.TIME.value:
if isinstance(right_value, int):
# timestamp
right_value = datetime.fromtimestamp(right_value)
else:
# iso format
right_value = datetime.fromisoformat(right_value)
right_value = int(right_value.timestamp())
else:
raise ValueError(f"Unsupported metadata field type: {field_type}")
return right_value
def convert_preset_filed(self) -> (str, str):
""" convert preset field to mysql field in knowledge file table"""
if self.metadata_field == "document_id":
return "id"
elif self.metadata_field == "document_name":
return "file_name"
elif self.metadata_field == "upload_time":
return "create_time"
elif self.metadata_field == "update_time":
return "update_time"
elif self.metadata_field == "uploader":
return "user_name"
elif self.metadata_field == "updater":
return "updater_name"
else:
raise ValueError(f"Unsupported preset metadata field: {self.metadata_field}")
def get_knowledge_file_filter(self, field_info: Dict, parent_node: BaseNode, is_preset: bool) -> (str, List[Dict]):
""" get knowledge file filter field info for mysql """
right_value = self.right_value
if self.right_value_type == 'ref' and self.right_value:
right_value = parent_node.get_other_node_variable(self.right_value)
field_type = field_info.get('field_type')
right_value = self.convert_right_value(field_type, right_value, is_preset)
if is_preset:
field_key = self.convert_preset_filed()
else:
field_key = f"JSON_UNQUOTE(JSON_EXTRACT(`user_metadata`, '$.{self.metadata_field}.field_value'))"
key_info = {}
if self.comparison_operation == ComparisonType.EQUAL:
key_info['comparison'] = '='
key_info['value'] = right_value
elif self.comparison_operation == ComparisonType.NOT_EQUAL:
key_info['comparison'] = '!='
key_info['value'] = right_value
elif self.comparison_operation == ComparisonType.CONTAINS:
key_info['comparison'] = 'like'
key_info['value'] = f"%{right_value}%"
elif self.comparison_operation == ComparisonType.NOT_CONTAINS:
key_info['comparison'] = 'not like'
key_info['value'] = f"%{right_value}%"
elif self.comparison_operation == ComparisonType.STARTS_WITH:
key_info['comparison'] = 'like'
key_info['value'] = f"{right_value}%"
elif self.comparison_operation == ComparisonType.ENDS_WITH:
key_info['comparison'] = 'like'
key_info['value'] = f"%{right_value}"
elif self.comparison_operation == ComparisonType.IS_EMPTY:
key_info['comparison'] = '='
key_info['value'] = 'null'
elif self.comparison_operation == ComparisonType.IS_NOT_EMPTY:
key_info['comparison'] = '!='
key_info['value'] = 'null'
elif self.comparison_operation == ComparisonType.GREATER_THAN:
key_info['comparison'] = '>'
key_info['value'] = right_value
elif self.comparison_operation == ComparisonType.GREATER_THAN_OR_EQUAL:
key_info['comparison'] = '>='
key_info['value'] = right_value
elif self.comparison_operation == ComparisonType.LESS_THAN:
key_info['comparison'] = '<'
key_info['value'] = right_value
elif self.comparison_operation == ComparisonType.LESS_THAN_OR_EQUAL:
key_info['comparison'] = '<='
key_info['value'] = right_value
else:
raise ValueError(f"Unsupported comparison operation: {self.comparison_operation}")
if not is_preset and self.comparison_operation in [ComparisonType.GREATER_THAN,
ComparisonType.GREATER_THAN_OR_EQUAL,
ComparisonType.LESS_THAN,
ComparisonType.LESS_THAN_OR_EQUAL]:
key_info['extra_filter'] = [{
'comparison': '!=',
'value': 'null'
}]
return {field_key: key_info}
class ConditionCases(BaseModel):
id: str = Field(default=None, description='Unique id for condition case')
conditions: List[ConditionOne] = Field(default_factory=list, description='List of conditions')
operator: Literal['and', 'or'] = Field(default='and', description='Logical operator to combine conditions')
enabled: bool = Field(default=False, description='Whether the condition case is enabled')
def get_knowledge_filter(self, knowledge: Knowledge, parent_node: BaseNode) -> (str, Dict):
""" if return is None, filter file is empty, don't need to retrieve from this knowledge """
if not self.enabled or not self.conditions:
return "", {}
metadata_field_info = {}
if knowledge.metadata_fields:
metadata_field_info = {one["field_name"]: one for one in knowledge.metadata_fields}
# Built-in metadata fields
preset_field_info = {
one.field_name: one.model_dump() for one in KNOWLEDGE_RAG_METADATA_SCHEMA if
one.field_name != "user_metadata"
}
all_filter_field = []
for condition in self.conditions:
if int(condition.knowledge_id) != knowledge.id:
continue
if field_info := preset_field_info.get(condition.metadata_field):
filter_field_info = condition.get_knowledge_file_filter(field_info, parent_node, True)
elif field_info := metadata_field_info.get(condition.metadata_field):
filter_field_info = condition.get_knowledge_file_filter(field_info, parent_node, False)
else:
logger.warning(f"condition field {condition.metadata_field} not in knowledge metadata fields")
raise ValueError(f"field {condition.metadata_field} not in knowledge metadata fields")
all_filter_field.append(filter_field_info)
if not all_filter_field:
return "", {}
file_ids = KnowledgeFileDao.filter_file_by_metadata_fields(knowledge.id, self.operator, all_filter_field)
if not file_ids:
# no file match the filter condition
logger.debug(f'knowledge {knowledge.id} no file match the filter condition')
return None, None
milvus_filter = f"document_id in {file_ids}"
es_filter = {"filter": [{"terms": {"metadata.document_id": file_ids}}]}
return milvus_filter, es_filter
class RagUtils(BaseNode):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self._knowledge_type = self.node_params.get('knowledge', {}).get('type', "knowledge")
self._knowledge_value = [
one['key'] for one in self.node_params.get('knowledge', {}).get('value', [])
]
self._advance_kwargs = self.node_params.get('advanced_retrieval_switch', {})
self._metadata_filter = ConditionCases(**self.node_params.get('metadata_filter', {}))
if self._advance_kwargs:
self._advance_kwargs = self.node_params.get('advanced_retrieval_switch', {})
self._knowledge_auth = self._advance_kwargs['user_auth']
self._max_chunk_size = int(self._advance_kwargs['max_chunk_size'])
self._keyword_weight = float(self._advance_kwargs['keyword_weight'])
self._vector_weight = float(self._advance_kwargs['vector_weight'])
self._rerank_flag = self._advance_kwargs['rerank_flag']
self._rerank_model_id = self._advance_kwargs['rerank_model']
else:
self._knowledge_auth = self.node_params.get('user_auth', False)
self._max_chunk_size = int(self.node_params.get('max_chunk_size', 15000))
self._keyword_weight = 0.5
self._vector_weight = 0.5
self._rerank_flag = False
self._rerank_model_id = ''
self._multi_milvus_retriever = None
self._multi_es_retriever = None
self._knowledge_vector_list = []
self._retriever_kwargs = {"k": 100, "param": {"ef": 110}}
self._rerank_model = None
self._knowledge_retriever_tool = None
def _run(self, unique_id: str) -> Dict[str, Any]:
raise NotImplementedError()
@staticmethod
def format_timestamp(timestamp: int) -> str:
try:
return datetime.fromtimestamp(timestamp).strftime('%Y-%m-%dT%H:%M:%S')
except Exception as e:
logger.error(f"Error formatting timestamp {timestamp}: {e}")
return str(timestamp)
def retrieve_question(self, question: str) -> List[Document]:
# 1: retrieve documents from multi retrievers
knowledge_retriever_tool = KnowledgeRetrieverTool(
vector_retriever=self._multi_milvus_retriever,
elastic_retriever=self._multi_es_retriever,
max_content=self._max_chunk_size,
rrf_weights=[self._vector_weight, self._keyword_weight],
rrf_remove_zero_score=True,
rerank=self._rerank_model,
sort_by_source_and_index=True
)
finally_docs = knowledge_retriever_tool.invoke(input={"query": question})
all_file_id = set([one.metadata.get("document_id") for one in finally_docs])
file_map = {}
if finally_docs:
if self._knowledge_type == 'knowledge':
file_info = KnowledgeFileDao.get_file_by_ids(list(all_file_id))
file_map = {one.id: one for one in file_info}
for one in finally_docs:
if "upload_time" in one.metadata:
one.metadata["upload_time"] = self.format_timestamp(one.metadata["upload_time"])
if "update_time" in one.metadata:
one.metadata["update_time"] = self.format_timestamp(one.metadata["update_time"])
file_id = one.metadata.get('document_id') or one.metadata.get('file_id')
if file_id and file_map.get(file_id):
for user_key, user_value in one.metadata.get('user_metadata', {}).items():
field_info = file_map[file_id].user_metadata.get(user_key)
if field_info and field_info.get('field_type') == MetadataFieldType.TIME.value:
one.metadata["user_metadata"][user_key] = self.format_timestamp(user_value)
return finally_docs
def init_user_question(self) -> List[str]:
# Convert all user questions to strings by default
ret = []
for one in self.node_params['user_question']:
ret.append(f"{self.get_other_node_variable(one)}")
return ret
def init_rerank_model(self):
if not self._rerank_flag or not self._rerank_model_id:
return
if self._rerank_model:
return
self._rerank_model = LLMService.get_bisheng_rerank_sync(model_id=self._rerank_model_id,
app_id=self.workflow_id,
app_name=self.workflow_name,
app_type=ApplicationTypeEnum.WORKFLOW,
user_id=self.user_id)
def init_multi_retriever(self):
if self._knowledge_type == "knowledge":
self.init_knowledge_retriever()
else:
self.init_file_retriever()
def init_knowledge_retriever(self):
""" retriever from knowledge base """
if not self._knowledge_vector_list:
self._knowledge_vector_list = KnowledgeRag.get_multi_knowledge_vectorstore_sync(
invoke_user_id=self.user_id,
knowledge_ids=self._knowledge_value,
user_name=self.user_info.user_name,
check_auth=self._knowledge_auth,
include_es=self._keyword_weight > 0,
include_milvus=self._vector_weight > 0,
)
all_milvus = []
all_milvus_filter = []
all_es = []
all_es_filter = []
self._multi_milvus_retriever = None
self._multi_es_retriever = None
for knowledge_id, knowledge_info in self._knowledge_vector_list.items():
knowledge = knowledge_info.get('knowledge')
milvus_vector = knowledge_info.get('milvus')
es_vector = knowledge_info.get('es')
milvus_filter, es_filter = self._metadata_filter.get_knowledge_filter(knowledge=knowledge,
parent_node=self)
if milvus_filter is None and es_filter is None:
continue
if milvus_vector:
all_milvus.append(milvus_vector)
milvus_filter = {"expr": milvus_filter} if milvus_filter else {}
logger.debug(f'retrieve milvus filter: {milvus_filter}')
all_milvus_filter.append(milvus_filter | self._retriever_kwargs)
if es_vector:
all_es.append(es_vector)
logger.debug(f'retrieve es filter: {es_filter}')
all_es_filter.append(es_filter | self._retriever_kwargs)
if all_milvus:
self._multi_milvus_retriever = MultiRetriever(
vectors=all_milvus,
search_kwargs=all_milvus_filter,
finally_k=self._retriever_kwargs["k"]
)
if all_es:
self._multi_es_retriever = MultiRetriever(
vectors=all_es,
search_kwargs=all_es_filter,
finally_k=self._retriever_kwargs["k"]
)
def init_file_retriever(self):
""" retriever from file user upload """
file_ids = []
for one in self._knowledge_value:
file_metadata = self.get_other_node_variable(one)
if not file_metadata:
# No corresponding temporary file data found, User did not upload file
continue
file_ids.append(file_metadata[0]['document_id'])
if not file_ids:
self._multi_es_retriever = None
self._multi_milvus_retriever = None
return
embeddings = LLMService.get_knowledge_default_embedding(self.user_id)
if not embeddings:
raise Exception('No knowledge base defaults configuredembeddingModels')
# vectorstore use different collection_name for different embedding model
tmp_collection_name = self.get_milvus_collection_name(getattr(embeddings, 'model_id'))
milvus_vector = KnowledgeRag.init_milvus_vectorstore(collection_name=tmp_collection_name, embeddings=embeddings)
milvus_extra = {"expr": f"document_id in {file_ids}"}
self._multi_milvus_retriever = milvus_vector.as_retriever(search_kwargs=self._retriever_kwargs | milvus_extra)
es_extra = {"filter": [{"terms": {"metadata.document_id": file_ids}}]}
es_vector = KnowledgeRag.init_es_vectorstore_sync(index_name=self.tmp_collection_name)
self._multi_es_retriever = es_vector.as_retriever(search_kwargs=self._retriever_kwargs | es_extra)
================================================
FILE: src/backend/bisheng/workflow/common/node.py
================================================
import copy
from enum import Enum
from typing import Optional, Any, List
from pydantic import BaseModel, Field, field_validator
class NodeType(Enum):
""" Node type """
START = "start"
END = "end"
INPUT = "input"
AGENT = "agent"
CODE = "code"
CONDITION = "condition"
LLM = "llm"
OUTPUT = "output"
FAKE_OUTPUT = "fake_output"
QA_RETRIEVER = "qa_retriever"
RAG = "rag"
REPORT = "report"
TOOL = "tool"
KNOWLEDGE_RETRIEVER = "knowledge_retriever"
NOTE = 'note' # Notes node Knowledge is used to display annotations, not actual execution nodes
class NodeParams(BaseModel):
key: str = Field(default="", description="Variablekey")
label: Optional[str] = Field("", description="Variable description text")
value: Optional[Any] = Field(None, description="Value of the variable")
# Variable type -> Detailed reference for data format https://dataelem.feishu.cn/wiki/IfBvwwvfFiHjuQkjFJgcxzoGnxb
type: Optional[str] = Field("", description="Variable type")
help: Optional[str] = Field("", description="Variable Help Text")
tab: Optional[str] = Field("", description="Variable belongs totab, empty to show all")
placeholder: Optional[str] = Field("", description="Variable placeholder text")
required: Optional[bool] = Field(False, description="Required?")
options: Optional[Any] = Field(None, description="Variable options")
class NodeGroupParams(BaseModel):
name: Optional[str] = Field(default="", description="Group name")
params: List[NodeParams] = Field(..., description="Group params")
description: Optional[str] = Field(default="", description="Node description")
class BaseNodeData(BaseModel):
id: str = Field(default="", description="Unique id for node")
type: str = Field(..., description="Node type")
name: Optional[str] = Field(default="", description="Node name")
description: Optional[str] = Field(default="", description="Node description")
group_params: Optional[List[NodeGroupParams]] = Field(default=None, description="Node group params")
tab: Optional[dict] = Field({}, description="tab config")
tool_key: Optional[str] = Field("", description="unique tool id, only for tool node")
v: Optional[int] = Field(default=0, description="node version")
@field_validator('v', mode='before')
@classmethod
def convert_v_to_int(cls, v: str | int | None) -> int:
if isinstance(v, str):
return int(v)
elif v is None:
return 0
return v
def get_variable_info(self, variable_key: str) -> NodeParams | None:
for group_info in self.group_params:
for one in group_info.params:
if one.key == variable_key:
return copy.deepcopy(one)
return None
================================================
FILE: src/backend/bisheng/workflow/common/workflow.py
================================================
from enum import Enum
class WorkflowStatus(Enum):
WAITING = 'WAITING' # Waiting for asynchronous task scheduling
RUNNING = 'RUNNING'
SUCCESS = 'SUCCESS'
FAILED = 'FAILED'
INPUT = 'INPUT' # Status to be entered
INPUT_OVER = 'INPUT_OVER' # Entered status
================================================
FILE: src/backend/bisheng/workflow/edges/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/edges/edges.py
================================================
from typing import Optional, List, Any
from pydantic import BaseModel, Field
class EdgeBase(BaseModel):
id: str = Field(..., description="Unique id for edge")
source: str = Field(..., description="source node id")
sourceHandle: str = Field(..., description="source node handle")
sourceType: Optional[str] = Field("", description="source node type")
target: str = Field(..., description="target node id")
targetHandle: str = Field(..., description="target node handle")
targetType: Optional[str] = Field("", description="target node type")
class EdgeManage:
def __init__(self, edges: List[Any]):
self.edges: List[EdgeBase] = [EdgeBase(**one) for one in edges]
# source: [edges]
self.source_map = {}
# target: [edges]
self.target_map = {}
for one in self.edges:
if one.source not in self.source_map:
self.source_map[one.source] = []
self.source_map[one.source].append(one)
if one.target not in self.target_map:
self.target_map[one.target] = []
self.target_map[one.target].append(one)
def get_target_node(self, source: str) -> List[str] | None:
""" get target node id by source node id"""
if source not in self.source_map:
return None
return [one.target for one in self.source_map[source]]
def get_source_node(self, target: str) -> List[str] | None:
""" get source node id by target node id"""
if target not in self.target_map:
return None
return [one.source for one in self.target_map[target]]
def get_source_edges(self, target: str) -> List[EdgeBase] | None:
""" get source edges by target node id"""
if target not in self.target_map:
return None
return self.target_map[target]
def get_target_edges(self, source: str) -> List[EdgeBase] | None:
""" get target edges by source node id"""
if source not in self.source_map:
return None
return self.source_map[source]
def get_all_edges_nodes(self, start_node_id: str, end_node_id: str) -> List[List[str]]:
""" get all branch nodes from start node to end node """
branches = []
def get_node_branch(node_id, branch: List, node_map: dict):
# The nodes that have been traversed are no longer traversed, indicating that they are looped
if node_id in node_map or node_id == end_node_id:
branch.append(node_id)
branches.append(branch)
return branch
branch.append(node_id)
node_map[node_id] = True
next_nodes = self.get_target_node(node_id)
if not next_nodes:
branches.append(branch)
return branch
for one_node in next_nodes:
tmp_node_map = node_map.copy()
tmp_branch = branch.copy()
get_node_branch(one_node, tmp_branch, tmp_node_map)
return branch
get_node_branch(start_node_id, [], {})
return branches
def get_next_nodes(self, node_id: str, exclude: Optional[List[str]] = None) -> List[str] | None:
""" get all next nodes by node id"""
# Get direct downstream nodes
output_nodes = self.get_target_node(node_id)
if not output_nodes:
return []
if not exclude:
exclude = [node_id]
# Exclude specified nodes
for one in exclude:
if one in output_nodes:
output_nodes.remove(one)
exclude.extend(output_nodes)
for one in output_nodes:
next_nodes = self.get_next_nodes(one, exclude=exclude)
if next_nodes:
output_nodes.extend(next_nodes)
return output_nodes
================================================
FILE: src/backend/bisheng/workflow/graph/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/graph/graph_engine.py
================================================
import operator
from typing import Annotated, Any, Dict
from langgraph.checkpoint.memory import MemorySaver
from langgraph.constants import END, START
from langgraph.graph import StateGraph
from loguru import logger
from typing_extensions import TypedDict
from bisheng.utils.exceptions import IgnoreException
from bisheng.workflow.callback.base_callback import BaseCallback
from bisheng.workflow.callback.event import UserInputData
from bisheng.workflow.common.node import BaseNodeData, NodeType
from bisheng.workflow.common.workflow import WorkflowStatus
from bisheng.workflow.edges.edges import EdgeManage
from bisheng.workflow.graph.graph_state import GraphState
from bisheng.workflow.nodes.base import BaseNode
from bisheng.workflow.nodes.node_manage import NodeFactory
from bisheng.workflow.nodes.output.output_fake import OutputFakeNode
class TempState(TypedDict):
# not use, only for langgraph state graph
flag: Annotated[bool, operator.and_]
class GraphEngine:
def __init__(self,
user_id: int = None,
workflow_id: str = None,
workflow_name: str = '',
workflow_data: Dict = None,
async_mode: bool = False,
max_steps: int = 0,
callback: BaseCallback = None):
self.user_id = user_id
self.workflow_id = workflow_id
self.workflow_name = workflow_name
self.workflow_data = workflow_data
self.max_steps = max_steps
self.async_mode = async_mode
# Callbacks
self.callback = callback
# node_id: NodeInstance
self.nodes_map = {}
# record how many nodes fan in this node
self.nodes_fan_in = {} # node_id: [node_ids]
# record how many nodes next to this node
self.nodes_next_nodes = {} # node_id: {node_ids}
# node_id: 1; Represents fromstartThe longest path from a node to this node
self.node_level = {}
# List of mutually exclusive nodes, containingconditionNodes andoutputNode (selective interaction)
self.condition_nodes = []
self.edges = None
self.graph_state = GraphState()
# init langgraph state graph
self.graph_builder = StateGraph(TempState)
self.graph = None
self.graph_config = {'configurable': {'thread_id': '1'}, 'recursion_limit': 50}
self.status = WorkflowStatus.RUNNING.value
self.reason = '' # Failure Reason
self.build_edges()
self.build_nodes()
def build_edges(self):
# init edges
self.edges = EdgeManage(self.workflow_data.get('edges', []))
def add_node_edge(self, node_instance: BaseNode):
""" Link edges of nodes """
if node_instance.type == NodeType.END.value or node_instance.type == NodeType.FAKE_OUTPUT.value:
return
# get target nodes
target_node_ids = self.edges.get_target_node(node_instance.id)
source_node_ids = self.edges.get_source_node(node_instance.id)
# There are no linked nodes reporting errors
if not target_node_ids and not source_node_ids:
raise Exception(
f'node {node_instance.name} {node_instance.id} must have at least one edge')
# output Node followed by afake Nodes are used to handle interrupts
if node_instance.type == NodeType.OUTPUT.value:
fake_node = self.nodes_map[f'{node_instance.id}_fake']
if self.async_mode:
self.graph_builder.add_node(fake_node.id, fake_node.arun)
else:
self.graph_builder.add_node(fake_node.id, fake_node.run)
self.graph_builder.add_edge(node_instance.id, fake_node.id)
self.graph_builder.add_conditional_edges(
fake_node.id, node_instance.route_node,
{node_id: node_id
for node_id in target_node_ids})
return
# condition And output Need to connect behind the node langgraphright of privacy edge_condition
if node_instance.type == NodeType.CONDITION.value:
self.graph_builder.add_conditional_edges(
node_instance.id, node_instance.route_node,
{node_id: node_id
for node_id in target_node_ids})
return
# Links Totargetnode_amb
for node_id in target_node_ids:
if node_id not in self.nodes_map:
raise Exception(f'target node {node_id} not found')
if self.nodes_fan_in.get(node_id) and len(self.nodes_fan_in.get(node_id)) > 1:
# need wait all fan in node exec over
continue
self.graph_builder.add_edge(node_instance.id, node_id)
def build_more_fan_in_node(self):
for node_id, source_ids in self.nodes_fan_in.items():
if not source_ids or len(source_ids) <= 1:
continue
# There are multiple fan-in nodes to determine if this node needs to wait
wait_nodes, no_wait_nodes = self.parse_fan_in_node(node_id)
logger.debug(f'node {node_id} wait nodes {wait_nodes}, no wait nodes {no_wait_nodes}')
if wait_nodes:
self.graph_builder.add_edge(wait_nodes, node_id)
if no_wait_nodes:
for one in no_wait_nodes:
self.graph_builder.add_edge(one, node_id)
def parse_fan_in_node(self, node_id: str):
source_ids = self.nodes_fan_in.get(node_id)
# Whether the hierarchy of all precursor nodes is less than or equal to this node
all_source_node_prev = True
for one in source_ids:
if self.node_level[one] > self.node_level[node_id]:
all_source_node_prev = False
break
# in the precursor node Contains The downstream node of this node does not need to wait, it needs to be excludedoutputAndconditionNode, because the two nodes are connected to this node through the conditional edge
if not all_source_node_prev:
return [], [one for one in source_ids if not one.startswith(('output_', 'condition_'))]
# Determine if there is aconditionNode oroutputNode (selective interaction) to this node Two unique paths
all_branches = []
for one in self.condition_nodes:
if node_id == one:
continue
branches = self.edges.get_all_edges_nodes(one, node_id)
for branch in branches:
if node_id not in branch:
continue
branch.remove(node_id)
branch.remove(one)
all_branches.append(branch)
def judge_not_same_branch():
# Determine if there are two unique paths in all edges
for i in range(len(all_branches)):
for j in range(i + 1, len(all_branches)):
if not (set(all_branches[i]) & set(all_branches[j])):
return True
return False
# Explain that it is a mutually exclusive ending node, there is no need to wait
if judge_not_same_branch():
return [], [one for one in source_ids if not one.startswith(('output_', 'condition_'))]
# Explain that it is not a mutually exclusive closing node, and you need to wait for all the predecessor nodes to finish executing before executing
wait_nodes = []
for one in source_ids:
if one.startswith('output_'):
one = f'{one}_fake'
wait_nodes.append(one)
return wait_nodes, []
def build_node_level(self, start_node: str):
""" Calculate hierarchy for all nodes """
# Hierarchy of marker nodes
def mark_node_level(node_id, node_map: dict, level: int):
# The nodes that have been traversed are no longer traversed, indicating that they are looped
if node_id in node_map:
return
self.node_level[node_id] = max(self.node_level.get(node_id, 0), level)
node_map[node_id] = True
next_nodes = self.edges.get_target_node(node_id)
if not next_nodes:
return
for one_node in next_nodes:
tmp_node_map = node_map.copy()
mark_node_level(one_node, tmp_node_map, level + 1)
return
mark_node_level(start_node, {}, 0)
def init_nodes(self, nodes):
""" return node id """
start_node = None
end_nodes = []
interrupt_nodes = []
for node in nodes:
node_data = BaseNodeData(**node.get('data', {}))
if not node_data.id:
raise Exception('node must have attribute id')
if node_data.type == NodeType.NOTE.value:
continue
node_instance = NodeFactory.instance_node(node_type=node_data.type,
node_data=node_data,
user_id=self.user_id,
workflow_id=self.workflow_id,
graph_state=self.graph_state,
target_edges=self.edges.get_target_edges(
node_data.id),
max_steps=self.max_steps,
callback=self.callback,
workflow_name=self.workflow_name)
if node_instance.is_condition_node():
self.condition_nodes.append(node_instance.id)
self.nodes_map[node_data.id] = node_instance
self.nodes_fan_in[node_instance.id] = self.edges.get_source_node(node_instance.id)
if node_instance.type not in [NodeType.START.value]:
self.nodes_next_nodes[node_instance.id] = self.edges.get_next_nodes(
node_instance.id)
# add node into langgraph
if self.async_mode:
self.graph_builder.add_node(node_instance.id, node_instance.arun)
else:
self.graph_builder.add_node(node_instance.id, node_instance.run)
# find special node
if node_instance.type == NodeType.START.value:
start_node = node_instance.id
elif node_instance.type == NodeType.END.value:
end_nodes.append(node_instance.id)
elif node_instance.type == NodeType.INPUT.value:
# Node that needs to abort receiving user input
interrupt_nodes.append(node_instance.id)
elif node_instance.type == NodeType.OUTPUT.value:
# Node that needs to abort receiving user input
fake_node = OutputFakeNode(id=f'{node_instance.id}_fake',
output_node=node_instance,
type=NodeType.FAKE_OUTPUT.value)
self.nodes_map[fake_node.id] = fake_node
interrupt_nodes.append(fake_node.id)
return start_node, end_nodes, interrupt_nodes
def build_nodes(self):
nodes = self.workflow_data.get('nodes', [])
if not nodes:
raise Exception('workflow must have at least one node')
start_node, end_nodes, interrupt_nodes = self.init_nodes(nodes)
if not start_node:
raise Exception('workflow must have start node')
self.graph_builder.add_edge(START, start_node)
if end_nodes:
for end_node in end_nodes:
self.graph_builder.add_edge(end_node, END)
# Calculate hierarchy of nodes
self.build_node_level(start_node)
# Link other nodes
for node_id, node_instance in self.nodes_map.items():
self.add_node_edge(node_instance)
# Handle nodes with multiple fan-in nodes
self.build_more_fan_in_node()
# compile langgraph
self.graph = self.graph_builder.compile(checkpointer=MemorySaver(),
interrupt_before=interrupt_nodes)
self.graph_config['recursion_limit'] = max(
(len(nodes) - len(end_nodes) - 1) * self.max_steps, 1) + len(end_nodes) + 1
# import datetime
# with open(f"./bisheng/data/graph/graph_{datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S')}.png",
# 'wb') as f:
# f.write(self.graph.get_graph().draw_mermaid_png())
def _run(self, input_data: Any):
try:
self.status = WorkflowStatus.RUNNING.value
for _ in self.graph.stream(input_data, config=self.graph_config):
pass
self.judge_status()
except IgnoreException as e:
logger.warning(f'graph ignore error: {e}')
self.status = WorkflowStatus.FAILED.value
self.reason = str(e)
except Exception as e:
logger.exception('graph run error')
self.status = WorkflowStatus.FAILED.value
self.reason = str(e)
async def _arun(self, input_data: Any):
try:
self.status = WorkflowStatus.RUNNING.value
async for _ in self.graph.astream(input_data, config=self.graph_config):
pass
self.judge_status()
except IgnoreException as e:
logger.warning(f'graph ignore error: {e}')
self.status = WorkflowStatus.FAILED.value
self.reason = str(e)
except Exception as e:
logger.exception('graph arun error')
self.status = WorkflowStatus.FAILED.value
self.reason = str(e)
def run(self):
self._run({'flag': True})
async def arun(self):
await self._arun({'flag': True})
def continue_run(self, data: Any = None):
"""
Receive user input
data：{node_id: {key: value}}
"""
# After receiving user input, proceed
if data is None:
data = {}
# Assign user input to the corresponding node
for node_id, node_params in data.items():
node_instance = self.nodes_map[node_id]
node_instance.handle_input(node_params)
# Resumegraph
self._run(None)
async def acontinue_run(self, data: Any = None):
"""
Receive user input
data：{node_id: {key: value}}
"""
# After receiving user input, proceed
if data is None:
data = {}
# Assign user input to the corresponding node
for node_id, node_params in data.items():
node_instance = self.nodes_map[node_id]
node_instance.handle_input(node_params)
# Resumegraph
await self._arun(None)
def judge_status(self):
# Judgment Status
snapshot = self.graph.get_state(self.graph_config)
next_nodes = snapshot.next
# Explain that the execution has been completed
if len(next_nodes) == 0:
self.status = WorkflowStatus.SUCCESS.value
return
# Determine what needs to be donenodetype, setting the corresponding engine state
for node_id in next_nodes:
node_instance = self.nodes_map[node_id]
if node_instance.type == NodeType.INPUT.value:
input_schema = node_instance.get_input_schema()
if input_schema:
# Callback events requiring user input
self.status = WorkflowStatus.INPUT.value
self.callback.on_user_input(
UserInputData(node_id=node_id, name=node_instance.name, input_schema=input_schema))
return
elif node_instance.type == NodeType.FAKE_OUTPUT.value:
intput_schema = node_instance.get_input_schema()
if intput_schema:
# output Node requires user input
self.status = WorkflowStatus.INPUT.value
return
def stop(self):
for _, node_instance in self.nodes_map.items():
node_instance.stop()
================================================
FILE: src/backend/bisheng/workflow/graph/graph_state.py
================================================
from typing import Any, Dict, Optional, List
from langchain.memory import ConversationBufferWindowMemory
from langchain_core.messages import AIMessage, HumanMessage, get_buffer_string, BaseMessage
from pydantic import BaseModel, Field
class GraphState(BaseModel):
""" All nodes Global State Management """
# Store Chat History
history_memory: Optional[ConversationBufferWindowMemory] = None
# Global Variable Pool
variables_pool: Dict[str, Dict[str, Any]] = Field(default_factory=dict, description='Global Variable Pool: {node_id: {key: value}}')
def get_history_memory(self, count: int) -> str:
""" Get chat history
'Cause it's not.1Right.1, so rewrite buffer_as_str"""
if not count:
count = self.history_memory.k
if count == 0:
return ''
messages = self.history_memory.chat_memory.messages[-count:]
return get_buffer_string(
messages,
human_prefix=self.history_memory.human_prefix,
ai_prefix=self.history_memory.ai_prefix,
)
def get_history_list(self, count: int) -> List[BaseMessage]:
return self.history_memory.buffer_as_messages[-count:]
def save_context(self, content: str, msg_sender: str) -> None:
""" Save chat
workflow In special cases, the process will have multiple rounds of interaction, so it is not one-to-one, reworking the message structure"""
if msg_sender == 'human':
self.history_memory.chat_memory.add_messages([HumanMessage(content=content)])
elif msg_sender == 'AI':
self.history_memory.chat_memory.add_messages([AIMessage(content=content)])
def set_variable(self, node_id: str, key: str, value: Any):
""" Put the data generated by the node into a global variable """
if node_id not in self.variables_pool:
self.variables_pool[node_id] = {}
self.variables_pool[node_id][key] = value
def get_variable(self, node_id: str, key: str, count: Optional[int] = None) -> Any:
""" Get data from global variables """
if node_id not in self.variables_pool:
return None
if key == 'chat_history':
return self.get_history_memory(count=count)
return self.variables_pool[node_id].get(key)
def get_variable_by_str(self, contact_key: str, history_count: Optional[int] = None) -> Any:
"""
Get data from global variables
contact_key: node_id.key#index #indexNot necessarily
"""
tmp_list = contact_key.split('.', 1)
node_id = tmp_list[0]
var_key = tmp_list[1]
variable_val_index = None
if var_key.find('#') != -1:
var_key, variable_val_index = var_key.split('#')
variable_val = self.get_variable(node_id, var_key, history_count)
# Handling of array variables
if variable_val_index:
if isinstance(variable_val, list):
variable_val_index = int(variable_val_index)
if len(variable_val) <= variable_val_index:
raise Exception(f'variable {contact_key} index out of range')
return variable_val[variable_val_index]
elif isinstance(variable_val, dict):
return variable_val.get(variable_val_index)
else:
raise Exception(f'variable {contact_key} is not a list or dict, not support #index')
return variable_val
def set_variable_by_str(self, contact_key: str, value: Any):
tmp_list = contact_key.split('.', 1)
node_id = tmp_list[0]
var_key = tmp_list[1]
if var_key.find('#') != -1:
var_key, variable_val_index = var_key.split('#')
old_value = self.get_variable(node_id, var_key)
if not old_value:
old_value = {}
old_value[variable_val_index] = value
value = old_value
self.set_variable(node_id, var_key, value)
def get_all_variables(self) -> Dict[str, Any]:
""" Get all variables,keyarenode_id.keyformat; """
ret = {}
for node_id, node_variables in self.variables_pool.items():
for key, value in node_variables.items():
ret[f'{node_id}.{key}'] = self.get_variable(node_id, key)
# get preset_question and custom_variables all keys
if key in ['preset_question', 'custom_variables']:
for k, v in value.items():
ret[f'{node_id}.{key}#{k}'] = v
return ret
================================================
FILE: src/backend/bisheng/workflow/graph/workflow.py
================================================
import json
import time
from typing import Dict
from bisheng.workflow.callback.base_callback import BaseCallback
from bisheng.workflow.common.workflow import WorkflowStatus
from bisheng.workflow.graph.graph_engine import GraphEngine
class Workflow:
def __init__(self,
workflow_id: str,
workflow_name: str = '',
user_id: int = None,
workflow_data: Dict = None,
async_mode: bool = False,
max_steps: int = 0,
timeout: int = 0,
callback: BaseCallback = None):
# Unique identifier of the run, unique saved to the databaseID
self.workflow_id = workflow_id
self.user_id = user_id
# Timeout, how long has the user input not been received terminatedworkflowRun (in minutes)
self.timeout = timeout
self.current_time = None
self.graph_engine = GraphEngine(user_id=user_id,
async_mode=async_mode,
workflow_id=workflow_id,
workflow_name=workflow_name or workflow_id,
workflow_data=workflow_data,
max_steps=max_steps,
callback=callback)
def save_user_input_history(self, input_data: dict | None):
if not input_data:
return
user_input_str = ''
for _, msg in input_data.items():
# Under the special handling of session input,keyRemove
if len(msg) == 1 and 'user_input' in msg:
user_input_str += msg['user_input']
continue
user_input_str += '\n' + json.dumps(msg, ensure_ascii=False)
self.graph_engine.graph_state.save_context(content=user_input_str, msg_sender='human')
def run(self, input_data: dict = None) -> (str, str):
"""
params:
input_data: user input data If not empty, executecontinue
return: workflow_status, reason
"""
# Implementationworkflow
if input_data is not None:
self.graph_engine.continue_run(input_data)
else:
# First run time
self.current_time = time.time()
self.graph_engine.run()
while self.graph_engine.status == WorkflowStatus.RUNNING.value:
self.graph_engine.continue_run()
return self.graph_engine.status, self.graph_engine.reason
async def arun(self, input_data: dict = None) -> (str, str):
"""
params:
input_data: user input data If not empty, executecontinue
return: workflow_status, reason
"""
# Implementationworkflow
if input_data is not None:
await self.graph_engine.acontinue_run(input_data)
else:
# First run time
self.current_time = time.time()
await self.graph_engine.arun()
while self.graph_engine.status == WorkflowStatus.RUNNING.value:
await self.graph_engine.acontinue_run()
return self.graph_engine.status, self.graph_engine.reason
def stop(self):
self.graph_engine.stop()
def status(self):
return self.graph_engine.status
def reason(self):
return self.graph_engine.reason
================================================
FILE: src/backend/bisheng/workflow/nodes/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/agent/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/agent/agent.py
================================================
import typing
from typing import Any, Dict
from langchain_core.messages import HumanMessage
from langchain_core.retrievers import BaseRetriever
from langchain_core.runnables import RunnableConfig
from langgraph.prebuilt import create_react_agent
from loguru import logger
from pydantic import BaseModel, field_validator, Field
from bisheng.common.constants.enums.telemetry import ApplicationTypeEnum
from bisheng.knowledge.domain.knowledge_rag import KnowledgeRag
from bisheng.llm.domain.services import LLMService
from bisheng.tool.domain.services.executor import ToolExecutor
from bisheng.workflow.callback.event import StreamMsgOverData
from bisheng.workflow.callback.llm_callback import LLMNodeCallbackHandler
from bisheng.workflow.nodes.base import BaseNode
from bisheng.workflow.nodes.prompt_template import PromptTemplateParser
from bisheng_langchain.gpts.assistant import ConfigurableAssistant
from bisheng_langchain.gpts.load_tools import load_tools
agent_executor_dict = {
'ReAct': 'get_react_agent_executor',
'function call': 'get_openai_functions_agent_executor',
}
class SqlAgentParams(BaseModel):
""" SQL Agent Param Model """
database_engine: typing.Optional[str] = Field("mysql",
description="Database type, supportmysql, db2, postgres, gaussdb, oracle")
db_username: str
db_password: str
db_address: str
db_name: str
open: bool = False
@field_validator("database_engine")
@classmethod
def validate_database_engine(cls, v):
# Convert to lowercase
if v:
v = v.lower()
if v not in ['mysql', 'db2', 'postgres', 'gaussdb', 'oracle', 'postgresql']:
raise ValueError(
"Unsupported database engine. Supported engines are: MySQL, DB2, PostgreSql, GaussDB, Oracle.")
return v
class AgentNode(BaseNode):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
# Determine if it is a single or batch
self._tab = self.node_data.tab['value']
# analyzingprompt
self._system_prompt = PromptTemplateParser(template=self.node_params['system_prompt'])
self._system_variables = self._system_prompt.extract()
self._user_prompt = PromptTemplateParser(template=self.node_params['user_prompt'])
self._user_variables = self._user_prompt.extract()
self._image_prompt = self.node_params.get('image_prompt', [])
self._batch_variable_list = []
self._system_prompt_list = []
self._user_prompt_list = []
self._tool_invoke_list = []
self._log_reasoning_content = []
# Chat Message
self._chat_history_flag = self.node_params['chat_history_flag']['value'] > 0
self._chat_history_num = self.node_params['chat_history_flag']['value']
self._llm = LLMService.get_bisheng_llm_sync(model_id=self.node_params['model_id'],
temperature=self.node_params.get('temperature', 0.3),
app_id=self.workflow_id,
app_name=self.workflow_name,
app_type=ApplicationTypeEnum.WORKFLOW,
user_id=self.user_id)
# Whether to output the results to the user
self._output_user = self.node_params.get('output_user', False)
# tools
self._tools = self.node_params['tool_list']
# knowledge
# self._knowledge_ids = self.node_params['knowledge_id']
# Determine whether it is a knowledge base or a temporary file list
self._knowledge_type = self.node_params['knowledge_id']['type']
self._knowledge_ids = [
one['key'] for one in self.node_params['knowledge_id']['value']
]
# Supported or notnl2sql
self._sql_agent_params = self.node_params.get('sql_agent', None)
self._sql_agent = SqlAgentParams.model_validate(self.node_params['sql_agent']) if (
self._sql_agent_params and self._sql_agent_params.get("open", False)) else None
self._sql_address = ''
if self._sql_agent and self._sql_agent.open:
self._sql_address = self._init_sql_address()
# agent
self._agent_executor_type = 'React'
self._agent = None
def _init_agent(self, system_prompt: str):
# Get a list of configured helper models
assistant_llm = LLMService.sync_get_assistant_llm()
if not assistant_llm.llm_list:
raise Exception('Assistant reasoning model list is empty')
default_llm = [
one for one in assistant_llm.llm_list if one.model_id == self.node_params['model_id']
]
if not default_llm:
raise Exception('The selected inference model is not in the list of assistant inference models')
default_llm = default_llm[0]
self._agent_executor_type = default_llm.agent_executor_type
knowledge_retriever = {
'max_content': default_llm.knowledge_max_content,
'sort_by_source_and_index': default_llm.knowledge_sort_index
}
func_tools = self._init_tools()
knowledge_tools = self._init_knowledge_tools(knowledge_retriever)
sql_agent_tools = self.init_sql_agent_tool()
func_tools.extend(knowledge_tools)
func_tools.extend(sql_agent_tools)
if self._agent_executor_type == 'ReAct':
self._agent = ConfigurableAssistant(
agent_executor_type=agent_executor_dict.get(self._agent_executor_type),
tools=func_tools,
llm=self._llm,
assistant_message=system_prompt,
)
else:
self._agent = create_react_agent(self._llm, func_tools, prompt=system_prompt, checkpointer=False)
def _init_tools(self):
if self._tools:
tool_ids = [int(one['key']) for one in self._tools]
return ToolExecutor.init_by_tool_ids_sync(tool_ids, app_id=self.workflow_id, app_name=self.workflow_name,
app_type=ApplicationTypeEnum.WORKFLOW,
user_id=self.user_id)
else:
return []
def init_sql_agent_tool(self):
if not self._sql_address:
return []
tool_params = {
'sql_agent': {
'llm': self._llm,
'sql_address': self._sql_address
}
}
return load_tools(tool_params=tool_params, llm=self._llm)
def _init_knowledge_tools(self, knowledge_retriever: dict):
if not self._knowledge_ids:
return []
tools = []
for index, knowledge_id in enumerate(self._knowledge_ids):
if self._knowledge_type == 'knowledge':
knowledge_tool = ToolExecutor.init_knowledge_tool_sync(self.user_id, knowledge_id,
llm=self._llm,
**knowledge_retriever)
tools.append(knowledge_tool)
else:
file_metadata_list = self.get_other_node_variable(knowledge_id)
if not file_metadata_list:
# Do not retrieve if no file has been uploaded
continue
description = ''
for one in file_metadata_list:
description += f'<{one.get("document_name")}>:<{one.get("abstract")}>; '
tool_init_params = {
"name": f'{knowledge_id.split(".")[-1].replace("#", "")}_knowledge_{index}',
"description": description,
"vector_retriever": self.init_file_milvus(file_metadata_list[0]),
"elastic_retriever": self.init_file_es(file_metadata_list[0]),
"llm": self._llm,
**knowledge_retriever
}
tmp_file_tool = ToolExecutor.init_tmp_knowledge_tool_sync(**tool_init_params)
tools.append(tmp_file_tool)
return tools
def init_file_milvus(self, file_metadata: Dict) -> BaseRetriever:
""" Initialize the temporary file selected by the usermilvus """
embeddings = LLMService.get_knowledge_default_embedding(self.user_id)
if not embeddings:
raise Exception('No default configuredembeddingModels')
file_ids = [file_metadata['document_id']]
collection_name = self.get_milvus_collection_name(getattr(embeddings, 'model_id'))
vector_client = KnowledgeRag.init_milvus_vectorstore(collection_name=collection_name, embeddings=embeddings)
return vector_client.as_retriever(
search_kwargs={"expr": f'document_id in {file_ids}'})
def init_file_es(self, file_metadata: Dict):
es_client = KnowledgeRag.init_es_vectorstore_sync(index_name=self.tmp_collection_name)
return es_client.as_retriever(
search_kwargs={"filter": [{"term": {"metadata.document_id": file_metadata['document_id']}}]})
def _init_sql_address(self) -> str:
""" Inisialisasi SQL Database Address """
if not self._sql_agent:
return ''
if self._sql_agent.database_engine == 'mysql':
try:
pass
except ImportError:
raise ImportError('Please install pymysql and sqlalchemy to use mysql database')
return f'mysql+pymysql://{self._sql_agent.db_username}:{self._sql_agent.db_password}@{self._sql_agent.db_address}/{self._sql_agent.db_name}?charset=utf8mb4'
elif self._sql_agent.database_engine == 'db2':
try:
pass
except ImportError:
raise ImportError('Please install ibm_db and ibm_db_sa to use db2 database')
return f'db2+ibm_db://{self._sql_agent.db_username}:{self._sql_agent.db_password}@{self._sql_agent.db_address}/{self._sql_agent.db_name}'
elif self._sql_agent.database_engine in ['postgres', 'postgresql']:
try:
pass
except ImportError:
raise ImportError('Please install psycopg2 and sqlalchemy to use postgresql database')
return f'postgresql+psycopg2://{self._sql_agent.db_username}:{self._sql_agent.db_password}@{self._sql_agent.db_address}/{self._sql_agent.db_name}'
elif self._sql_agent.database_engine == 'gaussdb':
try:
pass
except ImportError:
raise ImportError('Please install psycopg2 and opengauss_sqlalchemy to use gaussdb database')
return f'opengauss+psycopg2://{self._sql_agent.db_username}:{self._sql_agent.db_password}@{self._sql_agent.db_address}/{self._sql_agent.db_name}'
elif self._sql_agent.database_engine == 'oracle':
try:
pass
except ImportError:
raise ImportError('Please install oracledb and sqlalchemy to use oracle database')
return f'oracle+oracledb://{self._sql_agent.db_username}:{self._sql_agent.db_password}@{self._sql_agent.db_address}?service_name={self._sql_agent.db_name}'
else:
raise ValueError(f'Unsupported database engine: {self._sql_agent.database_engine}')
def _run(self, unique_id: str):
ret = {}
variable_map = {}
self._batch_variable_list = []
self._system_prompt_list = []
self._user_prompt_list = []
self._tool_invoke_list = []
self._log_reasoning_content = []
for one in self._system_variables:
variable_map[one] = self.get_other_node_variable(one)
system_prompt = self._system_prompt.format(variable_map)
self._system_prompt_list.append(system_prompt)
self._init_agent(system_prompt)
if self._tab == 'single':
self._tool_invoke_list.append([])
ret['output'], reasoning_content = self._run_once(None, unique_id, 'output', self._tool_invoke_list[0])
self._log_reasoning_content.append(reasoning_content)
if self._output_user:
self.callback_manager.on_stream_over(StreamMsgOverData(node_id=self.id,
name=self.name,
msg=ret['output'],
reasoning_content=reasoning_content,
unique_id=unique_id,
output_key='output'))
else:
for index, one in enumerate(self.node_params['batch_variable']):
self._batch_variable_list.append(self.get_other_node_variable(one))
output_key = self.node_params['output'][index]['key']
self._tool_invoke_list.append([])
ret[output_key], reasoning_content = self._run_once(one, unique_id, output_key,
self._tool_invoke_list[index])
self._log_reasoning_content.append(reasoning_content)
if self._output_user:
self.callback_manager.on_stream_over(StreamMsgOverData(node_id=self.id,
name=self.name,
msg=ret[output_key],
reasoning_content=reasoning_content,
unique_id=unique_id,
output_key=output_key))
logger.debug('agent_over result={}', ret)
if self._output_user:
# Nonstream Mode, processing results
for k, v in ret.items():
answer = v
self.graph_state.save_context(content=answer, msg_sender='AI')
return ret
def parse_log(self, unique_id: str, result: dict) -> Any:
ret = []
index = 0
for k, v in result.items():
one_ret = [
{"key": "system_prompt", "value": self._system_prompt_list[0], "type": "params"},
{"key": "user_prompt", "value": self._user_prompt_list[index], "type": "params"},
]
if self._batch_variable_list:
one_ret.insert(0,
{"key": "batch_variable", "value": self._batch_variable_list[index], "type": "variable"})
# Handler Call Log
one_ret.extend(self.parse_tool_log(self._tool_invoke_list[index]))
if self._log_reasoning_content[index]:
one_ret.append(
{"key": "Thinking about content", "value": self._log_reasoning_content[index], "type": "params"})
one_ret.append({"key": f'{self.id}.{k}', "value": v, "type": "variable"})
ret.append(one_ret)
index += 1
return ret
def parse_tool_log(self, tool_invoke_list: list) -> list:
ret = []
tool_invoke_info = {}
for one in tool_invoke_list:
if one['run_id'] not in tool_invoke_info:
tool_invoke_info[one['run_id']] = {}
if one['type'] == 'start':
tool_invoke_info[one['run_id']].update({
'name': one['name'],
'input': one['input']
})
elif one['type'] == 'end':
tool_invoke_info[one['run_id']].update({
'output': one['output']
})
elif one['type'] == 'error':
tool_invoke_info[one['run_id']].update({
'output': f'Error: {one["error"]}'
})
if tool_invoke_info:
for one in tool_invoke_info.values():
# knowledge_retriever_tool belong into rag logic，not show in tool log
if one["name"] == "knowledge_retriever_tool":
continue
ret.append({
"key": one["name"],
"value": f"Tool Input:\n {one['input']}, Tool Output:\n {one['output']}",
"type": "tool"
})
return ret
def _run_once(self, input_variable: str = None, unique_id: str = None, output_key: str = None,
tool_invoke_list: list = None) -> (str, str):
"""
params:
input_variable: Input variables, if yesbatchthen you need to pass in a variablekey, otherwiseNone
unique_id: Node Execute Uniqueid
output_key: Output Variableskey
tool_invoke_list: Tool Call Log
return:
0: Output results to user
1: Process of model thinking
"""
# Description is a variable that references a batch, The value of the variable needs to be replaced with the variable selected by the user
special_variable = f'{self.id}.batch_variable'
variable_map = {}
for one in self._user_variables:
if input_variable and one == special_variable:
variable_map[one] = self.get_other_node_variable(input_variable)
continue
variable_map[one] = self.get_other_node_variable(one)
user = self._user_prompt.format(variable_map)
self._user_prompt_list.append(user)
chat_history = []
if self._chat_history_flag:
chat_history = self.graph_state.get_history_list(self._chat_history_num)
llm_callback = LLMNodeCallbackHandler(callback=self.callback_manager,
unique_id=unique_id,
node_id=self.id,
node_name=self.name,
output=self._output_user,
output_key=output_key,
tool_list=tool_invoke_list,
cancel_llm_end=True)
config = RunnableConfig(callbacks=[llm_callback])
human_message = HumanMessage(content=[{
'type': 'text',
'text': user
}])
human_message = self.contact_file_into_prompt(human_message, self._image_prompt)
chat_history.append(human_message)
logger.debug(f'agent invoke chat_history: {chat_history}')
if self._agent_executor_type == 'ReAct':
result = self._agent.invoke({
'input': chat_history[-1].content,
'chat_history': chat_history[:-1],
}, config=config)
output = result['agent_outcome'].return_values['output']
if isinstance(output, dict):
output = list(output.values())[0]
return output, llm_callback.reasoning_content
else:
result = self._agent.invoke({'messages': chat_history}, config=config)
result = result['messages']
return result[-1].content, llm_callback.reasoning_content
================================================
FILE: src/backend/bisheng/workflow/nodes/base.py
================================================
import base64
import copy
import uuid
from abc import ABC, abstractmethod
from typing import Any, Dict, List
from langchain_core.messages import HumanMessage
from bisheng.core.cache.utils import file_download
from bisheng.user.domain.models.user import UserDao
from bisheng.utils.exceptions import IgnoreException
from bisheng.workflow.callback.base_callback import BaseCallback
from bisheng.workflow.callback.event import NodeEndData, NodeStartData
from bisheng.workflow.common.node import BaseNodeData, NodeType
from bisheng.workflow.edges.edges import EdgeBase
from bisheng.workflow.graph.graph_state import GraphState
from bisheng.workflow.nodes.prompt_template import PromptTemplateParser
class BaseNode(ABC):
def __init__(self, node_data: BaseNodeData, workflow_id: str, user_id: int,
graph_state: GraphState, target_edges: List[EdgeBase], max_steps: int,
callback: BaseCallback, **kwargs: Any):
self.id = node_data.id
self.type = node_data.type
self.name = node_data.name
self.description = node_data.description
self.target_edges = target_edges
# Execute Unique Identification of User
self.user_id = user_id
# Global State Management
self.workflow_id = workflow_id
self.workflow_name = kwargs.get('workflow_name')
self.graph_state = graph_state
# Data of all nodes
self.node_data = node_data
# Parameters required for storage nodes Directly usable parameters after processing
self.node_params = {}
# The value of the other node variables required to store the node
self.other_node_variable = {}
# Used to determine if the maximum number of runs has been exceeded
self.current_step = 0
self.max_steps = max_steps
# Callbacks to handle various events during node execution
self.callback_manager = callback
# Storing Temporary Data milvus Collection Name And es Collection Name workflow_id As partition key
# samecollectionMedium vector data must be the sameembedding_modelGenerated, so the collection name needs to containembedding_model_id
self.tmp_collection_name = 'tmp_workflow_data_new'
self.stop_flag = False
self.exec_unique_id = None
self.user_info = None
# Parse Simple Parameters
self.init_data()
def init_data(self):
""" Unified parameter processing, nodes with special needs can be processed when initializing themselves """
if not self.node_data.group_params:
return
for one in self.node_data.group_params:
for param_info in one.params:
self.node_params[param_info.key] = copy.deepcopy(param_info.value)
def init_user_info(self):
if self.user_info:
return
self.user_info = UserDao.get_user(int(self.user_id))
@abstractmethod
def _run(self, unique_id: str) -> Dict[str, Any]:
"""
Run node The returned results are stored in the global variable management and can be used by other nodes
:return:
"""
raise NotImplementedError
def parse_log(self, unique_id: str, result: dict) -> Any:
"""
Returns the node operation log, the default return is empty
params:
result: Node Run Result
return: The outermost layer is the rounds, and inside are the logs for each round.
