Showing preview only (348K chars total). Download the full file or copy to clipboard to get everything.
Repository: zai-org/GLM-4
Branch: main
Commit: 3f79b543647b
Files: 69
Total size: 328.3 KB
Directory structure:
gitextract_0zacjmse/
├── .github/
│ ├── ISSUE_TEMPLATE/
│ │ ├── bug_report.yaml
│ │ └── feature-request.yaml
│ └── PULL_REQUEST_TEMPLATE.md
├── .gitignore
├── .pre-commit-config.yaml
├── LICENSE
├── README.md
├── README_20240605.md
├── README_zh.md
├── README_zh_240605.md
├── demo/
│ ├── composite_demo/
│ │ ├── .gitignore
│ │ ├── README.md
│ │ ├── README_en.md
│ │ ├── browser/
│ │ │ ├── .gitignore
│ │ │ ├── package.json
│ │ │ ├── src/
│ │ │ │ ├── browser.ts
│ │ │ │ ├── config.ts
│ │ │ │ ├── server.ts
│ │ │ │ ├── types.ts
│ │ │ │ └── utils.ts
│ │ │ └── tsconfig.json
│ │ ├── requirements.txt
│ │ └── src/
│ │ ├── client.py
│ │ ├── clients/
│ │ │ ├── hf.py
│ │ │ ├── openai.py
│ │ │ └── vllm.py
│ │ ├── conversation.py
│ │ ├── main.py
│ │ ├── tools/
│ │ │ ├── browser.py
│ │ │ ├── cogview.py
│ │ │ ├── config.py
│ │ │ ├── interface.py
│ │ │ ├── python.py
│ │ │ └── tool_registry.py
│ │ └── utils.py
│ └── intel_device_demo/
│ ├── itrex/
│ │ ├── README.md
│ │ ├── README_en.md
│ │ ├── itrex_cli_demo.py
│ │ └── requirements.txt
│ └── openvino/
│ ├── README.md
│ ├── README_en.md
│ ├── convert.py
│ ├── openvino_cli_demo.py
│ └── requirements.txt
├── finetune/
│ ├── .gitignore
│ ├── README.md
│ ├── README_zh.md
│ ├── configs/
│ │ ├── ds_zero_2.json
│ │ ├── ds_zero_3.json
│ │ ├── lora.yaml
│ │ └── sft.yaml
│ ├── finetune.py
│ ├── finetune_vision.py
│ └── requirements.txt
├── inference/
│ ├── README.md
│ ├── README_zh.md
│ ├── glm4v_api_request.py
│ ├── glm4v_server.py
│ ├── requirements.txt
│ ├── trans_batch_demo.py
│ ├── trans_cli_demo.py
│ ├── trans_cli_vision_demo.py
│ ├── trans_stress_test.py
│ ├── trans_web_demo.py
│ ├── trans_web_vision_demo.py
│ ├── vllm_cli_demo.py
│ └── vllm_cli_vision_demo.py
├── pyproject.toml
└── resources/
└── WECHAT.md
================================================
FILE CONTENTS
================================================
================================================
FILE: .github/ISSUE_TEMPLATE/bug_report.yaml
================================================
name: "\U0001F41B Bug Report"
description: Submit a bug report to help us improve GLM-4-9B / 提交一个 Bug 问题报告来帮助我们改进 GLM-4-9B
body:
- type: textarea
id: system-info
attributes:
label: System Info / 系統信息
description: Your operating environment / 您的运行环境信息
placeholder: Includes Cuda version, Transformers version, Python version, operating system, hardware information (if you suspect a hardware problem)... / 包括Cuda版本，Transformers版本，Python版本，操作系统，硬件信息(如果您怀疑是硬件方面的问题)...
validations:
required: true
- type: textarea
id: who-can-help
attributes:
label: Who can help? / 谁可以帮助到您？
description: |
Your issue will be replied to more quickly if you can figure out the right person to tag with @
All issues are read by one of the maintainers, so if you don't know who to tag, just leave this blank and our maintainer will ping the right person.
Please tag fewer than 3 people.
如果您能找到合适的标签 @，您的问题会更快得到回复。
所有问题都会由我们的维护者阅读，如果您不知道该标记谁，只需留空，我们的维护人员会找到合适的开发组成员来解决问题。
标记的人数应该不超过 3 个人。
If it's not a bug in these three subsections, you may not specify the helper. Our maintainer will find the right person in the development group to solve the problem.
如果不是这三个子版块的bug，您可以不指明帮助者，我们的维护人员会找到合适的开发组成员来解决问题。
placeholder: "@Username ..."
- type: checkboxes
id: information-scripts-examples
attributes:
label: Information / 问题信息
description: 'The problem arises when using: / 问题出现在'
options:
- label: "The official example scripts / 官方的示例脚本"
- label: "My own modified scripts / 我自己修改的脚本和任务"
- type: textarea
id: reproduction
validations:
required: true
attributes:
label: Reproduction / 复现过程
description: |
Please provide a code example that reproduces the problem you encountered, preferably with a minimal reproduction unit.
If you have code snippets, error messages, stack traces, please provide them here as well.
Please format your code correctly using code tags. See https://help.github.com/en/github/writing-on-github/creating-and-highlighting-code-blocks#syntax-highlighting
Do not use screenshots, as they are difficult to read and (more importantly) do not allow others to copy and paste your code.
请提供能重现您遇到的问题的代码示例,最好是最小复现单元。
如果您有代码片段、错误信息、堆栈跟踪，也请在此提供。
请使用代码标签正确格式化您的代码。请参见 https://help.github.com/en/github/writing-on-github/creating-and-highlighting-code-blocks#syntax-highlighting
请勿使用截图，因为截图难以阅读，而且（更重要的是）不允许他人复制粘贴您的代码。
placeholder: |
Steps to reproduce the behavior/复现Bug的步骤:
1.
2.
3.
- type: textarea
id: expected-behavior
validations:
required: true
attributes:
label: Expected behavior / 期待表现
description: "A clear and concise description of what you would expect to happen. /简单描述您期望发生的事情。"
================================================
FILE: .github/ISSUE_TEMPLATE/feature-request.yaml
================================================
name: "\U0001F680 Feature request"
description: Submit a request for a new GLM-4-9B feature / 提交一个新的 GLM-4-9B 的功能建议
labels: [ "feature" ]
body:
- type: textarea
id: feature-request
validations:
required: true
attributes:
label: Feature request / 功能建议
description: |
A brief description of the functional proposal. Links to corresponding papers and code are desirable.
对功能建议的简述。最好提供对应的论文和代码链接
- type: textarea
id: motivation
validations:
required: true
attributes:
label: Motivation / 动机
description: |
Your motivation for making the suggestion. If that motivation is related to another GitHub issue, link to it here.
您提出建议的动机。如果该动机与另一个 GitHub 问题有关，请在此处提供对应的链接。
- type: textarea
id: contribution
validations:
required: true
attributes:
label: Your contribution / 您的贡献
description: |
Your PR link or any other link you can help with.
您的PR链接或者其他您能提供帮助的链接。
================================================
FILE: .github/PULL_REQUEST_TEMPLATE.md
================================================
# Contribution Guide
We welcome your contributions to this repository. To ensure elegant code style and better code quality, we have prepared the following contribution guidelines.
## What We Accept
+ This PR fixes a typo or improves the documentation (if this is the case, you may skip the other checks).
+ This PR fixes a specific issue — please reference the issue number in the PR description. Make sure your code strictly follows the coding standards below.
+ This PR introduces a new feature — please clearly explain the necessity and implementation of the feature. Make sure your code strictly follows the coding standards below.
## Code Style Guide
Good code style is an art. We have prepared a `pyproject.toml` and a `pre-commit` hook to enforce consistent code formatting across the project. You can clean up your code following the steps below:
1. Install the required dependencies:
```shell
pip install ruff pre-commit
```
2. Then, run the following command:
```shell
pre-commit run --all-files
```
If your code complies with the standards, you should not see any errors.
## Naming Conventions
- Please use **English** for naming; do not use Pinyin or other languages. All comments should also be in English.
- Follow **PEP8** naming conventions strictly, and use underscores to separate words. Avoid meaningless names such as `a`, `b`, `c`.
================================================
FILE: .gitignore
================================================
*venv
*.DS_Store
*.idea/
dataset
test*
================================================
FILE: .pre-commit-config.yaml
================================================
repos:
- repo: https://github.com/astral-sh/ruff-pre-commit
rev: v0.4.5
hooks:
- id: ruff
args: [--fix, --respect-gitignore, --config=pyproject.toml]
- id: ruff-format
args: [--config=pyproject.toml]
- repo: https://github.com/pre-commit/pre-commit-hooks
rev: v4.5.0
hooks:
- id: trailing-whitespace
- id: end-of-file-fixer
- id: check-yaml
- id: check-toml
- id: check-case-conflict
- id: check-merge-conflict
- id: debug-statements
================================================
FILE: LICENSE
================================================
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright 2025 Zhipu AI
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: README.md
================================================
# GLM-4-0414 Model Series
<p align="center">
👋 Join our <a href="https://discord.gg/8cnQKdAprg" target="_blank">Discord</a>, <a href="https://x.com/Zai_org" target="_blank">X</a> and <a href="resources/WECHAT.md" target="_blank"> WeChat (Chinese) </a>
</p>
<p align="center">
📍The open-source models released this time can be experienced for free at <a href="https://chat.z.ai">Z.ai</a>; for GLM commercial model services, please visit <a href="https://bigmodel.cn">bigmodel.cn</a>.
</p>
Read this in [中文](README_zh.md)
## Project Updates
- 🔥 **News**: ```2025/07/02```: We are releasing the [GLM-4.1V-9B-Thinking](https://huggingface.co/collections/THUDM/glm-41v-thinking-6862bbfc44593a8601c2578d) series VLM, check [this github repo](https://github.com/THUDM/GLM-4.1V-Thinking) to get more information.
- **News**: ```2025/04/14```: We are releasing the [GLM-4-32B-0414](https://huggingface.co/collections/THUDM/glm-4-0414-67f3cbcb34dd9d252707cb2e) series models, scaled up to 32B parameters, including models with capabilities for dialogue, reasoning, and rumination.
- **News**: ``2024/06/18``: We have released our [Technical Report](https://arxiv.org/pdf/2406.12793), feel free to check it out.
- **News**: ``2024/06/05``: We released the `GLM-4-9B` series of open-source models. Details can be found [here](README_20240605.md).
## Model Introduction
The GLM family welcomes new members, the **GLM-4-32B-0414** series models, featuring 32 billion parameters. Its performance is comparable to OpenAI’s GPT series and DeepSeek’s V3/R1 series. It also supports very user-friendly local deployment features. GLM-4-32B-Base-0414 was pre-trained on 15T of high-quality data, including substantial reasoning-type synthetic data. This lays the foundation for subsequent reinforcement learning extensions. In the post-training stage, we employed human preference alignment for dialogue scenarios. Additionally, using techniques like rejection sampling and reinforcement learning, we enhanced the model’s performance in instruction following, engineering code, and function calling, thus strengthening the atomic capabilities required for agent tasks. GLM-4-32B-0414 achieves good results in engineering code, Artifact generation, function calling, search-based Q&A, and report generation. In particular, on several benchmarks, such as code generation or specific Q&A tasks, GLM-4-32B-Base-0414 achieves comparable performance with those larger models like GPT-4o and DeepSeek-V3-0324 (671B).