[
[
{
"key": "xxx",
"value": "xxx",
"type": "tool" # tool: Tool Type Logs, variable: Log of global variables, params: Log of node parameter type,keySHOW: keymain body
}
]
]
"""
return []
def get_other_node_variable(self, variable_key: str) -> Any:
""" Get the variable values of other nodes from the global variable """
value = self.graph_state.get_variable_by_str(variable_key)
self.other_node_variable[variable_key] = value
return value
def get_input_schema(self) -> Any:
""" Returns the form description the user needs to enter """
return None
def is_condition_node(self) -> bool:
""" Whether it is a mutually exclusive node """
return self.node_data.type == NodeType.CONDITION.value
def get_milvus_collection_name(self, embedding_model_id: str) -> str:
return f"{self.tmp_collection_name}_{embedding_model_id}"
def handle_input(self, user_input: dict) -> Any:
# Update the data entered by the user to the number of nodes
self.node_params.update(user_input)
def route_node(self, state: dict) -> str:
"""
counterpart'slanggraphright of privacycondition_edgeright of privacyfunction, only special nodes need
:return: node_ambid
"""
raise NotImplementedError
def get_next_node_id(self, source_handle: str) -> list[str]:
next_nodes = []
for one in self.target_edges:
if one.sourceHandle == source_handle:
next_nodes.append(one.target)
return next_nodes
def parse_msg_with_variables(self, msg: str) -> (str, list[str]):
"""
params:
msg: user input msg with node variables
return:
0: new msg after replaced variable
1: list of variables node_id.xxxx
"""
msg_template = PromptTemplateParser(template=msg)
variables = msg_template.extract()
if len(variables) > 0:
var_map = {}
for one in variables:
var_map[one] = self.get_other_node_variable(one)
msg = msg_template.format(var_map)
return msg, variables
@staticmethod
def get_file_base64_data(file_path: str) -> str:
if file_path.startswith(('http', "https")):
file_path, _ = file_download(file_path)
with open(file_path, "rb") as f:
file_data = f.read()
base64_data = base64.b64encode(file_data).decode('utf-8')
return base64_data
def contact_file_into_prompt(self, human_message: HumanMessage, variable_list: List[str]) -> HumanMessage:
if not variable_list:
if isinstance(human_message.content, list):
human_message.content = human_message.content[0].get('text')
return human_message
for image_variable in variable_list:
image_value = self.get_other_node_variable(image_variable)
if not image_value:
continue
for file_path in image_value:
base64_image = self.get_file_base64_data(file_path)
human_message.content.append({
"type": "image",
"source_type": "base64",
"mime_type": "image/jpeg",
"data": base64_image,
})
return human_message
def run(self, state: dict) -> Any:
"""
Run node entry
:return:
"""
if self.stop_flag:
raise IgnoreException('stop by user')
if self.current_step >= self.max_steps:
raise IgnoreException(f'{self.name} -- has run more than the maximum number of times.')
exec_id = uuid.uuid4().hex
self.exec_unique_id = exec_id
self.callback_manager.on_node_start(
data=NodeStartData(unique_id=exec_id, node_id=self.id, name=self.name))
reason = None
log_data = None
try:
result = self._run(exec_id)
log_data = self.parse_log(exec_id, result)
# Store node output in global variables
if result:
for key, value in result.items():
self.graph_state.set_variable(self.id, key, value)
self.current_step += 1
except Exception as e:
reason = str(e)
raise e
finally:
# The end log of the output node is created byfakeNode Output, Because it is necessary to wait for the user to complete the input before the log can be displayed correctly
if reason or self.type != NodeType.OUTPUT.value:
self.callback_manager.on_node_end(data=NodeEndData(
unique_id=exec_id, node_id=self.id, name=self.name, reason=reason, log_data=log_data,
input_data=self.other_node_variable))
return state
async def arun(self, state: dict) -> Any:
return self.run(state)
def stop(self):
self.stop_flag = True
================================================
FILE: src/backend/bisheng/workflow/nodes/code/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/code/code.py
================================================
from typing import Any
from bisheng.workflow.nodes.base import BaseNode
from bisheng.workflow.nodes.code.code_parse import CodeParser
class CodeNode(BaseNode):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self._code_input = self.node_params['code_input']
self._code = self.node_params['code']
self._code_output = self.node_params['code_output']
self._code_parser = CodeParser(self._code)
self._parse_code()
def handle_input(self, user_input: dict) -> Any:
self.node_params.update(user_input)
self._code_input = self.node_params['code_input']
self._code_output = self.node_params['code_output']
self._code = self.node_params['code']
def _parse_code(self):
try:
self._code_parser.parse_code()
except Exception as e:
raise Exception(f"CodeNode {self.name} exec code error: " + str(e))
def _run(self, unique_id: str):
main_params = self._parse_code_input()
main_ret = self._code_parser.exec_method('main', **main_params)
main_ret = self._parse_code_output(main_ret)
return main_ret
def parse_log(self, unique_id: str, result: dict):
return [[
{"key": "code_input", "value": self._parse_code_input(), "type": "params"},
{"key": "code_output", "value": result, "type": "params"}
]]
def _parse_code_input(self) -> dict:
ret = {}
for one in self._code_input:
if one["type"] == "ref":
ret[one['key']] = self.get_other_node_variable(one['value'])
else:
ret[one['key']] = one['value']
return ret
def _parse_code_output(self, result: dict) -> dict:
if not isinstance(result, dict):
raise Exception(f"CodeNode {self.name} main function output must be dict")
ret = {}
for one in self._code_output:
if one["key"] not in result:
raise Exception(f"CodeNode {self.name} main function output must have key {one['key']}")
ret[one['key']] = result.get(one['key'])
return ret
================================================
FILE: src/backend/bisheng/workflow/nodes/code/code_parse.py
================================================
import ast
import importlib
import inspect
from typing import Union, Type, Dict, Any
class CodeParser:
"""
A parser for Python source code, extracting code details.
"""
def __init__(self, code: Union[str, Type]) -> None:
"""
Initializes the parser with the provided code.
"""
if isinstance(code, type):
if not inspect.isclass(code):
raise ValueError("The provided code must be a class.")
# If the code is a class, get its source code
code = inspect.getsource(code)
self.code = code
self.exec_globals = {}
self.exec_locals = {}
self.data: Dict[str, Any] = {
"imports": [],
}
self.handlers = {
ast.Import: self.parse_imports,
ast.ImportFrom: self.parse_imports,
ast.FunctionDef: self.parse_functions,
ast.ClassDef: self.parse_classes,
ast.Assign: self.parse_global_vars,
}
def parse_code(self) -> Dict[str, Any]:
"""
Runs all parsing operations and returns the resulting data.
"""
tree = self.get_tree()
for node in ast.walk(tree):
self.parse_node(node)
return self.data
def get_tree(self):
"""
Parses the provided code to validate its syntax.
It tries to parse the code into an abstract syntax tree (AST).
"""
return ast.parse(self.code)
def parse_node(self, node: Union[ast.stmt, ast.AST]) -> None:
"""
Parses an AST node and updates the data
dictionary with the relevant information.
"""
if handler := self.handlers.get(type(node)): # type: ignore
handler(node) # type: ignore
def parse_imports(self, node: Union[ast.Import, ast.ImportFrom]) -> None:
"""
Extracts "imports" from the code, including aliases.
"""
if isinstance(node, ast.Import):
for alias in node.names:
if alias.asname:
self.data["imports"].append(f"{alias.name} as {alias.asname}")
else:
self.data["imports"].append(alias.name)
# Actual Import Module
try:
self.exec_globals[alias.asname or alias.name] = importlib.import_module(alias.name)
except ModuleNotFoundError as e:
raise ModuleNotFoundError(f"Module {alias.name} not found. Please install it and try again.") from e
elif isinstance(node, ast.ImportFrom):
try:
imported_module = importlib.import_module(node.module)
for alias in node.names:
if alias.asname:
self.data["imports"].append((node.module, f"{alias.name} as {alias.asname}"))
else:
self.data["imports"].append((node.module, alias.name))
self.exec_globals[alias.name] = getattr(imported_module, alias.name)
except ModuleNotFoundError:
raise ModuleNotFoundError(f"Module {node.module} not found. Please install it and try again")
def parse_functions(self, node: ast.FunctionDef) -> None:
"""
Extracts "functions" from the code.
"""
compiled_func = compile(ast.Module(body=[node], type_ignores=[]), "", "exec")
exec(compiled_func, self.exec_globals, self.exec_locals)
def parse_classes(self, node: ast.ClassDef) -> None:
compiled_class = compile(ast.Module(body=[node], type_ignores=[]), "", "exec")
exec(compiled_class, self.exec_globals, self.exec_locals)
self.exec_globals[node.name] = self.exec_locals[node.name]
def parse_global_vars(self, node: ast.Assign) -> None:
"""
Extracts global variables from the code.
"""
global_var = {
"targets": [t.id if hasattr(t, "id") else ast.dump(t) for t in node.targets],
"value": ast.unparse(node.value),
}
if isinstance(node.value, ast.Constant):
for one in global_var["targets"]:
self.exec_globals[one] = global_var["value"]
def exec_method(self, method_name: str, *args, **kwargs):
"""
Executes the method with the provided arguments and keyword arguments.
"""
method = self.exec_locals.get(method_name)
if not method:
raise AttributeError(f"Method {method_name} not found.")
return method(*args, **kwargs)
def init_class(self, class_name: str, *args, **kwargs):
"""
Initializes the class with the provided arguments and keyword arguments.
"""
class_ = self.exec_globals.get(class_name)
if not class_:
raise AttributeError(f"Class {class_name} not found.")
return class_(*args, **kwargs)
================================================
FILE: src/backend/bisheng/workflow/nodes/condition/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/condition/condition.py
================================================
from typing import Any
from bisheng.workflow.nodes.base import BaseNode
from bisheng.workflow.nodes.condition.conidition_case import ConditionCases
from loguru import logger
class ConditionNode(BaseNode):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self._next_node_id = None
self._condition_cases = [ConditionCases(**one) for one in self.node_params['condition']]
# Store calculated variables and values
self._variable_key_value = {}
def _run(self, unique_id: str):
self._variable_key_value = {}
next_node_ids = None
for one in self._condition_cases:
flag = one.evaluate_conditions(self)
self._variable_key_value.update(one.variable_key_value)
if flag:
next_node_ids = self.get_next_node_id(one.id)
logger.info(f'Condition node {self.id} pass condition {next_node_ids}')
break
# Guaranteed logic, using default routing
if next_node_ids is None:
self._next_node_id = self.get_next_node_id('right_handle')
else:
self._next_node_id = next_node_ids
def parse_log(self, unique_id: str, result: dict) -> Any:
return [[
{
"key": k,
"value": v,
"type": "variable"
} for k, v in self._variable_key_value.items()
]]
def route_node(self, state: dict) -> str:
return self._next_node_id
================================================
FILE: src/backend/bisheng/workflow/nodes/condition/conidition_case.py
================================================
import re
from typing import List, Optional, Dict
from loguru import logger
from pydantic import ConfigDict, BaseModel, Field
from bisheng.workflow.common.condition import ComparisonType, LogicType
from bisheng.workflow.nodes.base import BaseNode
class ConditionOne(BaseModel):
id: str = Field(..., description='Unique id for condition')
left_var: str = Field(..., description='Left variable')
comparison_operation: str = Field(..., description='Compare type')
right_value_type: str = Field(..., description='Right value type')
right_value: str = Field(..., description='Right value')
variable_key_value: Dict = Field(default={}, description='variable key value')
def evaluate(self, node_instance: BaseNode) -> bool:
left_value = node_instance.get_other_node_variable(self.left_var)
self.variable_key_value[self.left_var] = left_value
right_value = self.right_value
if self.right_value_type == 'ref' and self.right_value:
right_value = node_instance.get_other_node_variable(self.right_value)
self.variable_key_value[self.right_value] = right_value
logger.debug(f'condition evaluate ope: {self.comparison_operation},'
f' left value: {left_value}, right value: {right_value}')
return self.compare_two_value(left_value, right_value)
def compare_two_value(self, left_value: str, right_value: str) -> bool:
if self.comparison_operation == ComparisonType.EQUAL:
return left_value == right_value
elif self.comparison_operation == ComparisonType.NOT_EQUAL:
return left_value != right_value
elif self.comparison_operation == ComparisonType.CONTAINS:
return left_value.find(right_value) != -1
elif self.comparison_operation == ComparisonType.NOT_CONTAINS:
return left_value.find(right_value) == -1
elif self.comparison_operation == ComparisonType.IS_EMPTY:
return left_value == '' or left_value is None
elif self.comparison_operation == ComparisonType.IS_NOT_EMPTY:
return left_value != '' and left_value is not None
elif self.comparison_operation == ComparisonType.STARTS_WITH:
return left_value.startswith(right_value)
elif self.comparison_operation == ComparisonType.ENDS_WITH:
return left_value.endswith(right_value)
elif self.comparison_operation == ComparisonType.GREATER_THAN:
return float(left_value) > float(right_value)
elif self.comparison_operation == ComparisonType.GREATER_THAN_OR_EQUAL:
return float(left_value) >= float(right_value)
elif self.comparison_operation == ComparisonType.LESS_THAN:
return float(left_value) < float(right_value)
elif self.comparison_operation == ComparisonType.LESS_THAN_OR_EQUAL:
return float(left_value) <= float(right_value)
elif self.comparison_operation == ComparisonType.REGEX:
right = re.compile(right_value)
return right.search(left_value) is not None
else:
raise Exception('not support comparison operator: %s' % self.comparison_operation)
class ConditionCases(BaseModel):
model_config = ConfigDict(arbitrary_types_allowed=True)
id: str = Field(..., description='Unique id for case')
operator: Optional[str] = Field('and', description='Operator for case')
conditions: Optional[List[ConditionOne]] = Field(None, description='Conditions for case')
variable_key_value: Dict = Field(default={}, description='variable key value')
def evaluate_conditions(self, node_instance: BaseNode) -> bool:
# Normally onlyelseNoconditions
if not self.conditions:
return True
for condition in self.conditions:
flag = condition.evaluate(node_instance)
self.variable_key_value.update(condition.variable_key_value)
if self.operator == LogicType.AND:
if not flag:
return False
else:
if flag:
return True
return True if self.operator == LogicType.AND else False
================================================
FILE: src/backend/bisheng/workflow/nodes/end/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/end/end.py
================================================
from bisheng.workflow.nodes.base import BaseNode
class EndNode(BaseNode):
def _run(self, unique_id: str):
return None
================================================
FILE: src/backend/bisheng/workflow/nodes/input/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/input/const.py
================================================
from bisheng.common.schemas.rag_schema import RagMetadataFieldSchema
# Enter the metadata definition for the temporary upload file of the node
InputFileMetadata = [
RagMetadataFieldSchema(field_name="document_id", field_type="text", kwargs={"nullable": False, "max_length": 1024}),
RagMetadataFieldSchema(field_name="document_name", field_type="text",
kwargs={"nullable": True, "max_length": 32767}),
RagMetadataFieldSchema(field_name="abstract", field_type="text", kwargs={"nullable": True, "max_length": 65535}),
RagMetadataFieldSchema(field_name="chunk_index", field_type="int64", kwargs={"nullable": True}),
RagMetadataFieldSchema(field_name="bbox", field_type="text", kwargs={"nullable": True, "max_length": 65535}),
RagMetadataFieldSchema(field_name="page", field_type="int64", kwargs={"nullable": True}),
RagMetadataFieldSchema(field_name="knowledge_id", field_type="text",
kwargs={"nullable": False, "max_length": 1024}),
RagMetadataFieldSchema(field_name="upload_time", field_type="int64", kwargs={"nullable": True}),
RagMetadataFieldSchema(field_name="update_time", field_type="int64", kwargs={"nullable": True}),
RagMetadataFieldSchema(field_name="uploader", field_type="text", kwargs={"nullable": True, "max_length": 1024}),
RagMetadataFieldSchema(field_name="updater", field_type="text", kwargs={"nullable": True, "max_length": 1024}),
RagMetadataFieldSchema(field_name="user_metadata", field_type="json", kwargs={"nullable": True})
]
================================================
FILE: src/backend/bisheng/workflow/nodes/input/input.py
================================================
import copy
import json
import time
from enum import Enum
from typing import Any, List, Dict
from urllib.parse import urlparse, unquote
from json_repair import json_repair
from langchain_core.messages import SystemMessage, HumanMessage
from loguru import logger
from bisheng.api.services.knowledge import KnowledgeService
from bisheng.api.services.knowledge_imp import read_chunk_text
from bisheng.api.v1.schemas import FileProcessBase
from bisheng.chat.types import IgnoreException
from bisheng.common.constants.enums.telemetry import ApplicationTypeEnum
from bisheng.common.errcode.knowledge import KnowledgeFileNotSupportedError
from bisheng.core.cache.utils import file_download
from bisheng.knowledge.domain.knowledge_rag import KnowledgeRag
from bisheng.llm.domain.services import LLMService
from bisheng.utils import generate_uuid
from bisheng.workflow.callback.event import GuideQuestionData
from bisheng.workflow.nodes.base import BaseNode
from bisheng.workflow.nodes.input.const import InputFileMetadata
class ParseModeEnum(str, Enum):
KEEP_RAW = 'keep_raw'
EXTRACT_TEXT = 'extract_text'
INGEST_TO_KNOWLEDGE_BASE = 'ingest_to_temp_kb'
class InputNode(BaseNode):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
# Node Current Version
self._current_v = 2
# Whether the record is a conversation or a form
self._tab = self.node_data.tab['value']
# Record what type of variable this is
self._node_params_map = {}
new_node_params = {}
# The maximum length of the input file in the dialog box, more than this length will be truncated
if self.node_data.v < self._current_v:
raise IgnoreException(f'{self.name} -- workflow node is update')
self._original_node_params = copy.deepcopy(self.node_params)
# save user set file key -> file key info
self._file_key_map = {}
if self.is_dialog_input():
new_node_params['user_input'] = self.node_params['user_input']
new_node_params['dialog_files_content'] = self.node_params.get('dialog_files_content', [])
else:
for value_info in self.node_params['form_input']:
value_key = value_info['key']
# The file key needs to be re-generated to avoid parse type not ingest to knowledge base
if value_info["type"] == "file":
value_key = f"file_{generate_uuid()[:8]}"
self._file_key_map[value_info['key']] = value_info
new_node_params[value_key] = value_info['value']
self._node_params_map[value_key] = value_info
self.node_params = new_node_params
self._image_ext = ['png', 'jpg', 'jpeg', 'bmp']
self._vector_client = None
self._es_client = None
def is_dialog_input(self):
""" Whether the input is in the form of a conversation """
if self._tab == 'dialog_input':
return True
elif self._tab == 'form_input':
return False
raise IgnoreException(f'{self.name} -- workflow node is update')
def get_input_schema(self) -> Any:
if self.is_dialog_input():
try:
self.handle_recommended_questions()
except Exception as e:
logger.exception("handle recommended questions error")
user_input_info = self.node_data.get_variable_info('user_input')
user_input_info.value = [
self.node_data.get_variable_info('dialog_files_content'),
self.node_data.get_variable_info('dialog_file_accept'),
self.node_data.get_variable_info('user_input_file'),
]
return user_input_info
form_input_info = self.node_data.get_variable_info('form_input')
form_variables = copy.deepcopy(self._node_params_map)
res = []
for one_key, one in form_variables.items():
one['key'] = one_key
one['value'], _ = self.parse_msg_with_variables(one['value'])
res.append(one)
form_input_info.value = res
return form_input_info
def handle_recommended_questions(self):
recommended_questions_flag = self._original_node_params.get('recommended_questions_flag', False)
if not recommended_questions_flag:
return
recommended_llm = self._original_node_params.get('recommended_llm', 0)
recommended_system_prompt = self._original_node_params.get('recommended_system_prompt', '')
recommended_history_num = self._original_node_params.get('recommended_history_num', 3)
if not recommended_llm or not recommended_system_prompt or not recommended_history_num:
logger.debug(f"{self.name} recommended questions config incomplete")
return
chat_history = self.graph_state.get_history_memory(recommended_history_num)
if not chat_history:
logger.debug(f"{self.name} recommended questions chat history is empty")
return
recommended_system_prompt, _ = self.parse_msg_with_variables(recommended_system_prompt)
llm_obj = LLMService.get_bisheng_llm_sync(model_id=recommended_llm,
app_id=self.workflow_id,
app_name=self.workflow_name,
app_type=ApplicationTypeEnum.WORKFLOW,
user_id=self.user_id)
user_prompt = f"# Current Conversation Context\n{chat_history}"
result = llm_obj.invoke([SystemMessage(content=recommended_system_prompt), HumanMessage(content=user_prompt)])
result = result.content
try:
result = json.loads(result)
except json.decoder.JSONDecodeError:
logger.debug("received non-json response from LLM, try json repair")
try:
result = json_repair.loads(result, skip_json_loads=True)
except Exception as e:
logger.error(f"json repair failed: {e}")
return
logger.debug(f"received response from LLM, result is {result}")
if not isinstance(result, dict):
return
questions = []
for key, value in result.items():
if isinstance(value, list):
questions = value
break
if not questions:
return
self.callback_manager.on_guide_question(data=GuideQuestionData(node_id=self.id, name=self.name,
unique_id=generate_uuid(),
guide_question=questions[:3]))
def _parse_upload_file_variables(self, key_info: Dict, key_value: Dict) -> Dict:
"""
parse upload_file variables
Documented metadataData, full-text files, minio file paths, image files path
"""
# Compatible processing of historical versions of nodes
if self.node_data.v <= self._current_v:
if self.is_dialog_input():
key_value.pop("dialog_file_paths", None)
return key_value
file_parse_mode = key_info.get('file_parse_mode', ParseModeEnum.INGEST_TO_KNOWLEDGE_BASE)
ret = {}
if file_parse_mode == ParseModeEnum.KEEP_RAW:
if key_info.get("file_type") in ["image", "all"]:
ret[key_info['image_file']] = key_value.get(key_info['image_file'], [])
ret[key_info['file_path']] = key_value.get(key_info['file_path'], [])
elif file_parse_mode == ParseModeEnum.EXTRACT_TEXT:
ret[key_info['file_content']] = key_value.get(key_info['file_content'], "")
else:
ret[key_info['key']] = key_value.get(key_info['key'], [])
return ret
def _run(self, unique_id: str):
if self.is_dialog_input():
key_info = {
"key": "dialog_files",
"file_content": "dialog_files_content",
"file_path": "dialog_file_paths",
"image_file": "dialog_image_files",
"file_type": self._original_node_params.get("dialog_file_accept"),
"file_parse_mode": self._original_node_params.get("file_parse_mode", ParseModeEnum.EXTRACT_TEXT),
"file_content_size": self._original_node_params.get("dialog_files_content_size", 15000)
}
# Input in the form of a dialog
result = self.parse_upload_file("dialog_files", key_info, self.node_params.get('dialog_files_content', []))
res = {
'user_input': self.node_params['user_input'],
}
result.pop("dialog_files", None)
res.update(self._parse_upload_file_variables(key_info, result))
self.graph_state.save_context(content=f'{res.get("dialog_files_content", "")}\n{res["user_input"]}',
msg_sender='human')
return res
ret = {}
human_input = ""
# The corresponding file upload needs to be processed in the form
for key, value in self.node_params.items():
key_info = self._node_params_map[key]
label, _ = self.parse_msg_with_variables(key_info.get('value')) if key_info.get('value') else key
if key_info['type'] == 'file':
new_params = self.parse_upload_file(key, key_info, value)
ret.update(self._parse_upload_file_variables(key_info, new_params))
if new_params[key_info['key']]:
content = ""
for one in new_params[key_info['key']]:
content += f"{one.get('document_name', '')},"
human_input += f"{label}: {content.rstrip(',')}\n"
else:
ret[key] = value
human_input += f"{label}: {value}\n"
self.graph_state.save_context(content=f'{human_input}', msg_sender='human')
return ret
def parse_log(self, unique_id: str, result: dict) -> Any:
ret = []
for k, v in result.items():
if (self._node_params_map.get(k) and self._node_params_map[k]['type'] == 'file') or (
self._file_key_map.get(k) and self._file_key_map[k]['type'] == 'file'):
continue
ret.append({"key": f'{self.id}.{k}', "value": v, "type": "variable"})
return [ret]
def get_upload_file_path_content(self, file_url: str) -> (list, list):
"""
params:
file_url: upload to minio share url
return:
0: chunks list
1: metadata list
"""
filepath, file_name = file_download(file_url)
texts = []
metadatas = []
try:
file_rule = FileProcessBase(knowledge_id=0)
texts, metadatas, _, _ = read_chunk_text(self.user_id, filepath, file_name, file_rule.separator,
file_rule.separator_rule,
file_rule.chunk_size, 0, None,
file_rule.retain_images, file_rule.enable_formula,
file_rule.force_ocr,
file_rule.filter_page_header_footer, file_rule.excel_rule)
except KnowledgeFileNotSupportedError as e:
logger.warning('input node file type is not support')
pass
return texts, metadatas
def init_vector_clients(self):
if self._vector_client is None:
embedding = LLMService.get_knowledge_default_embedding(self.user_id)
if not embedding:
raise Exception('No default configured embedding Models')
milvus_collection_name = self.get_milvus_collection_name(getattr(embedding, 'model_id'))
self._vector_client = KnowledgeRag.init_milvus_vectorstore(milvus_collection_name, embedding,
metadata_schemas=InputFileMetadata)
self._es_client = KnowledgeRag.init_es_vectorstore_sync(self.tmp_collection_name,
metadata_schemas=InputFileMetadata)
def parse_upload_file(self, key: str, key_info: dict, value: List[str]) -> dict | None:
"""
parse upload_file
Documented metadataData, full-text files, minio file paths, image files path
"""
# Parsing the file. need return values
all_metadata = []
all_file_content = ''
original_file_path = []
image_files_path = []
if not value:
logger.warning(f"{self.id}.{key} value is None")
return {
key_info['key']: all_metadata,
key_info['file_content']: all_file_content,
key_info['file_path']: original_file_path,
key_info['image_file']: image_files_path
}
file_parse_mode = key_info.get('file_parse_mode', ParseModeEnum.INGEST_TO_KNOWLEDGE_BASE)
file_content_max_size = int(key_info.get('file_content_size', 15000))
file_id = generate_uuid()
file_content_length = 0
for one_file_url in value:
url_obj = urlparse(one_file_url)
file_name = unquote(url_obj.path.split('/')[-1])
# get file original name
file_name = KnowledgeService.get_upload_file_original_name(file_name)
all_metadata.append({
"document_id": file_id,
"document_name": file_name,
"knowledge_id": self.workflow_id,
"upload_time": int(time.time()),
"bbox": '', # Temporary files cannot be traced because the source files are not persisted
})
file_ext = file_name.split('.')[-1].lower()
logger.debug(f"{self.id}.{key} file_parse_mode is keep_raw")
original_file_path.append(one_file_url)
if file_ext in self._image_ext:
image_files_path.append(one_file_url)
if file_parse_mode == ParseModeEnum.KEEP_RAW:
continue
texts, metadatas = self.get_upload_file_path_content(one_file_url)
if file_content_length < file_content_max_size:
file_content = "\n".join(texts)
file_content = file_content[:file_content_max_size - file_content_length]
file_content_length += len(file_content)
all_file_content += f"[file name]: {file_name}\n[file content begin]\n{file_content}\n[file content end]\n"
if not texts:
logger.debug(f"{self.id}.{key} extract file text is empty")
continue
if file_parse_mode == ParseModeEnum.EXTRACT_TEXT:
continue
self.init_vector_clients()
new_metadata = []
# A file corresponding to the same variable, placed in a file_id mile
for one in metadatas:
metadata = one.model_dump()
metadata.update(all_metadata[-1])
new_metadata.append(metadata)
# Uploaded to milvus And es
logger.debug(f'workflow_add_vectordb file={key} file_name={file_name} file_id={file_id}')
self._vector_client.add_texts(texts=texts, metadatas=new_metadata)
logger.debug(f'workflow_add_es file={key} file_name={file_name} file_id={file_id}')
self._es_client.add_texts(texts=texts, metadatas=new_metadata)
logger.debug(f'workflow_record_file_metadata file={key} file_name={file_name}')
all_metadata[-1] = new_metadata[0]
# Documentation metadata, other nodes according to metadataData to retrieve corresponding files
return {
key_info['key']: all_metadata,
key_info['file_content']: all_file_content,
key_info['file_path']: original_file_path,
key_info['image_file']: image_files_path
}
================================================
FILE: src/backend/bisheng/workflow/nodes/knowledge_retriever/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/knowledge_retriever/knowledge_retriever.py
================================================
from typing import Any
from loguru import logger
from bisheng.workflow.common.knowledge import RagUtils
class KnowledgeRetriever(RagUtils):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self._output_keys = [one.get("key") for one in self.node_params.get('retrieved_result', [])]
def _run(self, unique_id: str):
try:
self.user_questions = self.init_user_question()
self.init_user_info()
self.init_multi_retriever()
ret = {}
for index, question in enumerate(self.user_questions):
output_key = self._output_keys[index]
if question is None:
question = ""
try:
self.init_rerank_model()
question_answer = self.retrieve_question(question)
question_answer = [{
"text": one.page_content,
"metadata": {
"chunk_index": one.metadata.get('chunk_index'),
"knowledge_id": one.metadata.get('knowledge_id'),
"document_id": one.metadata.get('document_id'),
"document_name": one.metadata.get('document_name'),
"upload_time": one.metadata.get('upload_time'),
"update_time": one.metadata.get('update_time'),
"uploader": one.metadata.get('uploader'),
"updater": one.metadata.get('updater'),
"user_metadata": one.metadata.get('user_metadata'),
}
} for one in question_answer]
except Exception as e:
question_answer = str(e)
ret[output_key] = question_answer
except Exception as e:
logger.exception(f"KnowledgeRetriever node run error: {e}")
ret = {
one: str(e) for one in self._output_keys
}
return ret
def parse_log(self, unique_id: str, result: dict) -> Any:
ret = []
for index, question in enumerate(self.user_questions):
output_key = self._output_keys[index]
one_ret = [
{'key': f'{self.id}.user_question', 'value': question, "type": "variable"},
{'key': f'{self.id}.{output_key}', 'value': result[output_key], "type": "variable"},
]
ret.append(one_ret)
return ret
================================================
FILE: src/backend/bisheng/workflow/nodes/llm/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/llm/llm.py
================================================
from typing import Any
from langchain_core.messages import HumanMessage, SystemMessage
from langchain_core.runnables import RunnableConfig
from loguru import logger
from bisheng.common.constants.enums.telemetry import ApplicationTypeEnum
from bisheng.llm.domain.services import LLMService
from bisheng.workflow.callback.llm_callback import LLMNodeCallbackHandler
from bisheng.workflow.nodes.base import BaseNode
from bisheng.workflow.nodes.prompt_template import PromptTemplateParser
class LLMNode(BaseNode):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
# Determine if it is a single or batch
self._tab = self.node_data.tab['value']
# Whether to output the results to the user
self._output_user = self.node_params.get('output_user', False)
self._image_prompt = self.node_params.get('image_prompt', [])
# Inisialisasiprompt
self._system_prompt = PromptTemplateParser(template=self.node_params['system_prompt'])
self._system_variables = self._system_prompt.extract()
self._user_prompt = PromptTemplateParser(template=self.node_params['user_prompt'])
self._user_variables = self._user_prompt.extract()
# Data required to store logs
self._system_prompt_list = []
self._user_prompt_list = []
self._batch_variable_list = []
self._log_reasoning_content = []
# InisialisasillmObjects
self._llm = LLMService.get_bisheng_llm_sync(model_id=self.node_params['model_id'],
temperature=self.node_params.get('temperature', 0.3),
app_id=self.workflow_id,
app_name=self.workflow_name,
app_type=ApplicationTypeEnum.WORKFLOW,
user_id=self.user_id)
def _run(self, unique_id: str):
self._system_prompt_list = []
self._user_prompt_list = []
self._batch_variable_list = []
self._log_reasoning_content = []
result = {}
if self._tab == 'single':
result['output'], reasoning_content = self._run_once(None, unique_id, 'output')
self._log_reasoning_content.append(reasoning_content)
else:
for index, one in enumerate(self.node_params['batch_variable']):
self._batch_variable_list.append(self.get_other_node_variable(one))
output_key = self.node_params['output'][index]['key']
result[output_key], reasoning_content = self._run_once(one, unique_id, output_key)
self._log_reasoning_content.append(reasoning_content)
if self._output_user:
for k, v in result.items():
self.graph_state.save_context(content=v, msg_sender='AI')
return result
def parse_log(self, unique_id: str, result: dict) -> Any:
ret = []
index = 0
for k, v in result.items():
one_ret = [
{"key": "system_prompt", "value": self._system_prompt_list[index], "type": "params"},
{"key": "user_prompt", "value": self._user_prompt_list[index], "type": "params"},
]
if self._log_reasoning_content[index]:
one_ret.append({"key": "Thinking about content", "value": self._log_reasoning_content[index], "type": "params"})
one_ret.append({"key": f'{self.id}.{k}', "value": v, "type": "variable"})
if self._batch_variable_list:
one_ret.insert(0, {"key": f"{self.id}.batch_variable", "value": self._batch_variable_list[index],
"type": "variable"})
index += 1
ret.append(one_ret)
return ret
def _run_once(self,
input_variable: str = None,
unique_id: str = None,
output_key: str = None) -> (str, str):
# Description is a variable that references a batch, The value of the variable needs to be replaced with the variable selected by the user
special_variable = f'{self.id}.batch_variable'
variable_map = {}
for one in self._system_variables:
if input_variable and one == special_variable:
variable_map[one] = self.get_other_node_variable(input_variable)
continue
variable_map[one] = self.get_other_node_variable(one)
system = self._system_prompt.format(variable_map)
self._system_prompt_list.append(system)
variable_map = {}
for one in self._user_variables:
if input_variable and one == special_variable:
variable_map[one] = self.get_other_node_variable(input_variable)
continue
variable_map[one] = self.get_other_node_variable(one)
user = self._user_prompt.format(variable_map)
self._user_prompt_list.append(user)
logger.debug(
f'outputkey={output_key} workflow llm node prompt: system: {system}\nuser: {user}')
llm_callback = LLMNodeCallbackHandler(callback=self.callback_manager,
unique_id=unique_id,
node_id=self.id,
node_name=self.name,
output=self._output_user,
output_key=output_key)
config = RunnableConfig(callbacks=[llm_callback])
inputs = []
if system:
inputs.append(SystemMessage(content=system))
human_message = HumanMessage(content=[{
'type': 'text',
'text': user
}])
human_message = self.contact_file_into_prompt(human_message, self._image_prompt)
inputs.append(human_message)
logger.debug(f'llm invoke chat_history: {inputs} {self._image_prompt}')
result = self._llm.invoke(inputs, config=config)
return result.content, llm_callback.reasoning_content
================================================
FILE: src/backend/bisheng/workflow/nodes/node_manage.py
================================================
from bisheng.workflow.common.node import NodeType
from bisheng.workflow.nodes.agent.agent import AgentNode
from bisheng.workflow.nodes.code.code import CodeNode
from bisheng.workflow.nodes.condition.condition import ConditionNode
from bisheng.workflow.nodes.end.end import EndNode
from bisheng.workflow.nodes.input.input import InputNode
from bisheng.workflow.nodes.knowledge_retriever.knowledge_retriever import KnowledgeRetriever
from bisheng.workflow.nodes.llm.llm import LLMNode
from bisheng.workflow.nodes.output.output import OutputNode
from bisheng.workflow.nodes.qa_retriever.qa_retriever import QARetrieverNode
from bisheng.workflow.nodes.rag.rag import RagNode
from bisheng.workflow.nodes.report.report import ReportNode
from bisheng.workflow.nodes.start.start import StartNode
from bisheng.workflow.nodes.tool.tool import ToolNode
NODE_CLASS_MAP = {
NodeType.START.value: StartNode,
NodeType.END.value: EndNode,
NodeType.INPUT.value: InputNode,
NodeType.OUTPUT.value: OutputNode,
NodeType.TOOL.value: ToolNode,
NodeType.RAG.value: RagNode,
NodeType.REPORT.value: ReportNode,
NodeType.QA_RETRIEVER.value: QARetrieverNode,
NodeType.CONDITION.value: ConditionNode,
NodeType.AGENT.value: AgentNode,
NodeType.CODE.value: CodeNode,
NodeType.LLM.value: LLMNode,
NodeType.KNOWLEDGE_RETRIEVER.value: KnowledgeRetriever,
}
class NodeFactory:
@classmethod
def get_node_class(cls, node_type: str) -> 'BaseNode':
return NODE_CLASS_MAP.get(node_type)
@classmethod
def instance_node(cls, node_type: str, **kwargs) -> 'BaseNode':
node_class = cls.get_node_class(node_type)
if node_class is None:
raise Exception(f'Unknown node type:{node_type}')
return node_class(**kwargs)
================================================
FILE: src/backend/bisheng/workflow/nodes/output/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/output/output.py
================================================
import json
from typing import Any
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync
from bisheng.workflow.callback.event import OutputMsgChooseData, OutputMsgData, OutputMsgInputData
from bisheng.workflow.nodes.base import BaseNode
from bisheng.workflow.nodes.prompt_template import PromptTemplateParser
class OutputNode(BaseNode):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
# minio
self._minio_client = get_minio_storage_sync()
# Interaction Type
self._output_type = self.node_params['output_result']['type']
self._output_result = self.node_params['output_result']['value']
# Result of user processing
self._handled_output_result = self._output_result
# user input msg
if 'output_msg' in self.node_params:
_original_output_msg = self.node_params['output_msg']
else:
_original_output_msg = self.node_params['message']
self._output_msg = _original_output_msg['msg']
self._output_files = _original_output_msg['files']
# Message Content After Variable Replacement
self._parsed_output_msg = ''
self._parsed_files = []
self._source_documents = []
# Non-selective interaction, then the next node is wiredtarget. Selective interactions, which need to be judged based on user input
self._next_node_id = [one.target for one in self.target_edges]
def handle_input(self, user_input: dict) -> Any:
# Needs to be depositedstate，
self.graph_state.save_context(content=json.dumps(user_input, ensure_ascii=False), msg_sender='human')
self._handled_output_result = user_input['output_result']
self.graph_state.set_variable(self.id, 'output_result', user_input['output_result'])
def get_input_schema(self) -> Any:
# Explain that no interaction is required
if self._output_type not in ['input', 'choose']:
return None
group_params = self.node_data.dict(include={'group_params'})
return group_params['group_params']
def is_condition_node(self) -> bool:
return self._output_type == 'choose'
def route_node(self, state: dict) -> str | list[str]:
# Selective interaction requires judging the next node based on the user's input
if self._output_type == 'choose':
return self.get_next_node_id(self._handled_output_result)
return self._next_node_id
def _run(self, unique_id: str):
self._source_documents = []
self.parse_output_msg()
self.send_output_msg(unique_id)
res = {
'message': self._parsed_output_msg,
'output_result': self._handled_output_result
}
return res
def parse_log(self, unique_id: str, result: dict) -> Any:
ret = [
{
"key": "output_msg",
"value": self._parsed_output_msg,
"type": "params"
}
]
if self._output_type == 'input':
ret.append({
"key": "output_result",
"value": self._handled_output_result,
"type": "key"
})
return [ret]
def parse_output_msg(self):
""" Paddingmsgvariable in, get the file'sshareAddress: """
self._parsed_output_msg = self.parse_template_msg(self._output_msg)
if self._parsed_files:
return
for one in self._output_files:
one['path'] = self._minio_client.get_share_link_sync(one['path'])
self._parsed_files.append(one)
def send_output_msg(self, unique_id: str):
""" SendoutputNode's Message """
msg_params = {
'name': self.name,
'unique_id': unique_id,
'node_id': self.id,
'msg': self._parsed_output_msg,
'files': self._parsed_files,
'output_key': '',
'source_documents': self._source_documents
}
# where interaction is required, there isgroup_params
if self._output_type == 'input':
msg_params['key'] = 'output_result'
msg_params['input_msg'] = self.parse_template_msg(self._output_result)
self.callback_manager.on_output_input(data=OutputMsgInputData(**msg_params))
elif self._output_type == 'choose':
msg_params['key'] = 'output_result'
msg_params['options'] = self.node_data.get_variable_info('output_result').options
self.callback_manager.on_output_choose(data=OutputMsgChooseData(**msg_params))
else:
self.graph_state.save_context(content=self._parsed_output_msg,
msg_sender='AI')
self.callback_manager.on_output_msg(OutputMsgData(**msg_params))
def parse_template_msg(self, msg: str):
msg_template = PromptTemplateParser(template=msg)
variables = msg_template.extract()
if len(variables) > 0:
var_map = {}
for one in variables:
node_id = one.split('.')[0]
# CiteqaDemonstrate traceability when using the Knowledge Base node
if node_id.startswith('qa_retriever'):
self._source_documents = self.graph_state.get_variable(node_id, '$retrieved_result$')
var_map[one] = self.get_other_node_variable(one)
msg = msg_template.format(var_map)
return msg
================================================
FILE: src/backend/bisheng/workflow/nodes/output/output_fake.py
================================================
from bisheng.workflow.callback.event import NodeEndData
from pydantic import ConfigDict, BaseModel, Field
from bisheng.workflow.nodes.output.output import OutputNode
class OutputFakeNode(BaseModel):
""" Used to processoutputInterrupt to determine if user input is required """
model_config = ConfigDict(arbitrary_types_allowed=True)
id: str
output_node: OutputNode
type: str
def run(self, state: dict):
""" Do nothing, just use it to deal withoutputInterrupt to determine if user input is required """
self.output_node.callback_manager.on_node_end(data=NodeEndData(
unique_id=self.output_node.exec_unique_id,
node_id=self.output_node.id,
name=self.output_node.name,
reason=None,
log_data=self.output_node.parse_log(self.output_node.exec_unique_id, {})))
return state
async def arun(self, state: dict):
return self.run(state)
def get_input_schema(self):
return self.output_node.get_input_schema()
def stop(self):
pass
================================================
FILE: src/backend/bisheng/workflow/nodes/prompt_template.py
================================================
import re
REGEX = re.compile(r'\{\{#([#a-zA-Z_][#a-zA-Z0-9_]{0,29})#\}\}')
WITH_VARIABLE_TMPL_REGEX = re.compile(
r'\{\{#([#a-zA-Z_][a-zA-Z0-9_]{0,29}|[#a-zA-Z0-9_]{1,50}\.[#a-zA-Z0-9_\.]{1,100})#\}\}')
class PromptTemplateParser:
"""
Rules:
1. Template variables must be enclosed in `{{}}`.
2. The template variable Key can only be: letters + numbers + underscore, with a maximum length of 16 characters,
and can only start with letters and underscores.
3. The template variable Key cannot contain new lines or spaces, and must comply with rule 2.
4. In addition to the above, 3 types of special template variable Keys are accepted:
`{{#histories#}}` `{{#query#}}` `{{#context#}}`. No other `{{##}}` template variables are allowed.
"""
def __init__(self, template: str, with_variable_tmpl: bool = True):
self.template = template
self.with_variable_tmpl = with_variable_tmpl
self.regex = WITH_VARIABLE_TMPL_REGEX if with_variable_tmpl else REGEX
self.variable_keys = self.extract()
def extract(self) -> list:
# Regular expression to match the template rules
return re.findall(self.regex, self.template)
def format(self, inputs: dict, remove_template_variables: bool = True) -> str:
def replacer(match):
key = match.group(1)
value = inputs.get(key, match.group(0)) # return original matched string if key not found
if not isinstance(value, str):
value = str(value)
if remove_template_variables:
return PromptTemplateParser.remove_template_variables(
value, self.with_variable_tmpl)
return value
prompt = re.sub(self.regex, replacer, self.template)
return re.sub(r'<\|.*?\|>', '', prompt)
@classmethod
def remove_template_variables(cls, text: str, with_variable_tmpl: bool = False):
return re.sub(WITH_VARIABLE_TMPL_REGEX if with_variable_tmpl else REGEX, r'{\1}', text)
================================================
FILE: src/backend/bisheng/workflow/nodes/qa_retriever/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/qa_retriever/qa_retriever.py
================================================
import json
from typing import Any
from bisheng.interface.initialize.loading import instantiate_vectorstore
from bisheng.interface.vector_store.custom import MilvusWithPermissionCheck
from bisheng.user.domain.models.user import UserDao
from bisheng.workflow.nodes.base import BaseNode
from bisheng_langchain.chains.retrieval.retrieval_chain import RetrievalChain
class QARetrieverNode(BaseNode):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
# Initialize input
self._user_question = self.node_params.get('user_question', '')
self._qa_knowledge_id = self.node_params.get('qa_knowledge_id', [])
self._score = self.node_params.get('score', 0.6)
# Inisialisasiretriever, Running Initialization
self._retriever = None
def _init_retriever(self):
if self._retriever:
return
# Vector database client initialization, currently usingMilvus, more rational use of more generic factory methods
params = {}
params['search_kwargs'] = {'k': 1, 'score_threshold': self._score}
params['search_type'] = 'similarity_score_threshold'
params['collection_name'] = self._qa_knowledge_id # [{"key":"", "label":""}]
params['user_name'] = UserDao.get_user(self.user_id).user_name
params['_is_check_auth'] = False
knowledge_retriever = instantiate_vectorstore(
node_type='MilvusWithPermissionCheck',
class_object=MilvusWithPermissionCheck,
params=params,
)
self._retriever = RetrievalChain(retriever=knowledge_retriever)
def _run(self, unique_id: str):
self._init_retriever()
question = self.get_other_node_variable(self._user_question)
result = self._retriever.invoke({'query': question})
# qa have a result; turn out to bedocument
if result['result']:
# the source documents that store the retrieval results,keyLeft and right plus$As source documentkeyGo to inquiry
self.graph_state.set_variable(self.id, '$retrieved_result$', result['result'][0])
result_str = json.loads(result['result'][0].metadata['extra'])['answer']
else:
result_str = ''
self.graph_state.set_variable(self.id, '$retrieved_result$', None)
return {
'retrieved_result': result_str
}
def parse_log(self, unique_id: str, result: dict) -> Any:
return [[
{
"key": "user_question",
"value": self.get_other_node_variable(self._user_question),
"type": "params"
},
{
"key": "retrieved_result",
"value": result['retrieved_result'],
"type": "params"
}
]]
================================================
FILE: src/backend/bisheng/workflow/nodes/rag/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/rag/rag.py
================================================
import json
import time
from typing import List, Any
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain_core.documents import Document
from langchain_core.prompts import (ChatPromptTemplate, HumanMessagePromptTemplate,
SystemMessagePromptTemplate)
from langchain_core.runnables import RunnableConfig
from loguru import logger
from bisheng.chat.types import IgnoreException
from bisheng.common.constants.enums.telemetry import ApplicationTypeEnum
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync
from bisheng.llm.domain.services import LLMService
from bisheng.workflow.callback.event import OutputMsgData, StreamMsgOverData
from bisheng.workflow.callback.llm_callback import LLMNodeCallbackHandler
from bisheng.workflow.common.knowledge import RagUtils
from bisheng.workflow.nodes.prompt_template import PromptTemplateParser
class RagNode(RagUtils):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
# analyzingprompt
self._system_prompt = PromptTemplateParser(template=self.node_params['system_prompt'])
self._system_variables = self._system_prompt.extract()
self._user_prompt = PromptTemplateParser(template=self.node_params['user_prompt'])
self._user_variables = self._user_prompt.extract()
self._qa_prompt = None
self._llm = LLMService.get_bisheng_llm_sync(model_id=self.node_params['model_id'],
temperature=self.node_params.get('temperature', 0.3),
app_id=self.workflow_id,
app_name=self.workflow_name,
app_type=ApplicationTypeEnum.WORKFLOW,
user_id=self.user_id)
self._minio_client = get_minio_storage_sync()
# Whether to output the results to the user
self._output_user = self.node_params.get('output_user', False)
self._output_keys = [one.get("key") for one in self.node_params.get('output_user_input', [])]
# Run Log Data
self._log_source_documents = {}
self._log_system_prompt = []
self._log_user_prompt = []
self._log_reasoning_content = {}
self._milvus = None
self._es = None
def _run(self, unique_id: str):
ret = {}
self.init_user_info()
self._log_source_documents = {}
self._log_system_prompt = []
self._log_user_prompt = []
self._log_reasoning_content = {}
self.init_qa_prompt()
self.user_questions = self.init_user_question()
for index, question in enumerate(self.user_questions):
output_key = self._output_keys[index]
if question is None:
question = ''
question_answer = self.rag_one_question(question, output_key, unique_id)
ret[output_key] = question_answer
return ret
def rag_one_question(self, question: str, output_key: str, unique_id: str) -> str:
try:
self.init_multi_retriever()
self.init_rerank_model()
source_documents = self.retrieve_question(question)
except Exception as e:
logger.exception(f'RagNode retrieve_question error: ')
source_documents = [Document(page_content=str(e), metadata={})]
qa_chain = create_stuff_documents_chain(llm=self._llm, prompt=self._qa_prompt)
inputs = {
"context": source_documents,
}
if "question" in self._qa_prompt.input_variables:
inputs["question"] = question
# and one of the reasons thatragIt needs to be traced, so it can't be used universally.llm callbackto return the message. Need to getsource_documentReturn message content after
llm_callback = LLMNodeCallbackHandler(callback=self.callback_manager,
unique_id=unique_id,
node_id=self.id,
node_name=self.name,
output=self._output_user,
output_key=output_key,
cancel_llm_end=True)
result = qa_chain.invoke(inputs, config=RunnableConfig(callbacks=[llm_callback]))
if self._output_user:
self.graph_state.save_context(content=result, msg_sender='AI')
if llm_callback.output_len == 0:
self.callback_manager.on_output_msg(
OutputMsgData(node_id=self.id,
name=self.name,
msg=result,
unique_id=unique_id,
output_key=output_key,
source_documents=source_documents))
else:
# If there is a streaming output, the streaming end event is triggered, Because of the need tosource_documentSo do a streaming end event here
self.callback_manager.on_stream_over(StreamMsgOverData(
node_id=self.id,
name=self.name,
msg=result,
reasoning_content=llm_callback.reasoning_content,
unique_id=unique_id,
source_documents=source_documents,
output_key=output_key,
))
self._log_reasoning_content[output_key] = llm_callback.reasoning_content
self._log_source_documents[output_key] = source_documents
return result
def parse_log(self, unique_id: str, result: dict) -> Any:
ret = []
index = 0
user_question_list = self.init_user_question()
# Determine if the search results exceed a certain length, The reason iswsSending a message that exceeds a certain length will result in an error
source_documents = [[d.page_content for d in one] for one in self._log_source_documents.values()]
tmp_retrieved_type = 'variable'
tmp_retrieved_result = json.dumps(source_documents, indent=2, ensure_ascii=False)
if len(tmp_retrieved_result.encode('utf-8')) >= 50 * 1024: # Lebih dari50kbLog data storage file
tmp_retrieved_type = 'file'
tmp_object_name = f'/workflow/source_document/{time.time()}.txt'
self._minio_client.put_object_tmp_sync(tmp_object_name, tmp_retrieved_result.encode('utf-8'))
tmp_retrieved_result = self._minio_client.get_share_link_sync(tmp_object_name,
self._minio_client.tmp_bucket)
for key, val in result.items():
if tmp_retrieved_type != 'file':
tmp_retrieved_result = json.dumps([one.page_content for one in self._log_source_documents[key]],
indent=2, ensure_ascii=False)
one_ret = [
{'key': f'{self.id}.user_question', 'value': user_question_list[index], "type": "variable"},
{'key': f'{self.id}.retrieved_result', 'value': tmp_retrieved_result, "type": tmp_retrieved_type},
{'key': 'system_prompt', 'value': self._log_system_prompt[0], "type": "params"},
{'key': 'user_prompt', 'value': self._log_user_prompt[0], "type": "params"},
]
if self._log_reasoning_content[key]:
one_ret.append({'key': 'Thinking about content', 'value': self._log_reasoning_content[key], "type": "params"})
one_ret.append({'key': f'{self.id}.{key}', 'value': val, 'type': 'variable'})
index += 1
ret.append(one_ret)
return ret
def init_user_question(self) -> List[str]:
# Convert all user questions to strings by default
ret = []
for one in self.node_params['user_question']:
ret.append(f"{self.get_other_node_variable(one)}")
return ret
def init_qa_prompt(self):
variable_map = {}
for one in self._user_variables:
if one == f'{self.id}.user_question':
variable_map[one] = '$$question$$'
elif one == f'{self.id}.retrieved_result':
variable_map[one] = '$$context$$'
else:
variable_map[one] = self.get_other_node_variable(one)
if variable_map.get(f'{self.id}.retrieved_result') is None:
raise IgnoreException('User prompts must contain retrieved_result Variables')
user_prompt = self._user_prompt.format(variable_map)
log_user_prompt = user_prompt.replace('$$question$$', '{user_question}').replace('$$context$$',
'{retrieved_result}')
user_prompt = (user_prompt.replace('{', '{{').replace('}', '}}')
.replace('$$question$$', '{question}').replace('$$context$$', '{context}'))
self._log_user_prompt.append(log_user_prompt)
variable_map = {}
for one in self._system_variables:
variable_map[one] = self.get_other_node_variable(one)
system_prompt = self._system_prompt.format(variable_map)
system_prompt.replace('{', '{{').replace('}', '}}')
self._log_system_prompt.append(system_prompt)
messages_general = [
SystemMessagePromptTemplate.from_template(system_prompt),
HumanMessagePromptTemplate.from_template(user_prompt),
]
self._qa_prompt = ChatPromptTemplate.from_messages(messages_general)
================================================
FILE: src/backend/bisheng/workflow/nodes/report/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/report/docx_replace.py
================================================
import re
from io import BytesIO
from typing import List, Dict, Any, IO
from docx import Document
from docx.enum.style import WD_STYLE_TYPE
from docx.oxml import OxmlElement
from docx.shared import Pt, Inches, RGBColor
from docx.table import _Cell
from docx.text.paragraph import Paragraph
class DocxReplacer:
"""
Docx File Content Replacement Tool
Supports complex format placeholder replacement, including text, tables, images, titles, etc.
"""
def __init__(self, template_path: str | IO[bytes]):
self.template_path = template_path
self.doc = Document(template_path)
self.placeholder_pattern = re.compile(r'\{\{([^}]+)\}\}')
self._init_style()
def check_style(self, style_name: str, **kwargs):
all_style = self.doc.styles
if style_name not in all_style:
style = self.doc.styles.add_style(style_name, WD_STYLE_TYPE.PARAGRAPH)
font = style.font
for key, value in kwargs.items():
setattr(font, key, value)
else:
style = all_style[style_name]
return style
def _init_style(self):
self.check_style("Heading 1", size=304800)
self.check_style("Heading 2", size=254000)
self.check_style("Heading 3", size=203200)
self.check_style("Heading 4", size=177800)
self.check_style("Heading 5", size=152400)
self.check_style("Heading 6", bold=True)
def replace_and_save(self, variables: Dict[str, List[Dict[str, Any]]], output_path: str):
"""
Replace the placeholders and save the document.
Args:
Variables: A dictionary of variables, with the format of {"var_name": [{"type": "text", "content": "..."}]}
Output_path: The path of the output file
"""
self._process_paragraphs(self.doc.paragraphs, variables)
for table in self.doc.tables:
self._process_table(table, variables)
for section in self.doc.sections:
if section.header:
self._process_paragraphs(section.header.paragraphs, variables)
if section.footer:
self._process_paragraphs(section.footer.paragraphs, variables)
self.doc.save(output_path)
def _process_table(self, table, variables: Dict[str, List[Dict[str, Any]]]):
for row in table.rows:
for cell in row.cells:
self._process_paragraphs(cell.paragraphs, variables)
for nested_table in cell.tables:
self._process_table(nested_table, variables)
def _process_paragraphs(self, paragraphs: List[Paragraph], variables: Dict[str, List[Dict[str, Any]]]):
i = 0
while i < len(paragraphs):
paragraph = paragraphs[i]
text = paragraph.text
matches = list(self.placeholder_pattern.finditer(text))
if matches:
insert_index = self._get_paragraph_index(paragraph)
self._replace_paragraph_placeholders(paragraph, matches, variables, insert_index)
i += 1
def _get_paragraph_index(self, paragraph: Paragraph) -> int:
parent = paragraph._element.getparent()
return parent.index(paragraph._element)
def _replace_paragraph_placeholders(
self,
paragraph: Paragraph,
matches: List[re.Match],
variables: Dict[str, List[Dict[str, Any]]],
insert_index: int
):
parent = paragraph._element.getparent()
text = paragraph.text
segments = []
last_end = 0
for match in matches:
var_name = match.group(1)
start, end = match.span()
if start > last_end:
segments.append({
'type': 'text_segment',
'content': text[last_end:start],
'paragraph': paragraph
})
if var_name in variables:
segments.append({
'type': 'variable',
'content': variables[var_name],
'paragraph': paragraph
})
else:
segments.append({
'type': 'text_segment',
'content': match.group(0),
'paragraph': paragraph
})
last_end = end
if last_end < len(text):
segments.append({
'type': 'text_segment',
'content': text[last_end:],
'paragraph': paragraph
})
original_format = self._extract_paragraph_format(paragraph)
original_run_format = self._extract_run_format(paragraph.runs[0] if paragraph.runs else None)
parent.remove(paragraph._element)
current_insert_index = insert_index
current_paragraph = None
for segment in segments:
if segment['type'] == 'text_segment':
if current_paragraph is None:
current_paragraph = self._insert_paragraph_at_index(
parent, current_insert_index, original_format
)
current_insert_index += 1
run = current_paragraph.add_run(segment['content'])
self._apply_run_format(run, original_run_format)
elif segment['type'] == 'variable':
for item in segment['content']:
item_type = item.get('type')
if item_type == 'text':
if current_paragraph is None:
current_paragraph = self._insert_paragraph_at_index(
parent, current_insert_index, original_format
)
current_insert_index += 1
run = current_paragraph.add_run(item['content'])
self._apply_run_format(run, item)
elif item_type in ['table', 'image', 'heading']:
if current_paragraph is not None and current_paragraph.text.strip():
current_paragraph = None
if item_type == 'table':
self._insert_table_at_index(parent, current_insert_index, item)
elif item_type == 'image':
self._insert_image_at_index(parent, current_insert_index, item, original_format)
elif item_type == 'heading':
self._insert_heading_at_index(parent, current_insert_index, item)
current_insert_index += 1
current_paragraph = None
def _extract_paragraph_format(self, paragraph: Paragraph) -> Dict[str, Any]:
return {
'alignment': paragraph.alignment,
'left_indent': paragraph.paragraph_format.left_indent,
'right_indent': paragraph.paragraph_format.right_indent,
'first_line_indent': paragraph.paragraph_format.first_line_indent,
'space_before': paragraph.paragraph_format.space_before,
'space_after': paragraph.paragraph_format.space_after,
'line_spacing': paragraph.paragraph_format.line_spacing,
}
def _extract_run_format(self, run) -> Dict[str, Any]:
if run is None:
return {}
return {
'bold': run.bold,
'italic': run.italic,
'underline': run.underline,
'font_name': run.font.name,
'font_size': run.font.size,
'font_color': run.font.color.rgb if run.font.color.rgb else None,
}
def _insert_paragraph_at_index(self, parent, index: int, format_dict: Dict[str, Any]) -> Paragraph:
p_element = OxmlElement('w:p')
parent.insert(index, p_element)
paragraph = Paragraph(p_element, self.doc)
# 应用格式
if format_dict.get('alignment') is not None:
paragraph.alignment = format_dict['alignment']
if format_dict.get('left_indent') is not None:
paragraph.paragraph_format.left_indent = format_dict['left_indent']
if format_dict.get('right_indent') is not None:
paragraph.paragraph_format.right_indent = format_dict['right_indent']
if format_dict.get('first_line_indent') is not None:
paragraph.paragraph_format.first_line_indent = format_dict['first_line_indent']
if format_dict.get('space_before') is not None:
paragraph.paragraph_format.space_before = format_dict['space_before']
if format_dict.get('space_after') is not None:
paragraph.paragraph_format.space_after = format_dict['space_after']
if format_dict.get('line_spacing') is not None:
paragraph.paragraph_format.line_spacing = format_dict['line_spacing']
return paragraph
def _apply_run_format(self, run, format_data: Dict[str, Any]):
if format_data.get('bold'):
run.bold = True
if format_data.get('italic'):
run.italic = True
if format_data.get('underline'):
run.underline = True
if format_data.get('font_size'):
if isinstance(format_data['font_size'], int):
run.font.size = Pt(format_data['font_size'])
else:
run.font.size = format_data['font_size']
if format_data.get('font_name'):
run.font.name = format_data['font_name']
if format_data.get('color'):
if isinstance(format_data['color'], tuple) and len(format_data['color']) == 3:
run.font.color.rgb = RGBColor(*format_data['color'])
if format_data.get('font_color'):
run.font.color.rgb = format_data['font_color']
def _insert_table_at_index(self, parent, index: int, item: Dict[str, Any]):
data = item['content']
rows = len(data)
cols = len(data[0]) if rows > 0 else 0
if rows == 0 or cols == 0:
return
table = self.doc.add_table(rows=0, cols=cols)
if item.get('style'):
table.style = item['style']
for row_data in data:
row = table.add_row()
for col_idx, cell_content in enumerate(row_data):
if col_idx < len(row.cells):
cell = row.cells[col_idx]
self._fill_cell(cell, cell_content)
tbl_element = table._element
parent.remove(tbl_element)
parent.insert(index, tbl_element)
def _fill_cell(self, cell: _Cell, cell_content: Any):
"""
Filling Cell Content
Cell content format:
1. Single element (dictionary): {"type": "text", "content": "..."}
2. Composite elements (list): [{"type": "text", ...}, {"type": "image", ...}]
Args:
cell: The cell object
cell_content: The cell content, a dictionary or a list
"""
if cell.paragraphs:
default_paragraph = cell.paragraphs[0]
for run in default_paragraph.runs:
run.text = ''
else:
default_paragraph = cell.add_paragraph()
if isinstance(cell_content, dict):
if 'type' not in cell_content or 'content' not in cell_content:
raise ValueError(
f"The cell element must contain the `type` and `content` fields, but got：{cell_content}")
cell_content = [cell_content]
elif isinstance(cell_content, list):
for element in cell_content:
if not isinstance(element, dict) or 'type' not in element or 'content' not in element:
raise ValueError(
f"The cell element must contain the `type` and `content` fields, but got：{element}")
else:
raise ValueError(f"Not supported data type：{type(cell_content)}")
current_paragraph = default_paragraph
for element in cell_content:
element_type = element['type']
if element_type == 'text':
run = current_paragraph.add_run(element['content'])
self._apply_run_format(run, element)
elif element_type == 'image':
if current_paragraph.text.strip():
current_paragraph = cell.add_paragraph()
self._add_image_to_paragraph(current_paragraph, element)
current_paragraph = cell.add_paragraph()
elif element_type == 'paragraph':
current_paragraph = cell.add_paragraph()
if element.get('alignment'):
current_paragraph.alignment = element['alignment']
if isinstance(element['content'], str):
run = current_paragraph.add_run(element['content'])
self._apply_run_format(run, element)
elif isinstance(element['content'], list):
for text_item in element['content']:
if not isinstance(text_item, dict) or 'type' not in text_item:
raise ValueError(f"Paragraph content elements must include a `type` field; got：{text_item}")
if text_item['type'] == 'text':
run = current_paragraph.add_run(text_item['content'])
self._apply_run_format(run, text_item)
if element.get('alignment'):
for cell_paragraph in cell.paragraphs:
cell_paragraph.alignment = element['alignment']
def _add_image_to_paragraph(self, paragraph: Paragraph, image_data: Dict[str, Any]):
run = paragraph.add_run()
try:
width = Inches(image_data.get('width', 2))
height = Inches(image_data.get('height')) if image_data.get('height') else None
if isinstance(image_data['content'], str):
# local file path
if height:
run.add_picture(image_data['content'], width=width, height=height)
else:
run.add_picture(image_data['content'], width=width)
elif isinstance(image_data['content'], bytes):
# bytes data
image_stream = BytesIO(image_data['content'])
if height:
run.add_picture(image_stream, width=width, height=height)
else:
run.add_picture(image_stream, width=width)
except Exception as e:
paragraph.add_run(f"Image add failed: {str(e)}]")
# set alignment
if image_data.get('alignment'):
paragraph.alignment = image_data['alignment']
def _insert_image_at_index(self, parent, index: int, item: Dict[str, Any], paragraph_format: Dict[str, Any]):
paragraph = self._insert_paragraph_at_index(parent, index, paragraph_format)
self._add_image_to_paragraph(paragraph, item)
def _insert_heading_at_index(self, parent, index: int, item: Dict[str, Any]):
p_element = OxmlElement('w:p')
parent.insert(index, p_element)
paragraph = Paragraph(p_element, self.doc)
level = item.get('level', 1)
paragraph.style = f'Heading {level}'
run = paragraph.add_run(item['content'])
self._apply_run_format(run, item)
def extract_variables(self) -> List[str]:
variables = []
seen = set()
for paragraph in self.doc.paragraphs:
vars_in_paragraph = self._extract_vars_from_text(paragraph.text)
for var in vars_in_paragraph:
if var not in seen:
variables.append(var)
seen.add(var)
for table in self.doc.tables:
vars_in_table = self._extract_vars_from_table(table)
for var in vars_in_table:
if var not in seen:
variables.append(var)
seen.add(var)
for section in self.doc.sections:
if section.header:
for paragraph in section.header.paragraphs:
vars_in_paragraph = self._extract_vars_from_text(paragraph.text)
for var in vars_in_paragraph:
if var not in seen:
variables.append(var)
seen.add(var)
if section.footer:
for paragraph in section.footer.paragraphs:
vars_in_paragraph = self._extract_vars_from_text(paragraph.text)
for var in vars_in_paragraph:
if var not in seen:
variables.append(var)
seen.add(var)
return variables
def _extract_vars_from_text(self, text: str) -> List[str]:
matches = self.placeholder_pattern.findall(text)
return matches
def _extract_vars_from_table(self, table) -> List[str]:
variables = []
seen = set()
for row in table.rows:
for cell in row.cells:
for paragraph in cell.paragraphs:
vars_in_paragraph = self._extract_vars_from_text(paragraph.text)
for var in vars_in_paragraph:
if var not in seen:
variables.append(var)
seen.add(var)
for nested_table in cell.tables:
nested_vars = self._extract_vars_from_table(nested_table)
for var in nested_vars:
if var not in seen:
variables.append(var)
seen.add(var)
return variables
================================================
FILE: src/backend/bisheng/workflow/nodes/report/report.py
================================================
import io
import tempfile
from uuid import uuid4
from loguru import logger
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync
from bisheng.utils import generate_uuid
from bisheng.workflow.callback.event import OutputMsgData
from bisheng.workflow.nodes.base import BaseNode
from bisheng.workflow.nodes.report.docx_replace import DocxReplacer
from bisheng.workflow.nodes.report.text_classification import TextClassificationReport
class ReportNode(BaseNode):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self._report_info = self.node_params["report_info"]
self._version_key = self._report_info["version_key"].split("_")[0]
self._object_name = f"workflow/report/{self._version_key}.docx"
self._file_name = self._report_info["file_name"] if self._report_info["file_name"] else "tmp_report.docx"
if not self._file_name.endswith(".docx"):
self._file_name += ".docx"
self._minio_client = get_minio_storage_sync()
def _run(self, unique_id: str):
"""Master Execution Process"""
with tempfile.TemporaryDirectory() as temp_dir:
logger.debug("=== Download report template ===")
template_content = self._download_template()
docx_replacer = DocxReplacer(io.BytesIO(template_content))
template_variables = docx_replacer.extract_variables()
if not template_variables:
logger.debug("=== Report template not found variables ===")
finally_document = template_content
else:
logger.debug("=== Report template found variables ===")
# Get workflow variables
logger.debug("=== Get workflow variables ===")
workflow_variables = {}
text_classification = TextClassificationReport(temp_dir)
for one_variables in template_variables:
variable_value = self.get_other_node_variable(one_variables)
if type(variable_value) != str:
variable_value = str(variable_value)
variable_value = text_classification.get_all_classified_data(variable_value)
workflow_variables[one_variables] = variable_value
logger.debug("=== replace report template variables ===")
output_docx_path = f"{temp_dir}/{generate_uuid()}.docx"
docx_replacer.replace_and_save(workflow_variables, output_docx_path)
with open(output_docx_path, "rb") as f:
finally_document = f.read()
# SAVE AND SHARE
logger.info("=== Walking Tongs7: Save and share documents ===")
share_url = self._save_and_share_document(finally_document)
# Send Output Message
self._send_output_message(unique_id, share_url)
logger.info("=== Report Generation Complete ===")
def _download_template(self) -> bytes:
"""Download sample"""
if not self._minio_client.object_exists_sync(self._minio_client.bucket, self._object_name):
raise Exception(f"Template file does not exists!: {self._object_name}")
template_content = self._minio_client.get_object_sync(self._minio_client.bucket, self._object_name)
logger.info(f"Template downloaded successfully, size: {len(template_content)} byte")
return template_content
def _save_and_share_document(self, document_content: bytes) -> str:
"""Save the document and get a share link"""
# Generate unique file path
tmp_object_name = f"workflow/report/{uuid4().hex}/{self._file_name}"
# Uploaded toMinIO
self._minio_client.put_object_tmp_sync(tmp_object_name, document_content)
# Get share link
share_url = self._minio_client.get_share_link_sync(tmp_object_name, self._minio_client.tmp_bucket)
logger.info(f"Document saved successfully: {tmp_object_name}")
logger.info(f"Share Links: {share_url}")
return share_url
def _send_output_message(self, unique_id: str, share_url: str):
"""Send Output Message"""
self.callback_manager.on_output_msg(
OutputMsgData(
unique_id=unique_id,
node_id=self.id,
name=self.name,
msg="",
files=[{"path": share_url, "name": self._file_name}],
output_key="",
)
)
================================================
FILE: src/backend/bisheng/workflow/nodes/report/text_classification.py
================================================
# Convert the string to the format required by the docx replacement tool.
import os
import re
import tempfile
from dataclasses import dataclass
from enum import Enum
from typing import List, Dict, Tuple, Any
from urllib.parse import urlparse, unquote
from uuid import uuid4
import pandas as pd
import requests
from charset_normalizer import detect
from docx.enum.text import WD_ALIGN_PARAGRAPH
from loguru import logger
from openpyxl import load_workbook
from bisheng.core.storage.minio.minio_manager import get_minio_storage_sync
from bisheng.utils import md5_hash
class ResourceType(Enum):
"""Resource Type Enumeration"""
IMAGE = "image"
TABLE = "table"
TEXT = "text"
HEADING = "heading"
BOLD_TEXT = "bold_text"
@dataclass
class MatchPattern:
"""Match Pattern Definition"""
name: str
resource_type: ResourceType
pattern: str
flags: int
handler_method: str
# priority by order in the list
Patterns = [
MatchPattern(
name="markdown_table",
resource_type=ResourceType.TABLE,
pattern=r"(\s*\|[^\r\n]*\|[^\r\n]*(?:\r?\n\s*\|[^\r\n]*\|[^\r\n]*)+)",
flags=re.MULTILINE,
handler_method="_handle_markdown_table",
),
MatchPattern(
name="http_excel_csv",
resource_type=ResourceType.TABLE,
pattern=r"(https?://[^\s\u4e00-\u9fff]*\.(?:xlsx?|csv)(?:\?[^\s\u4e00-\u9fff]*)?)",
flags=re.IGNORECASE,
handler_method="_handle_http_excel_csv",
),
MatchPattern(
name="local_excel_csv",
resource_type=ResourceType.TABLE,
pattern=r"([^\s]*[/\\][^\s]*\.(?:xlsx?|csv))",
flags=re.IGNORECASE,
handler_method="_handle_http_excel_csv",
),
MatchPattern(
name="markdown_image",
resource_type=ResourceType.IMAGE,
pattern=r"!\[([^\]]*)\]\(([^)]+\.(?:png|jpg|jpeg|bmp|gif|webp)(?:\?[^)]*)?)\)",
flags=re.IGNORECASE,
handler_method="_handle_http_image",
),
MatchPattern(
name="http_image",
resource_type=ResourceType.IMAGE,
pattern=r"""
(?: # 非捕获组开始
(?: # URL协议
https?:// # http或https
| # 或
// # 协议相对URL
| # 或
file:// # 文件协议
)? # 协议可选
)? # 整个协议部分可选
(?: # 域名或路径
(?: # 域名部分
(?:www\.)? # 可选的www
[a-z0-9][a-z0-9-]* # 域名主体
(?:\.[a-z0-9]+)+ # 域名后缀
) # 或
| # 或
(?: # 本地路径
(?:[a-z]:)? # Windows驱动器(可选)
[\\/]? # 可选的路径分隔符
(?:[^<>()\[\]"'\s]+[\\/])* # 路径部分
)
)
[^<>()\[\]"\'\s]*? # 文件名(不含扩展名)
\. # 扩展名前的点
(?:png|jpg|jpeg|gif|bmp|webp|tif) # 图片扩展名
(?: # 可选的查询字符串
\? # 问号
[^<>()\[\]"'\s]* # 查询参数
)? # 查询字符串可选
(?: # 可选的片段标识符
\# # 井号
[^<>()\[\]"'\s]* # 片段
)? # 片段可选
""",
flags=re.IGNORECASE | re.VERBOSE,
handler_method="_handle_http_image",
),
MatchPattern(
name="heading",
resource_type=ResourceType.HEADING,
pattern=r'^(#{1,6})\s+(.+?)(?:\s+#+)?\s*$',
flags=re.IGNORECASE | re.MULTILINE,
handler_method="_handle_heading",
),
MatchPattern(
name="bold_text",
resource_type=ResourceType.BOLD_TEXT,
# 严格正则表达式
# (? List[Dict]:
string_data = [{"type": "text", "content": string_data}]
for pattern in Patterns:
handle_method = pattern.handler_method
handler = getattr(self, handle_method, None)
if not handler:
logger.warning(f"Handler method {handle_method} not found.")
continue
string_data = handler(pattern, string_data)
return string_data
def _download_file(self, file_url: str) -> str:
# Set request headers to simulate browser access
file_url_md5 = md5_hash(file_url)
if self.file_cache.get(file_url_md5):
return self.file_cache.get(file_url_md5)
url_obj = urlparse(file_url)
filename = unquote(url_obj.path.split('/')[-1])
if file_url.startswith(("https", "http")):
headers = {
"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 "
"(KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36"
}
response = requests.get(file_url, headers=headers, timeout=30, verify=False)
response.raise_for_status()
# Get filename
# accordingContent-TypeInferred extension
content_type = response.headers.get("Content-Type", "").lower()
if "image/png" in content_type:
filename = f"{uuid4().hex}.png"
elif "image/jpeg" in content_type or "image/jpg" in content_type:
filename = f"{uuid4().hex}.jpg"
elif "image/bmp" in content_type:
filename = f"{uuid4().hex}.bmp"
else:
content_disposition = response.headers.get('Content-Disposition')
if content_disposition:
filename = unquote(content_disposition).split('filename=')[-1].strip("\"'")
file_ext = os.path.splitext(filename)[1].lower()
filename = f"{uuid4().hex}{file_ext}"
file_content = response.content
else:
# maybe minio file path
minio_client = get_minio_storage_sync()
if not file_url.lstrip("/").startswith((minio_client.bucket, minio_client.tmp_bucket)):
raise ValueError(f"Invalid minio path {file_url}")
path_parts = url_obj.path.lstrip("/").split('/', 1)
if len(path_parts) != 2:
raise ValueError(f"Invalid minio path {file_url}")
bucket_name, object_name = path_parts
# Call Synchronized minio Method download
object_name = unquote(object_name)
file_content = minio_client.get_object_sync(bucket_name, object_name)
# Creating temp file
temp_file = os.path.join(self.static_resource_path, filename)
with open(temp_file, "wb") as f:
f.write(file_content)
self.file_cache[file_url_md5] = temp_file
return temp_file
def _handle_bold_text(self, pattern: MatchPattern, string_data: List[Dict]):
return self._base_handle_string_data(pattern, string_data, self._parse_bold_text_data)
def _parse_bold_text_data(self, bold_text_str: str) -> Dict:
# Remove the ** or __ markers
content = re.sub(r'(? Dict:
heading_level, heading_text = heading_str.split(" ", 1)
level = min(len(heading_level.strip()), 6)
return {"type": "heading", "content": heading_text, "level": level}
def _handle_http_image(self, pattern: MatchPattern, string_data: List[Dict]):
# Placeholder for actual implementation
return self._base_handle_string_data(pattern, string_data, self._parse_http_image_data)
def _parse_http_image_data(self, image_str: str) -> Dict:
# image_str:  or http://example.com/image.png or /local/path/image.png
try:
if image_str.startswith("!["):
new_image_str = image_str.split("](", 1)[1].rstrip(")")
local_file_path = self._download_file(new_image_str)
else:
local_file_path = self._download_file(image_str)
return {
"type": "image",
"content": local_file_path
}
except Exception as e:
logger.error(f"Failed to parse http image data: {str(e)}")
return {"type": "text", "content": image_str}
def _handle_http_excel_csv(self, pattern: MatchPattern, string_data: List[Dict]):
# Placeholder for actual implementation
return self._base_handle_string_data(pattern, string_data, self._parse_http_table_data)
def _parse_http_table_data(self, table_url: str) -> Dict:
# Placeholder for actual implementation
try:
if table_url.startswith(("http", "https")):
local_file_path = self._download_file(table_url)
else:
local_file_path = table_url
if local_file_path.endswith(".xlsx"):
table_data, table_alignment = self._parse_excel_data(local_file_path)
elif local_file_path.endswith(".csv"):
table_data, table_alignment = self._parse_csv_data(local_file_path)
else:
raise ValueError(f"Unsupported file format for table parsing: {local_file_path}")
return self._handle_table_data_to_dict(table_data, table_alignment)
except Exception as e:
logger.error(f"Failed to parse http table data: {str(e)}")
return {"type": "text", "content": table_url}
@staticmethod
def _convert_docx_alignment(alignment: str) -> Any:
if alignment == "left":
return WD_ALIGN_PARAGRAPH.LEFT
elif alignment == "center":
return WD_ALIGN_PARAGRAPH.CENTER
elif alignment == "right":
return WD_ALIGN_PARAGRAPH.RIGHT
else:
return WD_ALIGN_PARAGRAPH.LEFT
def _handle_table_data_to_dict(self, table_data: List[List[str]], table_alignment: List[str]) -> Dict:
# Convert table data and alignment into a dictionary format
content = []
for row in table_data:
one_row_data = []
for index, cell in enumerate(row):
# convert each cell to text type with image alignment
markdown_image_pattern = Patterns[3] # markdown_image pattern
http_image_pattern = Patterns[4] # http_image pattern
bold_text_pattern = Patterns[6] # bold_text pattern
cell_data = getattr(self, markdown_image_pattern.handler_method)(markdown_image_pattern,
[{"type": "text", "content": cell}])
cell_data = getattr(self, http_image_pattern.handler_method)(http_image_pattern, cell_data)
cell_data = getattr(self, bold_text_pattern.handler_method)(bold_text_pattern, cell_data)
cell_data[0]["alignment"] = self._convert_docx_alignment(
table_alignment[index] if index < len(table_alignment) else "left")
one_row_data.append(cell_data)
content.append(one_row_data)
return {"type": "table", "content": content}
@staticmethod
def _parse_excel_data(file_path: str) -> Tuple[List[List[str]], List[str]]:
# Use openpyxl read docx
workbook = load_workbook(file_path, data_only=True) # data_only=TrueGet Calculated Value
# Using the first worksheet
worksheet = workbook.active
table_data = []
max_col = 0
# Read all rows
for row in worksheet.iter_rows(values_only=True):
# Skip completely blank rows
if all(cell is None or str(cell).strip() == "" for cell in row):
continue
row_data = [str(cell) if cell is not None else "" for cell in row]
table_data.append(row_data)
max_col = max(max_col, len(row_data))
# Make sure all rows have the same number of columns
for row in table_data:
while len(row) < max_col:
row.append("")
# Generate alignment information (default left alignment)
alignments = ["left"] * max_col
logger.debug(f"ExcelFile parsed successfully: {len(table_data)}Parade x {max_col}column")
return table_data, alignments
@staticmethod
def _parse_csv_data(file_path: str) -> Tuple[List[List[str]], List[str]]:
# Auto-Detect Encoding
with open(file_path, 'rb') as f:
raw_data = f.read()
encoding_info = detect(raw_data)
encoding = encoding_info['encoding'] or 'utf-8'
# Use pandas read outCSVto better handle various formats
df = pd.read_csv(file_path, encoding=encoding)
# Convert to Tabular Data Format
table_data = []
# Add header
headers = [str(col) for col in df.columns]
table_data.append(headers)
# Add Data Row
for _, row in df.iterrows():
row_data = [str(cell) if pd.notna(cell) else "" for cell in row]
table_data.append(row_data)
# Generate alignment information (default left alignment)
alignments = ["left"] * len(headers)
logger.debug(f"CSVFile parsed successfully: {len(table_data)}Parade x {len(headers)}column")
return table_data, alignments
def _handle_markdown_table(self, pattern: MatchPattern, string_data: List[Dict]):
return self._base_handle_string_data(pattern, string_data, self._parse_table_data)
@staticmethod
def _base_handle_string_data(pattern: MatchPattern, string_data: List[Dict], parse_method) -> List[Dict]:
new_string_data = []
for one in string_data:
if one["type"] != "text" or not one["content"]:
new_string_data.append(one)
continue
content = one["content"]
match_flag = False
for one_match in re.finditer(pattern.pattern, one["content"], flags=pattern.flags):
match_flag = True
raw_text = one_match.group(0)
before, end = content.split(raw_text, 1)
new_string_data.append({
"type": "text",
"content": before
})
new_string_data.append(parse_method(raw_text))
content = end
if not match_flag:
new_string_data.append(one)
continue
if content:
new_string_data.append({
"type": "text",
"content": content
})
return new_string_data
def _parse_table_data(self, table_content: str) -> Dict:
"""Parse the table data while working with the in the table string"""
try:
# Analyze the table structure first
table_data, table_alignment = self._parse_markdown_table_from_content(table_content)
return self._handle_table_data_to_dict(table_data, table_alignment)
except Exception as e:
logger.error(f"Failed to parse table data: {str(e)}")
return {"type": "text", "content": table_content}
def _parse_markdown_table_from_content(self, table_content: str) -> Tuple[List[List[str]], List[str]]:
"""
Parsing from file contentsMarkdownTable Filter
Args:
table_content: table content in File contents
Returns:
tuple: (Table Data, Alignment Info List)
"""
# Keep all rows, including empty rows - Fully preserving the original table structure
lines = [line.strip() for line in table_content.strip().split("\n")]
table_rows = []
table_alignments = []
separator_found = False
for line in lines:
# Skip completely empty rows, but leave empty table rows with only vertical lines
if not line:
continue
# Check if it is a delimiter line
if self._is_table_separator_line(line):
table_alignments = self._parse_table_alignments(line)
separator_found = True
continue
# Parse Data Rows - Keep all table rows, including empty rows
cells = self._parse_table_row(line)
# Remove if cells Condition, leave empty table rows
table_rows.append(cells)
# Use default alignment if no separator found
if not separator_found and table_rows:
table_alignments = ["left"] * len(table_rows[0])
# Make sure all rows have the same number of columns
if table_rows:
max_cols = max(len(row) for row in table_rows)
for row in table_rows:
while len(row) < max_cols:
row.append("")
# Make sure the number of aligned messages matches the number of columns
while len(table_alignments) < max_cols:
table_alignments.append("left")
table_alignments = table_alignments[:max_cols]
logger.debug(
f"Successfully parsed table in content, size: {len(table_rows)}Parade x {len(table_alignments)}column")
return table_rows, table_alignments
@staticmethod
def _is_table_separator_line(line: str) -> bool:
"""Check if yesMarkdownTable Separator Row"""
content = line.strip().strip("|").strip()
if not content:
return False
cells = [cell.strip() for cell in content.split("|")]
# At least one cell must contain a delimiter character (-OR:）
has_separator_chars = False
for cell in cells:
if not cell:
continue
# Check for delimiter characters
if '-' in cell or ':' in cell:
has_separator_chars = True
# Check for anything else after removing the separator character
clean_cell = cell.replace("-", "").replace(":", "").strip()
if clean_cell:
return False
# Only lines containing delimiter characters can be considered delimiter lines
return has_separator_chars
@staticmethod
def _parse_table_alignments(separator_line: str) -> list:
"""Parse column alignment from delimiter rows"""
alignments = []
content = separator_line.strip().strip("|").strip()
cells = [cell.strip() for cell in content.split("|")]
for cell in cells:
if not cell:
alignments.append("left")
continue
if cell.startswith(":") and cell.endswith(":"):
alignments.append("center")
elif cell.endswith(":"):
alignments.append("right")
else:
alignments.append("left")
return alignments
@staticmethod
def _parse_table_row(line: str) -> list:
"""Parse Table Rows"""
content = line.strip()
if content.startswith("|"):
content = content[1:]
if content.endswith("|"):
content = content[:-1]
cells = []
current_cell = ""
escaped = False
for char in content:
if escaped:
current_cell += char
escaped = False
elif char == "\\":
escaped = True
current_cell += char
elif char == "|":
cells.append(current_cell.strip())
current_cell = ""
else:
current_cell += char
# Always add the last cell to ensure empty table rows are also parsed correctly
cells.append(current_cell.strip())
return cells
================================================
FILE: src/backend/bisheng/workflow/nodes/start/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/start/start.py
================================================
import datetime
import json
from typing import Any, Dict
from langchain.memory import ConversationBufferWindowMemory
from bisheng.chat.types import IgnoreException
from bisheng.database.models.group import GroupDao
from bisheng.database.models.role import RoleDao
from bisheng.database.models.user_group import UserGroupDao
from bisheng.user.domain.models.user import UserDao
from bisheng.user.domain.models.user_role import UserRoleDao
from bisheng.workflow.callback.event import GuideQuestionData, GuideWordData
from bisheng.workflow.nodes.base import BaseNode
class StartNode(BaseNode):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
# Initialize Current Time
self.node_params['current_time'] = datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')
# Initialize chat history
self.graph_state.history_memory = ConversationBufferWindowMemory(
k=self.node_params.get('chat_history', 10))
self._user_info = None
def _run(self, unique_id: str) -> Dict[str, Any]:
if self._user_info is None and self.user_id:
user_db = UserDao.get_user(int(self.user_id))
user_group = UserGroupDao.get_user_group(int(self.user_id))
groups = GroupDao.get_group_by_ids([group.group_id for group in user_group])
group_dict = {group.id: group.group_name for group in groups}
user_role = UserRoleDao.get_user_roles(int(self.user_id))
roles = RoleDao.get_role_by_ids([role.role_id for role in user_role])
user_roles = []
for one in roles:
if one.group_id not in group_dict.keys():
continue
user_roles.append(f"{group_dict[one.group_id]}-{one.role_name}")
if user_db:
self._user_info = json.dumps({
"user_name": user_db.user_name,
"user_group": list(group_dict.values()),
"user_role": user_roles,
}, ensure_ascii=False, indent=2)
if self.node_params['guide_word']:
self.callback_manager.on_guide_word(
data=GuideWordData(node_id=self.id, name=self.name, unique_id=unique_id,
guide_word=self.node_params['guide_word']))
if self.node_params['guide_question']:
self.callback_manager.on_guide_question(
data=GuideQuestionData(node_id=self.id, name=self.name, unique_id=unique_id,
guide_question=self.node_params['guide_question']))
if not self.node_data.v:
raise IgnoreException(f'{self.name} -- workflow node is update')
# convert preset_question to dict
new_preset_question = {}
for one in self.node_params['preset_question']:
new_preset_question[one['key']] = one['value']
# convert custom vars to dict
custom_vars = {}
for one in self.node_params.get('custom_variables', []):
custom_vars[one['key']] = one['value']
return {
'current_time': self.node_params['current_time'],
'chat_history': '',
'preset_question': new_preset_question,
'user_info': self._user_info,
'custom_variables': custom_vars
}
def parse_log(self, unique_id: str, result: dict) -> Any:
return [[
{
"key": "current_time",
"value": result['current_time'],
"type": "params"
},
{
"key": "preset_question",
"value": result['preset_question'],
"type": "params"
},
{
"key": "user_info",
"value": result['user_info'],
"type": "params"
}
]]
================================================
FILE: src/backend/bisheng/workflow/nodes/tool/__init__.py
================================================
================================================
FILE: src/backend/bisheng/workflow/nodes/tool/tool.py
================================================
from typing import Any
from bisheng.common.constants.enums.telemetry import ApplicationTypeEnum
from bisheng.tool.domain.models.gpts_tools import GptsToolsDao
from bisheng.tool.domain.services.executor import ToolExecutor
from bisheng.workflow.nodes.base import BaseNode
from bisheng.workflow.nodes.prompt_template import PromptTemplateParser
class ToolNode(BaseNode):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self._tool_key = self.node_data.tool_key
self._tool_info = GptsToolsDao.get_tool_by_tool_key(tool_key=self._tool_key)
if not self._tool_info:
raise Exception(f"Tools{self._tool_key}Does not exist")
self._tool = None
def _init_tool(self):
if self._tool:
return
self._tool = ToolExecutor.init_by_tool_id_sync(tool_id=self._tool_info.id, app_id=self.workflow_id,
app_name=self.workflow_name,
app_type=ApplicationTypeEnum.WORKFLOW, user_id=self.user_id)
def _run(self, unique_id: str):
self._init_tool()
tool_input = self.parse_tool_input()
output = self._tool.invoke(input=tool_input)
return {
"output": output
}
def parse_log(self, unique_id: str, result: dict) -> Any:
tool_input = self.parse_tool_input()
ret = [
{
"key": k,
"value": v,
"type": "params"
} for k, v in tool_input.items()
]
ret.append({
"key": "output",
"value": result.get("output", ''),
"type": "params"
})
return [ret]
def parse_tool_input(self) -> dict:
ret = {}
for key, val in self.node_params.items():
if key == "output":
continue
new_val = self.parse_template_msg(val)
if new_val == '' or new_val is None:
continue
ret[key] = new_val
return ret
def parse_template_msg(self, msg: str):
msg_template = PromptTemplateParser(template=msg)
variables = msg_template.extract()
if len(variables) > 0:
var_map = {}
for one in variables:
var_map[one] = self.get_other_node_variable(one)
msg = msg_template.format(var_map)
return msg
================================================
FILE: src/backend/bisheng_langchain/__init__.py
================================================
================================================
FILE: src/backend/bisheng_langchain/agents/__init__.py
================================================
from bisheng_langchain.agents.chatglm_functions_agent.base import ChatglmFunctionsAgent
from bisheng_langchain.agents.llm_functions_agent.base import LLMFunctionsAgent
__all__ = ['ChatglmFunctionsAgent', 'LLMFunctionsAgent']
================================================
FILE: src/backend/bisheng_langchain/agents/chatglm_functions_agent/__init__.py
================================================
================================================
FILE: src/backend/bisheng_langchain/agents/chatglm_functions_agent/base.py
================================================
import json
import re
from typing import Any, Dict, List, Optional, Sequence, Tuple, Union
from pydantic import model_validator, Field
from bisheng_langchain.chat_models.host_llm import HostChatGLM
from langchain.agents.agent import Agent, AgentOutputParser, BaseSingleActionAgent
from langchain.agents.structured_chat.output_parser import StructuredChatOutputParserWithRetries
from langchain.agents.structured_chat.prompt import FORMAT_INSTRUCTIONS, PREFIX, SUFFIX
from langchain.callbacks.base import BaseCallbackManager
from langchain.callbacks.manager import Callbacks
from langchain.prompts.chat import (ChatPromptTemplate, HumanMessagePromptTemplate,
SystemMessagePromptTemplate)
from langchain.schema import AgentAction, AgentFinish, BasePromptTemplate
from langchain.schema.language_model import BaseLanguageModel
from langchain.schema.messages import ChatMessage
from langchain.tools import BaseTool, StructuredTool
HUMAN_MESSAGE_TEMPLATE = '{input}\n\n{agent_scratchpad}'
def format_tool_to_chatglm_function(tool: BaseTool):
"""Format tool into the chatglm function API."""
if isinstance(tool, StructuredTool):
schema_ = tool.args_schema.schema()
# Bug with required missing for structured tools.
required = sorted(schema_['properties']) # BUG WORKAROUND
return {
'name': tool.name,
'description': tool.description,
'parameters': {
'type': 'object',
'properties': schema_['properties'],
'required': required,
},
}
else:
if tool.args_schema:
schema_ = tool.args_schema.schema()
parameters = {
'type': 'object',
'properties': schema_['properties'],
'required': list(schema_['properties'].keys()),
}
else:
parameters = {
# This is a hack to get around the fact that some tools
# do not expose an args_schema, and expect an argument
# which is a string.
# And Open AI does not support an array type for the
# parameters.
'type': 'object',
'properties': {
'__arg1': {
'type': 'string',
'title': '__arg1'
},
},
'required': ['__arg1'],
}
return {
'name': tool.name,
'description': tool.description,
'parameters': parameters,
}
class ChatglmFunctionsAgent(BaseSingleActionAgent):
"""Chatglm Functions Agent."""
llm: BaseLanguageModel
tools: Sequence[BaseTool]
prompt: BasePromptTemplate
output_parser: AgentOutputParser = Field(default_factory=StructuredChatOutputParserWithRetries)
"""Output parser for the agent."""
has_search: bool = False
history: List = []
def get_allowed_tools(self) -> List[str]:
"""Get allowed tools."""
return list([t.name for t in self.tools])
@model_validator(mode='before')
@classmethod
def validate_llm(cls, values: dict) -> dict:
if not isinstance(values['llm'], HostChatGLM):
raise ValueError('Only supported with ChatGLM3 models.')
return values
@model_validator(mode='before')
@classmethod
def validate_prompt(cls, values: dict) -> dict:
prompt: BasePromptTemplate = values['prompt']
if 'agent_scratchpad' not in prompt.input_variables:
raise ValueError('`agent_scratchpad` should be one of the variables in the prompt, '
f'got {prompt.input_variables}')
return values
@property
def input_keys(self) -> List[str]:
"""Get input keys. Input refers to user input here."""
return ['input']
@property
def observation_prefix(self) -> str:
"""Prefix to append the observation with."""
return 'Observation: '
@property
def llm_prefix(self) -> str:
"""Prefix to append the llm call with."""
return 'Thought:'
def _construct_scratchpad(self, intermediate_steps: List[Tuple[AgentAction, str]]) -> str:
agent_scratchpad = ''
for action, observation in intermediate_steps:
agent_scratchpad += action.log
agent_scratchpad += f'\n{self.observation_prefix}{observation}\n{self.llm_prefix}'
if not isinstance(agent_scratchpad, str):
raise ValueError('agent_scratchpad should be of type string.')
if agent_scratchpad:
return (f'This was your previous work '
f"(but I haven't seen any of it! I only see what "
f'you return as final answer):\n{agent_scratchpad}')
else:
return agent_scratchpad
@classmethod
def _validate_tools(cls, tools: Sequence[BaseTool]) -> None:
pass
@classmethod
def _get_default_output_parser(cls,
llm: Optional[BaseLanguageModel] = None,
**kwargs: Any) -> AgentOutputParser:
return StructuredChatOutputParserWithRetries.from_llm(llm=llm)
@property
def _stop(self) -> List[str]:
return ['Observation:']
def _tool_history(self, prompt: str):
ans = []
tools_json = []
for i, tool in enumerate(self.tools):
tool_config = format_tool_to_chatglm_function(tool)
tools_json.append(tool_config)
ans.append({
'role': 'system',
'content':
'Answer the following questions as best as you can. You have access to the following tools:',
'tools': tools_json
})
query = f"""{prompt.split("Human: ")[-1].strip()}"""
return ans, query
def _extract_observation(self, prompt: str):
return_json = prompt.split('Observation: ')[-1].split('\nThought:')[0]
self.history.append({'role': 'observation', 'content': return_json})
return
def _extract_tool(self):
tool_names = list([t.name for t in self.tools])
if self.history[-1]['metadata'] and len(self.history[-1]['metadata']) > 0:
metadata = self.history[-1]['metadata']
content = self.history[-1]['content']
if 'tool_call' in content:
for tool in tool_names:
if tool in metadata:
input_para = content.split("='")[-1].split("'")[0]
action_json = {'action': tool, 'action_input': input_para}
self.has_search = True
return f"""
Action:
```
{json.dumps(action_json, ensure_ascii=False)}
```"""
final_answer_json = {'action': 'Final Answer', 'action_input': self.history[-1]['content']}
self.has_search = False
return f"""
Action:
```
{json.dumps(final_answer_json, ensure_ascii=False)}
```"""
def get_full_inputs(self, intermediate_steps: List[Tuple[AgentAction, str]],
**kwargs: Any) -> Dict[str, Any]:
"""Create the full inputs for the LLMChain from intermediate steps."""
thoughts = self._construct_scratchpad(intermediate_steps)
new_inputs = {'agent_scratchpad': thoughts, 'stop': self._stop}
full_inputs = {**kwargs, **new_inputs}
return full_inputs
def plan(
self,
intermediate_steps: List[Tuple[AgentAction, str]],
callbacks: Callbacks = None,
**kwargs: Any,
) -> Union[AgentAction, AgentFinish]:
"""Given input, decided what to do.