**GLM-Z1-32B-0414** is a reasoning model with deep thinking capabilities. This was developed based on GLM-4-32B-0414 through cold start, extended reinforcement learning, and further training on tasks including mathematics, code, and logic. Compared to the base model, GLM-Z1-32B-0414 significantly improves mathematical abilities and the capability to solve complex tasks. During training, we also introduced general reinforcement learning based on pairwise ranking feedback, which enhances the model's general capabilities.
**GLM-Z1-Rumination-32B-0414** is a deep reasoning model with rumination capabilities (against OpenAI's Deep Research). Unlike typical deep thinking models, the rumination model is capable of deeper and longer thinking to solve more open-ended and complex problems (e.g., writing a comparative analysis of AI development in two cities and their future development plans). Z1-Rumination is trained through scaling end-to-end reinforcement learning with responses graded by the ground truth answers or rubrics and can make use of search tools during its deep thinking process to handle complex tasks. The model shows significant improvements in research-style writing and complex tasks.
Finally, **GLM-Z1-9B-0414** is a surprise. We employed all the aforementioned techniques to train a small model (9B). GLM-Z1-9B-0414 exhibits excellent capabilities in mathematical reasoning and general tasks. Its overall performance is top-ranked among all open-source models of the same size. Especially in resource-constrained scenarios, this model achieves an excellent balance between efficiency and effectiveness, providing a powerful option for users seeking lightweight deployment.
## Showcase
### Animation Generation
<table>
<tr>
<td style="text-align: center; font-size: 16px; font-weight: bold; padding: 10px; width: 420px;">
GLM-Z1-32B-0414
</td>
<td style="text-align: center; font-size: 16px; font-weight: bold; padding: 10px; width: 420px;">
GLM-4-32B-0414
</td>
</tr>
<tr>
<td style="vertical-align: top; padding: 10px; width: 420px;">
<video src="https://github.com/user-attachments/assets/849ff9fd-b54d-4c74-9ee5-3412e1a09e32"
style="width: 400px; height: 300px; object-fit: contain;" autoplay loop muted playsinline></video>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;">
write a Python program that shows a ball bouncing inside a spinning hexagon. The ball should be affected by gravity and friction, and it must bounce off the rotating walls realistically
</div>
</td>
<td style="vertical-align: top; padding: 10px; width: 420px;">
<video src="https://github.com/user-attachments/assets/8dccdb9d-cc44-4732-b438-74a4e3cb9dfb"
style="width: 400px; height: 300px; object-fit: contain;" autoplay loop muted playsinline></video>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;">
Use HTML to simulate the scenario of a small ball released from the center of a rotating hexagon. Consider the collision between the ball and the hexagon's edges, the gravity acting on the ball, and assume all collisions are perfectly elastic. (Prompt translated from Chinese)
</div>
</td>
</tr>
</table>
### Web Design
<table>
<tr>
<td style="text-align: center; font-size: 16px; font-weight: bold; padding: 10px; width: 420px;">
GLM-4-32B-0414
</td>
<td style="text-align: center; font-size: 16px; font-weight: bold; padding: 10px; width: 420px;">
GLM-4-32B-0414
</td>
</tr>
<tr>
<td style="vertical-align: top; padding: 10px; width: 420px;">
<img src="https://github.com/user-attachments/assets/bd9c1fc1-c784-4e8f-9c76-5f7389a715f1"/>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;">
Design a drawing board that supports custom function plotting, allowing adding and deleting custom functions, and assigning colors to functions. (Prompt translated from Chinese)
</div>
</td>
<td style="vertical-align: top; padding: 10px; width: 420px;">
<img src="https://github.com/user-attachments/assets/7ad12d52-9229-4278-8d1b-ffbf43e99070"/>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;"> Design a UI for a mobile machine learning platform, which should include interfaces for training tasks, storage management, and personal statistics. The personal statistics interface should use charts to display the user's resource usage over a period. Use Tailwind CSS to style the page, and display these 3 mobile interfaces tiled on a single HTML page. (Prompt translated from Chinese) </div>
</td>
</tr>
</table>
### SVG Generation
<table>
<tr>
<td style="text-align: center; font-size: 16px; font-weight: bold; padding: 10px; width: 420px;">
GLM-4-32B-0414
</td>
<td style="text-align: center; font-size: 16px; font-weight: bold; padding: 10px; width: 420px;">
GLM-4-32B-0414
</td>
</tr>
<tr>
<td style="vertical-align: top; padding: 10px; width: 420px;">
<img src="https://github.com/user-attachments/assets/9407e4c1-1876-4ab5-838c-839836fb418a"/>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;">
Create a misty Jiangnan scene using SVG. (Prompt translated from Chinese)
</div>
</td>
<td style="vertical-align: top; padding: 10px; width: 420px;">
<img src="https://github.com/user-attachments/assets/bcce8c5a-cedf-45c8-b666-ddb023d5b49c"/>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;"> Use SVG to illustrate the training process of an LLM. (Prompt translated from Chinese) </div>
</td>
</tr>
</table>
### Analysis and Research Report Writing
<td style="vertical-align: top; padding: 10px; width: 420px;">
<video src="https://github.com/user-attachments/assets/7939c8c5-0fcf-4bc4-be45-3964aad0e61c" style="width: 400px; height: 300px; object-fit: contain;" autoplay loop muted playsinline></video>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;">
Analysis of AI Development in Chinese Cities: A Comparative Study of Beijing and Hangzhou, Alongside an Investigation of International Cases of AI in Urban Governance. (Prompt translated from Chinese)
</div>
</td>
## Model List
### GLM-4-0414 Series Models
GLM-Z1-9B-0414 Open-Source Model [Try it Online](https://modelscope.cn/studios/ZhipuAI/GLM-Z1-9B-0414/summary)
| Model | Type | Seq Length* | Download |
|:--------------------------:|:---------:|:-----------:|:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|
| GLM-4-9B-0414 | Chat | 32K -> 128K | [🤗 Huggingface](https://huggingface.co/THUDM/GLM-4-9B-0414)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/GLM-4-9B-0414)<br> [🧩 Modelers](https://modelers.cn/models/zhipuai/GLM-4-9B-0414)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-9B-0414) |
| GLM-Z1-9B-0414 | Reasoning | 32K -> 128K | [🤗 Huggingface](https://huggingface.co/THUDM/GLM-4-Z1-9B-0414)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/GLM-Z1-9B-0414)<br> [🧩 Modelers](https://modelers.cn/models/zhipuai/GLM-Z1-9B-0414)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-Z1-9B-0414) |
| GLM-4-32B-Base-0414 | Base | 32K -> 128K | [🤗 Huggingface](https://huggingface.co/THUDM/GLM-4-32B-Base-0414)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/GLM-4-32B-Base-0414)<br> [🧩 Modelers](https://modelers.cn/models/zhipuai/GLM-4-32B-Base-0414)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-32B-Base-0414) |
| GLM-4-32B-0414 | Chat | 32K -> 128K | [🤗 Huggingface](https://huggingface.co/THUDM/GLM-4-32B-0414)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/GLM-4-32B-0414)<br> [🧩 Modelers](https://modelers.cn/models/zhipuai/GLM-4-32B-0414)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-32B-Base-0414) |
| GLM-Z1-32B-0414 | Reasoning | 32K -> 128K | [🤗 Huggingface](https://huggingface.co/THUDM/GLM-Z1-32B-0414)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/GLM-Z1-32B-0414)<br> [🧩 Modelers](https://modelers.cn/models/zhipuai/GLM-Z1-32B-0414)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-Z1-32B-0414) |
| GLM-Z1-Rumination-32B-0414 | Reasoning | 128K | [🤗 Huggingface](https://huggingface.co/THUDM/GLM-Z1-Rumination-32B-0414)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/GLM-Z1-Rumination-32B-0414)<br> [🧩 Modelers](https://modelers.cn/models/zhipuai/GLM-Z1-Rumination-32B-0414)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-Z1-Rumination-32B-0414) |
Due to its smaller model capacity, GLM-4-9B-0414 has not undergone the same agent capability enhancements as GLM-4-32B-0414. Instead, it has been optimized primarily for scenarios that require large-scale batch operations, such as translation tasks.
\* Models are natively trained with a 32K context. For requests where the total input + output length might exceed 32K tokens, we recommend activating YaRN for better extrapolation performance. See the [Model and Prompt Implementation](#model-and-prompt-implementation) section for details.
Below are the GLM-4 series models released on June 5, 2024. Details can be found [here](README_240605.md).
| Model | Type | Seq Length* | Download |
|:-----------------------------:|:---------:|:----------:|:-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|
| GLM-4-9B | Base | 8K | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b)<br> |
| GLM-4-9B-Chat | Chat | 128K | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-9B-Chat) |
| GLM-4-9B-Chat-HF | Chat | 128K | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat-hf)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-hf) |
| GLM-4-9B-Chat-1M | Chat | 1M | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat-1m)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-1m)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-9B-Chat-1M) |
| GLM-4-9B-Chat-1M-HF | Chat | 1M | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat-1m-hf)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-1m-hf) |
| GLM-4V-9B | Chat | 8K | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4v-9b)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4v-9b)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4V-9B) |
## Evaluation Results
### GLM-4-0414 Series
<div style="text-align: center;">
<img src="resources/Bench-32B.png" style="width: 80%;" />
</div>
| Model | IFEval | BFCL-v3 (Overall) | BFCL-v3 (MultiTurn) | TAU-Bench (Retail) | TAU-Bench (Airline) | SimpleQA | HotpotQA |
| ---------------- | ------ | ----------------- | ------------------- | ------------------ | ------------------- | -------- | -------- |
| Qwen2.5-Max | 85.6 | 50.9 | 30.5 | 58.3 | 22.0 | 79.0 | 52.8 |
| GPT-4o-1120 | 81.9 | 69.6 | 41.0 | 62.8 | 46.0 | 82.8 | 63.9 |
| DeepSeek-V3-0324 | 83.4 | 66.2 | 35.8 | 60.7 | 32.4 | 82.6 | 54.6 |
| DeepSeek-R1 | 84.3 | 57.5 | 12.4 | 33.0 | 37.3 | 83.9 | 63.1 |
| GLM-4-32B-0414 | 87.6 | 69.6 | 41.5 | 68.7 | 51.2 | 88.1 | 63.8 |
> For `SimpleQA` and `HotpotQA`, we sampled nearly 500 test cases from each test set, provided all models with basic `search` and `click` tools, ensured other settings remained consistent, and averaged the results over 3 runs.
| Model | Framework | [SWE-bench Verified](https://openai.com/index/introducing-swe-bench-verified/) | [SWE-bench Verified mini](https://github.com/mariushobbhahn/SWEBench-verified-mini) |
|---|---|---|---|
| GLM-4-32B-0414 | Moatless<sup>[1]</sup> | 33.8 | 38.0 |
| GLM-4-32B-0414 | Agentless<sup>[2]</sup> | 30.7 | 34.0 |
| GLM-4-32B-0414 | OpenHands<sup>[3]</sup> | 27.2 | 28.0 |
[1] [Moatless v0.0.3](https://github.com/aorwall/moatless-tools) used the following parameters: `response_format="react", thoughts_in_action=False, max_interations=30`. No retries on failed trajectories; other settings are default.