Args:
intermediate_steps: Steps the LLM has taken to date,
along with observations
callbacks: Callbacks to run.
**kwargs: User inputs.
Returns:
Action specifying what tool to use.
"""
full_inputs = self.get_full_inputs(intermediate_steps, **kwargs)
prompt = self.prompt.format_prompt(**full_inputs)
prompt = prompt.to_string()
if not self.has_search:
self.history, query = self._tool_history(prompt)
else:
self._extract_observation(prompt)
query = ''
self.history.append({'role': 'user', 'content': query})
chat_messages = []
for message in self.history:
additional_kwargs = dict()
for key in message.keys():
if key not in ['role', 'content']:
additional_kwargs[key] = message[key]
chat_messages.append(
ChatMessage(role=message['role'],
content=message['content'],
additional_kwargs=additional_kwargs))
predicted_message = self.llm.predict_messages(chat_messages, callbacks=callbacks)
self.history.append({
'role': 'assistant',
'content': predicted_message.content,
'metadata': predicted_message.additional_kwargs['metadata']
})
full_output = self._extract_tool()
return self.output_parser.parse(full_output)
async def aplan(
self,
intermediate_steps: List[Tuple[AgentAction, str]],
callbacks: Callbacks = None,
**kwargs: Any,
) -> Union[AgentAction, AgentFinish]:
"""Given input, decided what to do.
Args:
intermediate_steps: Steps the LLM has taken to date,
along with observations
callbacks: Callbacks to run.
**kwargs: User inputs.
Returns:
Action specifying what tool to use.
"""
full_inputs = self.get_full_inputs(intermediate_steps, **kwargs)
prompt = self.prompt.format_prompt(**full_inputs)
prompt = prompt.to_string()
if not self.has_search:
self.history, query = self._tool_history(prompt)
else:
self._extract_observation(prompt)
query = ''
self.history.append({'role': 'user', 'content': query})
chat_messages = []
for message in self.history:
additional_kwargs = dict()
for key in message.keys():
if key not in ['role', 'content']:
additional_kwargs[key] = message[key]
chat_messages.append(
ChatMessage(role=message['role'],
content=message['content'],
additional_kwargs=additional_kwargs))
predicted_message = await self.llm.apredict_messages(chat_messages, callbacks=callbacks)
self.history.append({
'role': 'assistant',
'content': predicted_message.content,
'metadata': predicted_message.additional_kwargs['metadata']
})
full_output = self._extract_tool()
return self.output_parser.parse(full_output)
@classmethod
def create_prompt(
cls,
tools: Sequence[BaseTool],
prefix: str = PREFIX,
suffix: str = SUFFIX,
human_message_template: str = HUMAN_MESSAGE_TEMPLATE,
format_instructions: str = FORMAT_INSTRUCTIONS,
input_variables: Optional[List[str]] = None,
memory_prompts: Optional[List[BasePromptTemplate]] = None,
) -> BasePromptTemplate:
tool_strings = []
for tool in tools:
args_schema = re.sub('}', '}}}}', re.sub('{', '{{{{', str(tool.args)))
tool_strings.append(f'{tool.name}: {tool.description}, args: {args_schema}')
formatted_tools = '\n'.join(tool_strings)
tool_names = ', '.join([tool.name for tool in tools])
format_instructions = format_instructions.format(tool_names=tool_names)
template = '\n\n'.join([prefix, formatted_tools, format_instructions, suffix])
if input_variables is None:
input_variables = ['input', 'agent_scratchpad']
_memory_prompts = memory_prompts or []
messages = [
SystemMessagePromptTemplate.from_template(template),
*_memory_prompts,
HumanMessagePromptTemplate.from_template(human_message_template),
]
return ChatPromptTemplate(input_variables=input_variables, messages=messages)
@classmethod
def from_llm_and_tools(
cls,
llm: BaseLanguageModel,
tools: Sequence[BaseTool],
callback_manager: Optional[BaseCallbackManager] = None,
output_parser: Optional[AgentOutputParser] = None,
prefix: str = PREFIX,
suffix: str = SUFFIX,
human_message_template: str = HUMAN_MESSAGE_TEMPLATE,
format_instructions: str = FORMAT_INSTRUCTIONS,
input_variables: Optional[List[str]] = None,
memory_prompts: Optional[List[BasePromptTemplate]] = None,
**kwargs: Any,
) -> Agent:
"""Construct an agent from an LLM and tools."""
cls._validate_tools(tools)
prompt = cls.create_prompt(
tools,
prefix=prefix,
suffix=suffix,
human_message_template=human_message_template,
format_instructions=format_instructions,
input_variables=input_variables,
memory_prompts=memory_prompts,
)
_output_parser = output_parser or cls._get_default_output_parser(llm=llm)
return cls(
llm=llm,
prompt=prompt,
tools=tools,
output_parser=_output_parser,
callback_manager=callback_manager,
**kwargs,
)
@property
def _agent_type(self) -> str:
raise ValueError
================================================
FILE: src/backend/bisheng_langchain/agents/chatglm_functions_agent/output_parser.py
================================================
from __future__ import annotations
import json
import logging
import re
from typing import Optional, Union
from pydantic import Field
from langchain.agents.agent import AgentOutputParser
from langchain.agents.structured_chat.prompt import FORMAT_INSTRUCTIONS
from langchain.output_parsers import OutputFixingParser
from langchain.schema import AgentAction, AgentFinish, OutputParserException
from langchain.schema.language_model import BaseLanguageModel
logger = logging.getLogger(__name__)
class StructuredChatOutputParser(AgentOutputParser):
"""Output parser for the structured chat agent."""
pattern = re.compile(r"```(?:json)?\n(.*?)```", re.DOTALL)
def get_format_instructions(self) -> str:
return FORMAT_INSTRUCTIONS
def parse(self, text: str) -> Union[AgentAction, AgentFinish]:
try:
action_match = self.pattern.search(text)
if action_match is not None:
response = json.loads(action_match.group(1).strip(), strict=False)
if isinstance(response, list):
# gpt turbo frequently ignores the directive to emit a single action
logger.warning("Got multiple action responses: %s", response)
response = response[0]
if response["action"] == "Final Answer":
return AgentFinish({"output": response["action_input"]}, text)
else:
return AgentAction(
response["action"], response.get("action_input", {}), text
)
else:
return AgentFinish({"output": text}, text)
except Exception as e:
raise OutputParserException(f"Could not parse LLM output: {text}") from e
@property
def _type(self) -> str:
return "structured_chat"
class StructuredChatOutputParserWithRetries(AgentOutputParser):
"""Output parser with retries for the structured chat agent."""
base_parser: AgentOutputParser = Field(default_factory=StructuredChatOutputParser)
"""The base parser to use."""
output_fixing_parser: Optional[OutputFixingParser] = None
"""The output fixing parser to use."""
def get_format_instructions(self) -> str:
return FORMAT_INSTRUCTIONS
def parse(self, text: str) -> Union[AgentAction, AgentFinish]:
try:
if self.output_fixing_parser is not None:
parsed_obj: Union[
AgentAction, AgentFinish
] = self.output_fixing_parser.parse(text)
else:
parsed_obj = self.base_parser.parse(text)
return parsed_obj
except Exception as e:
raise OutputParserException(f"Could not parse LLM output: {text}") from e
@classmethod
def from_llm(
cls,
llm: Optional[BaseLanguageModel] = None,
base_parser: Optional[StructuredChatOutputParser] = None,
) -> StructuredChatOutputParserWithRetries:
if llm is not None:
base_parser = base_parser or StructuredChatOutputParser()
output_fixing_parser = OutputFixingParser.from_llm(
llm=llm, parser=base_parser
)
return cls(output_fixing_parser=output_fixing_parser)
elif base_parser is not None:
return cls(base_parser=base_parser)
else:
return cls()
@property
def _type(self) -> str:
return "structured_chat_with_retries"
================================================
FILE: src/backend/bisheng_langchain/agents/chatglm_functions_agent/prompt.py
================================================
# flake8: noqa
PREFIX = """Respond to the human as helpfully and accurately as possible. You have access to the following tools:"""
FORMAT_INSTRUCTIONS = """Use a json blob to specify a tool by providing an action key (tool name) and an action_input key (tool input).
Valid "action" values: "Final Answer" or {tool_names}
Provide only ONE action per $JSON_BLOB, as shown:
```
{{{{
"action": $TOOL_NAME,
"action_input": $INPUT
}}}}
```
Follow this format:
Question: input question to answer
Thought: consider previous and subsequent steps
Action:
```
$JSON_BLOB
```
Observation: action result
... (repeat Thought/Action/Observation N times)
Thought: I know what to respond
Action:
```
{{{{
"action": "Final Answer",
"action_input": "Final response to human"
}}}}
```"""
SUFFIX = """Begin! Reminder to ALWAYS respond with a valid json blob of a single action. Use tools if necessary. Respond directly if appropriate. Format is Action:```$JSON_BLOB```then Observation:.
Thought:"""
================================================
FILE: src/backend/bisheng_langchain/agents/llm_functions_agent/__init__.py
================================================
================================================
FILE: src/backend/bisheng_langchain/agents/llm_functions_agent/base.py
================================================
"""Module implements an agent that uses OpenAI's APIs function enabled API."""
import json
from json import JSONDecodeError
from typing import Any, List, Optional, Sequence, Tuple, Union
from pydantic import model_validator
from bisheng_langchain.chat_models.host_llm import HostQwenChat
from bisheng_langchain.chat_models.proxy_llm import ProxyChatLLM
from langchain.agents import BaseSingleActionAgent
from langchain.callbacks.base import BaseCallbackManager
from langchain.callbacks.manager import Callbacks
from langchain.prompts.chat import (BaseMessagePromptTemplate, ChatPromptTemplate,
HumanMessagePromptTemplate, MessagesPlaceholder)
from langchain.schema import AgentAction, AgentFinish, BasePromptTemplate, OutputParserException
from langchain.schema.language_model import BaseLanguageModel
from langchain.schema.messages import AIMessage, BaseMessage, FunctionMessage, SystemMessage
from langchain.tools import BaseTool
from langchain.tools.convert_to_openai import format_tool_to_openai_function
from langchain_core.agents import AgentActionMessageLog
from langchain_openai import ChatOpenAI
def _convert_agent_action_to_messages(agent_action: AgentAction,
observation: str) -> List[BaseMessage]:
"""Convert an agent action to a message.
This code is used to reconstruct the original AI message from the agent action.
Args:
agent_action: Agent action to convert.
Returns:
AIMessage that corresponds to the original tool invocation.
"""
if isinstance(agent_action, AgentActionMessageLog):
return list(
agent_action.message_log) + [_create_function_message(agent_action, observation)]
else:
return [AIMessage(content=agent_action.log)]
def _create_function_message(agent_action: AgentAction, observation: str) -> FunctionMessage:
"""Convert agent action and observation into a function message.
Args:
agent_action: the tool invocation request from the agent
observation: the result of the tool invocation
Returns:
FunctionMessage that corresponds to the original tool invocation
"""
if not isinstance(observation, str):
try:
content = json.dumps(observation, ensure_ascii=False)
except Exception:
content = str(observation)
else:
content = observation
return FunctionMessage(
name=agent_action.tool,
content=content,
)
def _format_intermediate_steps(
intermediate_steps: List[Tuple[AgentAction, str]], ) -> List[BaseMessage]: # noqa
"""Format intermediate steps.
Args:
intermediate_steps: Steps the LLM has taken to date, along with observations
Returns:
list of messages to send to the LLM for the next prediction
"""
messages = []
for intermediate_step in intermediate_steps:
agent_action, observation = intermediate_step
messages.extend(_convert_agent_action_to_messages(agent_action, observation))
return messages
def _parse_ai_message(message: BaseMessage) -> Union[AgentAction, AgentFinish]:
"""Parse an AI message."""
if not isinstance(message, AIMessage):
raise TypeError(f'Expected an AI message got {type(message)}')
function_call = message.additional_kwargs.get('tool_calls', {})
if function_call:
function_name = function_call['name']
try:
_tool_input = json.loads(function_call['arguments'])
except JSONDecodeError:
raise OutputParserException(f'Could not parse tool input: {function_call} because '
f'the `arguments` is not valid JSON.')
# HACK HACK HACK:
# The code that encodes tool input into Open AI uses a special variable
# name called `__arg1` to handle old style tools that do not expose a
# schema and expect a single string argument as an input.
# We unpack the argument here if it exists.
# Open AI does not support passing in a JSON array as an argument.
if '__arg1' in _tool_input:
tool_input = _tool_input['__arg1']
else:
tool_input = _tool_input
content_msg = 'responded: {content}\n' if message.content else '\n'
log = f'\nInvoking: `{function_name}` with `{tool_input}`\n{content_msg}\n'
return AgentActionMessageLog(
tool=function_name,
tool_input=tool_input,
log=log,
message_log=[message],
)
return AgentFinish(return_values={'output': message.content}, log=message.content)
class LLMFunctionsAgent(BaseSingleActionAgent):
"""An Agent driven by function powered API.
Args:
llm: This should be an instance of ChatOpenAI, specifically a model
that supports using `functions`.
tools: The tools this agent has access to.
prompt: The prompt for this agent, should support agent_scratchpad as one
of the variables. For an easy way to construct this prompt, use
`OpenAIFunctionsAgent.create_prompt(...)`
"""
llm: BaseLanguageModel
tools: Sequence[BaseTool]
prompt: BasePromptTemplate
def get_allowed_tools(self) -> List[str]:
"""Get allowed tools."""
return list([t.name for t in self.tools])
@model_validator(mode='before')
@classmethod
def validate_llm(cls, values: dict) -> dict:
if ((not isinstance(values['llm'], ChatOpenAI))
and (not isinstance(values['llm'], HostQwenChat))
and (not isinstance(values['llm'], ProxyChatLLM))):
raise ValueError(
'Only supported with ChatOpenAI and HostQwenChat and ProxyChatLLM models.')
return values
@model_validator(mode='before')
@classmethod
def validate_prompt(cls, values: dict) -> dict:
prompt: BasePromptTemplate = values['prompt']
if 'agent_scratchpad' not in prompt.input_variables:
raise ValueError('`agent_scratchpad` should be one of the variables in the prompt, '
f'got {prompt.input_variables}')
return values
@property
def input_keys(self) -> List[str]:
"""Get input keys. Input refers to user input here."""
return ['input']
@property
def functions(self) -> List[dict]:
return [dict(format_tool_to_openai_function(t)) for t in self.tools]
def plan(
self,
intermediate_steps: List[Tuple[AgentAction, str]],
callbacks: Callbacks = None,
with_functions: bool = True,
**kwargs: Any,
) -> Union[AgentAction, AgentFinish]:
"""Given input, decided what to do.
Args:
intermediate_steps: Steps the LLM has taken to date, along with observations
**kwargs: User inputs.
Returns:
Action specifying what tool to use.
"""
agent_scratchpad = _format_intermediate_steps(intermediate_steps)
selected_inputs = {
k: kwargs[k]
for k in self.prompt.input_variables if k != 'agent_scratchpad'
}
full_inputs = dict(**selected_inputs, agent_scratchpad=agent_scratchpad)
prompt = self.prompt.format_prompt(**full_inputs)
messages = prompt.to_messages()
# print(messages)
if with_functions:
predicted_message = self.llm.predict_messages(
messages,
functions=self.functions,
callbacks=callbacks,
)
else:
predicted_message = self.llm.predict_messages(
messages,
callbacks=callbacks,
)
agent_decision = _parse_ai_message(predicted_message)
return agent_decision
async def aplan(
self,
intermediate_steps: List[Tuple[AgentAction, str]],
callbacks: Callbacks = None,
**kwargs: Any,
) -> Union[AgentAction, AgentFinish]:
"""Given input, decided what to do.
Args:
intermediate_steps: Steps the LLM has taken to date,
along with observations
**kwargs: User inputs.
Returns:
Action specifying what tool to use.
"""
agent_scratchpad = _format_intermediate_steps(intermediate_steps)
selected_inputs = {
k: kwargs[k]
for k in self.prompt.input_variables if k != 'agent_scratchpad'
}
full_inputs = dict(**selected_inputs, agent_scratchpad=agent_scratchpad)
prompt = self.prompt.format_prompt(**full_inputs)
messages = prompt.to_messages()
predicted_message = await self.llm.apredict_messages(messages,
functions=self.functions,
callbacks=callbacks)
agent_decision = _parse_ai_message(predicted_message)
return agent_decision
def return_stopped_response(
self,
early_stopping_method: str,
intermediate_steps: List[Tuple[AgentAction, str]],
**kwargs: Any,
) -> AgentFinish:
"""Return response when agent has been stopped due to max iterations."""
if early_stopping_method == 'force':
# `force` just returns a constant string
return AgentFinish({'output': 'Agent stopped due to iteration limit or time limit.'},
'')
elif early_stopping_method == 'generate':
# Generate does one final forward pass
agent_decision = self.plan(intermediate_steps, with_functions=False, **kwargs)
if type(agent_decision) == AgentFinish:
return agent_decision
else:
raise ValueError(f'got AgentAction with no functions provided: {agent_decision}')
else:
raise ValueError('early_stopping_method should be one of `force` or `generate`, '
f'got {early_stopping_method}')
@classmethod
def create_prompt(
cls,
system_message: Optional[SystemMessage] = SystemMessage(
content='You are a helpful AI assistant.'),
extra_prompt_messages: Optional[List[BaseMessagePromptTemplate]] = None,
) -> BasePromptTemplate:
"""Create prompt for this agent.
Args:
system_message: Message to use as the system message that will be the
first in the prompt.
extra_prompt_messages: Prompt messages that will be placed between the
system message and the new human input.
Returns:
A prompt template to pass into this agent.
"""
_prompts = extra_prompt_messages or []
messages: List[Union[BaseMessagePromptTemplate, BaseMessage]]
if system_message:
messages = [system_message]
else:
messages = []
messages.extend([
*_prompts,
HumanMessagePromptTemplate.from_template('{input}'),
MessagesPlaceholder(variable_name='agent_scratchpad'),
])
return ChatPromptTemplate(messages=messages)
@classmethod
def from_llm_and_tools(
cls,
llm: BaseLanguageModel,
tools: Sequence[BaseTool],
callback_manager: Optional[BaseCallbackManager] = None,
extra_prompt_messages: Optional[List[BaseMessagePromptTemplate]] = None,
system_message: Optional[SystemMessage] = SystemMessage(
content='You are a helpful AI assistant.'),
**kwargs: Any,
) -> BaseSingleActionAgent:
"""Construct an agent from an LLM and tools."""
if ((not isinstance(llm, ChatOpenAI)) and (not isinstance(llm, HostQwenChat))
and (not isinstance(llm, ProxyChatLLM))):
raise ValueError(
'Only supported with ChatOpenAI and HostQwenChat and ProxyChatLLM models.')
prompt = cls.create_prompt(
extra_prompt_messages=extra_prompt_messages,
system_message=system_message,
)
return cls(
llm=llm,
prompt=prompt,
tools=tools,
callback_manager=callback_manager,
**kwargs,
)
================================================
FILE: src/backend/bisheng_langchain/autogen_role/__init__.py
================================================
from bisheng_langchain.autogen_role.assistant import AutoGenAssistant
from bisheng_langchain.autogen_role.custom import AutoGenCustomRole
from bisheng_langchain.autogen_role.groupchat_manager import AutoGenGroupChatManager
from bisheng_langchain.autogen_role.user import AutoGenCoder, AutoGenUser
__all__ = ['AutoGenAssistant', 'AutoGenGroupChatManager',
'AutoGenUser', 'AutoGenCoder',
'AutoGenCustomRole']
================================================
FILE: src/backend/bisheng_langchain/autogen_role/assistant.py
================================================
"""Chain that runs an arbitrary python function."""
import logging
from typing import Callable, Dict, Optional
import httpx
from autogen import AssistantAgent
from langchain.base_language import BaseLanguageModel
logger = logging.getLogger(__name__)
class AutoGenAssistant(AssistantAgent):
"""Assistant agent, designed to solve a task with LLM.
"""
DEFAULT_SYSTEM_MESSAGE = """You are a helpful AI assistant.
Solve tasks using your coding and language skills.
In the following cases, suggest python code (in a python coding block) or shell script (in a sh coding block) for the user to execute.
1. When you need to collect info, use the code to output the info you need, for example, browse or search the web, download/read a file, print the content of a webpage or a file, get the current date/time, check the operating system. After sufficient info is printed and the task is ready to be solved based on your language skill, you can solve the task by yourself.
2. When you need to perform some task with code, use the code to perform the task and output the result. Finish the task smartly.
Solve the task step by step if you need to. If a plan is not provided, explain your plan first. Be clear which step uses code, and which step uses your language skill.
When using code, you must indicate the script type in the code block. The user cannot provide any other feedback or perform any other action beyond executing the code you suggest. The user can't modify your code. So do not suggest incomplete code which requires users to modify. Don't use a code block if it's not intended to be executed by the user.
If you want the user to save the code in a file before executing it, put # filename: inside the code block as the first line. Don't include multiple code blocks in one response. Do not ask users to copy and paste the result. Instead, use 'print' function for the output when relevant. Check the execution result returned by the user.
If the result indicates there is an error, fix the error and output the code again. Suggest the full code instead of partial code or code changes. If the error can't be fixed or if the task is not solved even after the code is executed successfully, analyze the problem, revisit your assumption, collect additional info you need, and think of a different approach to try.
When you find an answer, verify the answer carefully. Include verifiable evidence in your response if possible.
Reply "TERMINATE" in the end when everything is done.
""" # noqa
def __init__(
self,
name: str,
model_name: Optional[str] = 'gpt-4-0613', # when llm_flag=True, need to set
openai_api_key: Optional[str] = '', # when llm_flag=True, need to set
openai_api_base: Optional[str] = '', # when llm_flag=True, need to set
openai_proxy: Optional[str] = '', # when llm_flag=True, need to set
temperature: Optional[float] = 0, # when llm_flag=True, need to set
api_type: Optional[str] = None, # when llm_flag=True, need to set
api_version: Optional[str] = None, # when llm_flag=True, need to set
llm: Optional[BaseLanguageModel] = None,
system_message: Optional[
str] = DEFAULT_SYSTEM_MESSAGE, # agent system message, llm or group chat manage will use # noqa
is_termination_msg: Optional[Callable[[Dict], bool]] = None,
**kwargs,
):
is_termination_msg = (is_termination_msg if is_termination_msg is not None else
(lambda x: x.get('content') == 'TERMINATE'))
config_list = [
{
'model': model_name,
'api_key': openai_api_key,
'api_base': openai_api_base,
'api_type': api_type,
'api_version': api_version,
},
]
if openai_proxy:
config_list[0]['http_client'] = httpx.Client(proxies=openai_proxy)
config_list[0]['http_async_client'] = httpx.AsyncClient(proxies=openai_proxy)
if llm:
llm_config = llm
else:
llm_config = {
'seed': 42, # change the seed for different trials
'temperature': temperature,
'config_list': config_list,
'request_timeout': 120,
}
super().__init__(
name,
llm_config=llm_config,
system_message=system_message,
is_termination_msg=is_termination_msg,
max_consecutive_auto_reply=None,
human_input_mode='NEVER',
code_execution_config=False,
)
================================================
FILE: src/backend/bisheng_langchain/autogen_role/custom.py
================================================
from typing import Any, Awaitable, Callable, Dict, List, Optional, Union
from autogen import Agent, ConversableAgent
class AutoGenCustomRole(ConversableAgent):
"""Custom agent that can use langchain agent and chain."""
def __init__(
self,
name: str,
system_message: str,
func: Callable[..., str],
coroutine: Optional[Callable[..., Awaitable[str]]] = None,
**kwargs,
):
super().__init__(name=name,
system_message=system_message,
human_input_mode='NEVER',
code_execution_config=False,
llm_config=False,
**kwargs)
self.func = func
self.coroutine = coroutine
self.register_reply(Agent, AutoGenCustomRole.generate_custom_reply)
self.register_reply(Agent, AutoGenCustomRole.a_generate_custom_reply, is_async=True)
def generate_custom_reply(
self,
messages: Optional[List[Dict]] = None,
sender: Optional[Agent] = None,
config: Optional[Any] = None,
**kwargs,
) -> Union[str, Dict, None]:
if messages is None:
messages = self._oai_messages[sender]
message = messages[-1]
if 'content' in message:
query = message['content']
reply = self.func(query)
if isinstance(reply, dict):
reply = list(reply.values())
if reply:
reply = str(reply[0])
else:
reply = ''
return True, reply
return False, None
async def a_generate_custom_reply(
self,
messages: Optional[List[Dict]] = None,
sender: Optional[Agent] = None,
config: Optional[Any] = None,
**kwargs,
) -> Union[str, Dict, None]:
if messages is None:
messages = self._oai_messages[sender]
message = messages[-1]
if 'content' in message:
query = message['content']
if self.coroutine:
reply = await self.coroutine(query)
else:
reply = self.func(query)
if isinstance(reply, dict):
reply = list(reply.values())
if reply:
reply = str(reply[0])
else:
reply = ''
return True, reply
return False, None
================================================
FILE: src/backend/bisheng_langchain/autogen_role/groupchat_manager.py
================================================
"""Chain that runs an arbitrary python function."""
import logging
from typing import List, Optional
import httpx
from autogen import Agent, GroupChat, GroupChatManager
from langchain.base_language import BaseLanguageModel
from .user import AutoGenUser
logger = logging.getLogger(__name__)
class AutoGenGroupChatManager(GroupChatManager):
"""A chat manager agent that can manage a group chat of multiple agents.
"""
def __init__(
self,
agents: List[Agent],
max_round: int = 50,
llm: Optional[BaseLanguageModel] = None,
model_name: Optional[str] = 'gpt-4-0613',
openai_api_key: Optional[str] = '',
openai_api_base: Optional[str] = '',
openai_proxy: Optional[str] = '',
temperature: Optional[float] = 0,
api_type: Optional[str] = None, # when llm_flag=True, need to set
api_version: Optional[str] = None, # when llm_flag=True, need to set
name: Optional[str] = 'chat_manager',
system_message: Optional[str] = 'Group chat manager.',
**kwargs,
):
if not any(isinstance(agent, AutoGenUser) for agent in agents):
raise Exception('chat_manager must contains AutoGenUser')
groupchat = GroupChat(agents=agents, messages=[], max_round=max_round)
config_list = [
{
'model': model_name,
'api_key': openai_api_key,
'api_base': openai_api_base,
'api_type': api_type,
'api_version': api_version,
},
]
if openai_proxy:
config_list[0]['http_client'] = httpx.Client(proxies=openai_proxy)
config_list[0]['http_async_client'] = httpx.AsyncClient(proxies=openai_proxy)
if llm:
llm_config = llm
else:
llm_config = {
'seed': 42, # change the seed for different trials
'temperature': temperature,
'config_list': config_list,
'request_timeout': 120,
}
super().__init__(
groupchat=groupchat,
llm_config=llm_config,
name=name,
system_message=system_message,
)
================================================
FILE: src/backend/bisheng_langchain/autogen_role/user.py
================================================
"""Chain that runs an arbitrary python function."""
import logging
from typing import Callable, Dict, Optional
import httpx
from autogen import UserProxyAgent
from langchain.base_language import BaseLanguageModel
logger = logging.getLogger(__name__)
class AutoGenUserProxyAgent(UserProxyAgent):
"""A proxy agent for the user, that can execute code and provide feedback to the other agents.
"""
def __init__(
self,
name: str,
max_consecutive_auto_reply: Optional[int] = None,
human_input_mode: Optional[str] = 'ALWAYS', # hmean feedback input
code_execution_flag: Optional[bool] = True, # code execution
function_map: Optional[Dict[str, Callable]] = None, # function call
llm_flag: Optional[bool] = False, # llm call
model_name: Optional[str] = 'gpt-4-0613', # when llm_flag=True, need to set
openai_api_key: Optional[str] = '', # when llm_flag=True, need to set
openai_api_base: Optional[str] = '', # when llm_flag=True, need to set
openai_proxy: Optional[str] = '', # when llm_flag=True, need to set
temperature: Optional[float] = 0, # when llm_flag=True, need to set
api_type: Optional[str] = None, # when llm_flag=True, need to set
api_version: Optional[str] = None, # when llm_flag=True, need to set
llm: Optional[BaseLanguageModel] = None,
system_message: Optional[
str] = '', # agent system message, llm or group chat manage will use
is_termination_msg: Optional[Callable[[Dict], bool]] = None,
**kwargs,
):
is_termination_msg = (is_termination_msg if is_termination_msg is not None else
(lambda x: x.get('content') == 'TERMINATE'))
if code_execution_flag:
code_execution_config = {
'work_dir': '/tmp/autogen_coding', # code save path
'use_docker': False,
}
else:
code_execution_config = False
if llm_flag:
config_list = [
{
'model': model_name,
'api_key': openai_api_key,
'api_base': openai_api_base,
'api_type': api_type,
'api_version': api_version,
},
]
if openai_proxy:
config_list[0]['http_client'] = httpx.Client(proxies=openai_proxy)
config_list[0]['http_async_client'] = httpx.AsyncClient(proxies=openai_proxy)
if llm:
llm_config = llm
else:
llm_config = {
'seed': 42, # change the seed for different trials
'temperature': temperature,
'config_list': config_list,
'request_timeout': 120,
}
else:
llm_config = False
super().__init__(name,
is_termination_msg=is_termination_msg,
max_consecutive_auto_reply=max_consecutive_auto_reply,
human_input_mode=human_input_mode,
function_map=function_map,
code_execution_config=code_execution_config,
llm_config=llm_config,
system_message=system_message)
class AutoGenUser(UserProxyAgent):
"""A proxy agent for the user, that can provide feedback to the other agents.
"""
def __init__(
self,
name: str,
max_consecutive_auto_reply: Optional[int] = 10,
human_input_mode: Optional[str] = 'ALWAYS', # hmean feedback input
system_message: Optional[
str] = '', # agent system message, llm or group chat manage will use
is_termination_msg: Optional[Callable[[Dict], bool]] = None,
**kwargs,
):
is_termination_msg = (is_termination_msg if is_termination_msg is not None else
(lambda x: x.get('content') == 'TERMINATE'))
code_execution_config = False
llm_config = False
super().__init__(name,
is_termination_msg=is_termination_msg,
max_consecutive_auto_reply=max_consecutive_auto_reply,
human_input_mode=human_input_mode,
code_execution_config=code_execution_config,
llm_config=llm_config,
system_message=system_message)
class AutoGenCoder(UserProxyAgent):
"""A proxy agent for the coder, that can execute code to the other agents.
"""
def __init__(
self,
name: str,
function_map: Optional[Dict[str, Callable]] = None, # function call
system_message: Optional[
str] = '', # agent system message, llm or group chat manage will use
is_termination_msg: Optional[Callable[[Dict], bool]] = None,
**kwargs,
):
is_termination_msg = (is_termination_msg if is_termination_msg is not None else
(lambda x: x.get('content') == 'TERMINATE'))
code_execution_config = {
'work_dir': '/tmp/autogen_coding', # code save path
'use_docker': False,
}
llm_config = False
super().__init__(name,
is_termination_msg=is_termination_msg,
human_input_mode='NEVER',
function_map=function_map,
code_execution_config=code_execution_config,
llm_config=llm_config,
system_message=system_message)
================================================
FILE: src/backend/bisheng_langchain/chains/__init__.py
================================================
from bisheng_langchain.chains.autogen.auto_gen import AutoGenChain
from bisheng_langchain.chains.combine_documents.stuff import StuffDocumentsChain
from bisheng_langchain.chains.conversational_retrieval.base import ConversationalRetrievalChain
from bisheng_langchain.chains.retrieval.retrieval_chain import RetrievalChain
from bisheng_langchain.chains.router.multi_rule import MultiRuleChain
from bisheng_langchain.chains.router.rule_router import RuleBasedRouter
from bisheng_langchain.chains.transform import TransformChain
from bisheng_langchain.chains.qa_generation.base import QAGenerationChain
from bisheng_langchain.chains.qa_generation.base_v2 import QAGenerationChainV2
from .loader_output import LoaderOutputChain
__all__ = [
'StuffDocumentsChain', 'LoaderOutputChain', 'AutoGenChain', 'RuleBasedRouter',
'MultiRuleChain', 'RetrievalChain', 'ConversationalRetrievalChain', 'TransformChain',
'QAGenerationChain', 'QAGenerationChainV2'
]
================================================
FILE: src/backend/bisheng_langchain/chains/autogen/__init__.py
================================================
================================================
FILE: src/backend/bisheng_langchain/chains/autogen/auto_gen.py
================================================
"""Chain that runs an arbitrary python function."""
import functools
import json
import logging
from typing import Any, Dict, List, Optional
from autogen import ConversableAgent
from bisheng_langchain.autogen_role import AutoGenGroupChatManager, AutoGenUser
from langchain.callbacks.manager import AsyncCallbackManagerForChainRun, CallbackManagerForChainRun
from langchain.chains.base import Chain
logger = logging.getLogger(__name__)
class AutoGenChain(Chain):
"""Chain that print the loader output.
"""
user_proxy_agent: AutoGenUser
recipient: ConversableAgent
input_key: str = 'chat_topic' #: :meta private:
output_key: str = 'chat_content' #: :meta private:
@staticmethod
@functools.lru_cache
def _log_once(msg: str) -> None:
"""Log a message once.
:meta private:
"""
logger.warning(msg)
@property
def input_keys(self) -> List[str]:
"""Expect input keys.
:meta private:
"""
return [self.input_key]
@property
def output_keys(self) -> List[str]:
"""Return output keys.
:meta private:
"""
return [self.output_key]
def _call(
self,
inputs: Dict[str, str],
run_manager: Optional[CallbackManagerForChainRun] = None,
) -> Dict[str, str]:
message = inputs[self.input_key]
# io_output = io.StringIO()
# with contextlib.redirect_stdout(io_output):
global_chat_messages = []
self.user_proxy_agent.initiate_chat(self.recipient, message=message,
global_chat_messages=global_chat_messages,
run_manager=run_manager)
# chat_content = io_output.getvalue()
chat_content = json.dumps(
global_chat_messages, indent=2, ensure_ascii=False)
output = {self.output_key: chat_content}
return output
async def _acall(
self,
inputs: Dict[str, Any],
run_manager: Optional[AsyncCallbackManagerForChainRun] = None,
) -> Dict[str, Any]:
message = inputs[self.input_key]
# io_output = io.StringIO()
# with contextlib.redirect_stdout(io_output):
global_chat_messages = []
self.recipient.stop = False
await self.user_proxy_agent.a_initiate_chat(self.recipient,
message=message,
global_chat_messages=global_chat_messages,
run_manager=run_manager)
# chat_content = io_output.getvalue()
output = {self.output_key: global_chat_messages[-1].get('message'),
'intermediate_steps': global_chat_messages}
return output
async def stop(self):
self.recipient.stop = True
self.user_proxy_agent.event.set()
self.user_proxy_agent.event.clear()
async def reset(self):
if isinstance(self.recipient, AutoGenGroupChatManager):
self.recipient.reset()
async def input(self, input: str):
self.user_proxy_agent.input = input
self.user_proxy_agent.event.set()
self.user_proxy_agent.event.clear()
================================================
FILE: src/backend/bisheng_langchain/chains/combine_documents/__init__.py
================================================
================================================
FILE: src/backend/bisheng_langchain/chains/combine_documents/stuff.py
================================================
from typing import Any, List, Tuple
from langchain.callbacks.manager import Callbacks
from langchain.chains.combine_documents.stuff import StuffDocumentsChain as StuffDocumentsChainOld
from langchain.docstore.document import Document
class StuffDocumentsChain(StuffDocumentsChainOld):
token_max: int = -1
def combine_docs(self,
docs: List[Document],
callbacks: Callbacks = None,
**kwargs: Any) -> Tuple[str, dict]:
"""Stuff all documents into one prompt and pass to LLM.
Args:
docs: List of documents to join together into one variable
callbacks: Optional callbacks to pass along
**kwargs: additional parameters to use to get inputs to LLMChain.
Returns:
The first element returned is the single string output. The second
element returned is a dictionary of other keys to return.
"""
inputs = self._get_inputs(docs, **kwargs)
# print('inputs:', len(inputs['context']))
# print('prompt_length:', self.prompt_length(docs, **kwargs))
if self.token_max > 0:
inputs[self.document_variable_name] = inputs[
self.document_variable_name][:self.token_max]
# Call predict on the LLM.
return self.llm_chain.predict(callbacks=callbacks, **inputs), {}
async def acombine_docs(self,
docs: List[Document],
callbacks: Callbacks = None,
**kwargs: Any) -> Tuple[str, dict]:
"""Stuff all documents into one prompt and pass to LLM.
Args:
docs: List of documents to join together into one variable
callbacks: Optional callbacks to pass along
**kwargs: additional parameters to use to get inputs to LLMChain.
Returns:
The first element returned is the single string output. The second
element returned is a dictionary of other keys to return.
"""
inputs = self._get_inputs(docs, **kwargs)
if self.token_max > 0:
inputs[self.document_variable_name] = inputs[
self.document_variable_name][:self.token_max]
# Call predict on the LLM.
return await self.llm_chain.apredict(callbacks=callbacks, **inputs), {}
================================================
FILE: src/backend/bisheng_langchain/chains/conversational_retrieval/__init__.py
================================================
================================================
FILE: src/backend/bisheng_langchain/chains/conversational_retrieval/base.py
================================================
from __future__ import annotations
import inspect
from typing import Any, Dict, List, Optional, Tuple, Union
from langchain.callbacks.manager import AsyncCallbackManagerForChainRun, CallbackManagerForChainRun
from langchain.chains.conversational_retrieval.base import \
ConversationalRetrievalChain as BaseConversationalRetrievalChain
from langchain_core.messages import BaseMessage
# Depending on the memory type and configuration, the chat history format may differ.
# This needs to be consolidated.
CHAT_TURN_TYPE = Union[Tuple[str, str], BaseMessage]
_ROLE_MAP = {'human': 'Human: ', 'ai': 'Assistant: '}
def _get_chat_history(chat_history: List[CHAT_TURN_TYPE]) -> str:
buffer = ''
for dialogue_turn in chat_history:
if isinstance(dialogue_turn, BaseMessage):
role_prefix = _ROLE_MAP.get(dialogue_turn.type, f'{dialogue_turn.type}: ')
buffer += f'\n{role_prefix}{dialogue_turn.content}'
elif isinstance(dialogue_turn, tuple):
human = 'Human: ' + dialogue_turn[0]
ai = 'Assistant: ' + dialogue_turn[1]
buffer += '\n' + '\n'.join([human, ai])
else:
raise ValueError(f'Unsupported chat history format: {type(dialogue_turn)}.'
f' Full chat history: {chat_history} ')
return buffer
class ConversationalRetrievalChain(BaseConversationalRetrievalChain):
"""ConversationalRetrievalChain is a chain you can use to have a conversation with a character from a series."""
def _call(
self,
inputs: Dict[str, Any],
run_manager: Optional[CallbackManagerForChainRun] = None,
) -> Dict[str, Any]:
_run_manager = run_manager or CallbackManagerForChainRun.get_noop_manager()
question = inputs['question']
get_chat_history = self.get_chat_history or _get_chat_history
chat_history_str = get_chat_history(inputs['chat_history'])
if chat_history_str:
# callbacks = _run_manager.get_child()
new_question = self.question_generator.run(question=question,
chat_history=chat_history_str)
else:
new_question = question
accepts_run_manager = ('run_manager' in inspect.signature(self._get_docs).parameters)
if accepts_run_manager:
docs = self._get_docs(new_question, inputs, run_manager=_run_manager)
else:
docs = self._get_docs(new_question, inputs) # type: ignore[call-arg]
output: Dict[str, Any] = {}
if self.response_if_no_docs_found is not None and len(docs) == 0:
output[self.output_key] = self.response_if_no_docs_found
else:
new_inputs = inputs.copy()
if self.rephrase_question:
new_inputs['question'] = new_question
new_inputs['chat_history'] = chat_history_str
answer = self.combine_docs_chain.run(input_documents=docs,
callbacks=_run_manager.get_child(),
**new_inputs)
output[self.output_key] = answer
if self.return_source_documents:
output['source_documents'] = docs
if self.return_generated_question:
output['generated_question'] = new_question
return output
async def _acall(
self,
inputs: Dict[str, Any],
run_manager: Optional[AsyncCallbackManagerForChainRun] = None,
) -> Dict[str, Any]:
_run_manager = run_manager or AsyncCallbackManagerForChainRun.get_noop_manager()
question = inputs['question']
get_chat_history = self.get_chat_history or _get_chat_history
chat_history_str = get_chat_history(inputs['chat_history'])
if chat_history_str:
# callbacks = _run_manager.get_child()
new_question = await self.question_generator.arun(question=question,
chat_history=chat_history_str)
else:
new_question = question
accepts_run_manager = ('run_manager' in inspect.signature(self._aget_docs).parameters)
if accepts_run_manager:
docs = await self._aget_docs(new_question, inputs, run_manager=_run_manager)
else:
docs = await self._aget_docs(new_question, inputs) # type: ignore[call-arg]
output: Dict[str, Any] = {}
if self.response_if_no_docs_found is not None and len(docs) == 0:
output[self.output_key] = self.response_if_no_docs_found
else:
new_inputs = inputs.copy()
if self.rephrase_question:
new_inputs['question'] = new_question
new_inputs['chat_history'] = chat_history_str
answer = await self.combine_docs_chain.arun(input_documents=docs,
callbacks=_run_manager.get_child(),
**new_inputs)
output[self.output_key] = answer
if self.return_source_documents:
output['source_documents'] = docs
if self.return_generated_question:
output['generated_question'] = new_question
return output
================================================
FILE: src/backend/bisheng_langchain/chains/loader_output.py
================================================
"""Chain that runs an arbitrary python function."""
import functools
import logging
import json
from typing import Any, Awaitable, Callable, Dict, List, Optional
from langchain.callbacks.manager import (
AsyncCallbackManagerForChainRun,
CallbackManagerForChainRun,
)
from langchain.chains.base import Chain
from langchain.docstore.document import Document
logger = logging.getLogger(__name__)
class LoaderOutputChain(Chain):
"""Chain that print the loader output.
"""
documents: List[Document]
input_key: str = "begin" #: :meta private:
output_key: str = "text" #: :meta private:
@staticmethod
@functools.lru_cache
def _log_once(msg: str) -> None:
"""Log a message once.
:meta private:
"""
logger.warning(msg)
@property
def input_keys(self) -> List[str]:
"""Expect input keys.
:meta private:
"""
return [self.input_key]
@property
def output_keys(self) -> List[str]:
"""Return output keys.
:meta private:
"""
return [self.output_key]
def _call(
self,
inputs: Dict[str, str],
run_manager: Optional[CallbackManagerForChainRun] = None,
) -> Dict[str, str]:
contents = [doc.page_content for doc in self.documents]
contents = '\n\n'.join(contents)
# contents = json.dumps(contents, indent=2, ensure_ascii=False)
output = {self.output_key: contents}
return output
async def _acall(
self,
inputs: Dict[str, Any],
run_manager: Optional[AsyncCallbackManagerForChainRun] = None,
) -> Dict[str, Any]:
contents = [doc.page_content for doc in self.documents]
contents = '\n\n'.join(contents)
# contents = json.dumps(contents, indent=2, ensure_ascii=False)
output = {self.output_key: contents}
return output
================================================
FILE: src/backend/bisheng_langchain/chains/qa_generation/__init__.py
================================================
================================================
FILE: src/backend/bisheng_langchain/chains/qa_generation/base.py
================================================
from __future__ import annotations
import json
import re
import logging
from langchain.docstore.document import Document
from typing import Any, Dict, List, Optional
from langchain_core.callbacks import CallbackManagerForChainRun
from langchain_core.language_models import BaseLanguageModel
from langchain_core.prompts import BasePromptTemplate, ChatPromptTemplate
from pydantic import Field
from langchain_text_splitters import RecursiveCharacterTextSplitter, TextSplitter
from langchain.chains.base import Chain
from langchain.chains.llm import LLMChain
from langchain.chains.qa_generation.prompt import PROMPT_SELECTOR, CHAT_PROMPT, PROMPT
logger = logging.getLogger(__name__)
def parse_json(input_str: str) -> str:
match = re.search(r'```(json)?(.*)```', input_str, re.DOTALL)
if match is None:
out_str = input_str
else:
out_str = match.group(2)
out_str = out_str.strip()
out_str = out_str.replace('```', '')
return out_str
class QAGenerationChain(Chain):
"""Base class for question-answer generation chains."""
documents: List[Document]
llm_chain: LLMChain
"""LLM Chain that generates responses from user input and context."""
k: Optional[int] = None
"""Number of questions to generate."""
text_splitter: TextSplitter = Field(
default=RecursiveCharacterTextSplitter(
separators=["\n\n", "\n", " ", ""],
chunk_size=1000,
chunk_overlap=100,
)
)
"""Text splitter that splits the input into chunks."""
input_key: str = "begin"
"""Key of the input to the chain."""
output_key: str = "questions"
"""Key of the output of the chain."""
@classmethod
def from_llm(
cls,
documents: List[Document],
llm: BaseLanguageModel,
k: Optional[int] = None,
chunk_size: int = 512,
prompt: Optional[ChatPromptTemplate] = CHAT_PROMPT,
**kwargs: Any,
) -> QAGenerationChain:
"""
Create a QAGenerationChain from a language model.
Args:
llm: a language model
prompt: a prompt template
**kwargs: additional arguments
Returns:
a QAGenerationChain class
"""
_prompt = PROMPT_SELECTOR.get_prompt(llm) if prompt is None else prompt
chain = LLMChain(llm=llm, prompt=_prompt)
text_splitter = RecursiveCharacterTextSplitter(
separators=["\n\n", "\n", " ", ""],
chunk_size=chunk_size,
chunk_overlap=50,
)
return cls(documents=documents, llm_chain=chain, k=k, text_splitter=text_splitter, **kwargs)
@property
def _chain_type(self) -> str:
raise NotImplementedError
@property
def input_keys(self) -> List[str]:
return [self.input_key]
@property
def output_keys(self) -> List[str]:
return [self.output_key]
def _call(
self,
inputs: Dict[str, Any],
run_manager: Optional[CallbackManagerForChainRun] = None,
) -> Dict[str, List]:
contents = [doc.page_content for doc in self.documents]
contents = '\n\n'.join(contents)
docs = self.text_splitter.create_documents([contents])
# len(qa) = min(len(docs), self.k)
logger.info(f"Split {len(docs)} documents. Gen qa num: min({len(docs)}, {self.k}).")
qa = ''
qa_i = 0
for doc in docs:
try:
results = self.llm_chain.generate([{"text": doc.page_content}], run_manager=run_manager)
res = results.generations[0]
qa += res[0].text
qa_i += 1
except Exception as e:
logger.error(f"Failed to parse response Error: {e}")
continue
if self.k is not None and qa_i >= self.k:
break
return {self.output_key: qa}
async def _acall(
self,
inputs: Dict[str, Any],
run_manager: Optional[CallbackManagerForChainRun] = None,
) -> Dict[str, List]:
output = self._call(inputs, run_manager)
return output
================================================
FILE: src/backend/bisheng_langchain/chains/qa_generation/base_v2.py
================================================
from __future__ import annotations
import re
import json
import logging
import typing as t
import warnings
from typing import Any, Dict, List, Optional
from collections import defaultdict, namedtuple
from dataclasses import dataclass
from langchain_core.callbacks import CallbackManagerForChainRun
from langchain_core.language_models import BaseLanguageModel
from langchain_core.prompts import HumanMessagePromptTemplate, PromptTemplate
try:
from llama_index.core.node_parser import SimpleNodeParser
from llama_index.core.schema import Document as LlamaindexDocument
from llama_index.core.schema import BaseNode
except ImportError:
raise ImportError(
"llama_index must be installed to use this function. "
"Please, install it with `pip install llama_index`."
)
import numpy as np
import numpy.testing as npt
import pandas as pd
from langchain.prompts import ChatPromptTemplate
from langchain.docstore.document import Document
# from langchain.schema.document import Document as LangchainDocument
from langchain.chains.base import Chain
from numpy.random import default_rng
from tqdm import tqdm
from .prompt_v2 import (
SEED_QUESTION_CHAT_PROMPT,
SCORE_CONTEXT_CHAT_PROMPT,
FILTER_QUESTION_CHAT_PROMPT,
ANSWER_FORMULATE,
)
from .base import parse_json
logger = logging.getLogger(__name__)
def load_as_score(text):
"""
validate and returns given text as score
"""
pattern = r"^[\d.]+$"
if not re.match(pattern, text):
warnings.warn("Invalid score")
score = 0.0
else:
score = eval(text)
return score
def load_as_json(text):
"""
validate and return given text as json
"""
try:
return json.loads(parse_json(text))
except ValueError as e:
warnings.warn(f"Invalid json: {e}")
return {}
DEFAULT_TRAIN_DISTRIBUTION = {
"simple": 1.0,
"reasoning": 0.0,
"multi_context": 0.0,
"conditional": 0.0,
}
DataRow = namedtuple(
"DataRow",
[
"question",
"ground_truth_context",
"ground_truth",
"question_type",
"episode_done",
],
)
@dataclass
class TrainDataset:
"""
TrainDataset class
"""
train_data: t.List[DataRow]
def to_pandas(self) -> pd.DataFrame:
data_samples = []
for data in self.train_data:
data = {
"question": data.question,
"ground_truth_context": data.ground_truth_context,
"ground_truth": data.ground_truth,
"question_type": data.question_type,
"episode_done": data.episode_done,
}
data_samples.append(data)
return pd.DataFrame.from_records(data_samples)
class TrainsetGenerator:
"""
Ragas Train Set Generator
Attributes
----------
generator_llm: LangchainLLM
LLM used for all the generator operations in the TrainGeneration paradigm.
critique_llm: LangchainLLM
LLM used for all the filtering and scoring operations in TrainGeneration
paradigm.
chunk_size: int
The chunk size of nodes created from data.
train_distribution : dict
Distribution of different types of questions to be generated from given
set of documents. Defaults to {"easy":0.1, "reasoning":0.4, "conversation":0.5}
"""
def __init__(
self,
generator_llm: BaseLanguageModel,
critic_llm: BaseLanguageModel,
trainset_distribution: t.Optional[t.Dict[str, float]] = None,
chunk_size: int = 1024,
seed: int = 42,
prompt: Optional[ChatPromptTemplate] = SEED_QUESTION_CHAT_PROMPT,
filter_lowquality_context: bool = False,
filter_lowquality_question: bool = False,
answer_prompt: Optional[HumanMessagePromptTemplate] = ANSWER_FORMULATE,
) -> None:
self.generator_llm = generator_llm
self.critic_llm = critic_llm
trainset_distribution = trainset_distribution or DEFAULT_TRAIN_DISTRIBUTION
npt.assert_almost_equal(
1,
sum(trainset_distribution.values()),
err_msg="Sum of distribution should be 1",
)
probs = np.cumsum(list(trainset_distribution.values()))
types = trainset_distribution.keys()
self.trainset_distribution = dict(zip(types, probs))
self.chunk_size = chunk_size
self.threshold = 5.0
self.rng = default_rng(seed)
self.prompt = prompt
self.filter_lowquality_context = filter_lowquality_context
self.filter_lowquality_question = filter_lowquality_question
if answer_prompt is None:
answer_prompt = ANSWER_FORMULATE
self.answer_prompt = answer_prompt
@classmethod
def from_default(
cls,
llm: BaseLanguageModel,
chunk_size: int = 512,
trainset_distribution: dict = DEFAULT_TRAIN_DISTRIBUTION,
prompt: Optional[ChatPromptTemplate] = SEED_QUESTION_CHAT_PROMPT,
filter_lowquality_context: bool = False,
filter_lowquality_question: bool = False,
answer_prompt: Optional[PromptTemplate] = ANSWER_FORMULATE,
):
generator_llm = llm
critic_llm = llm
return cls(
generator_llm=generator_llm,
critic_llm=critic_llm,
chunk_size=chunk_size,
trainset_distribution=trainset_distribution,
prompt=prompt,
filter_lowquality_context=filter_lowquality_context,
filter_lowquality_question=filter_lowquality_question,
answer_prompt=answer_prompt,
)
def _get_evolve_type(self) -> str:
"""
Decides question evolution type based on probability
"""
prob = self.rng.uniform(0, 1)
return next(
(
key
for key in self.trainset_distribution.keys()
if prob <= self.trainset_distribution[key]
),
"simple",
)
def _filter_context(self, context: str) -> bool:
"""
context: str
The input context
Checks if the context is has information worthy of framing a question
"""
prompt = SCORE_CONTEXT_CHAT_PROMPT.format_prompt(context=context)
results = self.critic_llm(prompt.to_messages())
output = results.content
score = load_as_score(output)
print('context score:', score)
return score >= self.threshold
def _seed_question(self, context: str) -> str:
if self.prompt is None:
prompt = SEED_QUESTION_CHAT_PROMPT.format_prompt(context=context)
else:
prompt = self.prompt.format_prompt(context=context)
results = self.generator_llm(prompt.to_messages())
return results.content
def _filter_question(self, question: str) -> bool:
prompt = FILTER_QUESTION_CHAT_PROMPT.format_prompt(question=question)
results = self.critic_llm(prompt.to_messages())
results = results.content
json_results = load_as_json(results)
print('filter question:', question, json_results)
return json_results.get("verdict") != "No"
def _qc_template(self, prompt, question, context) -> str:
human_prompt = prompt.format(question=question, context=context)
prompt = ChatPromptTemplate.from_messages([human_prompt])
results = self.generator_llm(prompt.messages)
return results.content
def _generate_answer(self, question: str, context: t.List[str]) -> t.List[str]:
return [
self._qc_template(self.answer_prompt, qstn, context[i])
for i, qstn in enumerate(question.split("\n"))
]
def _remove_nodes(
self, available_indices: t.List[BaseNode], node_idx: t.List
) -> t.List[BaseNode]:
for idx in node_idx:
available_indices.remove(idx)
return available_indices
def _generate_doc_nodes_map(
self, document_nodes: t.List[BaseNode]
) -> t.Dict[str, t.List[BaseNode]]:
doc_nodes_map: t.Dict[str, t.List[BaseNode]] = defaultdict(list)
for node in document_nodes:
if node.ref_doc_id:
doc_nodes_map[node.ref_doc_id].append(node)
return doc_nodes_map # type: ignore
def _get_neighbour_node(
self, node: BaseNode, related_nodes: t.List[BaseNode]
) -> t.List[BaseNode]:
if len(related_nodes) < 2:
warnings.warn("No neighbors exists")
return [node]
idx = related_nodes.index(node)
ids = [idx - 1, idx] if idx == (len(related_nodes) - 1) else [idx, idx + 1]
return [related_nodes[idx] for idx in ids]
def generate(
self,
documents: t.List[LlamaindexDocument] | t.List[Document],
train_size: int,
) -> TrainDataset:
if not isinstance(documents[0], (LlamaindexDocument, Document)):
raise ValueError(
"Trainset Generatation only supports LlamaindexDocuments or Documents" # noqa
)
if isinstance(documents[0], Document):
# cast to LangchainDocument since its the only case here
documents = t.cast(t.List[Document], documents)
documents = [
LlamaindexDocument.from_langchain_format(doc) for doc in documents
]
# Convert documents into nodes
node_parser = SimpleNodeParser.from_defaults(
chunk_size=self.chunk_size, chunk_overlap=0, include_metadata=True
)
documents = t.cast(t.List[LlamaindexDocument], documents)
document_nodes: t.List[BaseNode] = node_parser.get_nodes_from_documents(
documents=documents
)
# # maximum 1 seed question per node
# if train_size > len(document_nodes):
# raise ValueError(
# """Maximum possible number of samples exceeded,
# reduce train_size or add more documents"""
# )
available_nodes = document_nodes
doc_nodes_map = self._generate_doc_nodes_map(document_nodes)
count_neighbours = sum(len(val) > 1 for _, val in doc_nodes_map.items())
if count_neighbours < len(documents) // 2:
warnings.warn("Most documents are too short")
count = 0
samples = []
pbar = tqdm(total=train_size)
while count < train_size and available_nodes != []:
print(count, train_size, len(available_nodes))
evolve_type = self._get_evolve_type()
curr_node = self.rng.choice(np.array(available_nodes), size=1)[0]
available_nodes = self._remove_nodes(available_nodes, [curr_node])
neighbor_nodes = doc_nodes_map[curr_node.source_node.node_id]
# Append multiple nodes randomly to remove chunking bias
size = self.rng.integers(1, 3)
nodes = (
self._get_neighbour_node(curr_node, neighbor_nodes)
if size > 1 and evolve_type != "multi_context"
else [curr_node]
)
text_chunk = " ".join([node.get_content() for node in nodes])
if self.filter_lowquality_context:
score = self._filter_context(text_chunk)
if not score:
continue
seed_question = self._seed_question(text_chunk)
question = seed_question
if self.filter_lowquality_question:
is_valid_question = self._filter_question(question)
else:
is_valid_question = True
if is_valid_question:
context = [text_chunk] * len(question.split("\n"))
is_conv = len(context) > 1
answer = self._generate_answer(question, context)
for i, (qstn, ctx, ans) in enumerate(
zip(question.split("\n"), context, answer)
):
episode_done = False if is_conv and i == 0 else True
samples.append(
DataRow(qstn, [ctx], [ans], evolve_type, episode_done)
)
count += 1
pbar.update(1)
return TrainDataset(train_data=samples)
class QAGenerationChainV2(Chain):
"""Base class for question-answer generation chains."""
documents: List[Document]
generator: TrainsetGenerator
"""LLM Chain that generates responses from user input and context."""
k: Optional[int] = None
"""Number of questions to generate."""
input_key: str = "begin"
"""Key of the input to the chain."""
output_key: str = "questions"
"""Key of the output of the chain."""
@classmethod
def from_llm(
cls,
documents: List[Document],
llm: BaseLanguageModel,
k: Optional[int] = None,
chunk_size: int = 512,
filter_lowquality_context: bool = False,
filter_lowquality_question: bool = False,
question_prompt: Optional[ChatPromptTemplate] = SEED_QUESTION_CHAT_PROMPT,
answer_prompt: Optional[HumanMessagePromptTemplate] = ANSWER_FORMULATE,
**kwargs: Any,
) -> QAGenerationChainV2:
"""
Create a QAGenerationChain from a language model.
Args:
llm: a language model
question_prompt: a prompt template for generate question
answer_prompt: a prompt template for generate answer
**kwargs: additional arguments
Returns:
a QAGenerationChain class
"""
generator = TrainsetGenerator.from_default(
llm,
chunk_size=chunk_size,
prompt=question_prompt,
answer_prompt=answer_prompt,
filter_lowquality_context=filter_lowquality_context,
filter_lowquality_question=filter_lowquality_question
)
return cls(documents=documents, generator=generator, k=k, **kwargs)
@property
def _chain_type(self) -> str:
raise NotImplementedError
@property
def input_keys(self) -> List[str]:
return [self.input_key]
@property
def output_keys(self) -> List[str]:
return [self.output_key]
def _call(
self,
inputs: Dict[str, Any],
run_manager: Optional[CallbackManagerForChainRun] = None,
) -> Dict[str, List]:
for doc in self.documents:
doc.metadata = {}
if self.k is None:
self.k = 1000
dataset = self.generator.generate(documents=self.documents, train_size=self.k)
df = dataset.to_pandas()
qa_pairs = df.to_dict("records")
qa = []
for pair in qa_pairs:
qa.append({
"question": pair["question"],
"answer": pair["ground_truth"][0],
"context": pair["ground_truth_context"][0],
})
qa = f'```json\n{json.dumps(qa, ensure_ascii=False, indent=4)}\n```'
return {self.output_key: qa}
async def _acall(
self,
inputs: Dict[str, Any],
run_manager: Optional[CallbackManagerForChainRun] = None,
) -> Dict[str, List]:
output = self._call(inputs, run_manager)
return output
================================================
FILE: src/backend/bisheng_langchain/chains/qa_generation/prompt.py
================================================
# flake8: noqa
from langchain.chains.prompt_selector import ConditionalPromptSelector, is_chat_model
from langchain_core.prompts.chat import (
ChatPromptTemplate,
HumanMessagePromptTemplate,
SystemMessagePromptTemplate,
)
from langchain_core.prompts.prompt import PromptTemplate
templ1 = """You are a smart assistant designed to help high school teachers come up with reading comprehension questions.
Given a piece of text, you must come up with a question and answer pair that can be used to test a student's reading comprehension abilities.
When coming up with this question/answer pair, you must respond in the following format and in same language as the text:
```
{{
"question": "$YOUR_QUESTION_HERE",
"answer": "$THE_ANSWER_HERE"
}}
```
Everything between the ``` must be valid json.
"""
templ2 = """Please come up with a question/answer pair, in the specified JSON format, for the following text:
----------------
{text}"""
CHAT_PROMPT = ChatPromptTemplate.from_messages(
[
SystemMessagePromptTemplate.from_template(templ1),
HumanMessagePromptTemplate.from_template(templ2),
]
)
templ = """You are a smart assistant designed to help high school teachers come up with reading comprehension questions.
Given a piece of text, you must come up with a question and answer pair that can be used to test a student's reading comprehension abilities.
When coming up with this question/answer pair, you must respond in the following format and in same language as the text:
```
{{
"question": "$YOUR_QUESTION_HERE",
"answer": "$THE_ANSWER_HERE"
}}
```
Everything between the ``` must be valid json.
Please come up with a question/answer pair, in the specified JSON format, for the following text:
----------------
{text}"""
PROMPT = PromptTemplate.from_template(templ)
PROMPT_SELECTOR = ConditionalPromptSelector(
default_prompt=PROMPT, conditionals=[(is_chat_model, CHAT_PROMPT)]
)
================================================
FILE: src/backend/bisheng_langchain/chains/qa_generation/prompt_v2.py
================================================
from langchain_core.prompts.chat import (
ChatPromptTemplate,
HumanMessagePromptTemplate,
SystemMessagePromptTemplate,
)
SEED_QUESTION_SYSTEM = SystemMessagePromptTemplate.from_template(
"""\
您的任务是遵循以下规则从给定的上下文中提出一个问题，规则如下：
1. 即使在没有给定上下文的情况下，问题也应该对人类有意义。
2. 应该可以从给定上下文中完全回答问题。
3. 问题应该来自包含重要信息的上下文部分。它也可以来自表格、段落、或者代码等。
4. 回答问题时不应包含任何链接。
5. 问题的难度应该是中等的。
6. 问题必须是合理的，并且必须能被人理解和回答。
7. 不要在问题中使用“提供的上下文”等短语。
8. 避免使用可以分解成多个问题的“和”字样来构建问题。
9. 如果上下文是中文，那么问题也应该是中文的。
Examples:
context:武汉达梦数据库股份有限公司 招股说明书 （申报稿） 1-1-226 表中作出恰当列报。 2、研发费用 2021年度、 2020年度、 2019 年度，达梦数据 研发费用金额分别 为11,786.99 万元、 9,660.26 万元、 6,255.86万元， 各年度研发费用占营 业收入的比例分别为 15.86 % 、 21.46 %、20.74 %。 由于研发投入金额及其占当期 营业收入的比例是 达梦数据 的关键 指标之一，可能存在因为核算不准 确而导致的错报风险。因此， 中天 运会计师 将研发费用的归集和核算 确定为关键审计事项。 针对研发费用的真实性与准确性，会计师执行的 重要审计程序主要包括： （1）了解与研发费用相关的关键内部控制，评价 这些控制的设计，确定其是否得到执行，并对相关内 部控制的运行有效性进行测试； （2）获取研发项目立项、审批资料，抽查重要研 发项目的过程文档，判断研发项目的真实性； （3）获取研发费用按项目、性质分类明细表，分
question:达梦2021年的研发费用占营业收入的比例是多少？
context:武汉达梦数据库股份有限公司 招股说明书 （申报稿） 1-1-329 （2）存货周转率 公司与同行业可比公司存货周转率对比情况如下： 公司简称 2021年度 2020年度 2019年度 中望软件 6.93 5.62 10.66 星环科技 3.38 3.21 2.24 金山办公 212.60 175.46 162.91 平均值 74.30 61.43 58.60 本公司 1.13 0.57 0.87 数据来源：可比公司招股说明书、定期报告。 报告期各期， 公司存货周转率显著低于同行业可比公司存货周转率平均水平， 主要是因为公司将未验收的数据及行业解决方案项目所发生的累 计成本均作为 存货核算。报告期各期末，公司存在 “湖北省司法行政数据中心项目 ”、“政法云 大数据中心基础设施服务及大数据中心软件采购 项目”等金额较大且实施周期较 长的数据及行业解决方案项目，导致年末存货金额较大。
question:达梦2021年的存货周转率相较于前一年有何变化？
""" # noqa: E501
)
SEED_QUESTION_HUMAN = HumanMessagePromptTemplate.from_template(
"""
context:{context}
question:
"""
)
SEED_QUESTION_CHAT_PROMPT = ChatPromptTemplate.from_messages(
[
SEED_QUESTION_SYSTEM,
SEED_QUESTION_HUMAN
]
)
SCORE_CONTEXT_SYSTEM = SystemMessagePromptTemplate.from_template(
"""Evaluate the provided context and assign a numerical score between 0 and 10 based on the following criteria:
1. Award a high score to context that thoroughly delves into and explains concepts.
2. Assign a lower score to context that contains excessive references, acknowledgments, external links, personal information, or other non-essential elements.
And you should only output the score.
Examples:
Context:
01-2022.04.30 贷方发生额共 计 2535.43 万元，户名；X 贸易有限公司；\n③根据用款企业提供的增值税纳税申报表来看，2021 年度用款企业年累计开票额为\n7826.48 万元，年累计应纳税合计 95.32 万元，年累计已纳税额 86.23 万元；截止至 2022 年 3 月，用款企业累计开票额为 1986.54 万元，累计应纳税合计19.54 万元，累计已纳税额\n20.23 万元。\n根据核算用款企业的银行流水及企业会计记账系统，剔除借款人往来转账款，估算用款 企业年营业额约在 6000 万元左右(纳税申报营业额)，全部营业收入约 20000 万元左右，借 款人所在 X 贸易有限公司综合毛
利润率约为 35%，净利润约 20%左右。\n\n| 资产种类 | 坐落 | 产权人 | 建筑面积 | 现价值 | 贷款余额 | 资产净值 |\n| --- | --- | --- | --- | --- | --- | --- |\n| 房产 | HN 省 YY 市 PP 小区 5#2-101 | A | 240.20 | 365.23 万 | 165.
Score: 4
Context:
认缴出资额 200 万元 实缴出资额 200 万元 持股比例 20% |\n| 企业所属商圈 | 无 | 是否为已准入商圈 | 是□ 否 ☑ |\n(1) 企业经营历史及现状说明\nX 贸易有限公司 (下称“用款企业”) 注册成立于 2015 年 11 月，统一社会信用代码1234567890ACBDEFGH，法定代表人 A，公司注册地址位于 M 市 N 区 JF 路 20 号 NJ 大厦 18 楼1807 室，实际办公地址位于 M 市 N 区 K 广场 C 座 19 楼 1901、1906、1908、1910、1912、1914，办公面积为 880.51 ㎡，经营场所为用款企业租赁房产，租赁期限，现阶段年租金 73 万余元。\n用款企业是著名品牌“XYZ”的运营公司，是
以经营短袜、连裤袜、 内衣、家居服、配饰为主要品类的亲体织物公司，致力于为年轻消费群体提供“一站式”多品类亲体织物购物 体验。 作为织物文化的传播者和输出者，用款企业秉承一贯的高品质与原创精神，依托中国 研发团队，创领多项核心技术，不断建立并升级健康织物行业标准，目前拥有实用新型专利 6 项，发明专利 1 项，注册商标 30 余个，为品牌的商标保护构建了全面的商标防御体系。\n“XYZ”品牌创立于 2006 年，于 2009 年正式进入中国市场，在成立 10 年的时间里，在全国共有 400 余家店面，运营主要有以下三种模式：\n①直营模式：目前用款企业
管控的直营店有 100 家左右，其中在 M 地区共有 9 家直营店，分别为 Y1 店、Y2 店、Y3 店、Y4 店、Y5 店、Y6 店、Y7 店、Y8 店、Y9 店。经查看用款企业相关财务系统并截屏 ，用款企业 2021 年度 、2022 年 1-4 月直营店营业收入合计分别为7623.45 万元、1987.23 万元，M 地区 9 家直营店收入合计分别为 1238.67 万元、302.54 万元。根据数据测算直营部分毛利润率65%。
Score: 7
""" # noqa: E501
)
SCORE_CONTEXT_HUMAN = HumanMessagePromptTemplate.from_template(
"""
Context:
{context}
Score:
""" # noqa: E501
)
SCORE_CONTEXT_CHAT_PROMPT = ChatPromptTemplate.from_messages(
[
SCORE_CONTEXT_SYSTEM,
SCORE_CONTEXT_HUMAN
]
)
FILTER_QUESTION_SYSTEM = SystemMessagePromptTemplate.from_template(
"""\
Determine if the given question can be clearly understood even when presented without any additional context. Specify reason and verdict is a valid json format.
Examples:
question: What is the discovery about space?
{{
"reason":"The question is too vague and does not specify which discovery about space it is referring to."
"verdit":"No"
}}
question: What caused the Great Depression?
{{
"reason":"The question is specific and refers to a well-known historical economic event, making it clear and answerable.",
"verdict":"Yes"
}}
question: What is the keyword that best describes the paper's focus in natural language understanding tasks?
{{
"reason": "The question mentions a 'paper' in it without referring it's name which makes it unclear without it",
"verdict": "No"
}}
question: Who wrote 'Romeo and Juliet'?
{{
"reason": "The question is clear and refers to a specific work by name therefore it is clear",
"verdict": "Yes"
}}
question: What did the study mention?
{{
"reason": "The question is vague and does not specify which study it is referring to",
"verdict": "No"
}}
question: What is the focus of the REPLUG paper?
{{
"reason": "The question refers to a specific work by it's name hence can be understood",
"verdict": "Yes"
}}
""" # noqa: E501
)
FILTER_QUESTION_HUMAN = HumanMessagePromptTemplate.from_template(
"""\
question:{question}
""" # noqa: E501
)
FILTER_QUESTION_CHAT_PROMPT = ChatPromptTemplate.from_messages(
[
FILTER_QUESTION_SYSTEM,
FILTER_QUESTION_HUMAN
]
)
ANSWER_FORMULATE = HumanMessagePromptTemplate.from_template(
"""\
Answer the question using the information from the given context.
context:{context}
question:{question}
answer:
""" # noqa: E501
)
================================================
FILE: src/backend/bisheng_langchain/chains/question_answering/__init__.py
================================================
"""Load question answering chains."""
from typing import Any, Mapping, Optional, Protocol
# from langchain.chains.combine_documents.stuff import StuffDocumentsChain
from bisheng_langchain.chains.combine_documents.stuff import StuffDocumentsChain
from langchain.callbacks.base import BaseCallbackManager
from langchain.callbacks.manager import Callbacks
from langchain.chains import ReduceDocumentsChain
from langchain.chains.combine_documents.base import BaseCombineDocumentsChain
from langchain.chains.combine_documents.map_reduce import MapReduceDocumentsChain
from langchain.chains.combine_documents.map_rerank import MapRerankDocumentsChain
from langchain.chains.combine_documents.refine import RefineDocumentsChain
from langchain.chains.llm import LLMChain
from langchain.chains.question_answering import map_reduce_prompt, refine_prompts, stuff_prompt
from langchain.chains.question_answering.map_rerank_prompt import PROMPT as MAP_RERANK_PROMPT
from langchain.schema.language_model import BaseLanguageModel
from langchain.schema.prompt_template import BasePromptTemplate
class LoadingCallable(Protocol):
"""Interface for loading the combine documents chain."""
def __call__(self, llm: BaseLanguageModel, **kwargs: Any) -> BaseCombineDocumentsChain:
"""Callable to load the combine documents chain."""
def _load_map_rerank_chain(
llm: BaseLanguageModel,
prompt: BasePromptTemplate = None,
verbose: bool = False,
document_variable_name: str = 'context',
rank_key: str = 'score',
answer_key: str = 'answer',
callback_manager: Optional[BaseCallbackManager] = None,
callbacks: Callbacks = None,
**kwargs: Any,
) -> MapRerankDocumentsChain:
"""Load the map rerank documents chain."""
if hasattr(MAP_RERANK_PROMPT.output_parser, 'default_output_key'):
MAP_RERANK_PROMPT.output_parser.default_output_key = 'answer'
prompt = prompt or MAP_RERANK_PROMPT
llm_chain = LLMChain(
llm=llm,
prompt=prompt,
verbose=verbose,
callback_manager=callback_manager,
callbacks=callbacks,
)
return MapRerankDocumentsChain(
llm_chain=llm_chain,
rank_key=rank_key,
answer_key=answer_key,
document_variable_name=document_variable_name,
verbose=verbose,
callback_manager=callback_manager,
**kwargs,
)
def _load_stuff_chain(
llm: BaseLanguageModel,
prompt: Optional[BasePromptTemplate] = None,
document_variable_name: str = 'context',
verbose: Optional[bool] = None,
callback_manager: Optional[BaseCallbackManager] = None,
callbacks: Callbacks = None,
**kwargs: Any,
) -> StuffDocumentsChain:
_prompt = prompt or stuff_prompt.PROMPT_SELECTOR.get_prompt(llm)
llm_chain = LLMChain(
llm=llm,
prompt=_prompt,
verbose=verbose,
callback_manager=callback_manager,
callbacks=callbacks,
)
# TODO: document prompt
return StuffDocumentsChain(
llm_chain=llm_chain,
document_variable_name=document_variable_name,
verbose=verbose,
callback_manager=callback_manager,
**kwargs,
)
def _load_map_reduce_chain(
llm: BaseLanguageModel,
question_prompt: Optional[BasePromptTemplate] = None,
combine_prompt: Optional[BasePromptTemplate] = None,
combine_document_variable_name: str = 'summaries',
map_reduce_document_variable_name: str = 'context',
collapse_prompt: Optional[BasePromptTemplate] = None,
reduce_llm: Optional[BaseLanguageModel] = None,
collapse_llm: Optional[BaseLanguageModel] = None,
verbose: Optional[bool] = None,
callback_manager: Optional[BaseCallbackManager] = None,
callbacks: Callbacks = None,
token_max: int = 3000,
**kwargs: Any,
) -> MapReduceDocumentsChain:
_question_prompt = (question_prompt
or map_reduce_prompt.QUESTION_PROMPT_SELECTOR.get_prompt(llm))
_combine_prompt = (combine_prompt or map_reduce_prompt.COMBINE_PROMPT_SELECTOR.get_prompt(llm))
map_chain = LLMChain(
llm=llm,
prompt=_question_prompt,
verbose=verbose,
callback_manager=callback_manager,
callbacks=callbacks,
)
_reduce_llm = reduce_llm or llm
reduce_chain = LLMChain(
llm=_reduce_llm,
prompt=_combine_prompt,
verbose=verbose,
callback_manager=callback_manager,
callbacks=callbacks,
)
# TODO: document prompt
combine_documents_chain = StuffDocumentsChain(
llm_chain=reduce_chain,
document_variable_name=combine_document_variable_name,
verbose=verbose,
callback_manager=callback_manager,
callbacks=callbacks,
)
if collapse_prompt is None:
collapse_chain = None
if collapse_llm is not None:
raise ValueError('collapse_llm provided, but collapse_prompt was not: please '
'provide one or stop providing collapse_llm.')
else:
_collapse_llm = collapse_llm or llm
collapse_chain = StuffDocumentsChain(
llm_chain=LLMChain(
llm=_collapse_llm,
prompt=collapse_prompt,
verbose=verbose,
callback_manager=callback_manager,
callbacks=callbacks,
),
document_variable_name=combine_document_variable_name,
verbose=verbose,
callback_manager=callback_manager,
)
reduce_documents_chain = ReduceDocumentsChain(
combine_documents_chain=combine_documents_chain,
collapse_documents_chain=collapse_chain,
token_max=token_max,
verbose=verbose,
)
return MapReduceDocumentsChain(
llm_chain=map_chain,
document_variable_name=map_reduce_document_variable_name,
reduce_documents_chain=reduce_documents_chain,
verbose=verbose,
callback_manager=callback_manager,
callbacks=callbacks,
**kwargs,
)
def _load_refine_chain(
llm: BaseLanguageModel,
question_prompt: Optional[BasePromptTemplate] = None,
refine_prompt: Optional[BasePromptTemplate] = None,
document_variable_name: str = 'context_str',
initial_response_name: str = 'existing_answer',
refine_llm: Optional[BaseLanguageModel] = None,
verbose: Optional[bool] = None,
callback_manager: Optional[BaseCallbackManager] = None,
callbacks: Callbacks = None,
**kwargs: Any,
) -> RefineDocumentsChain:
_question_prompt = (question_prompt or refine_prompts.QUESTION_PROMPT_SELECTOR.get_prompt(llm))
_refine_prompt = refine_prompt or refine_prompts.REFINE_PROMPT_SELECTOR.get_prompt(llm)
initial_chain = LLMChain(
llm=llm,
prompt=_question_prompt,
verbose=verbose,
callback_manager=callback_manager,
callbacks=callbacks,
)
_refine_llm = refine_llm or llm
refine_chain = LLMChain(
llm=_refine_llm,
prompt=_refine_prompt,
verbose=verbose,
callback_manager=callback_manager,
callbacks=callbacks,
)
return RefineDocumentsChain(
initial_llm_chain=initial_chain,
refine_llm_chain=refine_chain,
document_variable_name=document_variable_name,
initial_response_name=initial_response_name,
verbose=verbose,
callback_manager=callback_manager,
**kwargs,
)
def load_qa_chain(
llm: BaseLanguageModel,
chain_type: str = 'stuff',
verbose: Optional[bool] = None,
callback_manager: Optional[BaseCallbackManager] = None,
**kwargs: Any,
) -> BaseCombineDocumentsChain:
"""Load question answering chain.
Args:
llm: Language Model to use in the chain.
chain_type: Type of document combining chain to use. Should be one of "stuff",
"map_reduce", "map_rerank", and "refine".
verbose: Whether chains should be run in verbose mode or not. Note that this
applies to all chains that make up the final chain.
callback_manager: Callback manager to use for the chain.
Returns:
A chain to use for question answering.
"""
loader_mapping: Mapping[str, LoadingCallable] = {
'stuff': _load_stuff_chain,
'map_reduce': _load_map_reduce_chain,
'refine': _load_refine_chain,
'map_rerank': _load_map_rerank_chain,
}
if chain_type not in loader_mapping:
raise ValueError(f'Got unsupported chain type: {chain_type}. '
f'Should be one of {loader_mapping.keys()}')
return loader_mapping[chain_type](llm,
verbose=verbose,
callback_manager=callback_manager,
**kwargs)
================================================
FILE: src/backend/bisheng_langchain/chains/retrieval/__init__.py
================================================
================================================
FILE: src/backend/bisheng_langchain/chains/retrieval/retrieval_chain.py
================================================
import inspect
from typing import Any, Dict, List, Optional
from langchain.callbacks.manager import (
AsyncCallbackManagerForChainRun,
CallbackManagerForChainRun,
)
from langchain.chains.base import Chain
from langchain.schema import BaseRetriever, Document
from pydantic import Field
class RetrievalChain(Chain):
"""Chain to use to combine the documents."""
input_key: str = 'query' #: :meta private:
output_key: str = 'result' #: :meta private:
retriever: BaseRetriever = Field(exclude=True)
return_source_documents: bool = False
def _get_docs(
self,
question: str,
*,
run_manager: CallbackManagerForChainRun,
) -> List[Document]:
"""Get docs."""
return self.retriever.get_relevant_documents(question, callbacks=run_manager.get_child())
async def _aget_docs(
self,
question: str,
*,
run_manager: AsyncCallbackManagerForChainRun,
) -> List[Document]:
"""Get docs."""
return await self.retriever.aget_relevant_documents(question, callbacks=run_manager.get_child())
@property
def input_keys(self) -> List[str]:
"""Return the input keys.
:meta private:
"""
return [self.input_key]
@property
def output_keys(self) -> List[str]:
"""Return the output keys.
:meta private:
"""
_output_keys = [self.output_key]
if self.return_source_documents:
_output_keys = _output_keys + ["source_documents"]
return _output_keys
def _call(
self,
inputs: Dict[str, Any],
run_manager: Optional[CallbackManagerForChainRun] = None,
) -> Dict[str, Any]:
_run_manager = run_manager or CallbackManagerForChainRun.get_noop_manager()
question = inputs[self.input_key]
accepts_run_manager = "run_manager" in inspect.signature(self._get_docs).parameters
if accepts_run_manager:
docs = self._get_docs(question, run_manager=_run_manager)
else:
docs = self._get_docs(question) # type: ignore[call-arg]
if self.return_source_documents:
return {self.output_key: docs, "source_documents": docs}
else:
return {self.output_key: docs}
async def _acall(
self,
inputs: Dict[str, Any],
run_manager: Optional[AsyncCallbackManagerForChainRun] = None,
) -> Dict[str, Any]:
_run_manager = run_manager or AsyncCallbackManagerForChainRun.get_noop_manager()
question = inputs[self.input_key]
accepts_run_manager = "run_manager" in inspect.signature(self._aget_docs).parameters
if accepts_run_manager:
docs = await self._aget_docs(question, run_manager=_run_manager)
else:
docs = await self._aget_docs(question) # type: ignore[call-arg]
if self.return_source_documents:
return {self.output_key: docs, "source_documents": docs}
else:
return {self.output_key: docs}
================================================
FILE: src/backend/bisheng_langchain/chains/router/__init__.py
================================================
================================================
FILE: src/backend/bisheng_langchain/chains/router/multi_rule.py
================================================
from typing import List, Mapping
from langchain.chains.router.base import Chain, MultiRouteChain, RouterChain
class MultiRuleChain(MultiRouteChain):
router_chain: RouterChain
destination_chains: Mapping[str, Chain]
default_chain: Chain
output_variables: List[str]
@property
def output_keys(self) -> List[str]:
return self.output_variables
================================================
FILE: src/backend/bisheng_langchain/chains/router/rule_router.py
================================================
import asyncio
from typing import Any, Callable, Dict, List, Union
from langchain.callbacks.manager import Callbacks
from langchain.chains.router.base import Route, RouterChain
class RuleBasedRouter(RouterChain):
rule_function: Callable[..., str]
input_variables: List[str]
@property
def input_keys(self):
return self.input_variables
def _validate_outputs(self, outputs: Dict[str, Any]) -> None:
super()._validate_outputs(outputs)
if not isinstance(outputs['next_inputs'], dict):
raise ValueError
def _call(
self,
inputs: Union[Dict[str, Any], Any],
) -> Route:
result = self.rule_function(inputs)
if not result.get('destination') or not result:
return Route(None, result['next_inputs'])
return Route(result['destination'], result['next_inputs'])
def route(
self,
inputs: Union[Dict[str, Any], Any],
callbacks: Callbacks = None,
) -> Route:
result = self.rule_function(inputs)
if not result.get('destination') or not result:
return Route(None, result['next_inputs'])
return Route(result['destination'], result['next_inputs'])
async def aroute(
self,
inputs: Union[Dict[str, Any], Any],
callbacks: Callbacks = None,
) -> Route:
"""Route the inputs to the next chain based on the rule function."""
# 如果是异步function，那么就用await
if asyncio.iscoroutinefunction(self.rule_function):
result = await self.rule_function(inputs)
else:
result = self.rule_function(inputs)
if not result.get('destination') or not result:
return Route(None, result['next_inputs'])
return Route(result['destination'], result['next_inputs'])
================================================
FILE: src/backend/bisheng_langchain/chains/transform.py
================================================
"""Chain that runs an arbitrary python function."""
import functools
import inspect
import logging
from typing import Any, Awaitable, Callable, Dict, List, Optional
from langchain.chains.base import Chain
from langchain_core.callbacks import AsyncCallbackManagerForChainRun, CallbackManagerForChainRun
from pydantic import Field
logger = logging.getLogger(__name__)
class TransformChain(Chain):
"""Chain that transforms the chain output.
Example:
.. code-block:: python
from langchain.chains import TransformChain
transform_chain = TransformChain(input_variables=["text"],
output_variables["entities"], transform=func())
"""
input_variables: List[str]
"""The keys expected by the transform's input dictionary."""
output_variables: List[str]
"""The keys returned by the transform's output dictionary."""
transform_cb: Callable[[Dict[str, str]], Dict[str, str]] = Field(alias='transform')
"""The transform function."""
atransform_cb: Optional[Callable[[Dict[str, Any]],
Awaitable[Dict[str, Any]]]] = Field(None, alias='atransform')
"""The async coroutine transform function."""
@staticmethod
@functools.lru_cache
def _log_once(msg: str) -> None:
"""Log a message once.
:meta private:
"""
logger.warning(msg)
@property
def input_keys(self) -> List[str]:
"""Expect input keys.
:meta private:
"""
return self.input_variables
@property
def output_keys(self) -> List[str]:
"""Return output keys.
:meta private:
"""
return self.output_variables
def _call(
self,
inputs: Dict[str, str],
run_manager: Optional[CallbackManagerForChainRun] = None,
) -> Dict[str, str]:
new_arg_supported = inspect.signature(self.transform_cb).parameters.get('run_manager')
if new_arg_supported:
return self.transform_cb(inputs, run_manager)
else:
return self.transform_cb(inputs)
async def _acall(
self,
inputs: Dict[str, Any],
run_manager: Optional[AsyncCallbackManagerForChainRun] = None,
) -> Dict[str, Any]:
if self.atransform_cb is not None:
new_arg_supported = inspect.signature(self.transform_cb).parameters.get('run_manager')
if new_arg_supported:
return await self.atransform_cb(inputs, run_manager)
else:
return await self.atransform_cb(inputs)
else:
self._log_once("TransformChain's atransform is not provided, falling"
' back to synchronous transform')
return self._call(inputs, run_manager)
================================================
FILE: src/backend/bisheng_langchain/chat_models/__init__.py
================================================
from .host_llm import CustomLLMChat, HostBaichuanChat, HostChatGLM, HostLlama2Chat, HostQwenChat, HostQwen1_5Chat, HostYuanChat, HostYiChat
from .minimax import ChatMinimaxAI
from .proxy_llm import ProxyChatLLM
from .qwen import ChatQWen
from .wenxin import ChatWenxin
from .xunfeiai import ChatXunfeiAI
from .sensetime import SenseChat
__all__ = [
'ProxyChatLLM', 'ChatMinimaxAI', 'ChatWenxin', 'ChatXunfeiAI', 'HostChatGLM',
'HostBaichuanChat', 'HostLlama2Chat', 'HostQwenChat', 'CustomLLMChat', 'ChatQWen', 'SenseChat',
'HostYuanChat', 'HostYiChat', 'HostQwen1_5Chat'
]
================================================
FILE: src/backend/bisheng_langchain/chat_models/host_llm.py
================================================
"""proxy llm chat wrapper."""
from __future__ import annotations
import json
import logging
import sys
from typing import TYPE_CHECKING, Any, Callable, Dict, List, Mapping, Optional, Tuple, Union
import requests
from pydantic import ConfigDict, model_validator, Field
from bisheng_langchain.utils.requests import Requests
from langchain.callbacks.manager import AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun
from langchain.chat_models.base import BaseChatModel
from langchain.schema import ChatGeneration, ChatResult
from langchain.schema.messages import (AIMessage, BaseMessage, ChatMessage, FunctionMessage,
HumanMessage, SystemMessage)
from langchain.utils import get_from_dict_or_env
from langchain_core.language_models.llms import create_base_retry_decorator
# from requests.exceptions import HTTPError
# from .interface import MinimaxChatCompletion
# from .interface.types import ChatInput
if TYPE_CHECKING:
import tiktoken
logger = logging.getLogger(__name__)
def _import_tiktoken() -> Any:
try:
import tiktoken
except ImportError:
raise ValueError('Could not import tiktoken python package. '
'This is needed in order to calculate get_token_ids. '
'Please install it with `pip install tiktoken`.')
return tiktoken
def _create_retry_decorator(
llm: BaseHostChatLLM,
run_manager: Optional[Union[AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun]] = None,
) -> Callable[[Any], Any]:
errors = [requests.exceptions.ReadTimeout, ValueError]
return create_base_retry_decorator(error_types=errors,
max_retries=llm.max_retries,
run_manager=run_manager)
def _convert_dict_to_message(_dict: Mapping[str, Any]) -> BaseMessage:
role = _dict['role']
if role == 'user':
return HumanMessage(content=_dict['content'])
elif role == 'assistant':
content = _dict['content'] or '' # OpenAI returns None for tool invocations
additional_kwargs = {}
if _dict.get('function_call'):
additional_kwargs['function_call'] = dict(_dict['function_call'])
if 'metadata' in _dict:
additional_kwargs['metadata'] = _dict['metadata']
return AIMessage(content=content, additional_kwargs=additional_kwargs)
elif role == 'system':
return SystemMessage(content=_dict['content'])
elif role == 'function':
return FunctionMessage(content=_dict['content'], name=_dict['name'])
else:
return ChatMessage(content=_dict['content'], role=role)
def _convert_message_to_dict(message: BaseMessage) -> dict:
if isinstance(message, ChatMessage):
message_dict = {'role': message.role, 'content': message.content}
for key in message.additional_kwargs.keys():
message_dict[key] = message.additional_kwargs[key]
elif isinstance(message, HumanMessage):
message_dict = {'role': 'user', 'content': message.content}
elif isinstance(message, AIMessage):
message_dict = {'role': 'assistant', 'content': message.content}
if 'function_call' in message.additional_kwargs:
message_dict['function_call'] = message.additional_kwargs['function_call']
elif isinstance(message, SystemMessage):
message_dict = {'role': 'system', 'content': message.content}
elif isinstance(message, FunctionMessage):
message_dict = {
'role': 'function',
'content': message.content,
'name': message.name,
}
else:
raise ValueError(f'Got unknown type {message}')
if 'name' in message.additional_kwargs:
message_dict['name'] = message.additional_kwargs['name']
return message_dict
class BaseHostChatLLM(BaseChatModel):
"""Wrapper around base host Chat large language models.