[2] [Agentless v1.5.0](https://github.com/OpenAutoCoder/Agentless) used [BGE](https://github.com/FlagOpen/FlagEmbedding/blob/master/README.md) as the embedding model and [FAISS](https://github.com/facebookresearch/faiss) for similarity search. To speed up patch verification while maintaining performance, the timeout for running a single instance was changed from the default 300s to 180s.
[3] [OpenHands v0.29.1](https://github.com/All-Hands-AI/OpenHands/tree/main) did not use YaRN context extension but limited runs to a maximum of 60 iterations and summarized the history to prevent exceeding the 32K context limit. Summarization was configured as `llm_config="condenser", keep_first=1, max_size=32`. No retries on failed trajectories.
### GLM-Z1-0414 Series
<div style="text-align: center;">
<img src="resources/Bench-Z1-9B.png" style="width: 80%;" />
<img src="resources/Bench-Z1-32B.png" style="width: 80%;" />
</div>
## Model and Prompt Implementation
### Model Implementation
If you want to see our model implementation, please check the Pull Requests in the relevant repositories, which have been merged:
+ [vLLM Model Implementation](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/glm4.py)
+ [transformers Model Implementation](https://github.com/huggingface/transformers/blob/main/src/transformers/models/glm4/modeling_glm4.py)
+ [llama.cpp Model Implementation](https://github.com/ggml-org/llama.cpp/pull/12867)
### Handling Long Context (YaRN)
If the total input + output token count might exceed the model's native context length (mostly 32k for the GLM-4-0414 series), it is recommended to enable YaRN to achieve better long-context modeling capabilities. For supported frameworks, you can modify the corresponding `config.json`. Specifically, for GLM-Z1 series models, consider enabling YaRN (Rope Scaling) when the input length exceeds **8,192 tokens**.
```json
"rope_scaling": {
"factor": 4.0,
"original_max_position_embeddings": 32768,
"type": "yarn"
}
```
For most user requests, if the input + output token count does not exceed the native context length, no modifications are needed.
### Model Fine-tuning
You can find information about the computational resources required for model fine-tuning, as well as example fine-tuning scripts, in `finetune/README.md`.
To start a simple model fine-tuning example, run the following commands:
```shell
cd finetune
pip install -r ../inference/requirements.txt
pip install -r requirements.txt
# Use single GPU for Chat Fine-tune
python finetune.py data/AdvertiseGen/ THUDM/GLM-4-9B-0414 configs/lora.yaml
```
🎉 The script also supports fine-tuning with visual tracking using **SwanLab**. You can view the training logs of the example fine-tuning script on the [SwanLab Visualization Dashboard](https://swanlab.cn/@ShaohonChen/GLM4-Finetune/overview).
### Prompt Implementation
If you use the `apply_chat_template` method provided by the `transformers` library to construct prompts, here are the restrictions on `System Prompts` for different GLM-4-0414 models.
+ `GLM-4-32B-Base-0414`: Base model, no chat template.
+ `GLM-4-*-0414` / `GLM-Z1-*-0414`: If `tools` are provided, `apply_chat_template` will populate the tools into a fixed template within the `chat_template`, creating a separate `system` message with tool bindings prepended to the message list (`messages[0]`). All originally passed `messages` are automatically shifted one position back.
+ `GLM-Z1-Rumination-32B-0414`:
+ Does not support custom system prompts or custom tools. Your `tools` and `system` fields will be ignored by `apply_chat_template`. Using this model requires an external search engine or a custom retrieval API.
+ Supports four tools in total:
```
1. search
Description: Executes a search query and returns search results. Use this when you need to find information about a specific topic.
Parameters: query (string) - The search query string. Use English words unless it's a Chinese proper noun.
2. click
Description: Clicks on a link from the search results and navigates to the corresponding page. Use this when you need to view the detailed content of a specific search result.
Parameters: link_id (integer) - The ID of the link to click (from the sequence number in the search results).
3. open
Description: Opens a specific website. Gets the content of any website via URL.
Parameters: url (string) - The target website URL or domain name.
4. finish
Description: Completes the task. Use this when you have found the required information.
Parameters: None
```
+ The fixed template in `chat_template` uses English for the thought process. If you want to change to another language, you need to modify the following section (currently supports Chinese and English):
```
<Important Configuration>
- Language Used
* Search Keywords: English -> Change here to "Chinese" or another language
* Thinking: English -> Change here to "Chinese" or another language
```
To see the specific chat templates for the GLM-4-0414 series models, please check the `chat_template.jinja` file in the corresponding model repository.
## Citation
If you find our work helpful, please consider citing the following paper.
```bibtex
@misc{glm2024chatglm,
title={ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools},
author={Team GLM and Aohan Zeng and Bin Xu and Bowen Wang and Chenhui Zhang and Da Yin and Diego Rojas and Guanyu Feng and Hanlin Zhao and Hanyu Lai and Hao Yu and Hongning Wang and Jiadai Sun and Jiajie Zhang and Jiale Cheng and Jiayi Gui and Jie Tang and Jing Zhang and Juanzi Li and Lei Zhao and Lindong Wu and Lucen Zhong and Mingdao Liu and Minlie Huang and Peng Zhang and Qinkai Zheng and Rui Lu and Shuaiqi Duan and Shudan Zhang and Shulin Cao and Shuxun Yang and Weng Lam Tam and Wenyi Zhao and Xiao Liu and Xiao Xia and Xiaohan Zhang and Xiaotao Gu and Xin Lv and Xinghan Liu and Xinyi Liu and Xinyue Yang and Xixuan Song and Xunkai Zhang and Yifan An and Yifan Xu and Yilin Niu and Yuantao Yang and Yueyan Li and Yushi Bai and Yuxiao Dong and Zehan Qi and Zhaoyu Wang and Zhen Yang and Zhengxiao Du and Zhenyu Hou and Zihan Wang},
year={2024},
eprint={2406.12793},
archivePrefix={arXiv},
primaryClass={id='cs.CL' full_name='Computation and Language' is_active=True alt_name='cmp-lg' in_archive='cs' is_general=False description='Covers natural language processing. Roughly includes material in ACM Subject Class I.2.7. Note that work on artificial languages (programming languages, logics, formal systems) that does not explicitly address natural-language issues broadly construed (natural-language processing, computational linguistics, speech, text retrieval, etc.) is not appropriate for this area.'}
}
```
================================================
FILE: README_20240605.md
================================================
# GLM-4
<p align="center">
📄<a href="https://arxiv.org/pdf/2406.12793" target="_blank"> Report </a> • 🤗 <a href="https://huggingface.co/collections/THUDM/glm-4-665fcf188c414b03c2f7e3b7" target="_blank">HF Repo</a> • 🤖 <a href="https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat" target="_blank">ModelScope</a> • 🟣 <a href="https://wisemodel.cn/models/ZhipuAI/glm-4-9b-chat" target="_blank">WiseModel</a> • 🐦 <a href="https://twitter.com/thukeg" target="_blank">Twitter</a> • 👋 Join <a href="https://discord.gg/8cnQKdAprg" target="_blank">Discord</a> and <a href="resources/WECHAT.md" target="_blank">WeChat</a>
</p>
<p align="center">
📍Experience and use a larger-scale GLM business model on the <a href="https://open.bigmodel.cn/?utm_campaign=open&_channel_track_key=OWTVNma9">Zhipu AI Open Platform</a>
</p>
## Update
- 🔥🔥 **News**: ```2024/11/01```: Dependencies have been updated in this repository. Please update the dependencies in
`requirements.txt` to ensure the model runs correctly. The model weights
for [glm-4-9b-chat-hf](https://huggingface.co/THUDM/glm-4-9b-chat-hf) are compatible with `transformers>=4.46.2` and can
be implemented using the `GlmModel` class in the `transformers` library. Additionally, `tokenizer_chatglm.py`
in [glm-4-9b-chat](https://huggingface.co/THUDM/glm-4-9b-chat) and [glm-4v-9b](https://huggingface.co/THUDM/glm-4v-9b)
has been updated for the latest version of `transformers`. Please update the files on HuggingFace.
- 🔥 **News**: ```2024/10/27```: We have open-sourced [LongReward](https://github.com/THUDM/LongReward), a model that
uses AI feedback to enhance long-context large language models.
- 🔥 **News**: ```2024/10/25```: We have open-sourced the end-to-end Mandarin-English voice dialogue
model [GLM-4-Voice](https://github.com/THUDM/GLM-4-Voice).
- 🔥 **News**: ```2024/09/05```: We have open-sourced [longcite-glm4-9b](https://huggingface.co/THUDM/LongCite-glm4-9b),
a model enabling LLMs to produce fine-grained citations in long-context Q&A, along with the
dataset [LongCite-45k](https://huggingface.co/datasets/THUDM/LongCite-45k). Try it out online
at [Huggingface Space](https://huggingface.co/spaces/THUDM/LongCite).
- 🔥 **News**: ```2024/08/15```: We have
open-sourced [longwriter-glm4-9b](https://huggingface.co/THUDM/LongWriter-glm4-9b), a model capable of generating over
10,000 tokens in single-turn dialogue, along with the
dataset [LongWriter-6k](https://huggingface.co/datasets/THUDM/LongWriter-6k). Experience it online
at [Huggingface Space](https://huggingface.co/spaces/THUDM/LongWriter) or
the [ModelScope Community Space](https://modelscope.cn/studios/ZhipuAI/LongWriter-glm4-9b-demo).
- 🔥 **News**: ```2024/07/24```: We published the latest technical insights on long-text processing. Check out our
technical report on training the open-source GLM-4-9B model for long
texts [here](https://medium.com/@ChatGLM/glm-long-scaling-pre-trained-model-contexts-to-millions-caa3c48dea85).
- 🔥 **News**: ```2024/07/09```: The GLM-4-9B-Chat model is now compatible
with [Ollama](https://github.com/ollama/ollama) and [Llama.cpp](https://github.com/ggerganov/llama.cpp). See detailed
information in this [PR](https://github.com/ggerganov/llama.cpp/pull/8031).
- 🔥 **News**: ```2024/06/18```: We have released a [technical report](https://arxiv.org/pdf/2406.12793), available for
viewing.
- 🔥 **News**: ```2024/06/05```: We released the GLM-4-9B series of open-source models.
## Model Introduction
GLM-4-9B is the open-source version of the latest generation of pre-trained models in the GLM-4 series launched by Zhipu
AI. In the evaluation of data sets in semantics, mathematics, reasoning, code, and knowledge, **GLM-4-9B**
and its human preference-aligned version **GLM-4-9B-Chat** have shown superior performance beyond Llama-3-8B. In
addition to multi-round conversations, GLM-4-9B-Chat also has advanced features such as web browsing, code execution,
custom tool calls (Function Call), and long text reasoning (supporting up to 128K context).
This generation of models has added multi-language support, supporting 26 languages including Japanese, Korean,
and German. We have also launched the **GLM-4-9B-Chat-1M** model that supports 1M
context length (about 2 million Chinese characters) and the multimodal model GLM-4V-9B based on GLM-4-9B.
**GLM-4V-9B** possesses dialogue capabilities in both Chinese and English at a high resolution of 1120*1120.
In various multimodal evaluations, including comprehensive abilities in Chinese and English, perception & reasoning,
text recognition, and chart understanding, GLM-4V-9B demonstrates superior performance compared to
GPT-4-turbo-2024-04-09, Gemini 1.0 Pro, Qwen-VL-Max, and Claude 3 Opus.
## Model List
| Model | Type | Seq Length | Transformers Version | Download | Online Demo |
|:-------------------:|:----:|:----------:|:--------------------:|:-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|:------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|
| GLM-4-9B | Base | 8K | `4.44.0 - 4.45.0` | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/glm-4-9b) | / |
| GLM-4-9B-Chat | Chat | 128K | `>= 4.44.0` | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-9B-Chat) | [🤖 ModelScope CPU](https://modelscope.cn/studios/dash-infer/GLM-4-Chat-DashInfer-Demo/summary)<br> [🤖 ModelScope vLLM](https://modelscope.cn/studios/ZhipuAI/glm-4-9b-chat-vllm/summary) |
| GLM-4-9B-Chat-HF | Chat | 128K | `>= 4.46.0` | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat-hf)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-hf) | [🤖 ModelScope CPU](https://modelscope.cn/studios/dash-infer/GLM-4-Chat-DashInfer-Demo/summary)<br> [🤖 ModelScope vLLM](https://modelscope.cn/studios/ZhipuAI/glm-4-9b-chat-vllm/summary) |
| GLM-4-9B-Chat-1M | Chat | 1M | `>= 4.44.0` | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat-1m)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-1m)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-9B-Chat-1M) | / |
| GLM-4-9B-Chat-1M-HF | Chat | 1M | `>= 4.46.0` | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat-1m-hf)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-1m-hf) | / |
| GLM-4V-9B | Chat | 8K | `>= 4.46.0` | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4v-9b)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4v-9b)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4V-9B) | [🤖 ModelScope](https://modelscope.cn/studios/ZhipuAI/glm-4v-9b-Demo/summary) |
## BenchMarkß
### Typical Tasks
| Model | AlignBench | MT-Bench | IFEval | MMLU | C-Eval | GSM8K | MATH | HumanEval | NaturalCodeBench |
|:--------------------|:----------:|:--------:|:------:|:----:|:------:|:-----:|:----:|:---------:|:----------------:|
| Llama-3-8B-Instruct | 6.40 | 8.00 | 68.58 | 68.4 | 51.3 | 79.6 | 30.0 | 62.2 | 24.7 |
| ChatGLM3-6B | 5.18 | 5.50 | 28.1 | 66.4 | 69.0 | 72.3 | 25.7 | 58.5 | 11.3 |
| GLM-4-9B-Chat | 7.01 | 8.35 | 69.0 | 72.4 | 75.6 | 79.6 | 50.6 | 71.8 | 32.2 |
### Base Model
| Model | MMLU | C-Eval | GPQA | GSM8K | MATH | HumanEval |
|:--------------------|:----:|:------:|:----:|:-----:|:----:|:---------:|
| Llama-3-8B | 66.6 | 51.2 | - | 45.8 | - | 33.5 |
| Llama-3-8B-Instruct | 68.4 | 51.3 | 34.2 | 79.6 | 30.0 | 62.2 |
| ChatGLM3-6B-Base | 61.4 | 69.0 | 26.8 | 72.3 | 25.7 | 58.5 |
| GLM-4-9B | 74.7 | 77.1 | 34.3 | 84.0 | 30.4 | 70.1 |
> Since `GLM-4-9B` adds some math, reasoning, and code-related instruction data during pre-training, Llama-3-8B-Instruct
> is also included in the comparison range.
### Long Context
The [needle-in-the-haystack experiment](https://github.com/LargeWorldModel/LWM/blob/main/scripts/eval_needle.py) was
conducted with a context length of 1M, and the results are as follows:
The long text capability was further evaluated on LongBench-Chat, and the results are as follows:
<p align="center">
<img src="resources/longbench.png" alt="Description text" style="display: block; margin: auto; width: 65%;">
</p>
### Multi Language
The tests for GLM-4-9B-Chat and Llama-3-8B-Instruct are conducted on six multilingual datasets. The test results and the
corresponding languages selected for each dataset are shown in the table below:
| Dataset | Llama-3-8B-Instruct | GLM-4-9B-Chat | Languages |
|:------------|:-------------------:|:-------------:|:----------------------------------------------------------------------------------------------:|
| M-MMLU | 49.6 | 56.6 | all |
| FLORES | 25.0 | 28.8 | ru, es, de, fr, it, pt, pl, ja, nl, ar, tr, cs, vi, fa, hu, el, ro, sv, uk, fi, ko, da, bg, no |
| MGSM | 54.0 | 65.3 | zh, en, bn, de, es, fr, ja, ru, sw, te, th |
| XWinograd | 61.7 | 73.1 | zh, en, fr, jp, ru, pt |
| XStoryCloze | 84.7 | 90.7 | zh, en, ar, es, eu, hi, id, my, ru, sw, te |
| XCOPA | 73.3 | 80.1 | zh, et, ht, id, it, qu, sw, ta, th, tr, vi |
### Function Call
Tested
on [Berkeley Function Calling Leaderboard](https://github.com/ShishirPatil/gorilla/tree/main/berkeley-function-call-leaderboard).
| Model | Overall Acc. | AST Summary | Exec Summary | Relevance |
|:-----------------------|:------------:|:-----------:|:------------:|:---------:|
| Llama-3-8B-Instruct | 58.88 | 59.25 | 70.01 | 45.83 |
| gpt-4-turbo-2024-04-09 | 81.24 | 82.14 | 78.61 | 88.75 |
| ChatGLM3-6B | 57.88 | 62.18 | 69.78 | 5.42 |
| GLM-4-9B-Chat | 81.00 | 80.26 | 84.40 | 87.92 |
### Multi-Modal
GLM-4V-9B is a multimodal language model with visual understanding capabilities. The evaluation results of its related
classic tasks are as follows:
| | **MMBench-EN-Test** | **MMBench-CN-Test** | **SEEDBench_IMG** | **MMStar** | **MMMU** | **MME** | **HallusionBench** | **AI2D** | **OCRBench** |
|----------------------------|---------------------|---------------------|-------------------|------------|----------|---------|--------------------|----------|--------------|
| **gpt-4o-2024-05-13** | 83.4 | 82.1 | 77.1 | 63.9 | 69.2 | 2310.3 | 55 | 84.6 | 736 |
| **gpt-4-turbo-2024-04-09** | 81.0 | 80.2 | 73.0 | 56.0 | 61.7 | 2070.2 | 43.9 | 78.6 | 656 |
| **gpt-4-1106-preview** | 77.0 | 74.4 | 72.3 | 49.7 | 53.8 | 1771.5 | 46.5 | 75.9 | 516 |
| **InternVL-Chat-V1.5** | 82.3 | 80.7 | 75.2 | 57.1 | 46.8 | 2189.6 | 47.4 | 80.6 | 720 |
| **LLaVA-Next-Yi-34B** | 81.1 | 79 | 75.7 | 51.6 | 48.8 | 2050.2 | 34.8 | 78.9 | 574 |
| **Step-1V** | 80.7 | 79.9 | 70.3 | 50.0 | 49.9 | 2206.4 | 48.4 | 79.2 | 625 |
| **MiniCPM-Llama3-V2.5** | 77.6 | 73.8 | 72.3 | 51.8 | 45.8 | 2024.6 | 42.4 | 78.4 | 725 |
| **Qwen-VL-Max** | 77.6 | 75.7 | 72.7 | 49.5 | 52 | 2281.7 | 41.2 | 75.7 | 684 |
| **Gemini 1.0 Pro** | 73.6 | 74.3 | 70.7 | 38.6 | 49 | 2148.9 | 45.7 | 72.9 | 680 |
| **Claude 3 Opus** | 63.3 | 59.2 | 64 | 45.7 | 54.9 | 1586.8 | 37.8 | 70.6 | 694 |
| **GLM-4V-9B** | 81.1 | 79.4 | 76.8 | 58.7 | 47.2 | 2163.8 | 46.6 | 81.1 | 786 |
## Quick call
**For hardware configuration and system requirements, please check [here](basic_demo/README_en.md).**
### Use the following method to quickly call the GLM-4-9B-Chat language model
Use the transformers backend for inference:
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import os
os.environ[
'CUDA_VISIBLE_DEVICES'] = '0' # Set the GPU number. If inference with multiple GPUs, set multiple GPU numbers
MODEL_PATH = "THUDM/glm-4-9b-chat-hf"
device = "cuda" if torch.cuda.is_available() else "cpu"
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, trust_remote_code=True)
query = "你好"
inputs = tokenizer.apply_chat_template([{"role": "user", "content": query}],
add_generation_prompt=True,
tokenize=True,
return_tensors="pt",
return_dict=True
)
inputs = inputs.to(device)
model = AutoModelForCausalLM.from_pretrained(
MODEL_PATH,
torch_dtype=torch.bfloat16,
low_cpu_mem_usage=True,
trust_remote_code=True,
device_map="auto"
).eval()
gen_kwargs = {"max_length": 2500, "do_sample": True, "top_k": 1}
with torch.no_grad():
outputs = model.generate(**inputs, **gen_kwargs)
outputs = outputs[:, inputs['input_ids'].shape[1]:]
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```
Use the vLLM backend for inference:
```python
from transformers import AutoTokenizer
from vllm import LLM, SamplingParams
# GLM-4-9B-Chat
# If you encounter OOM, you can try to reduce max_model_len or increase tp_size
max_model_len, tp_size = 131072, 1
model_name = "THUDM/glm-4-9b-chat-hf"
prompt = [{"role": "user", "content": "你好"}]
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
llm = LLM(
model=model_name,
tensor_parallel_size=tp_size,
max_model_len=max_model_len,
trust_remote_code=True,
enforce_eager=True,
# if you encounter OOM in GLM-4-9B-Chat-1M, you can try to enable the following parameters
# enable_chunked_prefill=True,
# max_num_batched_tokens=8192
)
stop_token_ids = [151329, 151336, 151338]
sampling_params = SamplingParams(temperature=0.95, max_tokens=1024, stop_token_ids=stop_token_ids)
inputs = tokenizer.apply_chat_template(prompt, tokenize=False, add_generation_prompt=True)
outputs = llm.generate(prompts=inputs, sampling_params=sampling_params)
print(outputs[0].outputs[0].text)
```
### Use the following method to quickly call the GLM-4V-9B multimodal model
Use the transformers backend for inference:
```python
import torch
from PIL import Image
from transformers import AutoModelForCausalLM, AutoTokenizer
import os
os.environ[
'CUDA_VISIBLE_DEVICES'] = '0' # Set the GPU number. If inference with multiple GPUs, set multiple GPU numbers
MODEL_PATH = "THUDM/glm-4v-9b"
device = "cuda" if torch.cuda.is_available() else "cpu"
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, trust_remote_code=True)
query = '描述这张图片'
image = Image.open("your image").convert('RGB')
inputs = tokenizer.apply_chat_template([{"role": "user", "image": image, "content": query}],
add_generation_prompt=True, tokenize=True, return_tensors="pt",
return_dict=True) # chat mode
inputs = inputs.to(device)
model = AutoModelForCausalLM.from_pretrained(
MODEL_PATH,
torch_dtype=torch.bfloat16,
low_cpu_mem_usage=True,
trust_remote_code=True,
device_map="auto"
).eval()
gen_kwargs = {"max_length": 2500, "do_sample": True, "top_k": 1}
with torch.no_grad():
outputs = model.generate(**inputs, **gen_kwargs)
outputs = outputs[:, inputs['input_ids'].shape[1]:]
print(tokenizer.decode(outputs[0]))
```
Use the vLLM backend for inference:
```python
from PIL import Image
from vllm import LLM, SamplingParams
model_name = "THUDM/glm-4v-9b"
llm = LLM(model=model_name,
tensor_parallel_size=1,
max_model_len=8192,
trust_remote_code=True,
enforce_eager=True)
stop_token_ids = [151329, 151336, 151338]
sampling_params = SamplingParams(temperature=0.2,
max_tokens=1024,
stop_token_ids=stop_token_ids)
prompt = "What's the content of the image?"