"""
client: Optional[Any] #: :meta private:
"""Model name to use."""
model_name: str = Field('', alias='model')
temperature: float = 0.9
top_p: float = 0.95
do_sample: bool = False
"""Number of chat completions to generate for each prompt."""
max_tokens: int = 4096
"""What sampling temperature to use."""
model_kwargs: Optional[Dict[str, Any]] = Field(default_factory=dict)
"""Holds any model parameters valid for `create` call not explicitly specified."""
host_base_url: Optional[str] = None
is_ssl: Optional[bool] = False
headers: Optional[Dict[str, str]] = Field(default_factory=dict)
request_timeout: Optional[Union[float, Tuple[float, float]]] = 300
"""Timeout for requests to OpenAI completion API. Default is 600 seconds."""
max_retries: Optional[int] = 1
"""Maximum number of retries to make when generating."""
streaming: Optional[bool] = False
"""Whether to stream the results or not."""
n: Optional[int] = 1
"""Maximum number of tokens to generate."""
tiktoken_model_name: Optional[str] = None
"""The model name to pass to tiktoken when using this class.
Tiktoken is used to count the number of tokens in documents to constrain
them to be under a certain limit. By default, when set to None, this will
be the same as the embedding model name. However, there are some cases
where you may want to use this Embedding class with a model name not
supported by tiktoken. This can include when using Azure embeddings or
when using one of the many model providers that expose an OpenAI-like
API but with different models. In those cases, in order to avoid erroring
when tiktoken is called, you can specify a model name to use here."""
verbose: Optional[bool] = False
decoupled: Optional[bool] = False
model_config = ConfigDict(validate_by_name=True)
@model_validator(mode='before')
@classmethod
def validate_environment(cls, values: Dict) -> Dict:
"""Validate that api key and python package exists in environment."""
values['host_base_url'] = get_from_dict_or_env(values, 'host_base_url', 'HostBaseUrl')
model = values['model_name']
try:
if cls != CustomLLMChat:
url = values['host_base_url'].rsplit('/', 2)[0]
config_ep = f'{url}/v2/models/{model}/config'
config = requests.get(url=config_ep, json={}, timeout=5).json()
policy = config.get('model_transaction_policy', {})
values['decoupled'] = policy.get('decoupled', False)
# Host class should set below code
if values['decoupled']:
values[
'host_base_url'] = f"{values['host_base_url']}/{values['model_name']}/generate_stream"
else:
values[
'host_base_url'] = f"{values['host_base_url']}/{values['model_name']}/infer"
except Exception:
raise Exception(f'Update Decoupled status failed for model {model}')
try:
if values['headers']:
headers = values['headers']
else:
headers = {'Content-Type': 'application/json'}
if cls.is_ssl and "https" in values['host_base_url']:
import aiohttp
values['client'] = Requests(headers=headers,
aiosession=aiohttp.ClientSession(connector=aiohttp.TCPConnector(ssl=False)),
request_timeout=values['request_timeout'])
else:
values['client'] = Requests(headers=headers, request_timeout=values['request_timeout'])
except AttributeError:
raise ValueError('Try upgrading it with `pip install --upgrade requests`.')
return values
@property
def _default_params(self) -> Dict[str, Any]:
"""Get the default parameters for calling ChatMinimaxAI API."""
return {
'request_timeout': self.request_timeout,
'model': self.model_name,
'temperature': self.temperature,
'top_p': self.top_p,
'max_tokens': self.max_tokens,
'do_sample': self.do_sample,
**self.model_kwargs,
}
def completion_with_retry(self, **kwargs: Any) -> Any:
retry_decorator = _create_retry_decorator(self)
@retry_decorator
def _completion_with_retry(**kwargs: Any) -> Any:
self.client.headers = self.headers
messages = kwargs.get('messages')
temperature = kwargs.get('temperature')
top_p = kwargs.get('top_p')
max_tokens = kwargs.get('max_tokens')
do_sample = kwargs.get('do_sample')
params = {
'stream': False,
'messages': messages,
'model': self.model_name,
'top_p': top_p,
'temperature': temperature,
'max_tokens': max_tokens,
'do_sample': do_sample,
'function_call': kwargs.get('function_call', None),
'functions': kwargs.get('functions', [])
}
# print('messages:', messages)
# print('functions:', kwargs.get('functions', []))
if self.verbose:
logger.info('payload=%s', json.dumps(params, indent=2))
try:
resp = self.client.post(url=self.host_base_url, json=params)
if resp.text.startswith('data:'):
resp = json.loads(resp.text.replace('data:', ''))
else:
resp = resp.json()
except requests.exceptions.Timeout as exc:
raise ValueError(f'timeout in host llm infer, url=[{self.host_base_url}]') from exc
except Exception as e:
raise ValueError(f'exception in host llm infer: [{e}]') from e
if not resp.get('choices', []):
logger.info(resp)
raise ValueError(f'empty choices in llm chat result {resp}')
resp['usage'] = {}
return resp
return _completion_with_retry(**kwargs)
def _combine_llm_outputs(self, llm_outputs: List[Optional[dict]]) -> dict:
overall_token_usage: dict = {}
for output in llm_outputs:
if output is None:
# Happens in streaming
continue
token_usage = output['token_usage']
if token_usage is None:
continue
for k, v in token_usage.items():
if k in overall_token_usage:
overall_token_usage[k] += v
else:
overall_token_usage[k] = v
return {'token_usage': overall_token_usage, 'model_name': self.model_name}
def _generate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
'''用来处理同步请求'''
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs}
response = self.completion_with_retry(messages=message_dicts, **params)
return self._create_chat_result(response)
async def acompletion_with_retry(self, **kwargs: Any) -> Any:
"""Use tenacity to retry the async completion call."""
retry_decorator = _create_retry_decorator(self)
@retry_decorator
async def _acompletion_with_retry(**kwargs: Any) -> Any:
try:
async with self.client.apost(url=self.host_base_url, json=kwargs) as response:
if response.status != 200:
raise ValueError(f'Error: {response.status} contet: {response.text}')
async for txt in response.content.iter_any():
if b'\n' in txt:
for txt_ in txt.split(b'\n'):
yield txt_.decode('utf-8').strip()
else:
yield txt.decode('utf-8').strip()
except requests.exceptions.Timeout as exc:
raise ValueError(f'timeout in host llm infer, url=[{self.host_base_url}]') from exc
except Exception as e:
raise ValueError(f'exception in host llm infer: [{e}]') from e
text_haf = ''
async for response in _acompletion_with_retry(**kwargs):
is_error = False
if response:
if response.startswith('event:error'):
is_error = True
elif response.startswith('data:'):
text = response[len('data:'):].strip()
if text == '[DONE]':
break
try:
json.loads(text_haf + text)
yield (is_error, text_haf + text)
text_haf = ''
except Exception:
# 拆包了
if text_haf.startswith('{'):
text_haf = text
continue
logger.error(f'response_not_json response={response}')
if is_error:
break
elif response.startswith('{'):
yield (is_error, response)
else:
continue
async def _agenerate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
"""Generate chat completion with retry."""
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs}
if self.streaming and 'infer' not in self.host_base_url:
inner_completion = ''
role = 'assistant'
params['stream'] = True
function_call: Optional[dict] = None
async for is_error, stream_resp in self.acompletion_with_retry(messages=message_dicts,
**params):
output = json.loads(stream_resp)
if is_error:
logger.error(stream_resp)
raise ValueError(stream_resp)
choices = output.get('choices')
if choices:
for choice in choices:
role = choice['delta'].get('role', role)
token = choice['delta'].get('content', '')
inner_completion += token or ''
_function_call = choice['delta'].get('function_call')
if run_manager:
await run_manager.on_llm_new_token(token)
if _function_call:
if function_call is None:
function_call = _function_call
else:
function_call['arguments'] += _function_call['arguments']
message = _convert_dict_to_message({
'content': inner_completion,
'role': role if role is not None else 'assistant',
'function_call': function_call,
})
return ChatResult(generations=[ChatGeneration(message=message)])
else:
params['stream'] = False
response = [
response
async for _, response in self.acompletion_with_retry(messages=message_dicts,
**params)
]
response = json.loads(response[0])
return self._create_chat_result(response)
def _create_message_dicts(
self, messages: List[BaseMessage],
stop: Optional[List[str]]) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:
params = dict(self._client_params)
if stop is not None:
if 'stop' in params:
raise ValueError('`stop` found in both the input and default params.')
params['stop'] = stop
message_dicts = [_convert_message_to_dict(m) for m in messages]
return message_dicts, params
def _create_chat_result(self, response: Mapping[str, Any]) -> ChatResult:
generations = []
if 'choices' not in response:
raise Exception(f'LLM return error {response}')
for res in response['choices']:
message = _convert_dict_to_message(res['message'])
gen = ChatGeneration(message=message)
generations.append(gen)
llm_output = {'token_usage': response.get('usage'), 'model_name': self.model_name}
return ChatResult(generations=generations, llm_output=llm_output)
@property
def _identifying_params(self) -> Mapping[str, Any]:
"""Get the identifying parameters."""
return {**{'model_name': self.model_name}, **self._default_params}
@property
def _client_params(self) -> Mapping[str, Any]:
"""Get the parameters used for the client."""
minimaxai_creds: Dict[str, Any] = {
'model': self.model_name,
}
return {**minimaxai_creds, **self._default_params}
def _get_invocation_params(self,
stop: Optional[List[str]] = None,
**kwargs: Any) -> Dict[str, Any]:
"""Get the parameters used to invoke the model FOR THE CALLBACKS."""
return {
**super()._get_invocation_params(stop=stop, **kwargs),
**self._default_params,
'model': self.model_name,
'function': kwargs.get('functions'),
}
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'host_chat_llm'
def _get_encoding_model(self) -> Tuple[str, tiktoken.Encoding]:
tiktoken_ = _import_tiktoken()
if self.tiktoken_model_name is not None:
model = self.tiktoken_model_name
else:
model = self.model_name
# model chatglm-std, chatglm-lite
# Returns the number of tokens used by a list of messages.
try:
encoding = tiktoken_.encoding_for_model(model)
except KeyError:
logger.warning('Warning: model not found. Using cl100k_base encoding.')
model = 'cl100k_base'
encoding = tiktoken_.get_encoding(model)
return model, encoding
def get_token_ids(self, text: str) -> List[int]:
"""Get the tokens present in the text with tiktoken package."""
# tiktoken NOT supported for Python 3.7 or below
if sys.version_info[1] <= 7:
return super().get_token_ids(text)
_, encoding_model = self._get_encoding_model()
return encoding_model.encode(text)
def get_num_tokens_from_messages(self, messages: List[BaseMessage]) -> int:
"""Calculate num tokens for chatglm with tiktoken package.
todo: read chatglm document
Official documentation: https://github.com/openai/openai-cookbook/blob/
main/examples/How_to_format_inputs_to_ChatGPT_models.ipynb"""
if sys.version_info[1] <= 7:
return super().get_num_tokens_from_messages(messages)
model, encoding = self._get_encoding_model()
if model.startswith('chatglm'):
# every message follows {role/name}\n{content}\n
tokens_per_message = 4
# if there's a name, the role is omitted
tokens_per_name = -1
else:
raise NotImplementedError(
f'get_num_tokens_from_messages() is not presently implemented '
f'for model {model}.'
'See https://github.com/openai/openai-python/blob/main/chatml.md for '
'information on how messages are converted to tokens.')
num_tokens = 0
messages_dict = [_convert_message_to_dict(m) for m in messages]
for message in messages_dict:
num_tokens += tokens_per_message
for key, value in message.items():
num_tokens += len(encoding.encode(value))
if key == 'name':
num_tokens += tokens_per_name
# every reply is primed with assistant
num_tokens += 3
return num_tokens
class HostChatGLM(BaseHostChatLLM):
# chatglm2-12b, chatglm2-6b
model_name: str = Field('chatglm2-6b', alias='model')
temperature: float = 0.95
top_p: float = 0.7
max_tokens: int = 4096
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'chatglm'
class HostBaichuanChat(BaseHostChatLLM):
# Baichuan-7B-Chat, Baichuan-13B-Chat
model_name: str = Field('Baichuan-13B-Chat', alias='model')
temperature: float = 0.3
top_p: float = 0.85
max_tokens: int = 4096
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'baichuan_chat'
class HostQwenChat(BaseHostChatLLM):
# Qwen-7B-Chat
model_name: str = Field('Qwen-7B-Chat', alias='model')
temperature: float = 0
top_p: float = 1
max_tokens: int = 4096
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'qwen_chat'
class HostQwen1_5Chat(BaseHostChatLLM):
# Qwen-7B-Chat
model_name: str = Field('Qwen1.5-14B-Chat', alias='model')
temperature: float = 0
top_p: float = 1
max_tokens: int = 4096
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'qwen1.5_chat'
class HostLlama2Chat(BaseHostChatLLM):
# Llama-2-7b-chat-hf, Llama-2-13b-chat-hf, Llama-2-70b-chat-hf
model_name: str = Field('Llama-2-7b-chat-hf', alias='model')
temperature: float = 0.9
top_p: float = 0.6
max_tokens: int = 4096
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'llama2_chat'
class CustomLLMChat(BaseHostChatLLM):
# use custom llm chat api, api should compatiable with openai definition
model_name: str = Field('custom-llm-chat', alias='model')
temperature: float = 0.1
top_p: float = 0.1
max_tokens: int = 4096
host_base_url: str
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'custom_llm_chat'
class HostYuanChat(BaseHostChatLLM):
# use custom llm chat api, api should compatiable with openai definition
model_name: str = Field('Yuan2-2B-Janus-hf', alias='model')
temperature: float = 1
top_p: float = 0.9
max_tokens: int = 4096
host_base_url: str
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'yuan2'
class HostYiChat(BaseHostChatLLM):
# use custom llm chat api, api should compatiable with openai definition
model_name: str = Field('Yi-34B-Chat', alias='model')
temperature: float = 0.6
top_p: float = 0.8
max_tokens: int = 4096
host_base_url: str
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'yi_chat'
================================================
FILE: src/backend/bisheng_langchain/chat_models/interface/__init__.py
================================================
from .minimax import ChatCompletion as MinimaxChatCompletion
from .openai import ChatCompletion as OpenaiChatCompletion
from .wenxin import ChatCompletion as WenxinChatCompletion
from .xunfei import ChatCompletion as XunfeiChatCompletion
from .zhipuai import ChatCompletion as ZhipuaiChatCompletion
__all__ = [
'MinimaxChatCompletion', 'OpenaiChatCompletion', 'WenxinChatCompletion',
'XunfeiChatCompletion', 'ZhipuaiChatCompletion'
]
================================================
FILE: src/backend/bisheng_langchain/chat_models/interface/minimax.py
================================================
import json
import requests
from .types import ChatInput, ChatOutput, Choice, Message, Usage
from .utils import get_ts
class ChatCompletion(object):
def __init__(self, group_id, api_key, **kwargs):
ep_url = 'https://api.minimax.chat/v1/text/chatcompletion'
self.endpoint = f'{ep_url}?GroupId={group_id}'
self.headers = {
'Authorization': f'Bearer {api_key}',
'Content-Type': 'application/json'
}
def parseChunkDelta(self, chunk):
decoded_data = chunk.decode('utf-8')
parsed_data = json.loads(decoded_data[6:])
delta_content = parsed_data['choices'][0]['delta']
return delta_content
def __call__(self, inp: ChatInput, verbose=False):
messages = inp.messages
model = inp.model
top_p = 0.95 if inp.top_p is None else inp.top_p
temperature = 0.9 if inp.temperature is None else inp.temperature
stream = False if inp.stream is None else inp.stream
max_tokens = 1024 if inp.max_tokens is None else inp.max_tokens
if abs(temperature) <= 1e-6:
temperature = 1e-6
chat_messages = messages
system_prompt = ('MM智能助理是一款由MinMax自研的，没有调用其他产品接口的大型语言'
'模型。MiniMax是一家中国科技公司，一直致力于进行大模型相关的研究。\n----\n')
if messages[0].role == 'system':
system_prompt = messages[0].content
chat_messages = messages[1:]
new_messages = []
for m in chat_messages:
role = 'USER'
if m.role == 'system' or m.role == 'assistant':
role = 'BOT'
new_messages.append({'sender_type': role, 'text': m.content})
# role_meta is given, prompt must is not empty
system_info = {}
if system_prompt:
system_info = {
'prompt': system_prompt,
'role_meta': {
'user_name': '用户',
'bot_name': 'MM智能助理'
}
}
payload = {
'model': model,
'stream': stream,
'use_standard_sse': True,
'messages': new_messages,
'temperature': temperature,
'top_p': top_p,
'tokens_to_generate': max_tokens
}
payload.update(system_info)
if verbose:
print('payload', payload)
response = requests.post(self.endpoint,
headers=self.headers,
json=payload)
req_type = 'chat.completion'
status_message = 'success'
status_code = response.status_code
created = get_ts()
choices = []
usage = Usage()
if status_code == 200:
try:
info = json.loads(response.text)
if info['base_resp']['status_code'] == 0:
created = info['created']
# reply = info['reply']
choices = []
for s in info['choices']:
index = s['index']
finish_reason = s['finish_reason']
msg = Message(role='assistant', content=s['text'])
cho = Choice(index=index,
message=msg,
finish_reason=finish_reason)
choices.append(cho)
total_tokens = info['usage']['total_tokens']
usage = Usage(total_tokens=total_tokens)
else:
status_code = info['base_resp']['status_code']
status_message = info['base_resp']['status_msg']
except Exception as e:
status_code = 401
status_message = str(e)
else:
status_code = 400
status_message = 'requests error'
if status_code != 200:
raise Exception(status_message)
return ChatOutput(status_code=status_code,
status_message=status_message,
model=model,
object=req_type,
created=created,
choices=choices,
usage=usage)
================================================
FILE: src/backend/bisheng_langchain/chat_models/interface/openai.py
================================================
# import json
import openai
from .types import ChatInput, ChatOutput, Choice, Usage
from .utils import get_ts
class ChatCompletion(object):
def __init__(self, api_key, proxy=None, **kwargs):
openai.api_key = api_key
openai.proxy = proxy
def __call__(self, inp: ChatInput, verbose=False):
messages = inp.messages
model = inp.model
top_p = 0.7 if inp.top_p is None else inp.top_p
temperature = 0.97 if inp.temperature is None else inp.temperature
# stream = False if inp.stream is None else inp.stream
max_tokens = 1024 if inp.max_tokens is None else inp.max_tokens
stop = None
if inp.stop is not None:
stop = inp.stop.split('||')
new_messages = [m.dict() for m in messages]
created = get_ts()
payload = {
'model': model,
'messages': new_messages,
'temperature': temperature,
'top_p': top_p,
'stop': stop,
'max_tokens': max_tokens,
}
if inp.functions:
payload.update({'functions': inp.functions})
if verbose:
print('payload', payload)
req_type = 'chat.completion'
status_message = 'success'
choices = []
usage = Usage()
try:
resp = openai.ChatCompletion.create(**payload)
status_code = 200
choices = []
for choice in resp['choices']:
cho = Choice(**choice)
choices.append(cho)
usage = Usage(**resp['usage'])
except Exception as e:
status_code = 400
status_message = str(e)
if status_code != 200:
raise Exception(status_message)
return ChatOutput(status_code=status_code,
status_message=status_message,
model=model,
object=req_type,
created=created,
choices=choices,
usage=usage)
================================================
FILE: src/backend/bisheng_langchain/chat_models/interface/types.py
================================================
from typing import Union, List
from pydantic import BaseModel
class Message(BaseModel):
role: str
content: str
class Function(BaseModel):
name: str
description: str
parameters: dict
class ChatInput(BaseModel):
model: str
messages: List[Message] = []
top_p: float = None
temperature: float = None
n: int = 1
stream: bool = False
stop: str = None
max_tokens: int = 256
functions: List[Function] = []
function_call: str = None
class Choice(BaseModel):
index: int
message: Message = None
finish_reason: str = 'stop'
class Usage(BaseModel):
prompt_tokens: int = 0
completion_tokens: int = 0
total_tokens: int = 0
class ChatOutput(BaseModel):
status_code: int
status_message: str = 'success'
id: str = None
object: str = None
model: str = None
created: int = None
choices: List[Choice] = []
usage: Usage = None
class CompletionsInput(BaseModel):
model: str
prompt: str
top_p: float = None
temperature: float = None
n: int = 1
stream: bool = True
stop: str = None
max_tokens: int = 256
================================================
FILE: src/backend/bisheng_langchain/chat_models/interface/utils.py
================================================
import time
def get_ts():
return round(time.time() * 1000)
================================================
FILE: src/backend/bisheng_langchain/chat_models/interface/wenxin.py
================================================
import json
import requests
from .types import ChatInput, ChatOutput, Choice, Message, Usage
from .utils import get_ts
def get_access_token(api_key, sec_key):
url = (f'https://aip.baidubce.com/oauth/2.0/token?'
f'grant_type=client_credentials'
f'&client_id={api_key}&client_secret={sec_key}')
payload = json.dumps('')
headers = {
'Content-Type': 'application/json',
'Accept': 'application/json'
}
response = requests.request('POST', url, headers=headers, data=payload)
return response.json().get('access_token')
class ChatCompletion(object):
def __init__(self, api_key, sec_key, **kwargs):
self.api_key = api_key
self.sec_key = sec_key
self.ep_url = ('https://aip.baidubce.com/rpc/2.0/ai_custom/v1/'
'wenxinworkshop/chat/completions')
self.ep_url_pro = ('https://aip.baidubce.com/rpc/2.0/ai_custom/v1/'
'wenxinworkshop/chat/completions_pro')
self.ep_url_turbo = ('https://aip.baidubce.com/rpc/2.0/ai_custom/v1/'
'wenxinworkshop/chat/eb-instant')
# token = get_access_token(api_key, sec_key)
# self.endpoint = f"{self.ep_url}?access_token={token}"
self.headers = {'Content-Type': 'application/json'}
def __call__(self, inp: ChatInput, verbose=False):
messages = inp.messages
model = inp.model
top_p = 0.8 if inp.top_p is None else inp.top_p
temperature = 0.95 if inp.temperature is None else inp.temperature
stream = False if inp.stream is None else inp.stream
# max_tokens = 1024 if inp.max_tokens is None else inp.max_tokens
system_content = ''
new_messages = []
for m in messages:
role = m.role
if role == 'system':
system_content = m.content
continue
new_messages.append({'role': role, 'content': m.content})
if system_content:
new_messages[-1]['content'] = system_content + '\n' + new_messages[
-1]['content']
payload = {
'stream': stream,
'messages': new_messages,
'temperature': temperature,
'top_p': top_p
}
if verbose:
print('payload', payload)
token = get_access_token(self.api_key, self.sec_key)
endpoint = f'{self.ep_url}?access_token={token}'
if model == 'ernie-bot-turbo':
endpoint = f'{self.ep_url_turbo}?access_token={token}'
elif model == 'ernie-bot-4':
endpoint = f'{self.ep_url_pro}?access_token={token}'
response = requests.post(endpoint, headers=self.headers, json=payload)
req_type = 'chat.completion'
status_message = 'success'
status_code = response.status_code
created = get_ts()
choices = []
usage = Usage()
if status_code == 200:
try:
info = json.loads(response.text)
status_code = info.get('error_code', 200)
status_message = info.get('error_msg', status_message)
if status_code == 200:
created = info['created']
result = info['result']
finish_reason = 'default'
msg = Message(role='assistant', content=result)
choices = [
Choice(index=0,
message=msg,
finish_reason=finish_reason)
]
usage = Usage(**info['usage'])
except Exception as e:
status_code = 401
status_message = str(e)
else:
status_code = 400
status_message = 'requests error'
if status_code != 200:
raise Exception(status_message)
return ChatOutput(status_code=status_code,
status_message=status_message,
model=model,
object=req_type,
created=created,
choices=choices,
usage=usage)
================================================
FILE: src/backend/bisheng_langchain/chat_models/interface/xunfei.py
================================================
import base64
import hashlib
import hmac
import json
# import ssl
# import threading
from datetime import datetime
from time import mktime
from urllib.parse import urlencode, urlparse
from wsgiref.handlers import format_date_time
import websocket
from websocket import create_connection
import _thread as thread
from .types import ChatInput, ChatOutput, Choice, Message, Usage
from .utils import get_ts
class Ws_Param(object):
# 初始化
def __init__(self, APPID, APIKey, APISecret, gpt_url):
self.APPID = APPID
self.APIKey = APIKey
self.APISecret = APISecret
self.host = urlparse(gpt_url).netloc
self.path = urlparse(gpt_url).path
self.gpt_url = gpt_url
# 生成url
def create_url(self):
# 生成RFC1123格式的时间戳
now = datetime.now()
date = format_date_time(mktime(now.timetuple()))
# 拼接字符串
signature_origin = 'host: ' + self.host + '\n'
signature_origin += 'date: ' + date + '\n'
signature_origin += 'GET ' + self.path + ' HTTP/1.1'
# 进行hmac-sha256进行加密
signature_sha = hmac.new(
self.APISecret.encode('utf-8'),
signature_origin.encode('utf-8'),
digestmod=hashlib.sha256).digest()
signature_sha_base64 = base64.b64encode(
signature_sha).decode(encoding='utf-8')
authorization_origin = (
f'api_key="{self.APIKey}", '
f'algorithm="hmac-sha256", headers="host date request-line",'
f' signature="{signature_sha_base64}"')
authorization = base64.b64encode(
authorization_origin.encode('utf-8')).decode(encoding='utf-8')
# 将请求的鉴权参数组合为字典
v = {
'authorization': authorization,
'date': date,
'host': self.host
}
# 拼接鉴权参数，生成url
url = self.gpt_url + '?' + urlencode(v)
# 此处打印出建立连接时候的url,参考本demo的时候可取消上方打印的注释，
# 比对相同参数时生成的url与自己代码生成的url是否一致
return url
# 收到websocket错误的处理
def on_error(ws, error):
print('### error:', error)
# 收到websocket关闭的处理
def on_close(ws):
print('### closed ###')
# 收到websocket连接建立的处理
def on_open(ws):
thread.start_new_thread(run, (ws,))
def run(ws, *args):
data = json.dumps(gen_params(appid=ws.appid, question=ws.question))
ws.send(data)
# 收到websocket消息的处理
def on_message(ws, message):
print(message)
data = json.loads(message)
code = data['header']['code']
if code != 0:
print(f'请求错误: {code}, {data}')
ws.close()
else:
choices = data['payload']['choices']
status = choices['status']
content = choices['text'][0]['content']
print(content, end='')
if status == 2:
ws.close()
def gen_params(appid, question):
data = {
'header': {
'app_id': appid,
'uid': '1234'
},
'parameter': {
'chat': {
'domain': 'general',
'random_threshold': 0.5,
'max_tokens': 2048,
'auditing': 'default'
}
},
'payload': {
'message': {
'text': [
{'role': 'user', 'content': question}
]
}
}
}
return data
class ChatCompletion(object):
def __init__(self, appid, api_key, api_secret, **kwargs):
gpt_url1 = 'ws://spark-api.xf-yun.com/v1.1/chat'
gpt_url2 = 'ws://spark-api.xf-yun.com/v2.1/chat'
gpt_url3 = 'ws://spark-api.xf-yun.com/v3.1/chat'
self.wsParam1 = Ws_Param(appid, api_key, api_secret, gpt_url1)
self.wsParam2 = Ws_Param(appid, api_key, api_secret, gpt_url2)
self.wsParam3 = Ws_Param(appid, api_key, api_secret, gpt_url3)
websocket.enableTrace(False)
# todo: modify to the ws pool
# self.ws = websocket.WebSocket()
# self.ws.connect(self.wsUrl)
# self.mutex = threading.Lock()
self.header = {'app_id': appid, 'uid': 'elem'}
def __del__(self):
pass
# self.ws.close()
def __call__(self, inp: ChatInput, verbose=False):
messages = inp.messages
model = inp.model
# top_p = 0.7 if inp.top_p is None else inp.top_p
temperature = 0.5 if inp.temperature is None else inp.temperature
# stream = False if inp.stream is None else inp.stream
max_tokens = 1024 if inp.max_tokens is None else inp.max_tokens
# stop = None
# if inp.stop is not None:
# stop = inp.stop.split('||')
new_messages = []
for m in messages:
role = m.role
if role == 'system':
role = 'user'
new_messages.append({'role': role, 'content': m.content})
domain = 'general'
if model == 'spark-v2.0':
domain = 'generalv2'
elif model == 'spark-v3.0':
domain = 'generalv3'
created = get_ts()
payload = {
'header': self.header,
'payload': {'message': {'text': new_messages}},
'parameter': {
'chat': {
'domain': domain,
'temperature': temperature,
'max_tokens': max_tokens,
'auditing': 'default'
}
}
}
if verbose:
print('payload', payload)
req_type = 'chat.completion'
status_code = 200
status_message = 'success'
choices = []
usage = Usage()
texts = []
ws = None
try:
# self.mutex.acquire()
if model == 'spark-v2.0':
wsUrl = self.wsParam2.create_url()
elif model == 'spark-v3.0':
wsUrl = self.wsParam3.create_url()
else:
wsUrl = self.wsParam1.create_url()
ws = create_connection(wsUrl)
ws.send(json.dumps(payload))
texts = []
while True:
raw_data = ws.recv()
if not raw_data:
break
resp = json.loads(raw_data)
if resp['header']['code'] == 0:
texts.append(
resp['payload']['choices']['text'][0]['content'])
if resp['header']['code'] == 0 and resp['header']['status'] == 2:
usage_dict = resp['payload']['usage']['text']
usage_dict.pop('question_tokens')
usage = Usage(**usage_dict)
except Exception as e:
print('exception', e)
status_code = 401
status_message = str(e)
finally:
if ws:
ws.close()
# self.mutex.release()
if texts:
finish_reason = 'default'
msg = Message(role='assistant', content=''.join(texts))
cho = Choice(index=0, message=msg,
finish_reason=finish_reason)
choices.append(cho)
return ChatOutput(
status_code=status_code,
status_message=status_message,
model=model, object=req_type, created=created,
choices=choices, usage=usage)
================================================
FILE: src/backend/bisheng_langchain/chat_models/interface/zhipuai.py
================================================
# import json
import zhipuai
from .types import ChatInput, ChatOutput, Choice, Message, Usage
from .utils import get_ts
class ChatCompletion(object):
def __init__(self, api_key, **kwargs):
zhipuai.api_key = api_key
def __call__(self, inp: ChatInput, verbose=False):
messages = inp.messages
model = inp.model
top_p = 0.7 if inp.top_p is None else inp.top_p
temperature = 0.95 if inp.temperature is None else inp.temperature
# stream = False if inp.stream is None else inp.stream
# max_tokens = 1024 if inp.max_tokens is None else inp.max_tokens
new_messages = []
system_content = ''
for m in messages:
content = m.content
role = m.role
if role == 'system':
system_content += content
continue
new_messages.append({'role': role, 'content': content})
if system_content:
new_messages[-1]['content'] = (system_content +
new_messages[-1]['content'])
created = get_ts()
payload = {
'model': model,
'prompt': new_messages,
'temperature': temperature,
'top_p': top_p,
'request_id': str(created),
'incremental': False
}
if verbose:
print('payload', payload)
req_type = 'chat.completion'
status_message = 'success'
choices = []
usage = Usage()
try:
resp = zhipuai.model_api.invoke(**payload)
status_code = resp['code']
status_message = resp['msg']
if status_code == 200:
choices = []
for index, choice in enumerate(resp['data']['choices']):
finish_reason = 'default'
msg = Message(**choice)
cho = Choice(index=index,
message=msg,
finish_reason=finish_reason)
choices.append(cho)
usage = Usage(**resp['data']['usage'])
except Exception as e:
status_code = 400
status_message = str(e)
if status_code != 200:
raise Exception(status_message)
return ChatOutput(status_code=status_code,
status_message=status_message,
model=model,
object=req_type,
created=created,
choices=choices,
usage=usage)
================================================
FILE: src/backend/bisheng_langchain/chat_models/minimax.py
================================================
"""proxy llm chat wrapper."""
from __future__ import annotations
import logging
import sys
from typing import TYPE_CHECKING, Any, Callable, Dict, List, Mapping, Optional, Tuple, Union
# import requests
from langchain.callbacks.manager import AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun
from langchain.chat_models.base import BaseChatModel
from langchain.schema import ChatGeneration, ChatResult
from langchain.schema.messages import (AIMessage, BaseMessage, ChatMessage, FunctionMessage,
HumanMessage, SystemMessage)
from langchain.utils import get_from_dict_or_env
from pydantic import ConfigDict, model_validator, Field
from tenacity import (before_sleep_log, retry, retry_if_exception_type, stop_after_attempt,
wait_exponential)
from .interface import MinimaxChatCompletion
from .interface.types import ChatInput
if TYPE_CHECKING:
import tiktoken
logger = logging.getLogger(__name__)
def _import_tiktoken() -> Any:
try:
import tiktoken
except ImportError:
raise ValueError('Could not import tiktoken python package. '
'This is needed in order to calculate get_token_ids. '
'Please install it with `pip install tiktoken`.')
return tiktoken
def _create_retry_decorator(llm: ChatMinimaxAI) -> Callable[[Any], Any]:
min_seconds = 1
max_seconds = 20
# Wait 2^x * 1 second between each retry starting with
# 4 seconds, then up to 10 seconds, then 10 seconds afterwards
return retry(
reraise=True,
stop=stop_after_attempt(llm.max_retries),
wait=wait_exponential(multiplier=1, min=min_seconds, max=max_seconds),
retry=(retry_if_exception_type(Exception)),
before_sleep=before_sleep_log(logger, logging.WARNING),
)
def _convert_dict_to_message(_dict: Mapping[str, Any]) -> BaseMessage:
role = _dict['role']
if role == 'user':
return HumanMessage(content=_dict['content'])
elif role == 'assistant':
content = _dict['content'] or '' # OpenAI returns None for tool invocations
if _dict.get('function_call'):
additional_kwargs = {'function_call': dict(_dict['function_call'])}
else:
additional_kwargs = {}
return AIMessage(content=content, additional_kwargs=additional_kwargs)
elif role == 'system':
return SystemMessage(content=_dict['content'])
elif role == 'function':
return FunctionMessage(content=_dict['content'], name=_dict['name'])
else:
return ChatMessage(content=_dict['content'], role=role)
def _convert_message_to_dict(message: BaseMessage) -> dict:
if isinstance(message, ChatMessage):
message_dict = {'role': message.role, 'content': message.content}
elif isinstance(message, HumanMessage):
message_dict = {'role': 'user', 'content': message.content}
elif isinstance(message, AIMessage):
message_dict = {'role': 'assistant', 'content': message.content}
if 'function_call' in message.additional_kwargs:
message_dict['function_call'] = message.additional_kwargs['function_call']
elif isinstance(message, SystemMessage):
message_dict = {'role': 'system', 'content': message.content}
elif isinstance(message, FunctionMessage):
message_dict = {
'role': 'function',
'content': message.content,
'name': message.name,
}
else:
raise ValueError(f'Got unknown type {message}')
if 'name' in message.additional_kwargs:
message_dict['name'] = message.additional_kwargs['name']
return message_dict
class ChatMinimaxAI(BaseChatModel):
"""Wrapper around proxy Chat large language models.
To use, the environment variable ``ELEMAI_API_KEY`` set with your API key.
Example:
.. code-block:: python
from bisheng_langchain.chat_models import ChatMinimaxAI
chat_miniamaxai = ChatMinimaxAI(model_name="abab5.5-chat")
"""
client: Optional[Any] #: :meta private:
"""Model name to use."""
model_name: str = Field('abab5.5-chat', alias='model')
temperature: float = 0.9
top_p: float = 0.95
"""What sampling temperature to use."""
model_kwargs: Optional[Dict[str, Any]] = Field(default_factory=dict)
"""Holds any model parameters valid for `create` call not explicitly specified."""
minimaxai_api_key: Optional[str] = None
minimaxai_group_id: Optional[str] = None
headers: Optional[Dict[str, str]] = Field(default_factory=dict)
request_timeout: Optional[Union[float, Tuple[float, float]]] = None
"""Timeout for requests to OpenAI completion API. Default is 600 seconds."""
max_retries: Optional[int] = 6
"""Maximum number of retries to make when generating."""
streaming: Optional[bool] = False
"""Whether to stream the results or not."""
n: Optional[int] = 1
"""Number of chat completions to generate for each prompt."""
max_tokens: Optional[int] = 1024
"""Maximum number of tokens to generate."""
tiktoken_model_name: Optional[str] = None
"""The model name to pass to tiktoken when using this class.
Tiktoken is used to count the number of tokens in documents to constrain
them to be under a certain limit. By default, when set to None, this will
be the same as the embedding model name. However, there are some cases
where you may want to use this Embedding class with a model name not
supported by tiktoken. This can include when using Azure embeddings or
when using one of the many model providers that expose an OpenAI-like
API but with different models. In those cases, in order to avoid erroring
when tiktoken is called, you can specify a model name to use here."""
verbose: Optional[bool] = False
model_config = ConfigDict(validate_by_name=True)
@model_validator(mode='before')
@classmethod
def validate_environment(cls, values: Dict) -> Dict:
"""Validate that api key and python package exists in environment."""
values['minimaxai_api_key'] = get_from_dict_or_env(values, 'minimaxai_api_key',
'MINIMAXAI_API_KEY')
values['minimaxai_group_id'] = get_from_dict_or_env(values, 'minimaxai_group_id',
'MINIMAXAI_GROUP_ID')
api_key = values['minimaxai_api_key']
group_id = values['minimaxai_group_id']
try:
values['client'] = MinimaxChatCompletion(group_id, api_key)
except AttributeError:
raise ValueError('Try upgrading it with `pip install --upgrade requests`.')
return values
@property
def _default_params(self) -> Dict[str, Any]:
"""Get the default parameters for calling ChatMinimaxAI API."""
return {
'model': self.model_name,
'temperature': self.temperature,
'top_p': self.top_p,
'max_tokens': self.max_tokens,
**self.model_kwargs,
}
def completion_with_retry(self, **kwargs: Any) -> Any:
retry_decorator = _create_retry_decorator(self)
@retry_decorator
def _completion_with_retry(**kwargs: Any) -> Any:
messages = kwargs.get('messages')
temperature = kwargs.get('temperature')
top_p = kwargs.get('top_p')
max_tokens = kwargs.get('max_tokens')
params = {
'messages': messages,
'model': self.model_name,
'top_p': top_p,
'temperature': temperature,
'max_tokens': max_tokens
}
return self.client(ChatInput.parse_obj(params), self.verbose).dict()
return _completion_with_retry(**kwargs)
def _combine_llm_outputs(self, llm_outputs: List[Optional[dict]]) -> dict:
overall_token_usage: dict = {}
for output in llm_outputs:
if output is None:
# Happens in streaming
continue
token_usage = output['token_usage']
if token_usage is None:
continue
for k, v in token_usage.items():
if k in overall_token_usage:
overall_token_usage[k] += v
else:
overall_token_usage[k] = v
return {'token_usage': overall_token_usage, 'model_name': self.model_name}
def _generate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs}
response = self.completion_with_retry(messages=message_dicts, **params)
return self._create_chat_result(response)
async def _agenerate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
return self._generate(messages, stop, run_manager, **kwargs)
def _create_message_dicts(
self, messages: List[BaseMessage],
stop: Optional[List[str]]) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:
params = dict(self._client_params)
if stop is not None:
if 'stop' in params:
raise ValueError('`stop` found in both the input and default params.')
params['stop'] = stop
message_dicts = [_convert_message_to_dict(m) for m in messages]
return message_dicts, params
def _create_chat_result(self, response: Mapping[str, Any]) -> ChatResult:
generations = []
for res in response['choices']:
message = _convert_dict_to_message(res['message'])
gen = ChatGeneration(message=message)
generations.append(gen)
llm_output = {'token_usage': response['usage'], 'model_name': self.model_name}
return ChatResult(generations=generations, llm_output=llm_output)
@property
def _identifying_params(self) -> Mapping[str, Any]:
"""Get the identifying parameters."""
return {**{'model_name': self.model_name}, **self._default_params}
@property
def _client_params(self) -> Mapping[str, Any]:
"""Get the parameters used for the client."""
minimaxai_creds: Dict[str, Any] = {
'model': self.model_name,
}
return {**minimaxai_creds, **self._default_params}
def _get_invocation_params(self,
stop: Optional[List[str]] = None,
**kwargs: Any) -> Dict[str, Any]:
"""Get the parameters used to invoke the model FOR THE CALLBACKS."""
return {
**super()._get_invocation_params(stop=stop, **kwargs),
**self._default_params,
'model': self.model_name,
'function': kwargs.get('functions'),
}
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'minimaxai_chat'
def _get_encoding_model(self) -> Tuple[str, tiktoken.Encoding]:
tiktoken_ = _import_tiktoken()
if self.tiktoken_model_name is not None:
model = self.tiktoken_model_name
else:
model = self.model_name
# model chatglm-std, chatglm-lite
# Returns the number of tokens used by a list of messages.
try:
encoding = tiktoken_.encoding_for_model(model)
except KeyError:
logger.warning('Warning: model not found. Using cl100k_base encoding.')
model = 'cl100k_base'
encoding = tiktoken_.get_encoding(model)
return model, encoding
def get_token_ids(self, text: str) -> List[int]:
"""Get the tokens present in the text with tiktoken package."""
# tiktoken NOT supported for Python 3.7 or below
if sys.version_info[1] <= 7:
return super().get_token_ids(text)
_, encoding_model = self._get_encoding_model()
return encoding_model.encode(text)
def get_num_tokens_from_messages(self, messages: List[BaseMessage]) -> int:
"""Calculate num tokens for chatglm with tiktoken package.
todo: read chatglm document
Official documentation: https://github.com/openai/openai-cookbook/blob/
main/examples/How_to_format_inputs_to_ChatGPT_models.ipynb"""
if sys.version_info[1] <= 7:
return super().get_num_tokens_from_messages(messages)
model, encoding = self._get_encoding_model()
if model.startswith('chatglm'):
# every message follows {role/name}\n{content}\n
tokens_per_message = 4
# if there's a name, the role is omitted
tokens_per_name = -1
else:
raise NotImplementedError(
f'get_num_tokens_from_messages() is not presently implemented '
f'for model {model}.'
'See https://github.com/openai/openai-python/blob/main/chatml.md for '
'information on how messages are converted to tokens.')
num_tokens = 0
messages_dict = [_convert_message_to_dict(m) for m in messages]
for message in messages_dict:
num_tokens += tokens_per_message
for key, value in message.items():
num_tokens += len(encoding.encode(value))
if key == 'name':
num_tokens += tokens_per_name
# every reply is primed with assistant
num_tokens += 3
return num_tokens
================================================
FILE: src/backend/bisheng_langchain/chat_models/proxy_llm.py
================================================
"""proxy llm chat wrapper."""
from __future__ import annotations
import json
import logging
import sys
from typing import TYPE_CHECKING, Any, Callable, Dict, List, Mapping, Optional, Tuple, Union
from pydantic import ConfigDict, model_validator, Field
from bisheng_langchain.utils import requests
from langchain.callbacks.manager import AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun
from langchain.chat_models.base import BaseChatModel
from langchain.schema import ChatGeneration, ChatResult
from langchain.schema.messages import (AIMessage, BaseMessage, ChatMessage, FunctionMessage,
HumanMessage, SystemMessage)
from langchain.utils import get_from_dict_or_env
from tenacity import (before_sleep_log, retry, retry_if_exception_type, stop_after_attempt,
wait_exponential)
if TYPE_CHECKING:
import tiktoken
logger = logging.getLogger(__name__)
def _import_tiktoken() -> Any:
try:
import tiktoken
except ImportError:
raise ValueError('Could not import tiktoken python package. '
'This is needed in order to calculate get_token_ids. '
'Please install it with `pip install tiktoken`.')
return tiktoken
def _create_retry_decorator(llm: ProxyChatLLM) -> Callable[[Any], Any]:
min_seconds = 1
max_seconds = 20
# Wait 2^x * 1 second between each retry starting with
# 4 seconds, then up to 10 seconds, then 10 seconds afterwards
return retry(
reraise=True,
stop=stop_after_attempt(llm.max_retries),
wait=wait_exponential(multiplier=1, min=min_seconds, max=max_seconds),
retry=(retry_if_exception_type(Exception)),
before_sleep=before_sleep_log(logger, logging.WARNING),
)
def _convert_dict_to_message(_dict: Mapping[str, Any]) -> BaseMessage:
role = _dict['role']
if role == 'user':
return HumanMessage(content=_dict['content'])
elif role == 'assistant':
content = _dict['content'] or '' # OpenAI returns None for tool invocations
if _dict.get('function_call'):
additional_kwargs = {'function_call': dict(_dict['function_call'])}
else:
additional_kwargs = {}
return AIMessage(content=content, additional_kwargs=additional_kwargs)
elif role == 'system':
return SystemMessage(content=_dict['content'])
elif role == 'function':
return FunctionMessage(content=_dict['content'], name=_dict['name'])
else:
return ChatMessage(content=_dict['content'], role=role)
def _convert_message_to_dict(message: BaseMessage) -> dict:
if isinstance(message, ChatMessage):
message_dict = {'role': message.role, 'content': message.content}
elif isinstance(message, HumanMessage):
message_dict = {'role': 'user', 'content': message.content}
elif isinstance(message, AIMessage):
message_dict = {'role': 'assistant', 'content': message.content}
if 'function_call' in message.additional_kwargs:
message_dict['function_call'] = message.additional_kwargs['function_call']
elif isinstance(message, SystemMessage):
message_dict = {'role': 'system', 'content': message.content}
elif isinstance(message, FunctionMessage):
message_dict = {
'role': 'function',
'content': message.content,
'name': message.name,
}
else:
raise ValueError(f'Got unknown type {message}')
if 'name' in message.additional_kwargs:
message_dict['name'] = message.additional_kwargs['name']
return message_dict
class ProxyChatLLM(BaseChatModel):
"""Wrapper around proxy Chat large language models.
To use, the environment variable ``ELEMAI_API_KEY`` set with your API key.
Example:
.. code-block:: python
from bisheng_langchain.chat_models import ProxyChatLLM
proxy_chat_llm = ProxyChatLLM(model_name="chatglm_std")
"""
client: Optional[Any] #: :meta private:
"""Model name to use."""
model_name: str = Field('chatglm_std', alias='model')
temperature: float = 0.7
top_p: float = 0.9
"""What sampling temperature to use."""
model_kwargs: Optional[Dict[str, Any]] = Field(default_factory=dict)
"""Holds any model parameters valid for `create` call not explicitly specified."""
elemai_api_key: Optional[str] = None
elemai_base_url: Optional[str] = None
headers: Optional[Dict[str, str]] = Field(default_factory=dict)
request_timeout: Optional[Union[float, Tuple[float, float]]] = None
"""Timeout for requests to OpenAI completion API. Default is 600 seconds."""
max_retries: Optional[int] = 0
"""Maximum number of retries to make when generating."""
streaming: Optional[bool] = False
"""Whether to stream the results or not."""
n: Optional[int] = 1
"""Number of chat completions to generate for each prompt."""
max_tokens: Optional[int] = 2048
"""Maximum number of tokens to generate."""
tiktoken_model_name: Optional[str] = None
"""The model name to pass to tiktoken when using this class.