image = Image.open("your image").convert('RGB')
inputs = {
"prompt": prompt,
"multi_modal_data": {
"image": image
},
}
outputs = llm.generate(inputs, sampling_params=sampling_params)
for o in outputs:
generated_text = o.outputs[0].text
print(generated_text)
```
## Complete project list
If you want to learn more about the GLM-4-9B series open source models, this open source repository provides developers
with basic GLM-4-9B usage and development code through the following content
+ [basic_demo](basic_demo/README.md): Contains
+ Interaction code using transformers and vLLM backend
+ OpenAI API backend interaction code
+ Batch reasoning code
+ [composite_demo](composite_demo/README.md): Contains
+ Fully functional demonstration code for GLM-4-9B and GLM-4V-9B open source models, including All Tools capabilities,
long document interpretation, and multimodal capabilities.
+ [fintune_demo](finetune_demo/README.md): Contains
+ PEFT (LORA, P-Tuning) fine-tuning code
+ SFT fine-tuning code
+ [intel_device_demo](intel_device_demo/): Contains
+ OpenVINO deployment code
+ Intel® Extension for Transformers deployment code
## Friendly Links
+ [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory): Efficient open-source fine-tuning framework,
already supports GLM-4-9B-Chat language model fine-tuning.
+ [SWIFT](https://github.com/modelscope/swift): LLM/VLM training framework from ModelScope, supports
GLM-4-9B-Chat / GLM-4V-9b fine-tuning.
+ [Xorbits Inference](https://github.com/xorbitsai/inference): Performance-enhanced and comprehensive global inference
framework, easily deploy your own models or import cutting-edge open source models with one click.
+ [LangChain-ChatChat](https://github.com/chatchat-space/Langchain-Chatchat): RAG and Agent applications based on
language models such as Langchain and ChatGLM
+ [self-llm](https://github.com/datawhalechina/self-llm/tree/master/models/GLM-4): Datawhale's self-llm project, which
includes
the GLM-4-9B open source model cookbook.
+ [chatglm.cpp](https://github.com/li-plus/chatglm.cpp): Real-time inference on your laptop accelerated by quantization,
similar to llama.cpp.
+ [OpenVINO](https://github.com/openvinotoolkit): glm-4-9b-chat already supports the use of OpenVINO. The toolkit accelerates inference and has a greater inference speed improvement on Intel's GPU, GPU and NPU devices. For
specific usage, please refer to [OpenVINO notebooks](https://github.com/openvinotoolkit/openvino_notebooks/blob/latest/notebooks/llm-chatbot/llm-chatbot-generate-api.ipynb)
## License
+ The use of GLM-4 model weights must follow
the [Model License](https://huggingface.co/THUDM/glm-4-9b/blob/main/LICENSE).
+ The code in this open source repository follows the [Apache 2.0](LICENSE) license.
Please strictly follow the open source license.
## Reference
If you find our work helpful, please consider citing the following paper.
```
@misc{glm2024chatglm,
title={ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools},
author={Team GLM and Aohan Zeng and Bin Xu and Bowen Wang and Chenhui Zhang and Da Yin and Diego Rojas and Guanyu Feng and Hanlin Zhao and Hanyu Lai and Hao Yu and Hongning Wang and Jiadai Sun and Jiajie Zhang and Jiale Cheng and Jiayi Gui and Jie Tang and Jing Zhang and Juanzi Li and Lei Zhao and Lindong Wu and Lucen Zhong and Mingdao Liu and Minlie Huang and Peng Zhang and Qinkai Zheng and Rui Lu and Shuaiqi Duan and Shudan Zhang and Shulin Cao and Shuxun Yang and Weng Lam Tam and Wenyi Zhao and Xiao Liu and Xiao Xia and Xiaohan Zhang and Xiaotao Gu and Xin Lv and Xinghan Liu and Xinyi Liu and Xinyue Yang and Xixuan Song and Xunkai Zhang and Yifan An and Yifan Xu and Yilin Niu and Yuantao Yang and Yueyan Li and Yushi Bai and Yuxiao Dong and Zehan Qi and Zhaoyu Wang and Zhen Yang and Zhengxiao Du and Zhenyu Hou and Zihan Wang},
year={2024},
eprint={2406.12793},
archivePrefix={arXiv},
primaryClass={id='cs.CL' full_name='Computation and Language' is_active=True alt_name='cmp-lg' in_archive='cs' is_general=False description='Covers natural language processing. Roughly includes material in ACM Subject Class I.2.7. Note that work on artificial languages (programming languages, logics, formal systems) that does not explicitly address natural-language issues broadly construed (natural-language processing, computational linguistics, speech, text retrieval, etc.) is not appropriate for this area.'}
}
```
```
@misc{wang2023cogvlm,
title={CogVLM: Visual Expert for Pretrained Language Models},
author={Weihan Wang and Qingsong Lv and Wenmeng Yu and Wenyi Hong and Ji Qi and Yan Wang and Junhui Ji and Zhuoyi Yang and Lei Zhao and Xixuan Song and Jiazheng Xu and Bin Xu and Juanzi Li and Yuxiao Dong and Ming Ding and Jie Tang},
year={2023},
eprint={2311.03079},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
================================================
FILE: README_zh.md
================================================
# GLM-4-0414 系列模型
<p align="center">
👋 加入我们的 <a href="https://discord.gg/8cnQKdAprg" target="_blank">Discord</a>, <a href="https://x.com/Zai_org" target="_blank">X</a> 和 <a href="resources/WECHAT.md" target="_blank"> 微信 </a>
</p>
<p align="center">
📍本次开源模型可以在 <a href="https://chat.z.ai">Z.ai</a> 免费体验；使用 GLM 商业模型服务请到 <a href="https://bigmodel.cn">bigmodel.cn</a>。
</p>
Read this in [English](README)
## 项目更新
- 🔥 **News**：`2025/07/02`：我们正式发布 [GLM-4.1V-9B-Thinking](https://huggingface.co/collections/THUDM/glm-41v-thinking-6862bbfc44593a8601c2578d) 系列视觉理解模型，更多信息请查看 [GitHub 仓库](https://github.com/THUDM/GLM-4.1V-Thinking)。
- **News**: ```2025/04/14```: 我们发布 [GLM-4-32B-0414](https://huggingface.co/collections/THUDM/glm-4-0414-67f3cbcb34dd9d252707cb2e) 系列模型，规模提升至 32B，包含对话、推理、沉思多种能力的模型。
- **News**: ``2024/06/18``: 我们发布 [技术报告](https://arxiv.org/pdf/2406.12793), 欢迎查看。
- **News**: ``2024/06/05``: 我们发布 `GLM-4-9B` 系列开源模型，其内容可以在[这里](README_240605.md)查看。
## 模型介绍
GLM 家族迎来新一代开源模型 **GLM-4-32B-0414** 系列，320 亿参数，效果比肩 OpenAI 的 GPT 系列和 DeepSeek 的 V3/R1 系列，且支持非常友好的本地部署特性。GLM-4-32B-Base-0414 经过 15T 高质量数据的预训练，其中包含大量推理类的合成数据，这为后续的强化学习扩展打下了基础。在后训练阶段，除了针对对话场景进行了人类偏好对齐外，我们还通过拒绝采样和强化学习等技术强化了模型在指令遵循、工程代码、函数调用方面的效果，加强了智能体任务所需的原子能力。GLM-4-32B-0414 在工程代码、Artifacts 生成、函数调用、搜索问答及报告等方面都取得了不错的效果，部分 Benchmark 甚至可以媲美更大规模的 GPT-4o、DeepSeek-V3-0324（671B）等模型。
**GLM-Z1-32B-0414** 是具有**深度思考能力**的推理模型，这是在 GLM-4-32B-0414 的基础上，通过冷启动和扩展强化学习，以及在数学、代码和逻辑等任务上对模型的进一步训练得到的。相对于基础模型，GLM-Z1-32B-0414 显著提升了数理能力和解决复杂任务的能力。在训练的过程中，我们还引入了基于对战排序反馈的通用强化学习，进一步增强了模型的通用能力。
**GLM-Z1-Rumination-32B-0414** 是具有**沉思能力**的深度推理模型（对标 Open AI 的 Deep Research）。不同于一般的深度思考模型，沉思模型通过更长时间的深度思考来解决更开放和复杂的问题（例如：撰写两个城市AI发展对比情况，以及未来的发展规划），沉思模型在深度思考过程中结合搜索工具处理复杂任务，并经过利用多种规则型奖励来指导和扩展端到端强化学习训练得到。Z1-Rumination 在研究型写作和复杂检索任务上的能力得到了显著提升。
最后，**GLM-Z1-9B-0414** 是一个惊喜。我们沿用上述一系列技术，训练了一个保持开源传统的 9B 小尺寸模型。尽管规模更小，GLM-Z1-9B-0414 在数学推理和通用任务中依然展现出极为优秀的能力，其整体表现已处于同尺寸开源模型中的领先水平。特别是在资源受限的场景下，该模型在效率与效果之间实现了出色的平衡，为追求轻量化部署的用户提供了强有力的选择。
## 效果展示
### 动画绘制
<table>
<tr>
<td style="text-align: center; font-size: 16px; font-weight: bold; padding: 10px; width: 420px;">
GLM-Z1-32B-0414
</td>
<td style="text-align: center; font-size: 16px; font-weight: bold; padding: 10px; width: 420px;">
GLM-4-32B-0414
</td>
</tr>
<tr>
<td style="vertical-align: top; padding: 10px; width: 420px;">
<video src="https://github.com/user-attachments/assets/849ff9fd-b54d-4c74-9ee5-3412e1a09e32"
style="width: 400px; height: 300px; object-fit: contain;" autoplay loop muted playsinline></video>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;">
write a Python program that shows a ball bouncing inside a spinning hexagon. The ball should be affected by gravity and friction, and it must bounce off the rotating walls realistically
</div>
</td>
<td style="vertical-align: top; padding: 10px; width: 420px;">
<video src="https://github.com/user-attachments/assets/8dccdb9d-cc44-4732-b438-74a4e3cb9dfb"
style="width: 400px; height: 300px; object-fit: contain;" autoplay loop muted playsinline></video>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;">
用 HTML 模拟一个小球在从一个旋转中的六边形中心释放后的场景。考虑小球和六边形边框的碰撞和小球受到的重力，并假设碰撞都是完全弹性碰撞
</div>
</td>
</tr>
</table>
### 网页设计
<table>
<tr>
<td style="text-align: center; font-size: 16px; font-weight: bold; padding: 10px; width: 420px;">
GLM-4-32B-0414
</td>
<td style="text-align: center; font-size: 16px; font-weight: bold; padding: 10px; width: 420px;">
GLM-4-32B-0414
</td>
</tr>
<tr>
<td style="vertical-align: top; padding: 10px; width: 420px;">
<img src="https://github.com/user-attachments/assets/bd9c1fc1-c784-4e8f-9c76-5f7389a715f1"/>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;">
设计一个支持自定义函数绘制的绘图板，可以添加和删除自定义函数，并为函数指定颜色
</div>
</td>
<td style="vertical-align: top; padding: 10px; width: 420px;">
<img src="https://github.com/user-attachments/assets/7ad12d52-9229-4278-8d1b-ffbf43e99070"/>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;"> 给我设计一个移动端机器学习平台的 UI，其中要包括训练任务，存储管理，和个人统计信息界面。个人信息统计界面要用图表展示用户过去一段时间的各类资源使用情况。使用 Tailwind CSS 来美化页面，把这 3 个手机界面平铺展示到一个 HTML 页面中 </div>
</td>
</tr>
</table>
### SVG 生成
<table>
<tr>
<td style="text-align: center; font-size: 16px; font-weight: bold; padding: 10px; width: 420px;">
GLM-4-32B-0414
</td>
<td style="text-align: center; font-size: 16px; font-weight: bold; padding: 10px; width: 420px;">
GLM-4-32B-0414
</td>
</tr>
<tr>
<td style="vertical-align: top; padding: 10px; width: 420px;">
<img src="https://github.com/user-attachments/assets/9407e4c1-1876-4ab5-838c-839836fb418a"/>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;">
用SVG创作一幅烟雨江南
</div>
</td>
<td style="vertical-align: top; padding: 10px; width: 420px;">
<img src="https://github.com/user-attachments/assets/bcce8c5a-cedf-45c8-b666-ddb023d5b49c"/>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;"> 用 SVG 展示一个 LLM 的训练流程 </div>
</td>
</tr>
</table>
### 分析调研撰写
<td style="vertical-align: top; padding: 10px; width: 420px;">
<video src="https://github.com/user-attachments/assets/7939c8c5-0fcf-4bc4-be45-3964aad0e61c" style="width: 400px; height: 300px; object-fit: contain;" autoplay loop muted playsinline></video>
<div style="margin-top: 10px; font-size: 14px; color: #333; width: 400px;">
中国城市 AI 发展分析：北京与杭州的对比研究。同时调研国外城市用 AI 进行城市治理的案例。
</div>
</td>
## 模型列表
### GLM-4-0414 系列模型
GLM-Z1-9B-0414 开源模型 [在线体验](https://modelscope.cn/studios/ZhipuAI/GLM-Z1-9B-0414/summary)
| Model | Type | Seq Length* | Download |
|:--------------------------:|:---------:|:-----------:|:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|
| GLM-4-9B-0414 | Chat | 32K -> 128K | [🤗 Huggingface](https://huggingface.co/THUDM/GLM-4-9B-0414)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/GLM-4-9B-0414)<br> [🧩 Modelers](https://modelers.cn/models/zhipuai/GLM-4-9B-0414)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-9B-0414) |
| GLM-Z1-9B-0414 | Reasoning | 32K -> 128K | [🤗 Huggingface](https://huggingface.co/THUDM/GLM-4-Z1-9B-0414)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/GLM-Z1-9B-0414)<br> [🧩 Modelers](https://modelers.cn/models/zhipuai/GLM-Z1-9B-0414)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-Z1-9B-0414) |
| GLM-4-32B-Base-0414 | Base | 32K -> 128K | [🤗 Huggingface](https://huggingface.co/THUDM/GLM-4-32B-Base-0414)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/GLM-4-32B-Base-0414)<br> [🧩 Modelers](https://modelers.cn/models/zhipuai/GLM-4-32B-Base-0414)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-32B-Base-0414) |
| GLM-4-32B-0414 | Chat | 32K -> 128K | [🤗 Huggingface](https://huggingface.co/THUDM/GLM-4-32B-0414)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/GLM-4-32B-0414)<br> [🧩 Modelers](https://modelers.cn/models/zhipuai/GLM-4-32B-0414)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-32B-Base-0414) |
| GLM-Z1-32B-0414 | Reasoning | 32K -> 128K | [🤗 Huggingface](https://huggingface.co/THUDM/GLM-Z1-32B-0414)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/GLM-Z1-32B-0414)<br> [🧩 Modelers](https://modelers.cn/models/zhipuai/GLM-Z1-32B-0414)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-Z1-32B-0414) |
| GLM-Z1-Rumination-32B-0414 | Reasoning | 128K | [🤗 Huggingface](https://huggingface.co/THUDM/GLM-Z1-Rumination-32B-0414)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/GLM-Z1-Rumination-32B-0414)<br> [🧩 Modelers](https://modelers.cn/models/zhipuai/GLM-Z1-Rumination-32B-0414)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-Z1-Rumination-32B-0414) |
GLM-4-9B-0414 由于其较小的模型容量，我们未对其智能体能力进行类似 GLM-4-32B-0414 的强化，主要针对翻译等需要大批量调用的场景进行优化。
\* 模型原生采用 32K 上下文进行训练，对于输入 + 输出长度可能超过 32K 的请求，我们建议激活 YaRN 来获得较好的外推性能，详情见[部署章节](#%E6%A8%A1%E5%9E%8B%E5%92%8C%E6%8F%90%E7%A4%BA%E8%AF%8D%E5%AE%9E%E7%8E%B0)。
以下为 2024 年 6 月 5 日发布的 GLM-4 系列模型，其详细内容可以在[这里](README_zh_240605.md)查看。
| Model | Type | Seq Length* | Download |
|:-----------------------------:|:---------:|:----------:|:-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|
| GLM-4-9B | Base | 8K | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b)<br> |
| GLM-4-9B-Chat | Chat | 128K | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-9B-Chat) |
| GLM-4-9B-Chat-HF | Chat | 128K | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat-hf)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-hf) |
| GLM-4-9B-Chat-1M | Chat | 1M | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat-1m)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-1m)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-9B-Chat-1M) |
| GLM-4-9B-Chat-1M-HF | Chat | 1M | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat-1m-hf)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-1m-hf) |
| GLM-4V-9B | Chat | 8K | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4v-9b)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4v-9b)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4V-9B) |
## 评测结果
### GLM-4-0414 系列
<div style="text-align: center;">
<img src="resources/Bench-32B.png" style="width: 80%;" />
</div>
| 模型 | IFEval | BFCL-v3 (Overall) | BFCL-v3 (MultiTurn) | TAU-Bench (Retail) | TAU-Bench (Airline) | SimpleQA | HotpotQA |
| ---------------- | ------ | ----------------- | ------------------- | ------------------ | ------------------- | -------- | -------- |
| Qwen2.5-Max | 85.6 | 50.9 | 30.5 | 58.3 | 22.0 | 79.0 | 52.8 |
| GPT-4o-1120 | 81.9 | 69.6 | 41.0 | 62.8 | 46.0 | 82.8 | 63.9 |
| DeepSeek-V3-0324 | 83.4 | 66.2 | 35.8 | 60.7 | 32.4 | 82.6 | 54.6 |
| DeepSeek-R1 | 84.3 | 57.5 | 12.4 | 33.0 | 37.3 | 83.9 | 63.1 |
| GLM-4-32B-0414 | 87.6 | 69.6 | 41.5 | 68.7 | 51.2 | 88.1 | 63.8 |
> 对于 `SimpleQA` 和 `HotpotQA`，我们分别从测试集中采样了近500条测试样例，提供所有模型最基础的 `search` 和 `click` 工具，另外确保其余 Setting 保持一致后，3次评测取平均值
| 模型 | 框架 | [SWE-bench Verified](https://openai.com/index/introducing-swe-bench-verified/) | [SWE-bench Verified mini](https://github.com/mariushobbhahn/SWEBench-verified-mini) |
|---|--------------------------|---|-------------------------------------------------------------------------------------|
| GLM-4-32B-0414 | Moatless<sup>[1]</sup> | 33.8 | 38.0 |
| GLM-4-32B-0414 | Agentless<sup>[2]</sup> | 30.7 | 34.0 |
| GLM-4-32B-0414 | OpenHands<sup>[3]</sup> | 27.2 | 28.0 |
[1] [Moatless v0.0.3](https://github.com/aorwall/moatless-tools) 使用如下参数 `response_format="react", thoughts_in_action=False, max_interations=30`，未对失败轨迹进行重试，其余为默认配置
[2] [Agentless v1.5.0](https://github.com/OpenAutoCoder/Agentless) 其中的 Embedding 模型使用了 [BGE](https://github.com/FlagOpen/FlagEmbedding/blob/master/README_zh.md)，基于[FAISS](https://github.com/facebookresearch/faiss)进行相似性检索，为加快patch验证的速度同时尽可能保证效果，将运行单个实例的超时时间从默认的300s修改为180s
[3] [OpenHands v0.29.1](https://github.com/All-Hands-AI/OpenHands/tree/main) 未采用 YaRN 上下文扩展，而是限制了最大 60 个 iterations，并对 history 进行 summarization 以防止超出 32K 上下文限制，summarization 配置为 `llm_config="condenser", keep_first=1, max_size=32`，同样未对失败轨迹进行重试
### GLM-Z1-0414 系列
<div style="text-align: center;">
<img src="resources/Bench-Z1-9B.png" style="width: 80%;" />
<img src="resources/Bench-Z1-32B.png" style="width: 80%;" />
</div>
## 模型和提示词实现
### 模型实现
如果你想查看我们的模型实现，欢迎查看在相关仓库的模型实现 Pull Request，他们已经被合并。
+ [vLLM 模型实现](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/glm4.py)
+ [transformers 模型实现](https://github.com/huggingface/transformers/blob/main/src/transformers/models/glm4/modeling_glm4.py)