Tiktoken is used to count the number of tokens in documents to constrain
them to be under a certain limit. By default, when set to None, this will
be the same as the embedding model name. However, there are some cases
where you may want to use this Embedding class with a model name not
supported by tiktoken. This can include when using Azure embeddings or
when using one of the many model providers that expose an OpenAI-like
API but with different models. In those cases, in order to avoid erroring
when tiktoken is called, you can specify a model name to use here."""
model_config = ConfigDict(validate_by_name=True)
@model_validator(mode='before')
@classmethod
def validate_environment(cls, values: Dict) -> Dict:
"""Validate that api key and python package exists in environment."""
values['elemai_api_key'] = get_from_dict_or_env(values, 'elemai_api_key', 'ELEMAI_API_KEY')
values['elemai_base_url'] = get_from_dict_or_env(values, 'elemai_base_url',
'ELEMAI_BASE_URL')
elemai_api_key = values['elemai_api_key']
values['headers'] = {
'Authorization': f'Bearer {elemai_api_key}',
'Content-Type': 'application/json'
}
try:
values['client'] = requests.Requests(headers=values['headers'])
except AttributeError:
raise ValueError('Try upgrading it with `pip install --upgrade requests`.')
return values
@property
def _default_params(self) -> Dict[str, Any]:
"""Get the default parameters for calling ProxyChatLLM API."""
self.client.request_timeout = self.request_timeout
return {
'model': self.model_name,
'temperature': self.temperature,
'top_p': self.top_p,
'max_tokens': self.max_tokens,
**self.model_kwargs,
}
def completion_with_retry(self, **kwargs: Any) -> Any:
retry_decorator = _create_retry_decorator(self)
@retry_decorator
def _completion_with_retry(**kwargs: Any) -> Any:
messages = kwargs.get('messages')
temperature = kwargs.get('temperature')
top_p = kwargs.get('top_p')
max_tokens = kwargs.get('max_tokens')
params = {
'messages':
messages,
'model':
self.model_name,
'top_p':
top_p,
'temperature':
temperature,
'max_tokens':
max_tokens,
**({
'stop': kwargs.get('stop')
} if kwargs.get('stop', None) is not None else {}),
**({
'function_call': kwargs.get('function_call')
} if kwargs.get('function_call', None) is not None else {}),
**({
'functions': kwargs.get('functions')
} if kwargs.get('functions', None) is not None else {})
}
response = self.client.post(self.elemai_base_url, json=params)
return response.json()
rsp_dict = _completion_with_retry(**kwargs)
if 200 != rsp_dict.get('status_code', 200):
logger.error(f'proxy_llm_error resp={rsp_dict}')
raise Exception(rsp_dict)
return rsp_dict
def _combine_llm_outputs(self, llm_outputs: List[Optional[dict]]) -> dict:
overall_token_usage: dict = {}
for output in llm_outputs:
if output is None:
# Happens in streaming
continue
token_usage = output['token_usage']
for k, v in token_usage.items():
if k in overall_token_usage:
overall_token_usage[k] += v
else:
overall_token_usage[k] = v
return {'token_usage': overall_token_usage, 'model_name': self.model_name}
def _generate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs}
response = self.completion_with_retry(messages=message_dicts, **params)
return self._create_chat_result(response)
async def acompletion_with_retry(self, **kwargs: Any) -> Any:
"""Use tenacity to retry the async completion call."""
retry_decorator = _create_retry_decorator(self)
@retry_decorator
async def _acompletion_with_retry(**kwargs: Any) -> Any:
# Use OpenAI's async api https://github.com/openai/openai-python#async-api
async with self.client.apost(url=self.elemai_base_url, json=kwargs) as response:
async for txt in response.content.iter_any():
if b'\n' in txt:
for txt_ in txt.split(b'\n'):
yield txt_.decode('utf-8').strip()
else:
yield txt.decode('utf-8').strip()
async for response in _acompletion_with_retry(**kwargs):
if response:
yield json.loads(response)
async def _agenerate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs}
if self.streaming:
inner_completion = ''
role = 'assistant'
params['stream'] = True
function_call: Optional[dict] = None
async for stream_resp in self.acompletion_with_retry(messages=message_dicts, **params):
role = stream_resp['choices'][0]['delta'].get('role') or role
token = stream_resp['choices'][0]['delta'].get('content', '')
inner_completion += token or ''
_function_call = stream_resp['choices'][0]['delta'].get('function_call')
if _function_call:
if function_call is None:
function_call = _function_call
else:
function_call['arguments'] += _function_call['arguments']
if run_manager:
await run_manager.on_llm_new_token(token)
message = _convert_dict_to_message({
'content': inner_completion,
'role': role,
'function_call': function_call,
})
return ChatResult(generations=[ChatGeneration(message=message)])
else:
response = [
item
async for item in self.acompletion_with_retry(messages=message_dicts, **params)
]
return self._create_chat_result(response[0])
def _create_message_dicts(
self, messages: List[BaseMessage],
stop: Optional[List[str]]) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:
params = dict(self._client_params)
if stop is not None:
if 'stop' in params:
raise ValueError('`stop` found in both the input and default params.')
params['stop'] = stop
message_dicts = [_convert_message_to_dict(m) for m in messages]
return message_dicts, params
def _create_chat_result(self, response: Mapping[str, Any]) -> ChatResult:
generations = []
if 'choices' not in response:
raise Exception(f'LLM return error {response}')
for res in response['choices']:
message = _convert_dict_to_message(res['message'])
gen = ChatGeneration(message=message)
generations.append(gen)
llm_output = {'token_usage': response['usage'], 'model_name': self.model_name}
return ChatResult(generations=generations, llm_output=llm_output)
@property
def _identifying_params(self) -> Mapping[str, Any]:
"""Get the identifying parameters."""
return {**{'model_name': self.model_name}, **self._default_params}
@property
def _client_params(self) -> Mapping[str, Any]:
"""Get the parameters used for the elemai client."""
elemai_creds: Dict[str, Any] = {
'api_key': self.elemai_api_key,
'base_url': self.elemai_base_url,
'model': self.model_name,
}
return {**elemai_creds, **self._default_params}
def _get_invocation_params(self,
stop: Optional[List[str]] = None,
**kwargs: Any) -> Dict[str, Any]:
"""Get the parameters used to invoke the model FOR THE CALLBACKS."""
return {
**super()._get_invocation_params(stop=stop, **kwargs),
**self._default_params,
'model': self.model_name,
'function': kwargs.get('functions'),
}
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'proxy-chat'
def _get_encoding_model(self) -> Tuple[str, tiktoken.Encoding]:
tiktoken_ = _import_tiktoken()
if self.tiktoken_model_name is not None:
model = self.tiktoken_model_name
else:
model = self.model_name
# model chatglm-std, chatglm-lite
# Returns the number of tokens used by a list of messages.
try:
encoding = tiktoken_.encoding_for_model(model)
except KeyError:
logger.warning('Warning: model not found. Using cl100k_base encoding.')
model = 'cl100k_base'
encoding = tiktoken_.get_encoding(model)
return model, encoding
def get_token_ids(self, text: str) -> List[int]:
"""Get the tokens present in the text with tiktoken package."""
# tiktoken NOT supported for Python 3.7 or below
if sys.version_info[1] <= 7:
return super().get_token_ids(text)
_, encoding_model = self._get_encoding_model()
return encoding_model.encode(text)
def get_num_tokens_from_messages(self, messages: List[BaseMessage]) -> int:
"""Calculate num tokens for chatglm with tiktoken package.
todo: read chatglm document
Official documentation: https://github.com/openai/openai-cookbook/blob/
main/examples/How_to_format_inputs_to_ChatGPT_models.ipynb"""
if sys.version_info[1] <= 7:
return super().get_num_tokens_from_messages(messages)
model, encoding = self._get_encoding_model()
if model.startswith('chatglm'):
# every message follows {role/name}\n{content}\n
tokens_per_message = 4
# if there's a name, the role is omitted
tokens_per_name = -1
else:
raise NotImplementedError(
f'get_num_tokens_from_messages() is not presently implemented '
f'for model {model}.'
'See https://github.com/openai/openai-python/blob/main/chatml.md for '
'information on how messages are converted to tokens.')
num_tokens = 0
messages_dict = [_convert_message_to_dict(m) for m in messages]
for message in messages_dict:
num_tokens += tokens_per_message
for key, value in message.items():
num_tokens += len(encoding.encode(value))
if key == 'name':
num_tokens += tokens_per_name
# every reply is primed with assistant
num_tokens += 3
return num_tokens
================================================
FILE: src/backend/bisheng_langchain/chat_models/qwen.py
================================================
"""proxy llm chat wrapper."""
from __future__ import annotations
import copy
import json
import logging
import sys
from typing import TYPE_CHECKING, Any, Callable, Dict, List, Mapping, Optional, Tuple, Union
from pydantic import ConfigDict, model_validator, Field
from bisheng_langchain.utils.requests import Requests
# import requests
from langchain.callbacks.manager import AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun
from langchain.chat_models.base import BaseChatModel
from langchain.schema import ChatGeneration, ChatResult
from langchain.schema.messages import (AIMessage, BaseMessage, ChatMessage, FunctionMessage,
HumanMessage, SystemMessage, ToolMessage)
from langchain.utils import get_from_dict_or_env
from tenacity import (before_sleep_log, retry, retry_if_exception_type, stop_after_attempt,
wait_exponential)
if TYPE_CHECKING:
import tiktoken
logger = logging.getLogger(__name__)
def _import_tiktoken() -> Any:
try:
import tiktoken
except ImportError:
raise ValueError('Could not import tiktoken python package. '
'This is needed in order to calculate get_token_ids. '
'Please install it with `pip install tiktoken`.')
return tiktoken
def _create_retry_decorator(llm: ChatQWen) -> Callable[[Any], Any]:
min_seconds = 1
max_seconds = 20
# Wait 2^x * 1 second between each retry starting with
# 4 seconds, then up to 10 seconds, then 10 seconds afterwards
return retry(
reraise=True,
stop=stop_after_attempt(llm.max_retries),
wait=wait_exponential(multiplier=1, min=min_seconds, max=max_seconds),
retry=(retry_if_exception_type(Exception)),
before_sleep=before_sleep_log(logger, logging.WARNING),
)
def _convert_dict_to_message(_dict: Mapping[str, Any]) -> BaseMessage:
role = _dict['role']
if role == 'user':
return HumanMessage(content=_dict['content'])
elif role == 'assistant':
content = _dict['content'] or '' # OpenAI returns None for tool invocations
if _dict.get('function_call'):
additional_kwargs = {'function_call': dict(_dict['function_call'])}
else:
additional_kwargs = {}
if _dict.get("tool_calls"):
additional_kwargs = {'tool_calls': _dict['tool_calls']}
else:
additional_kwargs = {}
return AIMessage(content=content, additional_kwargs=additional_kwargs)
elif role == 'system':
return SystemMessage(content=_dict['content'])
elif role == 'function':
return FunctionMessage(content=_dict['content'], name=_dict['name'])
elif role == "tool":
additional_kwargs = {}
if "name" in _dict:
additional_kwargs["name"] = _dict["name"]
return ToolMessage(
content=_dict.get("content", ""),
tool_call_id=_dict.get("tool_call_id"),
additional_kwargs=additional_kwargs,
)
else:
return ChatMessage(content=_dict['content'], role=role)
def _convert_message_to_dict(message: BaseMessage) -> dict:
if isinstance(message, ChatMessage):
message_dict = {'role': message.role, 'content': message.content}
elif isinstance(message, HumanMessage):
message_dict = {'role': 'user', 'content': message.content}
elif isinstance(message, AIMessage):
message_dict = {'role': 'assistant', 'content': message.content}
if 'function_call' in message.additional_kwargs:
message_dict['function_call'] = message.additional_kwargs['function_call']
if "tool_calls" in message.additional_kwargs:
message_dict["tool_calls"] = message.additional_kwargs["tool_calls"]
elif isinstance(message, SystemMessage):
message_dict = {'role': 'system', 'content': message.content}
elif isinstance(message, FunctionMessage):
message_dict = {
'role': 'function',
'content': message.content,
'name': message.name,
}
elif isinstance(message, ToolMessage):
message_dict = {
"role": "tool",
"content": message.content,
"tool_call_id": message.tool_call_id,
}
else:
raise ValueError(f'Got unknown type {message}')
if 'name' in message.additional_kwargs:
message_dict['name'] = message.additional_kwargs['name']
return message_dict
url = 'https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generation/generation'
class ChatQWen(BaseChatModel):
"""Wrapper around proxy Chat large language models.
To use, the environment variable ``api_key`` set with your API key.
Example:
.. code-block:: python
from bisheng_langchain.chat_models import ChatQWen
chat_qwen = ChatQWen(model_name="qwen-turbo")
"""
client: Optional[Any] #: :meta private:
"""Model name to use."""
model_name: str = Field('qwen-turbo', alias='model')
temperature: float = 0.95
top_p: float = 0.8
repetition_penalty: float = 1.1
top_k: int = None
seed: Optional[int] = 1234
"""生成时，随机数的种子，用于控制模型生成的随机性。如果使用相同的种子，每次运行生成的结果都将相同；"""
"""What sampling temperature to use."""
model_kwargs: Optional[Dict[str, Any]] = Field(default_factory=dict)
"""Holds any model parameters valid for `create` call not explicitly specified."""
api_key: Optional[str] = None
request_timeout: Optional[Union[float, Tuple[float, float]]] = None
"""Timeout for requests to OpenAI completion API. Default is 600 seconds."""
max_retries: Optional[int] = 3
"""Maximum number of retries to make when generating."""
streaming: Optional[bool] = False
"""Whether to stream the results or not."""
max_tokens: Optional[int] = 1024
"""Maximum number of tokens to generate."""
result_format: Optional[str] = 'message'
"""compatibale with openai"""
tiktoken_model_name: Optional[str] = None
"""The model name to pass to tiktoken when using this class.
Tiktoken is used to count the number of tokens in documents to constrain
them to be under a certain limit. By default, when set to None, this will
be the same as the embedding model name. However, there are some cases
where you may want to use this Embedding class with a model name not
supported by tiktoken. This can include when using Azure embeddings or
when using one of the many model providers that expose an OpenAI-like
API but with different models. In those cases, in order to avoid erroring
when tiktoken is called, you can specify a model name to use here."""
verbose: Optional[bool] = False
model_config = ConfigDict(validate_by_name=True)
@model_validator(mode='before')
@classmethod
def validate_environment(cls, values: Dict) -> Dict:
"""Validate that api key and python package exists in environment."""
values['api_key'] = get_from_dict_or_env(values, 'api_key', 'QWEN_API_KEY')
api_key = values['api_key']
try:
header = {'Authorization': f'Bearer {api_key}', 'Content-Type': 'application/json'}
values['client'] = Requests(headers=header, )
except AttributeError:
raise ValueError('Try upgrading it with `pip install --upgrade requests`.')
return values
@property
def _default_params(self) -> Dict[str, Any]:
"""Get the default parameters for calling ChatWenxin API."""
return {
'temperature': self.temperature,
'top_p': self.top_p,
'top_k': self.top_k,
'max_tokens': self.max_tokens,
'seed': self.seed,
'result_format': self.result_format,
'repetition_penalty': self.repetition_penalty,
**self.model_kwargs,
}
def completion_with_retry(self, **kwargs: Any) -> Any:
retry_decorator = _create_retry_decorator(self)
self.client.headers.pop('Accept', '')
@retry_decorator
def _completion_with_retry(**kwargs: Any) -> Any:
messages = kwargs.pop('messages', '')
input, params = ChatQWen._build_input_parameters(self.model_name,
messages=messages,
**kwargs)
inp = {'input': input, 'parameters': params, 'model': self.model_name}
return self.client.post(url=url, json=inp).json()
rsp_dict = _completion_with_retry(**kwargs)
if 'code' in rsp_dict and rsp_dict['code'] == 'DataInspectionFailed':
output_res = {'choices': [{'finish_reason': 'stop', 'message': {'role': 'assistant', 'content': rsp_dict['message']}}]}
usage_res = {'total_tokens': 2, 'output_tokens': 1, 'input_tokens': 1}
return output_res, usage_res
elif 'output' not in rsp_dict:
logger.error(f'proxy_llm_error resp={rsp_dict}')
message = rsp_dict['message']
raise Exception(message)
else:
return rsp_dict['output'], rsp_dict.get('usage', '')
async def acompletion_with_retry(self, **kwargs: Any) -> Any:
"""Use tenacity to retry the async completion call."""
retry_decorator = _create_retry_decorator(self)
if self.streaming:
self.client.headers.update({'Accept': 'text/event-stream'})
else:
self.client.headers.pop('Accept', '')
@retry_decorator
async def _acompletion_with_retry(**kwargs: Any) -> Any:
messages = kwargs.pop('messages', '')
input, params = ChatQWen._build_input_parameters(self.model_name,
messages=messages,
**kwargs)
inp = {'input': input, 'parameters': params, 'model': self.model_name}
# Use OpenAI's async api https://github.com/openai/openai-python#async-api
async with self.client.apost(url=url, json=inp) as response:
async for line in response.content.iter_any():
if b'\n' in line:
for txt_ in line.split(b'\n'):
yield txt_.decode('utf-8').strip()
else:
yield line.decode('utf-8').strip()
async for response in _acompletion_with_retry(**kwargs):
is_error = False
if response:
if response.startswith('event:error'):
is_error = True
elif response.startswith('data:'):
yield (is_error, response[len('data:'):])
if is_error:
break
elif response.startswith('{'):
yield (is_error, response)
else:
continue
def _combine_llm_outputs(self, llm_outputs: List[Optional[dict]]) -> dict:
overall_token_usage: dict = {}
for output in llm_outputs:
if output is None:
# Happens in streaming
continue
token_usage = output['token_usage']
for k, v in token_usage.items():
if k in overall_token_usage:
overall_token_usage[k] += v
else:
overall_token_usage[k] = v
return {'token_usage': overall_token_usage, 'model_name': self.model_name}
def _generate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs}
output, usage = self.completion_with_retry(messages=message_dicts, **params)
return self._create_chat_result(output, usage)
async def _agenerate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs}
if self.streaming:
inner_completion = ''
role = 'assistant'
params['stream'] = True
tool_calls: Optional[list[dict]] = None
async for is_error, stream_resp in self.acompletion_with_retry(messages=message_dicts,
**params):
output = None
msg = json.loads(stream_resp)
if is_error:
logger.error(stream_resp)
raise ValueError(stream_resp)
if 'output' in msg:
output = msg['output']
choices = output.get('choices')
if choices:
for choice in choices:
role = choice['message'].get('role', role)
token = choice['message'].get('content', '')
inner_completion += token or ''
_tool_calls = choice['message'].get('tool_calls')
if run_manager:
await run_manager.on_llm_new_token(token)
if _tool_calls:
if tool_calls is None:
tool_calls = _tool_calls
else:
tool_calls[0]['arguments'] += _tool_calls[0]['arguments']
message = _convert_dict_to_message({
'content': inner_completion,
'role': role,
'tool_calls': tool_calls,
})
return ChatResult(generations=[ChatGeneration(message=message)])
else:
response = [
response
async for _, response in self.acompletion_with_retry(messages=message_dicts,
**params)
]
response = json.loads(response[0])
return self._create_chat_result(response.get('output'), response.get('usage'))
def _create_message_dicts(
self, messages: List[BaseMessage],
stop: Optional[List[str]]) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:
params = dict(self._client_params)
if stop is not None:
if 'stop' in params:
raise ValueError('`stop` found in both the input and default params.')
params['stop'] = stop
message_dicts = [_convert_message_to_dict(m) for m in messages]
return message_dicts, params
def _create_chat_result(self, response: Mapping[str, Any], usage) -> ChatResult:
generations = []
for res in response['choices']:
message = _convert_dict_to_message(res['message'])
gen = ChatGeneration(message=message)
generations.append(gen)
llm_output = {'token_usage': usage, 'model_name': self.model_name}
return ChatResult(generations=generations, llm_output=llm_output)
@property
def _identifying_params(self) -> Mapping[str, Any]:
"""Get the identifying parameters."""
return {**{'model_name': self.model_name}, **self._default_params}
@property
def _client_params(self) -> Mapping[str, Any]:
"""Get the parameters used for the client."""
return {**self._default_params}
def _get_invocation_params(self,
stop: Optional[List[str]] = None,
**kwargs: Any) -> Dict[str, Any]:
"""Get the parameters used to invoke the model FOR THE CALLBACKS."""
return {
**super()._get_invocation_params(stop=stop, **kwargs),
**self._default_params,
'model': self.model_name,
'function': kwargs.get('functions'),
}
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'qwen'
def _get_encoding_model(self) -> Tuple[str, tiktoken.Encoding]:
tiktoken_ = _import_tiktoken()
if self.tiktoken_model_name is not None:
model = self.tiktoken_model_name
else:
model = self.model_name
# model chatglm-std, chatglm-lite
# Returns the number of tokens used by a list of messages.
try:
encoding = tiktoken_.encoding_for_model(model)
except KeyError:
logger.warning('Warning: model not found. Using cl100k_base encoding.')
model = 'cl100k_base'
encoding = tiktoken_.get_encoding(model)
return model, encoding
def get_token_ids(self, text: str) -> List[int]:
"""Get the tokens present in the text with tiktoken package."""
# tiktoken NOT supported for Python 3.7 or below
if sys.version_info[1] <= 7:
return super().get_token_ids(text)
_, encoding_model = self._get_encoding_model()
return encoding_model.encode(text)
def get_num_tokens_from_messages(self, messages: List[BaseMessage]) -> int:
"""Calculate num tokens for chatglm with tiktoken package.
todo: read chatglm document
Official documentation: https://github.com/openai/openai-cookbook/blob/
main/examples/How_to_format_inputs_to_ChatGPT_models.ipynb"""
if sys.version_info[1] <= 7:
return super().get_num_tokens_from_messages(messages)
model, encoding = self._get_encoding_model()
if model.startswith('chatglm'):
# every message follows {role/name}\n{content}\n
tokens_per_message = 4
# if there's a name, the role is omitted
tokens_per_name = -1
else:
raise NotImplementedError(
f'get_num_tokens_from_messages() is not presently implemented '
f'for model {model}.'
'See https://github.com/openai/openai-python/blob/main/chatml.md for '
'information on how messages are converted to tokens.')
num_tokens = 0
messages_dict = [_convert_message_to_dict(m) for m in messages]
for message in messages_dict:
num_tokens += tokens_per_message
for key, value in message.items():
num_tokens += len(encoding.encode(value))
if key == 'name':
num_tokens += tokens_per_name
# every reply is primed with assistant
num_tokens += 3
return num_tokens
@classmethod
def _build_input_parameters(cls, model, messages, **kwargs):
parameters = {}
input = {}
if messages is not None:
msgs = copy.deepcopy(messages)
input = {'messages': msgs}
if model.startswith('qwen'):
enable_search = kwargs.pop('enable_search', False)
if enable_search:
parameters['enable_search'] = enable_search
elif model.startswith('bailian'):
customized_model_id = kwargs.pop('customized_model_id', None)
if customized_model_id is None:
raise ValueError('customized_model_id is required for %s' % model)
input['customized_model_id'] = customized_model_id
if 'incremental_output' not in kwargs and kwargs.get('stream'):
parameters['incremental_output'] = True
return input, {**parameters, **kwargs}
================================================
FILE: src/backend/bisheng_langchain/chat_models/sensetime.py
================================================
from __future__ import annotations
import ast
import json
import logging
import time
from typing import Any, Dict, List, Mapping, Optional, Tuple, Union
import jwt
from pydantic import ConfigDict, model_validator, Field
from bisheng_langchain.utils.requests import Requests
from langchain.callbacks.manager import AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun
from langchain.chat_models.base import BaseChatModel
from langchain.schema import ChatGeneration, ChatResult
from langchain.schema.messages import (AIMessage, BaseMessage, ChatMessage, FunctionMessage,
HumanMessage, SystemMessage)
from langchain.utils import get_from_dict_or_env
from tenacity import (before_sleep_log, retry, retry_if_exception_type, stop_after_attempt,
wait_exponential)
# if TYPE_CHECKING:
# import jwt
logger = logging.getLogger(__name__)
def _import_pyjwt() -> Any:
try:
import jwt
except ImportError:
raise ValueError('Could not import jwt python package. '
'This is needed in order to calculate get_token_ids. '
'Please install it with `pip install PyJWT`.')
return jwt
def encode_jwt_token(ak, sk):
headers = {'alg': 'HS256', 'typ': 'JWT'}
payload = {
'iss': ak,
'exp': int(time.time()) + 18000, # 填写您期望的有效时间，此处示例代表当前时间+300分钟
'nbf': int(time.time()) - 500 # 填写您期望的生效时间，此处示例代表当前时间-500秒
}
token = jwt.encode(payload, sk, headers=headers)
return token
def _create_retry_decorator(llm):
min_seconds = 1
max_seconds = 20
# Wait 2^x * 1 second between each retry starting with
# 4 seconds, then up to 10 seconds, then 10 seconds afterwards
return retry(
reraise=True,
stop=stop_after_attempt(llm.max_retries),
wait=wait_exponential(multiplier=1, min=min_seconds, max=max_seconds),
retry=(retry_if_exception_type(Exception)),
before_sleep=before_sleep_log(logger, logging.WARNING),
)
def _convert_dict_to_message(_dict: Mapping[str, Any]) -> BaseMessage:
role = _dict['role']
if role == 'user':
return HumanMessage(content=_dict['content'])
elif role == 'assistant':
content = _dict['content'] or '' # OpenAI returns None for tool invocations
if _dict.get('function_call'):
additional_kwargs = {'function_call': dict(_dict['function_call'])}
else:
additional_kwargs = {}
return AIMessage(content=content, additional_kwargs=additional_kwargs)
elif role == 'system':
return SystemMessage(content=_dict['content'])
elif role == 'function':
return FunctionMessage(content=_dict['content'], name=_dict['name'])
else:
return ChatMessage(content=_dict['content'], role=role)
def _convert_message_to_dict(message: BaseMessage) -> dict:
if isinstance(message, ChatMessage):
message_dict = {'role': message.role, 'content': message.content}
elif isinstance(message, HumanMessage):
message_dict = {'role': 'user', 'content': message.content}
elif isinstance(message, AIMessage):
message_dict = {'role': 'assistant', 'content': message.content}
elif isinstance(message, SystemMessage):
message_dict = {'role': 'system', 'content': message.content}
# raise ValueError(f"not support system role {message}")
elif isinstance(message, FunctionMessage):
raise ValueError(f'not support funciton {message}')
else:
raise ValueError(f'Got unknown type {message}')
# if "name" in message.additional_kwargs:
# message_dict["name"] = message.additional_kwargs["name"]
return message_dict
def _convert_message_to_dict2(message: BaseMessage) -> List[dict]:
if isinstance(message, ChatMessage):
message_dict = {'role': message.role, 'content': message.content}
elif isinstance(message, HumanMessage):
message_dict = {'role': 'user', 'content': message.content}
elif isinstance(message, AIMessage):
message_dict = {'role': 'assistant', 'content': message.content}
elif isinstance(message, SystemMessage):
raise ValueError(f'not support system role {message}')
elif isinstance(message, FunctionMessage):
raise ValueError(f'not support funciton {message}')
else:
raise ValueError(f'Got unknown type {message}')
return [message_dict]
url = 'https://api.sensenova.cn/v1/llm/chat-completions'
class SenseChat(BaseChatModel):
client: Optional[Any] #: :meta private:
model_name: str = Field(default='SenseChat', alias='model')
"""Model name to use."""
temperature: float = 0.8
top_p: float = 0.7
"""What sampling temperature to use."""
model_kwargs: Optional[Dict[str, Any]] = Field(default_factory=dict)
"""Holds any model parameters valid for `create` call not explicitly specified."""
access_key_id: Optional[str] = None
secret_access_key: Optional[str] = None
repetition_penalty: float = 1.05
request_timeout: Optional[Union[float, Tuple[float, float]]] = None
"""Timeout for requests to OpenAI completion API. Default is 600 seconds."""
max_retries: Optional[int] = 6
"""Maximum number of retries to make when generating."""
streaming: Optional[bool] = False
"""Whether to stream the results or not."""
n: Optional[int] = 1
"""Number of chat completions to generate for each prompt."""
max_tokens: Optional[int] = 1024
"""Maximum number of tokens to generate."""
tiktoken_model_name: Optional[str] = None
"""The model name to pass to tiktoken when using this class.
Tiktoken is used to count the number of tokens in documents to constrain
them to be under a certain limit. By default, when set to None, this will
be the same as the embedding model name. However, there are some cases
where you may want to use this Embedding class with a model name not
supported by tiktoken. This can include when using Azure embeddings or
when using one of the many model providers that expose an OpenAI-like
API but with different models. In those cases, in order to avoid erroring
when tiktoken is called, you can specify a model name to use here."""
verbose: Optional[bool] = False
model_config = ConfigDict(validate_by_name=True)
@model_validator(mode='before')
@classmethod
def validate_environment(cls, values: Dict) -> Dict:
"""Validate that api key and python package exists in environment."""
_import_pyjwt()
values['access_key_id'] = get_from_dict_or_env(values, 'access_key_id', 'ACCESS_KEY_ID')
values['secret_access_key'] = get_from_dict_or_env(values, 'secret_access_key',
'SECRET_ACCESS_KEY')
token = encode_jwt_token(values['access_key_id'], values['secret_access_key'])
if isinstance(token, bytes):
token = token.decode('utf-8')
try:
header = {
'Authorization': 'Bearer {}'.format(token),
'Content-Type': 'application/json'
}
values['client'] = Requests(headers=header, )
except AttributeError:
raise ValueError('Try upgrading it with `pip install --upgrade requests`.')
return values
@property
def _default_params(self) -> Dict[str, Any]:
"""Get the default parameters for calling ZhipuAI API."""
return {
'model': self.model_name,
'temperature': self.temperature,
'top_p': self.top_p,
'max_tokens': self.max_tokens,
**self.model_kwargs,
}
def completion_with_retry(self, **kwargs: Any) -> Any:
retry_decorator = _create_retry_decorator(self)
@retry_decorator
def _completion_with_retry(**kwargs: Any) -> Any:
messages = kwargs.get('messages')
temperature = kwargs.get('temperature')
top_p = kwargs.get('top_p')
# messages
params = {
'messages': messages,
'model': self.model_name,
'top_p': top_p,
'temperature': temperature,
'repetition_penalty': self.repetition_penalty,
'n': self.n,
'max_new_tokens': self.max_tokens,
'stream': False # self.streaming
}
token = encode_jwt_token(self.access_key_id, self.secret_access_key)
if isinstance(token, bytes):
token = token.decode('utf-8')
self.client.headers.update({'Authorization': 'Bearer {}'.format(token)})
response = self.client.post(url=url, json=params).json()
return response
rsp_dict = _completion_with_retry(**kwargs)
if 'error' in rsp_dict:
logger.error(f'sensechat_error resp={rsp_dict}')
message = rsp_dict['error']['message']
raise Exception(message)
else:
# return rsp_dict['data'], rsp_dict.get('usage', '')
return rsp_dict, rsp_dict.get('usage', '')
async def acompletion_with_retry(self, **kwargs: Any) -> Any:
"""Use tenacity to retry the async completion call."""
retry_decorator = _create_retry_decorator(self)
if self.streaming:
self.client.headers.update({'Accept': 'text/event-stream'})
else:
self.client.headers.pop('Accept', '')
@retry_decorator
async def _acompletion_with_retry(**kwargs: Any) -> Any:
messages = kwargs.pop('messages', '')
inp = {
'messages': messages,
'model': self.model_name,
'top_p': self.top_p,
'temperature': self.temperature,
'repetition_penalty': self.repetition_penalty,
'n': self.n,
'max_new_tokens': self.max_tokens,
'stream': True
}
# Use OpenAI's async api https://github.com/openai/openai-python#async-api
async with self.client.apost(url=url, json=inp) as response:
async for line in response.content.iter_any():
if b'\n' in line:
for txt_ in line.split(b'\n'):
yield txt_.decode('utf-8').strip()
else:
yield line.decode('utf-8').strip()
async for response in _acompletion_with_retry(**kwargs):
is_error = False
if response:
if response.startswith('event:error'):
is_error = True
elif response.startswith('data:'):
yield (is_error, response[len('data:'):])
if is_error:
break
elif response.startswith('{'):
yield (is_error, response)
else:
continue
def _combine_llm_outputs(self, llm_outputs: List[Optional[dict]]) -> dict:
overall_token_usage: dict = {}
for output in llm_outputs:
if output is None:
# Happens in streaming
continue
token_usage = output['token_usage']
for k, v in token_usage.items():
if k in overall_token_usage:
overall_token_usage[k] += v
else:
overall_token_usage[k] = v
return {'token_usage': overall_token_usage, 'model_name': self.model_name}
def _generate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs}
response, usage = self.completion_with_retry(messages=message_dicts, **params)
return self._create_chat_result(response)
async def _agenerate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs}
if self.streaming:
inner_completion = ''
role = 'user'
params['stream'] = True
function_call: Optional[dict] = None
async for is_error, stream_resp in self.acompletion_with_retry(messages=message_dicts,
**params):
if str(stream_resp).startswith('[DONE]'):
continue
output = json.loads(stream_resp)
if is_error:
logger.error(stream_resp)
raise ValueError(stream_resp)
if 'data' in output:
output = output['data']
choices = None
if 'choices' in output:
choices = output.get('choices')
if choices:
for choice in choices:
token = choice['delta']
inner_completion += token or ''
_function_call = ''
if run_manager:
await run_manager.on_llm_new_token(token)
if _function_call:
if function_call is None:
function_call = _function_call
else:
function_call['arguments'] += _function_call['arguments']
message = _convert_dict_to_message({
'content': inner_completion,
'role': role,
'function_call': function_call,
})
return ChatResult(generations=[ChatGeneration(message=message)])
else:
return self._generate(messages, stop, run_manager, **kwargs)
def _create_message_dicts(
self, messages: List[BaseMessage],
stop: Optional[List[str]]) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:
params = dict(self._client_params)
if stop is not None:
if 'stop' in params:
raise ValueError('`stop` found in both the input and default params.')
params['stop'] = stop
system_content = ''
message_dicts = []
for m in messages:
if m.type == 'system':
system_content += m.content
continue
message_dicts.extend(_convert_message_to_dict2(m))
if system_content:
message_dicts[-1]['content'] = system_content + message_dicts[-1]['content']
return message_dicts, params
def _create_chat_result(self, response: Mapping[str, Any]) -> ChatResult:
generations = []
def _norm_text(text):
if text[0] == '"' and text[-1] == '"':
out = ast.literal_eval(text)
else:
out = text
return out
for res in response['data']['choices']:
res['content'] = _norm_text(res['message'])
res['role'] = 'user'
message = _convert_dict_to_message(res)
gen = ChatGeneration(message=message)
generations.append(gen)
llm_output = {'token_usage': response['data']['usage'], 'model_name': self.model_name}
return ChatResult(generations=generations, llm_output=llm_output)
@property
def _identifying_params(self) -> Mapping[str, Any]:
"""Get the identifying parameters."""
return {**{'model_name': self.model_name}, **self._default_params}
@property
def _client_params(self) -> Mapping[str, Any]:
"""Get the parameters used for the openai client."""
zhipu_creds: Dict[str, Any] = {
'access_key_id': self.access_key_id,
'secret_access_key': self.secret_access_key,
'model': self.model_name,
}
return {**zhipu_creds, **self._default_params}
def _get_invocation_params(self,
stop: Optional[List[str]] = None,
**kwargs: Any) -> Dict[str, Any]:
"""Get the parameters used to invoke the model FOR THE CALLBACKS."""
return {
**super()._get_invocation_params(stop=stop, **kwargs),
**self._default_params,
'model': self.model_name,
'function': kwargs.get('functions'),
}
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'sense-chat'
================================================
FILE: src/backend/bisheng_langchain/chat_models/wenxin.py
================================================
"""proxy llm chat wrapper."""
from __future__ import annotations
import logging
import sys
from typing import TYPE_CHECKING, Any, Callable, Dict, List, Mapping, Optional, Tuple, Union
# import requests
from langchain.callbacks.manager import AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun
from langchain.chat_models.base import BaseChatModel
from langchain.schema import ChatGeneration, ChatResult
from langchain.schema.messages import (AIMessage, BaseMessage, ChatMessage, FunctionMessage,
HumanMessage, SystemMessage)
from langchain.utils import get_from_dict_or_env
from pydantic import ConfigDict, model_validator, Field
from tenacity import (before_sleep_log, retry, retry_if_exception_type, stop_after_attempt,
wait_exponential)
from .interface import WenxinChatCompletion
from .interface.types import ChatInput
if TYPE_CHECKING:
import tiktoken
logger = logging.getLogger(__name__)
def _import_tiktoken() -> Any:
try:
import tiktoken
except ImportError:
raise ValueError('Could not import tiktoken python package. '
'This is needed in order to calculate get_token_ids. '
'Please install it with `pip install tiktoken`.')
return tiktoken
def _create_retry_decorator(llm: ChatWenxin) -> Callable[[Any], Any]:
min_seconds = 1
max_seconds = 20
# Wait 2^x * 1 second between each retry starting with
# 4 seconds, then up to 10 seconds, then 10 seconds afterwards
return retry(
reraise=True,
stop=stop_after_attempt(llm.max_retries),
wait=wait_exponential(multiplier=1, min=min_seconds, max=max_seconds),
retry=(retry_if_exception_type(Exception)),
before_sleep=before_sleep_log(logger, logging.WARNING),
)
def _convert_dict_to_message(_dict: Mapping[str, Any]) -> BaseMessage:
role = _dict['role']
if role == 'user':
return HumanMessage(content=_dict['content'])
elif role == 'assistant':
content = _dict['content'] or '' # OpenAI returns None for tool invocations
if _dict.get('function_call'):
additional_kwargs = {'function_call': dict(_dict['function_call'])}
else:
additional_kwargs = {}
return AIMessage(content=content, additional_kwargs=additional_kwargs)
elif role == 'system':
return SystemMessage(content=_dict['content'])
elif role == 'function':
return FunctionMessage(content=_dict['content'], name=_dict['name'])
else:
return ChatMessage(content=_dict['content'], role=role)
def _convert_message_to_dict(message: BaseMessage) -> dict:
if isinstance(message, ChatMessage):
message_dict = {'role': message.role, 'content': message.content}
elif isinstance(message, HumanMessage):
message_dict = {'role': 'user', 'content': message.content}
elif isinstance(message, AIMessage):
message_dict = {'role': 'assistant', 'content': message.content}
if 'function_call' in message.additional_kwargs:
message_dict['function_call'] = message.additional_kwargs['function_call']
elif isinstance(message, SystemMessage):
message_dict = {'role': 'system', 'content': message.content}
elif isinstance(message, FunctionMessage):
message_dict = {
'role': 'function',
'content': message.content,
'name': message.name,
}
else:
raise ValueError(f'Got unknown type {message}')
if 'name' in message.additional_kwargs:
message_dict['name'] = message.additional_kwargs['name']
return message_dict
class ChatWenxin(BaseChatModel):
"""Wrapper around proxy Chat large language models.
To use, the environment variable ``ELEMAI_API_KEY`` set with your API key.
Example:
.. code-block:: python
from bisheng_langchain.chat_models import ChatWenxin
chat_miniamaxai = ChatWenxin(model_name="ernie-bot")
"""
client: Optional[Any] #: :meta private:
"""Model name to use."""
model_name: str = Field('ernie-bot', alias='model')
temperature: float = 0.95
top_p: float = 0.8
"""What sampling temperature to use."""
model_kwargs: Optional[Dict[str, Any]] = Field(default_factory=dict)
"""Holds any model parameters valid for `create` call not explicitly specified."""
wenxin_api_key: Optional[str] = None
wenxin_secret_key: Optional[str] = None
request_timeout: Optional[Union[float, Tuple[float, float]]] = None
"""Timeout for requests to OpenAI completion API. Default is 600 seconds."""
max_retries: Optional[int] = 6
"""Maximum number of retries to make when generating."""
streaming: Optional[bool] = False
"""Whether to stream the results or not."""
n: Optional[int] = 1
"""Number of chat completions to generate for each prompt."""
max_tokens: Optional[int] = 1024
"""Maximum number of tokens to generate."""
tiktoken_model_name: Optional[str] = None
"""The model name to pass to tiktoken when using this class.
Tiktoken is used to count the number of tokens in documents to constrain
them to be under a certain limit. By default, when set to None, this will
be the same as the embedding model name. However, there are some cases
where you may want to use this Embedding class with a model name not
supported by tiktoken. This can include when using Azure embeddings or
when using one of the many model providers that expose an OpenAI-like
API but with different models. In those cases, in order to avoid erroring
when tiktoken is called, you can specify a model name to use here."""
verbose: Optional[bool] = False
model_config = ConfigDict(validate_by_name=True)
@model_validator(mode='before')
@classmethod
def validate_environment(cls, values: Dict) -> Dict:
"""Validate that api key and python package exists in environment."""
values['wenxin_api_key'] = get_from_dict_or_env(values, 'wenxin_api_key', 'WENXIN_API_KEY')
values['wenxin_secret_key'] = get_from_dict_or_env(values, 'wenxin_secret_key',
'WENXIN_SECRET_KEY')
api_key = values['wenxin_api_key']
secret_key = values['wenxin_secret_key']
try:
values['client'] = WenxinChatCompletion(api_key, secret_key)
except AttributeError:
raise ValueError('Try upgrading it with `pip install --upgrade requests`.')
return values
@property
def _default_params(self) -> Dict[str, Any]:
"""Get the default parameters for calling ChatWenxin API."""
return {
'model': self.model_name,
'temperature': self.temperature,
'top_p': self.top_p,
'max_tokens': self.max_tokens,
**self.model_kwargs,
}
def completion_with_retry(self, **kwargs: Any) -> Any:
retry_decorator = _create_retry_decorator(self)
@retry_decorator
def _completion_with_retry(**kwargs: Any) -> Any:
messages = kwargs.get('messages')
temperature = kwargs.get('temperature')
top_p = kwargs.get('top_p')
max_tokens = kwargs.get('max_tokens')
params = {
'messages': messages,
'model': self.model_name,
'top_p': top_p,
'temperature': temperature,
'max_tokens': max_tokens
}
return self.client(ChatInput.parse_obj(params), self.verbose).dict()
return _completion_with_retry(**kwargs)
def _combine_llm_outputs(self, llm_outputs: List[Optional[dict]]) -> dict:
overall_token_usage: dict = {}
for output in llm_outputs:
if output is None:
# Happens in streaming
continue
token_usage = output['token_usage']
for k, v in token_usage.items():
if k in overall_token_usage:
overall_token_usage[k] += v
else:
overall_token_usage[k] = v
return {'token_usage': overall_token_usage, 'model_name': self.model_name}
def _generate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs}
response = self.completion_with_retry(messages=message_dicts, **params)
return self._create_chat_result(response)
async def _agenerate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
return self._generate(messages, stop, run_manager, **kwargs)
def _create_message_dicts(
self, messages: List[BaseMessage],
stop: Optional[List[str]]) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:
params = dict(self._client_params)
if stop is not None:
if 'stop' in params:
raise ValueError('`stop` found in both the input and default params.')
params['stop'] = stop
message_dicts = [_convert_message_to_dict(m) for m in messages]
return message_dicts, params
def _create_chat_result(self, response: Mapping[str, Any]) -> ChatResult:
generations = []
for res in response['choices']:
message = _convert_dict_to_message(res['message'])
gen = ChatGeneration(message=message)
generations.append(gen)
llm_output = {'token_usage': response['usage'], 'model_name': self.model_name}
return ChatResult(generations=generations, llm_output=llm_output)
@property
def _identifying_params(self) -> Mapping[str, Any]:
"""Get the identifying parameters."""
return {**{'model_name': self.model_name}, **self._default_params}
@property
def _client_params(self) -> Mapping[str, Any]:
"""Get the parameters used for the client."""
minimaxai_creds: Dict[str, Any] = {
'model': self.model_name,
}
return {**minimaxai_creds, **self._default_params}
def _get_invocation_params(self,
stop: Optional[List[str]] = None,
**kwargs: Any) -> Dict[str, Any]:
"""Get the parameters used to invoke the model FOR THE CALLBACKS."""
return {
**super()._get_invocation_params(stop=stop, **kwargs),
**self._default_params,
'model': self.model_name,
'function': kwargs.get('functions'),
}
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'ernie-bot-chat'
def _get_encoding_model(self) -> Tuple[str, tiktoken.Encoding]:
tiktoken_ = _import_tiktoken()
if self.tiktoken_model_name is not None:
model = self.tiktoken_model_name
else:
model = self.model_name
# model chatglm-std, chatglm-lite
# Returns the number of tokens used by a list of messages.
try:
encoding = tiktoken_.encoding_for_model(model)
except KeyError:
logger.warning('Warning: model not found. Using cl100k_base encoding.')
model = 'cl100k_base'
encoding = tiktoken_.get_encoding(model)
return model, encoding
def get_token_ids(self, text: str) -> List[int]:
"""Get the tokens present in the text with tiktoken package."""
# tiktoken NOT supported for Python 3.7 or below
if sys.version_info[1] <= 7:
return super().get_token_ids(text)
_, encoding_model = self._get_encoding_model()
return encoding_model.encode(text)
def get_num_tokens_from_messages(self, messages: List[BaseMessage]) -> int:
"""Calculate num tokens for chatglm with tiktoken package.
todo: read chatglm document
Official documentation: https://github.com/openai/openai-cookbook/blob/
main/examples/How_to_format_inputs_to_ChatGPT_models.ipynb"""
if sys.version_info[1] <= 7:
return super().get_num_tokens_from_messages(messages)
model, encoding = self._get_encoding_model()
if model.startswith('chatglm'):
# every message follows {role/name}\n{content}\n
tokens_per_message = 4
# if there's a name, the role is omitted
tokens_per_name = -1
else:
raise NotImplementedError(
f'get_num_tokens_from_messages() is not presently implemented '
f'for model {model}.'
'See https://github.com/openai/openai-python/blob/main/chatml.md for '
'information on how messages are converted to tokens.')
num_tokens = 0
messages_dict = [_convert_message_to_dict(m) for m in messages]
for message in messages_dict:
num_tokens += tokens_per_message
for key, value in message.items():
num_tokens += len(encoding.encode(value))
if key == 'name':
num_tokens += tokens_per_name
# every reply is primed with assistant
num_tokens += 3
return num_tokens
================================================
FILE: src/backend/bisheng_langchain/chat_models/xunfeiai.py
================================================
"""proxy llm chat wrapper."""
from __future__ import annotations
import logging
import sys
from typing import TYPE_CHECKING, Any, Callable, Dict, List, Mapping, Optional, Tuple, Union
# import requests
from langchain.callbacks.manager import AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun
from langchain.chat_models.base import BaseChatModel
from langchain.schema import ChatGeneration, ChatResult
from langchain.schema.messages import (AIMessage, BaseMessage, ChatMessage, FunctionMessage,
HumanMessage, SystemMessage)
from langchain.utils import get_from_dict_or_env
from pydantic import ConfigDict, model_validator, Field
from tenacity import (before_sleep_log, retry, retry_if_exception_type, stop_after_attempt,
wait_exponential)
from .interface import XunfeiChatCompletion
from .interface.types import ChatInput
if TYPE_CHECKING:
import tiktoken
logger = logging.getLogger(__name__)
def _import_tiktoken() -> Any:
try:
import tiktoken
except ImportError:
raise ValueError('Could not import tiktoken python package. '
'This is needed in order to calculate get_token_ids. '
'Please install it with `pip install tiktoken`.')
return tiktoken
def _create_retry_decorator(llm: ChatXunfeiAI) -> Callable[[Any], Any]:
min_seconds = 1
max_seconds = 20
# Wait 2^x * 1 second between each retry starting with
# 4 seconds, then up to 10 seconds, then 10 seconds afterwards
return retry(
reraise=True,
stop=stop_after_attempt(llm.max_retries),
wait=wait_exponential(multiplier=1, min=min_seconds, max=max_seconds),
retry=(retry_if_exception_type(Exception)),
before_sleep=before_sleep_log(logger, logging.WARNING),
)
def _convert_dict_to_message(_dict: Mapping[str, Any]) -> BaseMessage:
role = _dict['role']
if role == 'user':
return HumanMessage(content=_dict['content'])
elif role == 'assistant':
content = _dict[
'content'] or '' # OpenAI returns None for tool invocations
if _dict.get('function_call'):
additional_kwargs = {'function_call': dict(_dict['function_call'])}
else:
additional_kwargs = {}
return AIMessage(content=content, additional_kwargs=additional_kwargs)
elif role == 'system':
return SystemMessage(content=_dict['content'])
elif role == 'function':
return FunctionMessage(content=_dict['content'], name=_dict['name'])
else:
return ChatMessage(content=_dict['content'], role=role)
def _convert_message_to_dict(message: BaseMessage) -> dict:
if isinstance(message, ChatMessage):
message_dict = {'role': message.role, 'content': message.content}
elif isinstance(message, HumanMessage):
message_dict = {'role': 'user', 'content': message.content}
elif isinstance(message, AIMessage):
message_dict = {'role': 'assistant', 'content': message.content}
if 'function_call' in message.additional_kwargs:
message_dict['function_call'] = message.additional_kwargs[
'function_call']
elif isinstance(message, SystemMessage):
message_dict = {'role': 'system', 'content': message.content}
elif isinstance(message, FunctionMessage):
message_dict = {
'role': 'function',
'content': message.content,
'name': message.name,
}
else:
raise ValueError(f'Got unknown type {message}')
if 'name' in message.additional_kwargs:
message_dict['name'] = message.additional_kwargs['name']
return message_dict
class ChatXunfeiAI(BaseChatModel):
"""Wrapper around proxy Chat large language models.