+ [llama.cpp 模型实现](https://github.com/ggml-org/llama.cpp/pull/12867)
### 处理长上下文（YaRN）
如果模型的输出 + 输出 token 数可能超过模型的原生上下文长度（GLM-4-0414系列多数为32k），建议开启 YaRN 来获得更好的长上下文建模能力。对于支持的框架，你可以在对应的`config.json`中修改。具体地，对于 GLM-Z1 系列模型，当输入长度超过 **8,192 tokens** 时，考虑启用 YaRN（Rope Scaling）。
```json
"rope_scaling": {
"factor": 4.0,
"original_max_position_embeddings": 32768,
"type": "yarn"
}
```
对于多数用户请求，如果输出 + 输出 token 数不会超过原生上下文长度，则无需任何修改。
### 模型微调
可以在`finetune/README.md`下找到模型的微调所需算力信息和案例的微调脚本。
可以通过以下命令开启一个简单的模型微调案例
```shell
cd finetune
pip install -r ../inference/requirements.txt
pip install -r requirements.txt
# Use single GPU for Chat Fine-tune
python finetune.py data/AdvertiseGen/ THUDM/GLM-4-9B-0414 configs/lora.yaml
```
🎉 脚本也支持使用**SwanLab**进行微调可视化跟踪，可以访问[SwanLab可视化看板](https://swanlab.cn/@ShaohonChen/GLM4-Finetune/overview)获得案例微调脚本的训练日志。
### 提示词实现
如果你使用`transformers`库提供的`apply_chat_template`方法构建提示词。以下是对不同 GLM-4-0414 模型中 `系统提示词`的限制。
+ `GLM-4-32B-Base-0414`: 基座模型，无对话模板。
+ `GLM-4-*-0414` / `GLM-Z1-*-0414`: 如果传入`tools`，则由 `apply_chat_template` 填充工具到`chat_template`中的固定模板，单独作为一条带有`tools`绑定的 `system`字段信息并拼接于`messages[0]`。原本传入的所有 `messages` 自动往后移动一个位置。
+ `GLM-Z1-Rumination-32B-0414`:
+ 不支持自定义系统提示词，不支持自定义工具，你的所有 `tools` 和 `system` 字段会被 `apply_chat_template` 忽略。使用该模型需要外接搜索引擎或者自定义retrieval API。
+ 一共支持四个工具，分别是
```
1. search
描述: 执行搜索查询并返回搜索结果。当您需要查找有关特定主题的信息时使用此功能。
参数: query (字符串) - 搜索查询字符串，除非是中文专有名词，否则使用英文单词
2. click
描述: 点击搜索结果中的链接并导航到相应页面。当您需要查看特定搜索结果的详细内容时使用此功能。
参数: link_id (整数) - 要点击的链接ID（来自搜索结果中的序号）
3. open
描述: 打开特定网站。通过URL获取任何网站的内容。
参数: url (字符串) - 目标网站URL或域名
4. finish
描述: 完成任务。当您已找到所需信息时使用此功能。
参数: 无
```
+ `chat_template`中的固定模板使用英文思过程，如果要更换其他语言，需要修改以下部分（暂时支持中文和英文）
```
<重要配置>
- 采用语言
* 搜索关键词：英文 -> 在这里换成“中文”或者其他语言
* 思考：英文 -> 在这里换成“中文”或者其他语言
```
GLM-4-0414 系列模型的提示词构造可以前往对应的模型仓库中的 `chat_template.jinja` 查看具体的模型对话模板。
## 引用
如果你觉得我们的工作有帮助的话，请考虑引用下列论文。
```
@misc{glm2024chatglm,
title={ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools},
author={Team GLM and Aohan Zeng and Bin Xu and Bowen Wang and Chenhui Zhang and Da Yin and Diego Rojas and Guanyu Feng and Hanlin Zhao and Hanyu Lai and Hao Yu and Hongning Wang and Jiadai Sun and Jiajie Zhang and Jiale Cheng and Jiayi Gui and Jie Tang and Jing Zhang and Juanzi Li and Lei Zhao and Lindong Wu and Lucen Zhong and Mingdao Liu and Minlie Huang and Peng Zhang and Qinkai Zheng and Rui Lu and Shuaiqi Duan and Shudan Zhang and Shulin Cao and Shuxun Yang and Weng Lam Tam and Wenyi Zhao and Xiao Liu and Xiao Xia and Xiaohan Zhang and Xiaotao Gu and Xin Lv and Xinghan Liu and Xinyi Liu and Xinyue Yang and Xixuan Song and Xunkai Zhang and Yifan An and Yifan Xu and Yilin Niu and Yuantao Yang and Yueyan Li and Yushi Bai and Yuxiao Dong and Zehan Qi and Zhaoyu Wang and Zhen Yang and Zhengxiao Du and Zhenyu Hou and Zihan Wang},
year={2024},
eprint={2406.12793},
archivePrefix={arXiv},
primaryClass={id='cs.CL' full_name='Computation and Language' is_active=True alt_name='cmp-lg' in_archive='cs' is_general=False description='Covers natural language processing. Roughly includes material in ACM Subject Class I.2.7. Note that work on artificial languages (programming languages, logics, formal systems) that does not explicitly address natural-language issues broadly construed (natural-language processing, computational linguistics, speech, text retrieval, etc.) is not appropriate for this area.'}
}
```
================================================
FILE: README_zh_240605.md
================================================
# GLM-4
<p align="center">
📄<a href="https://arxiv.org/pdf/2406.12793" target="_blank"> Report </a> • 🤗 <a href="https://huggingface.co/collections/THUDM/glm-4-665fcf188c414b03c2f7e3b7" target="_blank">HF Repo</a> • 🤖 <a href="https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat" target="_blank">ModelScope</a> • 🟣 <a href="https://wisemodel.cn/models/ZhipuAI/glm-4-9b-chat" target="_blank">WiseModel</a> • 🐦 <a href="https://twitter.com/thukeg" target="_blank">Twitter</a> • 👋 加入我们的 <a href="https://discord.gg/8cnQKdAprg" target="_blank">Discord</a> 和 <a href="resources/WECHAT.md" target="_blank">微信</a>
</p>
<p align="center">
📍在 <a href="https://open.bigmodel.cn/?utm_campaign=open&_channel_track_key=OWTVNma9">智谱AI开放平台</a> 体验和使用更大规模的 GLM 商业模型。
</p>
Read this in [English](README_en.md)
## 项目更新
- 🔥🔥 **News**: ```2024/12/10```: 本仓库微调代码支持使用`Ascend NPU`进行微调。请更新微调代码并查看代码内注释。
- 🔥 **News**: ```2024/11/01```: 本仓库依赖进行升级，请更新`requirements.txt`中的依赖以保证正常运行模型。[glm-4-9b-chat-hf](https://huggingface.co/THUDM/glm-4-9b-chat-hf) 是适配 `transformers>=4.46.2` 的模型权重，使用 `transformers` 库中的 `GlmModel` 类实现。
同时，[glm-4-9b-chat](https://huggingface.co/THUDM/glm-4-9b-chat), [glm-4v-9b](https://huggingface.co/THUDM/glm-4v-9b) 中的 `tokenzier_chatglm.py` 已经更新以适配最新版本的 `transformers`库。请前往 HuggingFace 更新文件。
- 🔥 **News**: ```2024/10/27```: 我们开源了 [LongReward](https://github.com/THUDM/LongReward)，这是一个使用 AI 反馈改进长上下文大型语言模型。
- 🔥 **News**: ```2024/10/25```: 我们开源了端到端中英语音对话模型 [GLM-4-Voice](https://github.com/THUDM/GLM-4-Voice)。
- 🔥 **News**: ```2024/09/05``` 我们开源了使LLMs能够在长上下文问答中生成细粒度引用的模型 [longcite-glm4-9b](https://huggingface.co/THUDM/LongCite-glm4-9b) 以及数据集 [LongCite-45k](https://huggingface.co/datasets/THUDM/LongCite-45k), 欢迎在 [Huggingface Space](https://huggingface.co/spaces/THUDM/LongCite) 在线体验。
- 🔥**News**: ```2024/08/15```: 我们开源具备长文本输出能力(单轮对话大模型输出可超过1万token) 的模型 [longwriter-glm4-9b](https://huggingface.co/THUDM/LongWriter-glm4-9b) 以及数据集 [LongWriter-6k](https://huggingface.co/datasets/THUDM/LongWriter-6k), 欢迎在 [Huggingface Space](https://huggingface.co/spaces/THUDM/LongWriter) 或 [魔搭社区空间](https://modelscope.cn/studios/ZhipuAI/LongWriter-glm4-9b-demo) 在线体验。
- 🔥 **News**: ```2024/07/24```: 我们发布了与长文本相关的最新技术解读，关注 [这里](https://medium.com/@ChatGLM/glm-long-scaling-pre-trained-model-contexts-to-millions-caa3c48dea85) 查看我们在训练 GLM-4-9B 开源模型中关于长文本技术的技术报告。
- 🔥 **News**: ``2024/07/09``: GLM-4-9B-Chat 模型已适配 [Ollama](https://github.com/ollama/ollama), [Llama.cpp](https://github.com/ggerganov/llama.cpp)，您可以在 [PR](https://github.com/ggerganov/llama.cpp/pull/8031) 查看具体的细节。
- 🔥 **News**: ``2024/06/18``: 我们发布 [技术报告](https://arxiv.org/pdf/2406.12793), 欢迎查看。
- 🔥 **News**: ``2024/06/05``: 我们发布 GLM-4-9B 系列开源模型。
## 模型介绍
GLM-4-9B 是智谱 AI 推出的最新一代预训练模型 GLM-4 系列中的开源版本。 在语义、数学、推理、代码和知识等多方面的数据集测评中，
**GLM-4-9B** 及其人类偏好对齐的版本 **GLM-4-9B-Chat** 均表现出超越 Llama-3-8B 的卓越性能。除了能进行多轮对话，GLM-4-9B-Chat
还具备网页浏览、代码执行、自定义工具调用（Function Call）和长文本推理（支持最大 128K 上下文）等高级功能。本代模型增加了多语言支持，支持包括日语，韩语，德语在内的
26 种语言。我们还推出了支持 1M 上下文长度（约 200 万中文字符）的 **GLM-4-9B-Chat-1M** 模型和基于 GLM-4-9B 的多模态模型
GLM-4V-9B。**GLM-4V-9B** 具备 1120 * 1120 高分辨率下的中英双语多轮对话能力，在中英文综合能力、感知推理、文字识别、图表理解等多方面多模态评测中，GLM-4V-9B
表现出超越 GPT-4-turbo-2024-04-09、Gemini 1.0 Pro、Qwen-VL-Max 和 Claude 3 Opus 的卓越性能。
## Model List
| Model | Type | Seq Length | Transformers Version | Download | Online Demo |
|:-------------------:|:----:|:----------:|:--------------------:|:-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|:------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:|
| GLM-4-9B | Base | 8K | `4.44.0 - 4.45.0` | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/glm-4-9b) | / |
| GLM-4-9B-Chat | Chat | 128K | `>= 4.44.0` | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-9B-Chat) | [🤖 ModelScope CPU](https://modelscope.cn/studios/dash-infer/GLM-4-Chat-DashInfer-Demo/summary)<br> [🤖 ModelScope vLLM](https://modelscope.cn/studios/ZhipuAI/glm-4-9b-chat-vllm/summary) |
| GLM-4-9B-Chat-HF | Chat | 128K | `>= 4.46.0` | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat-hf)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-hf) | [🤖 ModelScope CPU](https://modelscope.cn/studios/dash-infer/GLM-4-Chat-DashInfer-Demo/summary)<br> [🤖 ModelScope vLLM](https://modelscope.cn/studios/ZhipuAI/glm-4-9b-chat-vllm/summary) |
| GLM-4-9B-Chat-1M | Chat | 1M | `>= 4.44.0` | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat-1m)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-1m)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4-9B-Chat-1M) | / |