To use, the environment variable ``ELEMAI_API_KEY`` set with your API key.
Example:
.. code-block:: python
from bisheng_langchain.chat_models import ChatXunfeiAI
chat_miniamaxai = ChatXunfeiAI(model_name="spark")
"""
client: Optional[Any] #: :meta private:
"""Model name to use."""
model_name: str = Field('spark-1.5', alias='model')
temperature: float = 0.5
top_p: float = 0.7
"""What sampling temperature to use."""
model_kwargs: Optional[Dict[str, Any]] = Field(default_factory=dict)
"""Holds any model parameters valid for `create` call not explicitly specified."""
xunfeiai_appid: Optional[str] = None
xunfeiai_api_key: Optional[str] = None
xunfeiai_api_secret: Optional[str] = None
request_timeout: Optional[Union[float, Tuple[float, float]]] = None
"""Timeout for requests to OpenAI completion API. Default is 600 seconds."""
max_retries: Optional[int] = 6
"""Maximum number of retries to make when generating."""
streaming: Optional[bool] = False
"""Whether to stream the results or not."""
n: Optional[int] = 1
"""Number of chat completions to generate for each prompt."""
max_tokens: Optional[int] = 1024
"""Maximum number of tokens to generate."""
tiktoken_model_name: Optional[str] = None
"""The model name to pass to tiktoken when using this class.
Tiktoken is used to count the number of tokens in documents to constrain
them to be under a certain limit. By default, when set to None, this will
be the same as the embedding model name. However, there are some cases
where you may want to use this Embedding class with a model name not
supported by tiktoken. This can include when using Azure embeddings or
when using one of the many model providers that expose an OpenAI-like
API but with different models. In those cases, in order to avoid erroring
when tiktoken is called, you can specify a model name to use here."""
verbose: Optional[bool] = False
model_config = ConfigDict(validate_by_name=True)
@model_validator(mode='before')
@classmethod
def validate_environment(cls, values: Dict) -> Dict:
"""Validate that api key and python package exists in environment."""
values['xunfeiai_appid'] = get_from_dict_or_env(
values, 'xunfeiai_appid', 'XUNFEIAI_APPID')
values['xunfeiai_api_key'] = get_from_dict_or_env(
values, 'xunfeiai_api_key', 'XUNFEIAI_API_KEY')
values['xunfeiai_api_secret'] = get_from_dict_or_env(
values, 'xunfeiai_api_secret', 'XUNFEIAI_API_SECRET')
appid = values['xunfeiai_appid']
api_key = values['xunfeiai_api_key']
api_secret = values['xunfeiai_api_secret']
try:
values['client'] = XunfeiChatCompletion(appid, api_key, api_secret)
except AttributeError:
raise ValueError(
'Try upgrading it with `pip install --upgrade requests`.')
return values
@property
def _default_params(self) -> Dict[str, Any]:
"""Get the default parameters for calling ChatXunfeiAI API."""
return {
'model': self.model_name,
'temperature': self.temperature,
'top_p': self.top_p,
'max_tokens': self.max_tokens,
**self.model_kwargs,
}
def completion_with_retry(self, **kwargs: Any) -> Any:
retry_decorator = _create_retry_decorator(self)
@retry_decorator
def _completion_with_retry(**kwargs: Any) -> Any:
messages = kwargs.get('messages')
temperature = kwargs.get('temperature')
top_p = kwargs.get('top_p')
max_tokens = kwargs.get('max_tokens')
params = {
'messages': messages,
'model': self.model_name,
'top_p': top_p,
'temperature': temperature,
'max_tokens': max_tokens
}
return self.client(ChatInput.parse_obj(params),
self.verbose).dict()
return _completion_with_retry(**kwargs)
def _combine_llm_outputs(self, llm_outputs: List[Optional[dict]]) -> dict:
overall_token_usage: dict = {}
for output in llm_outputs:
if output is None:
# Happens in streaming
continue
token_usage = output['token_usage']
for k, v in token_usage.items():
if k in overall_token_usage:
overall_token_usage[k] += v
else:
overall_token_usage[k] = v
return {
'token_usage': overall_token_usage,
'model_name': self.model_name
}
def _generate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs}
response = self.completion_with_retry(messages=message_dicts, **params)
return self._create_chat_result(response)
async def _agenerate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
return self._generate(messages, stop, run_manager, **kwargs)
def _create_message_dicts(
self, messages: List[BaseMessage], stop: Optional[List[str]]
) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:
params = dict(self._client_params)
if stop is not None:
if 'stop' in params:
raise ValueError(
'`stop` found in both the input and default params.')
params['stop'] = stop
message_dicts = [_convert_message_to_dict(m) for m in messages]
return message_dicts, params
def _create_chat_result(self, response: Mapping[str, Any]) -> ChatResult:
generations = []
for res in response['choices']:
message = _convert_dict_to_message(res['message'])
gen = ChatGeneration(message=message)
generations.append(gen)
llm_output = {
'token_usage': response['usage'],
'model_name': self.model_name
}
return ChatResult(generations=generations, llm_output=llm_output)
@property
def _identifying_params(self) -> Mapping[str, Any]:
"""Get the identifying parameters."""
return {**{'model_name': self.model_name}, **self._default_params}
@property
def _client_params(self) -> Mapping[str, Any]:
"""Get the parameters used for the client."""
minimaxai_creds: Dict[str, Any] = {
'model': self.model_name,
}
return {**minimaxai_creds, **self._default_params}
def _get_invocation_params(self,
stop: Optional[List[str]] = None,
**kwargs: Any) -> Dict[str, Any]:
"""Get the parameters used to invoke the model FOR THE CALLBACKS."""
return {
**super()._get_invocation_params(stop=stop, **kwargs),
**self._default_params,
'model': self.model_name,
'function': kwargs.get('functions'),
}
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'xunfeiai_chat'
def _get_encoding_model(self) -> Tuple[str, tiktoken.Encoding]:
tiktoken_ = _import_tiktoken()
if self.tiktoken_model_name is not None:
model = self.tiktoken_model_name
else:
model = self.model_name
# model chatglm-std, chatglm-lite
# Returns the number of tokens used by a list of messages.
try:
encoding = tiktoken_.encoding_for_model(model)
except KeyError:
logger.warning(
'Warning: model not found. Using cl100k_base encoding.')
model = 'cl100k_base'
encoding = tiktoken_.get_encoding(model)
return model, encoding
def get_token_ids(self, text: str) -> List[int]:
"""Get the tokens present in the text with tiktoken package."""
# tiktoken NOT supported for Python 3.7 or below
if sys.version_info[1] <= 7:
return super().get_token_ids(text)
_, encoding_model = self._get_encoding_model()
return encoding_model.encode(text)
def get_num_tokens_from_messages(self, messages: List[BaseMessage]) -> int:
"""Calculate num tokens for chatglm with tiktoken package.
todo: read chatglm document
Official documentation: https://github.com/openai/openai-cookbook/blob/
main/examples/How_to_format_inputs_to_ChatGPT_models.ipynb"""
if sys.version_info[1] <= 7:
return super().get_num_tokens_from_messages(messages)
model, encoding = self._get_encoding_model()
if model.startswith('spark'):
# every message follows {role/name}\n{content}\n
tokens_per_message = 4
# if there's a name, the role is omitted
tokens_per_name = -1
else:
raise NotImplementedError(
f'get_num_tokens_from_messages() is not presently implemented '
f'for model {model}.'
'See https://github.com/openai/openai-python/blob/main/chatml.md for '
'information on how messages are converted to tokens.')
num_tokens = 0
messages_dict = [_convert_message_to_dict(m) for m in messages]
for message in messages_dict:
num_tokens += tokens_per_message
for key, value in message.items():
num_tokens += len(encoding.encode(value))
if key == 'name':
num_tokens += tokens_per_name
# every reply is primed with assistant
num_tokens += 3
return num_tokens
================================================
FILE: src/backend/bisheng_langchain/chat_models/zhipuai.py
================================================
"""ZhipuAI chat wrapper."""
from __future__ import annotations
import ast
import logging
import sys
from typing import TYPE_CHECKING, Any, Callable, Dict, List, Mapping, Optional, Tuple, Union
from langchain.callbacks.manager import AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun
from langchain.chat_models.base import BaseChatModel
from langchain.schema import ChatGeneration, ChatResult
from langchain.schema.messages import (AIMessage, BaseMessage, ChatMessage, FunctionMessage,
HumanMessage, SystemMessage)
from langchain.utils import get_from_dict_or_env
from pydantic import ConfigDict, model_validator, Field
from tenacity import (before_sleep_log, retry, retry_if_exception_type, stop_after_attempt,
wait_exponential)
if TYPE_CHECKING:
import tiktoken
logger = logging.getLogger(__name__)
def _import_tiktoken() -> Any:
try:
import tiktoken
except ImportError:
raise ValueError('Could not import tiktoken python package. '
'This is needed in order to calculate get_token_ids. '
'Please install it with `pip install tiktoken`.')
return tiktoken
def _create_retry_decorator(llm: ChatZhipuAI) -> Callable[[Any], Any]:
min_seconds = 1
max_seconds = 20
# Wait 2^x * 1 second between each retry starting with
# 4 seconds, then up to 10 seconds, then 10 seconds afterwards
return retry(
reraise=True,
stop=stop_after_attempt(llm.max_retries),
wait=wait_exponential(multiplier=1, min=min_seconds, max=max_seconds),
retry=(retry_if_exception_type(Exception)),
before_sleep=before_sleep_log(logger, logging.WARNING),
)
def _convert_dict_to_message(_dict: Mapping[str, Any]) -> BaseMessage:
role = _dict['role']
if role == 'user':
return HumanMessage(content=_dict['content'])
elif role == 'assistant':
content = _dict['content'] or '' # OpenAI returns None for tool invocations
if _dict.get('function_call'):
additional_kwargs = {'function_call': dict(_dict['function_call'])}
else:
additional_kwargs = {}
return AIMessage(content=content, additional_kwargs=additional_kwargs)
elif role == 'system':
return SystemMessage(content=_dict['content'])
elif role == 'function':
return FunctionMessage(content=_dict['content'], name=_dict['name'])
else:
return ChatMessage(content=_dict['content'], role=role)
def _convert_message_to_dict(message: BaseMessage) -> dict:
if isinstance(message, ChatMessage):
message_dict = {'role': message.role, 'content': message.content}
elif isinstance(message, HumanMessage):
message_dict = {'role': 'user', 'content': message.content}
elif isinstance(message, AIMessage):
message_dict = {'role': 'assistant', 'content': message.content}
elif isinstance(message, SystemMessage):
message_dict = {'role': 'system', 'content': message.content}
# raise ValueError(f"not support system role {message}")
elif isinstance(message, FunctionMessage):
raise ValueError(f'not support funciton {message}')
else:
raise ValueError(f'Got unknown type {message}')
# if "name" in message.additional_kwargs:
# message_dict["name"] = message.additional_kwargs["name"]
return message_dict
def _convert_message_to_dict2(message: BaseMessage) -> List[dict]:
if isinstance(message, ChatMessage):
message_dict = {'role': message.role, 'content': message.content}
elif isinstance(message, HumanMessage):
message_dict = {'role': 'user', 'content': message.content}
elif isinstance(message, AIMessage):
message_dict = {'role': 'assistant', 'content': message.content}
elif isinstance(message, SystemMessage):
raise ValueError(f'not support system role {message}')
elif isinstance(message, FunctionMessage):
raise ValueError(f'not support funciton {message}')
else:
raise ValueError(f'Got unknown type {message}')
return [message_dict]
class ChatZhipuAI(BaseChatModel):
"""Wrapper around ZhipuAI Chat large language models.
To use, you should have the ``zhipuai`` python package installed, and the
environment variable ``ZHIPU_API_KEY`` set with your API key.
Example:
.. code-block:: python
from lib.zhipuai import ChatZhipuAI
chat_zhipu = ChatZhipu(model_name="chatglm-std")
"""
client: Optional[Any] #: :meta private:
model_name: str = Field(default='chatglm_std', alias='model')
"""Model name to use."""
temperature: float = 0.95
top_p: float = 0.7
"""What sampling temperature to use."""
model_kwargs: Optional[Dict[str, Any]] = Field(default_factory=dict)
"""Holds any model parameters valid for `create` call not explicitly specified."""
zhipuai_api_key: Optional[str] = None
request_timeout: Optional[Union[float, Tuple[float, float]]] = None
"""Timeout for requests to OpenAI completion API. Default is 600 seconds."""
max_retries: Optional[int] = 6
"""Maximum number of retries to make when generating."""
streaming: Optional[bool] = False
"""Whether to stream the results or not."""
n: Optional[int] = 1
"""Number of chat completions to generate for each prompt."""
max_tokens: Optional[int] = 1024
"""Maximum number of tokens to generate."""
tiktoken_model_name: Optional[str] = None
"""The model name to pass to tiktoken when using this class.
Tiktoken is used to count the number of tokens in documents to constrain
them to be under a certain limit. By default, when set to None, this will
be the same as the embedding model name. However, there are some cases
where you may want to use this Embedding class with a model name not
supported by tiktoken. This can include when using Azure embeddings or
when using one of the many model providers that expose an OpenAI-like
API but with different models. In those cases, in order to avoid erroring
when tiktoken is called, you can specify a model name to use here."""
verbose: Optional[bool] = False
model_config = ConfigDict(validate_by_name=True)
@model_validator(mode='before')
@classmethod
def validate_environment(cls, values: Dict) -> Dict:
"""Validate that api key and python package exists in environment."""
values['zhipuai_api_key'] = get_from_dict_or_env(values, 'zhipuai_api_key',
'ZHIPUAI_API_KEY')
try:
import zhipuai
zhipuai.api_key = values['zhipuai_api_key']
except ImportError:
raise ValueError('Could not import openai python package. '
'Please install it with `pip install zhipuai`.')
try:
values['client'] = zhipuai.model_api.invoke
except AttributeError:
raise ValueError('`openai` has no `ChatCompletion` attribute, this is likely '
'due to an old version of the zhipuai package. Try upgrading it '
'with `pip install --upgrade zhipuai`.')
return values
@property
def _default_params(self) -> Dict[str, Any]:
"""Get the default parameters for calling ZhipuAI API."""
return {
'model': self.model_name,
'temperature': self.temperature,
'top_p': self.top_p,
'max_tokens': self.max_tokens,
**self.model_kwargs,
}
def completion_with_retry(self, **kwargs: Any) -> Any:
retry_decorator = _create_retry_decorator(self)
@retry_decorator
def _completion_with_retry(**kwargs: Any) -> Any:
messages = kwargs.get('messages')
temperature = kwargs.get('temperature')
top_p = kwargs.get('top_p')
params = {
'prompt': messages,
'model': self.model_name,
'top_p': top_p,
'temperature': temperature,
'incremental': False
}
return self.client(**params)
return _completion_with_retry(**kwargs)
def _combine_llm_outputs(self, llm_outputs: List[Optional[dict]]) -> dict:
overall_token_usage: dict = {}
for output in llm_outputs:
if output is None:
# Happens in streaming
continue
token_usage = output['token_usage']
for k, v in token_usage.items():
if k in overall_token_usage:
overall_token_usage[k] += v
else:
overall_token_usage[k] = v
return {'token_usage': overall_token_usage, 'model_name': self.model_name}
def _generate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs}
response = self.completion_with_retry(messages=message_dicts, **params)
if response['code'] == 1301:
response = {'code': 200,
'msg': '操作成功',
'data': {'request_id': '0', 'task_id': '0', 'task_status': 'SUCCESS',
'choices': [{'role': 'assistant', 'content': '""'}],
'usage': {'prompt_tokens': 0, 'completion_tokens': 0, 'total_tokens': 0}},
'success': True}
return self._create_chat_result(response)
async def _agenerate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> ChatResult:
return self._generate(messages, stop, run_manager, **kwargs)
def _create_message_dicts(
self, messages: List[BaseMessage],
stop: Optional[List[str]]) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:
params = dict(self._client_params)
if stop is not None:
if 'stop' in params:
raise ValueError('`stop` found in both the input and default params.')
params['stop'] = stop
system_content = ''
message_dicts = []
for m in messages:
if m.type == 'system':
system_content += m.content
continue
message_dicts.extend(_convert_message_to_dict2(m))
if system_content:
message_dicts[-1]['content'] = system_content + message_dicts[-1]['content']
return message_dicts, params
def _create_chat_result(self, response: Mapping[str, Any]) -> ChatResult:
generations = []
# print('response', response)
def _norm_text(text):
if text[0] == '"' and text[-1] == '"':
out = ast.literal_eval(text)
else:
out = text
return out
for res in response['data']['choices']:
res['content'] = _norm_text(res['content'])
message = _convert_dict_to_message(res)
gen = ChatGeneration(message=message)
generations.append(gen)
llm_output = {'token_usage': response['data']['usage'], 'model_name': self.model_name}
return ChatResult(generations=generations, llm_output=llm_output)
@property
def _identifying_params(self) -> Mapping[str, Any]:
"""Get the identifying parameters."""
return {**{'model_name': self.model_name}, **self._default_params}
@property
def _client_params(self) -> Mapping[str, Any]:
"""Get the parameters used for the openai client."""
zhipu_creds: Dict[str, Any] = {
'api_key': self.zhipuai_api_key,
'model': self.model_name,
}
return {**zhipu_creds, **self._default_params}
def _get_invocation_params(self,
stop: Optional[List[str]] = None,
**kwargs: Any) -> Dict[str, Any]:
"""Get the parameters used to invoke the model FOR THE CALLBACKS."""
return {
**super()._get_invocation_params(stop=stop, **kwargs),
**self._default_params,
'model': self.model_name,
'function': kwargs.get('functions'),
}
@property
def _llm_type(self) -> str:
"""Return type of chat model."""
return 'zhipu-chat'
def _get_encoding_model(self) -> Tuple[str, tiktoken.Encoding]:
tiktoken_ = _import_tiktoken()
if self.tiktoken_model_name is not None:
model = self.tiktoken_model_name
else:
model = self.model_name
# model chatglm-std, chatglm-lite
# Returns the number of tokens used by a list of messages.
try:
encoding = tiktoken_.encoding_for_model(model)
except KeyError:
logger.warning('Warning: model not found. Using cl100k_base encoding.')
model = 'cl100k_base'
encoding = tiktoken_.get_encoding(model)
return model, encoding
def get_token_ids(self, text: str) -> List[int]:
"""Get the tokens present in the text with tiktoken package."""
# tiktoken NOT supported for Python 3.7 or below
if sys.version_info[1] <= 7:
return super().get_token_ids(text)
_, encoding_model = self._get_encoding_model()
return encoding_model.encode(text)
def get_num_tokens_from_messages(self, messages: List[BaseMessage]) -> int:
"""Calculate num tokens for chatglm with tiktoken package.
todo: read chatglm document
Official documentation: https://github.com/openai/openai-cookbook/blob/
main/examples/How_to_format_inputs_to_ChatGPT_models.ipynb"""
if sys.version_info[1] <= 7:
return super().get_num_tokens_from_messages(messages)
model, encoding = self._get_encoding_model()
if model.startswith('chatglm'):
# every message follows {role/name}\n{content}\n
tokens_per_message = 4
# if there's a name, the role is omitted
tokens_per_name = -1
else:
raise NotImplementedError(
f'get_num_tokens_from_messages() is not presently implemented '
f'for model {model}.'
'See https://github.com/openai/openai-python/blob/main/chatml.md for '
'information on how messages are converted to tokens.')
num_tokens = 0
messages_dict = [_convert_message_to_dict(m) for m in messages]
for message in messages_dict:
num_tokens += tokens_per_message
for key, value in message.items():
num_tokens += len(encoding.encode(value))
if key == 'name':
num_tokens += tokens_per_name
# every reply is primed with assistant
num_tokens += 3
return num_tokens
================================================
FILE: src/backend/bisheng_langchain/document_loaders/__init__.py
================================================
from .custom_kv import CustomKVLoader
from .elem_pdf import PDFWithSemanticLoader
from .elem_unstrcutured_loader import ElemUnstructuredLoader, ElemUnstructuredLoaderV0
from .universal_kv import UniversalKVLoader
__all__ = [
'PDFWithSemanticLoader',
'ElemUnstructuredLoader',
'ElemUnstructuredLoaderV0',
'UniversalKVLoader',
'CustomKVLoader',
]
================================================
FILE: src/backend/bisheng_langchain/document_loaders/custom_kv.py
================================================
# flake8: noqa
"""Loads PDF with semantic splilter."""
import base64
import json
import logging
import os
import re
import tempfile
from pathlib import Path
from time import sleep
from typing import List, Optional, Tuple, Union
from urllib.parse import quote_plus, unquote, urlparse
import cv2
import fitz
import numpy as np
from bisheng_langchain.utils.requests import Requests
from langchain_community.docstore.document import Document
from langchain_community.document_loaders.base import BaseLoader
from PIL import Image
logger = logging.getLogger(__name__)
def convert_base64(image):
image_binary = cv2.imencode('.jpg', image)[1].tobytes()
x = base64.b64encode(image_binary)
return x.decode('ascii').replace('\n', '')
def transpdf2png(pdf_file):
pdf_bytes = open(pdf_file, 'rb').read()
pdf = fitz.Document(stream=pdf_bytes, filetype='pdf')
dpis = [72, 144, 200]
pdf_images = dict()
for page in pdf:
pix = None
for dpi in dpis:
pix = page.get_pixmap(dpi=dpi)
if min(pix.width, pix.height) >= 1600: break
mode = 'RGBA' if pix.alpha else 'RGB'
img = Image.frombytes(mode, [pix.width, pix.height], pix.samples)
# RGB to BGR
img = np.array(img)[:, :, ::-1]
img_name = 'page_{:03d}'.format(page.number)
pdf_images[img_name] = img
return pdf_images
class CustomKVLoader(BaseLoader):
"""Extract key-value from pdf or image.
"""
def __init__(self,
file_path: str,
elm_api_base_url: str,
elm_api_key: str,
schemas: str,
elem_server_id: str,
task_type: str,
request_timeout: Optional[Union[float, Tuple[float, float]]] = 30) -> None:
"""Initialize with a file path."""
self.file_path = file_path
self.elm_api_base_url = elm_api_base_url
self.elm_api_key = elm_api_key
self.elem_server_id = elem_server_id
self.task_type = task_type
self.schemas = set(schemas.split('|'))
self.headers = {'Authorization': f'Bearer {elm_api_key}'}
self.requests = Requests(headers=self.headers, request_timeout=request_timeout)
if '~' in self.file_path:
self.file_path = os.path.expanduser(self.file_path)
# If the file is a web path, download it to a temporary file, and use that
if not os.path.isfile(self.file_path) and self._is_valid_url(self.file_path):
r = self.requests.get(self.file_path)
if r.status_code != 200:
raise ValueError('Check the url of your file; returned status code %s' %
r.status_code)
self.temp_dir = tempfile.TemporaryDirectory()
temp_file = Path(self.temp_dir.name) / unquote(
urlparse(self.file_path).path.split('/')[-1])
with open(temp_file, mode='wb') as f:
f.write(r.content)
self.file_path = str(temp_file)
elif not os.path.isfile(self.file_path):
raise ValueError('File path %s is not a valid file or url' % self.file_path)
super().__init__()
@staticmethod
def _is_valid_url(url: str) -> bool:
"""Check if the url is valid."""
parsed = urlparse(url)
return bool(parsed.netloc) and bool(parsed.scheme)
def load(self) -> List[Document]:
"""Load given path as pages."""
# mime_type = filetype.guess(self.file_path).mime
# if mime_type.endswith('pdf'):
# file_type = 'pdf'
# elif mime_type.startswith('image'):
# file_type = 'img'
# else:
# raise ValueError(f'file type {file_type} is not support.')
with open(self.file_path, 'rb') as file:
file = {'file': open(self.file_path, 'rb')}
result = {}
if self.task_type == 'extraction-job':
url = self.elm_api_base_url + '/task'
# 创建task
body = {'scene_id': self.elem_server_id}
elif self.task_type == 'logic-job':
url = self.elm_api_base_url + '/logic-job'
body = {'logic_service_id': self.elem_server_id}
resp = self.requests.post(url=url, json={}, data=body, files=file)
if resp.status_code == 200:
task_id = resp.json().get('data').get('task_id')
if not task_id:
logger.error(f'task_create_fail res={resp.text}')
return
# get status
status_url = url + f'/status?task_id={task_id}'
count = 0
while True:
status = self.requests.get(status_url).json()
if 1 == status.get('data').get('status'):
count += 1
sleep(2)
elif 3 == status.get('data').get('status'):
# 失败
logger.error(f'custom_kv type={self.task_type} resp={status}')
return []
else:
break
# get result
job_id = 'job_id' if self.task_type == 'logic-job' else 'task_id'
match = re.match(r'^(?:https?:\/\/)?(?:www\.)?([^\/\n]+)', self.elm_api_base_url)
detail_url = quote_plus(match.group() + f'/logic-job-detail/{task_id}')
result_url = url + f'/result?{job_id}={task_id}&detail_url={detail_url}'
result = self.requests.get(result_url).json()
# only for independent key
document_result = {}
try:
result = self.requests.get(result_url).json()
file_reuslt = result.get('data')
for result in file_reuslt:
independent = result.get('result').get('independent_list')
for element in independent:
if element.get('element_name') in self.schemas:
document_result[element.get('element_name')] = [
el.get('words') for el in element.get('entity_list')
]
except Exception as e:
logger.error(
f'task_result_error scene_id={self.elem_server_id} res={result} except={str(e)}'
)
raise Exception('custom_kv parse_error')
else:
logger.error(f'custom_kv=create_task resp={resp.text}')
raise Exception('custom_kv create task fail')
content = json.dumps(document_result)
doc = Document(page_content=content)
return [doc]
================================================
FILE: src/backend/bisheng_langchain/document_loaders/elem_html.py
================================================
================================================
FILE: src/backend/bisheng_langchain/document_loaders/elem_image.py
================================================
================================================
FILE: src/backend/bisheng_langchain/document_loaders/elem_pdf.py
================================================
# flake8: noqa
"""Loads PDF with semantic splilter."""
import io
import json
import os
import re
import time
from collections import Counter
from copy import deepcopy
from typing import List, Optional, Union
import fitz
import numpy as np
from bisheng_langchain.document_loaders.parsers import LayoutParser
from langchain_community.docstore.document import Document
from langchain_community.document_loaders.blob_loaders import Blob
from langchain_community.document_loaders.pdf import BasePDFLoader
from shapely import Polygon
from shapely import box as Rect
RE_MULTISPACE_INCLUDING_NEWLINES = re.compile(pattern=r'\s+', flags=re.DOTALL)
def merge_rects(bboxes):
x0 = np.min(bboxes[:, 0])
y0 = np.min(bboxes[:, 1])
x1 = np.max(bboxes[:, 2])
y1 = np.max(bboxes[:, 3])
return [x0, y0, x1, y1]
def norm_rect(bbox):
x0 = np.min([bbox[0], bbox[2]])
x1 = np.max([bbox[0], bbox[2]])
y0 = np.min([bbox[1], bbox[3]])
y1 = np.max([bbox[1], bbox[3]])
return np.asarray([x0, y0, x1, y1])
def find_max_continuous_seq(arr):
n = len(arr)
max_info = (0, 1)
for i in range(n):
m = 1
for j in range(i + 1, n):
if arr[j] - arr[j - 1] == 1:
m += 1
else:
break
if m > max_info[1]:
max_info = (i, m)
max_info = (max_info[0] + arr[0], max_info[1])
return max_info
def order_by_tbyx(block_info, th=10):
"""
block_info: [(b0, b1, b2, b3, text, x, y)+]
th: threshold of the position threshold
"""
# sort using y1 first and then x1
res = sorted(block_info, key=lambda b: (b[1], b[0]))
for i in range(len(res) - 1):
for j in range(i, 0, -1):
# restore the order using the
if (abs(res[j + 1][1] - res[j][1]) < th and (res[j + 1][0] < res[j][0])):
tmp = deepcopy(res[j])
res[j] = deepcopy(res[j + 1])
res[j + 1] = deepcopy(tmp)
else:
break
return res
def join_lines(texts, is_table=False):
if is_table:
return '\n'.join(texts)
flags = []
PUNC_SET = set(['.', ',', ';', '?', '!'])
for text in texts:
flags.append(np.all([t.isalnum() for t in text.rsplit(' ', 5)]))
if np.all(flags):
t0 = texts[0]
for t in texts[1:]:
if t0[-1] == '-':
t0 = t0[:-1] + t
elif t0[-1].isalnum() and t[0].isalnum():
t0 += ' ' + t
elif t0[-1] in PUNC_SET or t[0] in PUNC_SET:
t0 += ' ' + t
else:
t0 += t
return t0
else:
return ''.join(texts)
class Segment:
def __init__(self, seg):
self.whole = seg
self.segs = []
@staticmethod
def is_align(seg0, seg1, delta=5, mode=0):
# mode=0 edge align
# mode=1, edge align or center align
res = Segment.contain(seg0, seg1)
if not res:
return False
else:
if mode == 1:
r1 = seg1[0] - seg0[0] <= delta or seg0[1] - seg1[1] <= delta
c0 = (seg0[0] + seg0[1]) / 2
c1 = (seg1[0] + seg1[1]) / 2
r2 = abs(c1 - c0) <= delta
return r1 or r2
else:
return seg1[0] - seg0[0] <= delta or seg0[1] - seg1[1] <= delta
@staticmethod
def contain(seg0, seg1):
return seg0[0] <= seg1[0] and seg0[1] >= seg1[0]
@staticmethod
def overlap(seg0, seg1):
max_x0 = max(seg0[0], seg1[0])
min_x1 = min(seg0[1], seg1[1])
return max_x0 < min_x1
def _merge(self, segs):
x0s = [s[0] for s in segs]
x1s = [s[1] for s in segs]
return (np.min(x0s), np.max(x1s))
def add(self, seg):
if not self.segs:
self.segs.append(seg)
else:
overlaps = []
non_overlaps = []
for seg0 in self.segs:
if Segment.overlap(seg0, seg):
overlaps.append(seg0)
else:
non_overlaps.append(seg0)
if not overlaps:
self.segs.append(seg)
else:
overlaps.append(seg)
new_seg = self._merge(overlaps)
non_overlaps.append(new_seg)
self.segs = non_overlaps
def get_free_segment(self, incr_margin=True, margin_threshold=10):
sorted_segs = sorted(self.segs, key=lambda x: x[0])
n = len(sorted_segs)
free_segs = []
if incr_margin:
if n > 0:
seg_1st = sorted_segs[0]
if (seg_1st[0] - self.whole[0]) > margin_threshold:
free_segs.append((self.whole[0], seg_1st[0]))
seg_last = sorted_segs[-1]
if (self.whole[1] - seg_last[1]) > margin_threshold:
free_segs.append((seg_last[1], self.whole[1]))
for i in range(n - 1):
x0 = sorted_segs[i][1]
x1 = sorted_segs[i + 1][0]
free_segs.append((x0, x1))
return free_segs
class PDFWithSemanticLoader(BasePDFLoader):
"""Loads a PDF with pypdf and chunks at character level.
Loader also stores page numbers in metadata.
"""
def __init__(self,
file_path: str,
password: Optional[Union[str, bytes]] = None,
layout_api_key: str = None,
layout_api_url: str = None,
is_join_table: bool = True,
with_columns: bool = False,
support_rotate: bool = False,
text_elem_sep: str = '\n',
start: int = 0,
n: int = None,
html_output_file: str = None,
verbose: bool = False) -> None:
"""Initialize with a file path."""
self.layout_parser = LayoutParser(api_key=layout_api_key, api_base_url=layout_api_url)
self.with_columns = with_columns
self.is_join_table = is_join_table
self.support_rotate = support_rotate
self.start = start
self.n = n
self.html_output_file = html_output_file
self.verbose = verbose
self.text_elem_sep = text_elem_sep
super().__init__(file_path)
def _get_image_blobs(self, fitz_doc, pdf_reader, n=None, start=0):
blobs = []
pages = []
if not n:
n = fitz_doc.page_count
for pg in range(start, start + n):
bytes_img = None
page = fitz_doc.load_page(pg)
pages.append(page)
mat = fitz.Matrix(1, 1)
try:
pm = page.get_pixmap(matrix=mat, alpha=False)
bytes_img = pm.getPNGData()
except Exception:
# some pdf input cannot get render image from fitz
page = pdf_reader.get_page(pg)
pil_image = page.render().to_pil()
img_byte_arr = io.BytesIO()
pil_image.save(img_byte_arr, format='PNG')
bytes_img = img_byte_arr.getvalue()
blobs.append(Blob(data=bytes_img))
return blobs, pages
def _allocate_semantic(self, page, layout):
class_name = ['印章', '图片', '标题', '段落', '表格', '页眉', '页码', '页脚']
effective_class_inds = [3, 4, 5, 999]
non_conti_class_ids = [6, 7, 8]
TEXT_ID = 4
TABLE_ID = 5
textpage = page.get_textpage()
blocks = textpage.extractBLOCKS()
if self.support_rotate:
rotation_matrix = np.asarray(page.rotation_matrix).reshape((3, 2))
c1 = (rotation_matrix[0, 0] - 1) <= 1e-6
c2 = (rotation_matrix[1, 1] - 1) <= 1e-6
is_rotated = c1 and c2
# print('c1/c2', c1, c2)
if is_rotated:
new_blocks = []
for b in blocks:
bbox = np.asarray([b[0], b[1], b[2], b[3]])
aug_bbox = bbox.reshape((-1, 2))
padding = np.ones((len(aug_bbox), 1))
aug_bbox = np.hstack([aug_bbox, padding])
bb = np.dot(aug_bbox, rotation_matrix).reshape(-1)
bb = norm_rect(bb)
info = (bb[0], bb[1], bb[2], bb[3], b[4], b[5], b[6])
new_blocks.append(info)
blocks = new_blocks
if not self.with_columns:
blocks = order_by_tbyx(blocks)
# print('---ori blocks---')
# for b in blocks:
# print(b)
IMG_BLOCK_TYPE = 1
text_ploys = []
text_rects = []
texts = []
for b in blocks:
if b[-1] != IMG_BLOCK_TYPE:
text = re.sub(RE_MULTISPACE_INCLUDING_NEWLINES, ' ', b[4] or '').strip()
if text:
texts.append(text)
text_ploys.append(Rect(b[0], b[1], b[2], b[3]))
text_rects.append([b[0], b[1], b[2], b[3]])
text_rects = np.asarray(text_rects)
texts = np.asarray(texts)
semantic_polys = []
semantic_labels = []
layout_info = json.loads(layout.page_content)
for info in layout_info:
bbs = info['bbox']
coords = ((bbs[0], bbs[1]), (bbs[2], bbs[3]), (bbs[4], bbs[5]), (bbs[6], bbs[7]))
semantic_polys.append(Polygon(coords))
semantic_labels.append(info['category_id'])
# caculate containing overlap
sem_cnt = len(semantic_polys)
texts_cnt = len(text_ploys)
contain_matrix = np.zeros((sem_cnt, texts_cnt))
for i in range(sem_cnt):
for j in range(texts_cnt):
inter = semantic_polys[i].intersection(text_ploys[j]).area
contain_matrix[i, j] = inter * 1.0 / text_ploys[j].area
# print('----------------containing matrix--------')
# for r in contain_matrix.tolist():
# print([round(r_, 2) for r_ in r])
# print('---text---')
# for t in texts:
# print(t)
# merge continuous text block by the containing matrix
CONTRAIN_THRESHOLD = 0.70
contain_info = []
for i in range(sem_cnt):
ind = np.argwhere(contain_matrix[i, :] > CONTRAIN_THRESHOLD)[:, 0]
if len(ind) == 0: continue
label = semantic_labels[i]
if label in non_conti_class_ids:
n = len(ind)
contain_info.append((None, None, n, label, ind))
else:
start, n = find_max_continuous_seq(ind)
if n >= 1:
contain_info.append((start, start + n, n, label, None))
contain_info = sorted(contain_info, key=lambda x: x[2], reverse=True)
mask = np.zeros(texts_cnt)
new_block_info = []
for info in contain_info:
start, end, n, label, ind = info
if label in non_conti_class_ids and np.all(mask[ind] == 0):
rect = merge_rects(text_rects[ind])
ori_orders = [blocks[i][-2] for i in ind]
ts = texts[ind]
rs = text_rects[ind]
ord_ind = np.min(ori_orders)
mask[ind] = 1
new_block_info.append((rect[0], rect[1], rect[2], rect[3], ts, rs, ord_ind))
elif np.all(mask[start:end] == 0):
rect = merge_rects(text_rects[start:end])
ori_orders = [blocks[i][-2] for i in range(start, end)]
arg_ind = np.argsort(ori_orders)
# print('ori_orders', ori_orders, arg_ind)
ord_ind = np.min(ori_orders)
ts = texts[start:end]
rs = text_rects[start:end]
if label == TABLE_ID:
ts = ts[arg_ind]
rs = rs[arg_ind]
mask[start:end] = 1
new_block_info.append((rect[0], rect[1], rect[2], rect[3], ts, rs, ord_ind))
for i in range(texts_cnt):
if mask[i] == 0:
b = blocks[i]
r = np.asarray([b[0], b[1], b[2], b[3]])
ord_ind = b[-2]
new_block_info.append((b[0], b[1], b[2], b[3], [texts[i]], [r], ord_ind))
if self.with_columns:
new_blocks = sorted(new_block_info, key=lambda x: x[-1])
else:
new_blocks = order_by_tbyx(new_block_info)
# print('\n\n---new blocks---')
# for idx, b in enumerate(new_blocks):
# print(idx, b)
text_ploys = []
texts = []
for b in new_blocks:
texts.append(b[4])
text_ploys.append(Rect(b[0], b[1], b[2], b[3]))
# caculate overlap
sem_cnt = len(semantic_polys)
texts_cnt = len(text_ploys)
overlap_matrix = np.zeros((sem_cnt, texts_cnt))
for i in range(sem_cnt):
for j in range(texts_cnt):
inter = semantic_polys[i].intersection(text_ploys[j]).area
union = semantic_polys[i].union(text_ploys[j]).area
overlap_matrix[i, j] = (inter * 1.0) / union
# print('---overlap_matrix---')
# for r in overlap_matrix:
# print([round(r_, 3) for r_ in r])
# allocate label
OVERLAP_THRESHOLD = 0.2
texts_labels = []
DEF_SEM_LABEL = 999
for j in range(texts_cnt):
ind = np.argwhere(overlap_matrix[:, j] > OVERLAP_THRESHOLD)[:, 0]
if len(ind) == 0:
sem_label = DEF_SEM_LABEL
else:
c = Counter([semantic_labels[i] for i in ind])
items = c.most_common()
sem_label = items[0][0]
if len(items) > 1 and TEXT_ID in dict(items):
sem_label = TEXT_ID
texts_labels.append(sem_label)
# print(texts_labels)
# filter the unused element
filtered_blocks = []
for label, b in zip(texts_labels, new_blocks):
if label in effective_class_inds:
text = join_lines(b[4], label == TABLE_ID)
filtered_blocks.append((b[0], b[1], b[2], b[3], text, b[5], label))
# print('---filtered_blocks---')
# for b in filtered_blocks:
# print(b)
return filtered_blocks
def _divide_blocks_into_groups(self, blocks):
# support only pure two columns layout, each has same width
rects = np.asarray([[b[0], b[1], b[2], b[3]] for b in blocks])
min_x0 = np.min(rects[:, 0])
max_x1 = np.max(rects[:, 2])
root_seg = (min_x0, max_x1)
root_pc = (min_x0 + max_x1) / 2
root_offset = 20
center_seg = (root_pc - root_offset, root_pc + root_offset)
segment = Segment(root_seg)
for r in rects:
segment.add((r[0], r[2]))
COLUMN_THRESHOLD = 0.90
CENTER_GAP_THRESHOLD = 0.90
free_segs = segment.get_free_segment()
columns = []
if len(free_segs) == 1 and len(segment.segs) == 2:
free_seg = free_segs[0]
seg0 = segment.segs[0]
seg1 = segment.segs[1]
cover = seg0[1] - seg0[0] + seg1[1] - seg1[0]
c0 = cover / (root_seg[1] - root_seg[0])
c1 = Segment.contain(center_seg, free_seg)
if c0 > COLUMN_THRESHOLD and c1:
# two columns
columns.extend([seg0, seg1])
groups = [blocks]
if columns:
groups = [[] for _ in columns]
for b, r in zip(blocks, rects):
column_ind = 0
cand_seg = (r[0], r[2])
for i, seg in enumerate(columns):
if Segment.contain(seg, cand_seg):
column_ind = i
break
groups[i].append(b)
return groups
def _allocate_continuous(self, groups):
g_bound = []
groups = [g for g in groups if g]
for blocks in groups:
arr = [[b[0], b[1], b[2], b[3]] for b in blocks]
bboxes = np.asarray(arr)
g_bound.append(np.asarray(merge_rects(bboxes)))
LINE_FULL_THRESHOLD = 0.80
START_THRESHOLD = 0.8
SIMI_HEIGHT_THRESHOLD = 0.3
SIMI_WIDTH_THRESHOLD = 0.3
TEXT_ID = 4
TABLE_ID = 5
def _get_elem(blocks, is_first=True):
if not blocks:
return (None, None, None, None, None)
if is_first:
b1 = blocks[0]
b1_label = b1[-1]
r1 = b1[5][0]
r1_w = r1[2] - r1[0]
r1_h = r1[3] - r1[1]
return (b1, b1_label, r1, r1_w, r1_h)
else:
b0 = blocks[-1]
b0_label = b0[-1]
r0 = b0[5][-1]
r0_w = r0[2] - r0[0]
r0_h = r0[3] - r0[1]
return (b0, b0_label, r0, r0_w, r0_h)
b0, b0_label, r0, r0_w, r0_h = _get_elem(groups[0], False)
g0 = g_bound[0]
for i in range(1, len(groups)):
b1, b1_label, r1, r1_w, r1_h = _get_elem(groups[i], True)
g1 = g_bound[i]
# print('\n_allocate_continuous:')
# print(b0, b0_label, b1, b1_label)
if b0_label and b0_label == b1_label and b0_label == TEXT_ID:
c0 = r0_w / (g0[2] - g0[0])
c1 = (r1[0] - g1[0]) / r1_h
c2 = np.abs(r0_h - r1_h) / r1_h
# print('\n\n---conti texts---')
# print(b0_label, c0, c1, c2,
# b0, b0_label, r0, r0_w, r0_h,
# b1, b1_label, r1, r1_w, r1_h)
if (c0 > LINE_FULL_THRESHOLD and c1 < START_THRESHOLD
and c2 < SIMI_HEIGHT_THRESHOLD):
new_text = join_lines([b0[4], b1[4]])
new_block = (b0[0], b0[1], b0[2], b0[3], new_text, b0[5], b0[6])
groups[i - 1][-1] = new_block
groups[i].pop(0)
elif (self.is_join_table and b0_label and b0_label == b1_label
and b0_label == TABLE_ID):
c0 = (r1_w - r0_w) / r1_h
if c0 < SIMI_WIDTH_THRESHOLD:
new_text = join_lines([b0[4], b1[4]], True)
new_block = (b0[0], b0[1], b0[2], b0[3], new_text, b0[5], b0[6])
groups[i - 1][-1] = new_block
groups[i].pop(0)
b0, b0_label, r0, r0_w, r0_h = _get_elem(groups[i], False)
return groups
def save_to_html(self, groups, output_file):
styles = ['style="background-color: #EBEBEB;"', 'style="background-color: #ABBAEA;"']
idx = 0
table_style = 'style="border:1px solid black;"'
with open(output_file, 'w') as fout:
for blocks in groups:
for b in blocks:
if b[-1] == 3:
text = f'
{b[4]}
'
elif b[-1] == 4:
text = f'
{b[4]}
'
idx += 1
elif b[-1] == 5:
rows = b[4].split('\n')
content = []
for r in rows:
content.append(f'
"));
}
execAfterRender(vditor);
};
var insertRowAbove = function (vditor, range, cellElement) {
var rowHTML = "";
for (var m = 0; m < cellElement.parentElement.childElementCount; m++) {
if (cellElement.tagName === "TH") {
rowHTML += "
");
}
else {
rowHTML += "
");
}
}
if (cellElement.tagName === "TH") {
cellElement.parentElement.parentElement.insertAdjacentHTML("beforebegin", "
".concat(rowHTML, "
"));
range.insertNode(document.createElement("wbr"));
var theadHTML = cellElement.parentElement.innerHTML.replace(/
"));
}
execAfterRender(vditor);
};
var insertColumn = function (vditor, tableElement, cellElement, type) {
if (type === void 0) { type = "afterend"; }
var index = 0;
var previousElement = cellElement.previousElementSibling;
while (previousElement) {
index++;
previousElement = previousElement.previousElementSibling;
}
for (var i = 0; i < tableElement.rows.length; i++) {
if (i === 0) {
tableElement.rows[i].cells[index].insertAdjacentHTML(type, "
", 0);
});
return _this;
}
return Help;
}(MenuItem));
;// CONCATENATED MODULE: ./src/ts/toolbar/Indent.ts
var Indent_extends = (undefined && undefined.__extends) || (function () {
var extendStatics = function (d, b) {
extendStatics = Object.setPrototypeOf ||
({ __proto__: [] } instanceof Array && function (d, b) { d.__proto__ = b; }) ||
function (d, b) { for (var p in b) if (Object.prototype.hasOwnProperty.call(b, p)) d[p] = b[p]; };
return extendStatics(d, b);
};
return function (d, b) {
if (typeof b !== "function" && b !== null)
throw new TypeError("Class extends value " + String(b) + " is not a constructor or null");
extendStatics(d, b);
function __() { this.constructor = d; }
d.prototype = b === null ? Object.create(b) : (__.prototype = b.prototype, new __());
};
})();
var Indent = /** @class */ (function (_super) {
Indent_extends(Indent, _super);
function Indent(vditor, menuItem) {
var _this = _super.call(this, vditor, menuItem) || this;
_this.element.children[0].addEventListener((0,compatibility/* getEventName */.Le)(), function (event) {
event.preventDefault();
if (_this.element.firstElementChild.classList.contains(constants/* Constants.CLASS_MENU_DISABLED */.g.CLASS_MENU_DISABLED) ||
vditor.currentMode === "sv") {
return;
}
var range = (0,util_selection/* getEditorRange */.zh)(vditor);
var liElement = (0,hasClosest/* hasClosestByMatchTag */.lG)(range.startContainer, "LI");
if (liElement) {
listIndent(vditor, liElement, range);
}
});
return _this;
}
return Indent;
}(MenuItem));
;// CONCATENATED MODULE: ./src/ts/toolbar/Info.ts
var Info_extends = (undefined && undefined.__extends) || (function () {
var extendStatics = function (d, b) {
extendStatics = Object.setPrototypeOf ||
({ __proto__: [] } instanceof Array && function (d, b) { d.__proto__ = b; }) ||
function (d, b) { for (var p in b) if (Object.prototype.hasOwnProperty.call(b, p)) d[p] = b[p]; };
return extendStatics(d, b);
};
return function (d, b) {
if (typeof b !== "function" && b !== null)
throw new TypeError("Class extends value " + String(b) + " is not a constructor or null");
extendStatics(d, b);
function __() { this.constructor = d; }
d.prototype = b === null ? Object.create(b) : (__.prototype = b.prototype, new __());
};
})();
var Info = /** @class */ (function (_super) {
Info_extends(Info, _super);
function Info(vditor, menuItem) {
var _this = _super.call(this, vditor, menuItem) || this;
_this.element.children[0].addEventListener((0,compatibility/* getEventName */.Le)(), function (event) {
event.preventDefault();
vditor.tip.show("
================================================
FILE: README_CN.md
================================================
![](\"https://unpkg.com/vditor/dist/images/logo.png\"){kind=logo}
\n \n \n Vditor \u662F\u4E00\u6B3E\u6D4F\u89C8\u5668\u7AEF\u7684 Markdown \u7F16\u8F91\u5668\uFF0C\u652F\u6301\u6240\u89C1\u5373\u6240\u5F97\u3001\u5373\u65F6\u6E32\u67D3\uFF08\u7C7B\u4F3C Typora\uFF09\u548C\u5206\u5C4F\u9884\u89C8\u6A21\u5F0F\u3002\n \u5B83\u4F7F\u7528 TypeScript \u5B9E\u73B0\uFF0C\u652F\u6301\u539F\u751F JavaScript \u4EE5\u53CA Vue\u3001React\u3001Angular \u548C Svelte \u7B49\u6846\u67B6\u3002\n \n\n \n