| GLM-4-9B-Chat-1M-HF | Chat | 1M | `>= 4.46.0` | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4-9b-chat-1m-hf)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4-9b-chat-1m-hf) | / |
| GLM-4V-9B | Chat | 8K | `>= 4.46.0` | [🤗 Huggingface](https://huggingface.co/THUDM/glm-4v-9b)<br> [🤖 ModelScope](https://modelscope.cn/models/ZhipuAI/glm-4v-9b)<br> [🟣 WiseModel](https://wisemodel.cn/models/ZhipuAI/GLM-4V-9B) | [🤖 ModelScope](https://modelscope.cn/studios/ZhipuAI/glm-4v-9b-Demo/summary) |
## 评测结果
### 对话模型典型任务
| Model | AlignBench | MT-Bench | IFEval | MMLU | C-Eval | GSM8K | MATH | HumanEval | NaturalCodeBench |
|:--------------------|:----------:|:--------:|:------:|:----:|:------:|:-----:|:----:|:---------:|:----------------:|
| Llama-3-8B-Instruct | 6.40 | 8.00 | 68.6 | 68.4 | 51.3 | 79.6 | 30.0 | 62.2 | 24.7 |
| ChatGLM3-6B | 5.18 | 5.50 | 28.1 | 61.4 | 69.0 | 72.3 | 25.7 | 58.5 | 11.3 |
| GLM-4-9B-Chat | 7.01 | 8.35 | 69.0 | 72.4 | 75.6 | 79.6 | 50.6 | 71.8 | 32.2 |
### 基座模型典型任务
| Model | MMLU | C-Eval | GPQA | GSM8K | MATH | HumanEval |
|:--------------------|:----:|:------:|:----:|:-----:|:----:|:---------:|
| Llama-3-8B | 66.6 | 51.2 | - | 45.8 | - | 33.5 |
| Llama-3-8B-Instruct | 68.4 | 51.3 | 34.2 | 79.6 | 30.0 | 62.2 |
| ChatGLM3-6B-Base | 61.4 | 69.0 | 26.8 | 72.3 | 25.7 | 58.5 |
| GLM-4-9B | 74.7 | 77.1 | 34.3 | 84.0 | 30.4 | 70.1 |
> 由于 `GLM-4-9B` 在预训练过程中加入了部分数学、推理、代码相关的 instruction 数据，所以将 Llama-3-8B-Instruct 也列入比较范围。
### 长文本
在 1M 的上下文长度下进行[大海捞针实验](https://github.com/LargeWorldModel/LWM/blob/main/scripts/eval_needle.py)，结果如下：
在 LongBench-Chat 上对长文本能力进行了进一步评测，结果如下:
<p align="center">
<img src="resources/longbench.png" alt="描述文字" style="display: block; margin: auto; width: 65%;">
</p>
### 多语言能力
在六个多语言数据集上对 GLM-4-9B-Chat 和 Llama-3-8B-Instruct 进行了测试，测试结果及数据集对应选取语言如下表
| Dataset | Llama-3-8B-Instruct | GLM-4-9B-Chat | Languages |
|:------------|:-------------------:|:-------------:|:----------------------------------------------------------------------------------------------:|
| M-MMLU | 49.6 | 56.6 | all |
| FLORES | 25.0 | 28.8 | ru, es, de, fr, it, pt, pl, ja, nl, ar, tr, cs, vi, fa, hu, el, ro, sv, uk, fi, ko, da, bg, no |
| MGSM | 54.0 | 65.3 | zh, en, bn, de, es, fr, ja, ru, sw, te, th |
| XWinograd | 61.7 | 73.1 | zh, en, fr, jp, ru, pt |
| XStoryCloze | 84.7 | 90.7 | zh, en, ar, es, eu, hi, id, my, ru, sw, te |
| XCOPA | 73.3 | 80.1 | zh, et, ht, id, it, qu, sw, ta, th, tr, vi |
### 工具调用能力
我们在 [Berkeley Function Calling Leaderboard](https://github.com/ShishirPatil/gorilla/tree/main/berkeley-function-call-leaderboard)
上进行了测试并得到了以下结果：
| Model | Overall Acc. | AST Summary | Exec Summary | Relevance |
|:-----------------------|:------------:|:-----------:|:------------:|:---------:|
| Llama-3-8B-Instruct | 58.88 | 59.25 | 70.01 | 45.83 |
| gpt-4-turbo-2024-04-09 | 81.24 | 82.14 | 78.61 | 88.75 |
| ChatGLM3-6B | 57.88 | 62.18 | 69.78 | 5.42 |
| GLM-4-9B-Chat | 81.00 | 80.26 | 84.40 | 87.92 |
### 多模态能力
GLM-4V-9B 是一个多模态语言模型，具备视觉理解能力，其相关经典任务的评测结果如下：
| | **MMBench-EN-Test** | **MMBench-CN-Test** | **SEEDBench_IMG** | **MMStar** | **MMMU** | **MME** | **HallusionBench** | **AI2D** | **OCRBench** |
|----------------------------|---------------------|---------------------|-------------------|------------|----------|---------|--------------------|----------|--------------|
| **gpt-4o-2024-05-13** | 83.4 | 82.1 | 77.1 | 63.9 | 69.2 | 2310.3 | 55.0 | 84.6 | 736 |
| **gpt-4-turbo-2024-04-09** | 81.0 | 80.2 | 73.0 | 56.0 | 61.7 | 2070.2 | 43.9 | 78.6 | 656 |
| **gpt-4-1106-preview** | 77.0 | 74.4 | 72.3 | 49.7 | 53.8 | 1771.5 | 46.5 | 75.9 | 516 |
| **InternVL-Chat-V1.5** | 82.3 | 80.7 | 75.2 | 57.1 | 46.8 | 2189.6 | 47.4 | 80.6 | 720 |
| **LLaVA-Next-Yi-34B** | 81.1 | 79.0 | 75.7 | 51.6 | 48.8 | 2050.2 | 34.8 | 78.9 | 574 |
| **Step-1V** | 80.7 | 79.9 | 70.3 | 50.0 | 49.9 | 2206.4 | 48.4 | 79.2 | 625 |
| **MiniCPM-Llama3-V2.5** | 77.6 | 73.8 | 72.3 | 51.8 | 45.8 | 2024.6 | 42.4 | 78.4 | 725 |
| **Qwen-VL-Max** | 77.6 | 75.7 | 72.7 | 49.5 | 52.0 | 2281.7 | 41.2 | 75.7 | 684 |
| **Gemini 1.0 Pro** | 73.6 | 74.3 | 70.7 | 38.6 | 49.0 | 2148.9 | 45.7 | 72.9 | 680 |
| **Claude 3 Opus** | 63.3 | 59.2 | 64.0 | 45.7 | 54.9 | 1586.8 | 37.8 | 70.6 | 694 |
| **GLM-4V-9B** | 81.1 | 79.4 | 76.8 | 58.7 | 47.2 | 2163.8 | 46.6 | 81.1 | 786 |
## 快速调用
**硬件配置和系统要求，请查看[这里](basic_demo/README.md)。**
### 使用以下方法快速调用 GLM-4-9B-Chat 语言模型
使用 transformers 后端进行推理:
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0' # 设置 GPU 编号，如果单机单卡指定一个，单机多卡指定多个 GPU 编号
MODEL_PATH = "THUDM/glm-4-9b-chat-hf"
device = "cuda" if torch.cuda.is_available() else "cpu"
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, trust_remote_code=True)
query = "你好"
inputs = tokenizer.apply_chat_template([{"role": "user", "content": query}],
add_generation_prompt=True,
tokenize=True,
return_tensors="pt",
return_dict=True
)
inputs = inputs.to(device)
model = AutoModelForCausalLM.from_pretrained(
MODEL_PATH,
torch_dtype=torch.bfloat16,
low_cpu_mem_usage=True,
trust_remote_code=True,
device_map="auto"
).eval()
gen_kwargs = {"max_length": 2500, "do_sample": True, "top_k": 1}
with torch.no_grad():
outputs = model.generate(**inputs, **gen_kwargs)
outputs = outputs[:, inputs['input_ids'].shape[1]:]
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```
使用 vLLM 后端进行推理:
```python
from transformers import AutoTokenizer
from vllm import LLM, SamplingParams
# GLM-4-9B-Chat-1M
# max_model_len, tp_size = 1048576, 4
# 如果遇见 OOM 现象，建议减少max_model_len，或者增加tp_size
max_model_len, tp_size = 131072, 1
model_name = "THUDM/glm-4-9b-chat-hf"
prompt = [{"role": "user", "content": "你好"}]
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
llm = LLM(
model=model_name,
tensor_parallel_size=tp_size,
max_model_len=max_model_len,
trust_remote_code=True,
enforce_eager=True,
# GLM-4-9B-Chat-1M 如果遇见 OOM 现象，建议开启下述参数
# enable_chunked_prefill=True,
# max_num_batched_tokens=8192
)
stop_token_ids = [151329, 151336, 151338]
sampling_params = SamplingParams(temperature=0.95, max_tokens=1024, stop_token_ids=stop_token_ids)
inputs = tokenizer.apply_chat_template(prompt, tokenize=False, add_generation_prompt=True)
outputs = llm.generate(prompts=inputs, sampling_params=sampling_params)
print(outputs[0].outputs[0].text)
```
### 使用以下方法快速调用 GLM-4V-9B 多模态模型
使用 transformers 后端进行推理:
```python
import torch
from PIL import Image
from transformers import AutoModelForCausalLM, AutoTokenizer
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0' # 设置 GPU 编号，如果单机单卡指定一个，单机多卡指定多个 GPU 编号
MODEL_PATH = "THUDM/glm-4v-9b"
device = "cuda" if torch.cuda.is_available() else "cpu"
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, trust_remote_code=True)
query = '描述这张图片'
image = Image.open("your image").convert('RGB')
inputs = tokenizer.apply_chat_template([{"role": "user", "image": image, "content": query}],
add_generation_prompt=True, tokenize=True, return_tensors="pt",
return_dict=True) # chat mode
inputs = inputs.to(device)
model = AutoModelForCausalLM.from_pretrained(
MODEL_PATH,
torch_dtype=torch.bfloat16,
low_cpu_mem_usage=True,
trust_remote_code=True,
device_map="auto"
).eval()
gen_kwargs = {"max_length": 2500, "do_sample": True, "top_k": 1}
with torch.no_grad():
outputs = model.generate(**inputs, **gen_kwargs)
outputs = outputs[:, inputs['input_ids'].shape[1]:]
print(tokenizer.decode(outputs[0]))
```
使用 vLLM 后端进行推理:
```python
from PIL import Image
from vllm import LLM, SamplingParams
model_name = "THUDM/glm-4v-9b"
llm = LLM(model=model_name,
tensor_parallel_size=1,
max_model_len=8192,
trust_remote_code=True,
enforce_eager=True)
stop_token_ids = [151329, 151336, 151338]
sampling_params = SamplingParams(temperature=0.2,
max_tokens=1024,
stop_token_ids=stop_token_ids)
prompt = "What's the content of the image?"
image = Image.open("your image").convert('RGB')
inputs = {
"prompt": prompt,
"multi_modal_data": {
"image": image
},
}
outputs = llm.generate(inputs, sampling_params=sampling_params)
for o in outputs:
generated_text = o.outputs[0].text
print(generated_text)
```
## 完整项目列表
如果你想更进一步了解 GLM-4-9B 系列开源模型，本开源仓库通过以下内容为开发者提供基础的 GLM-4-9B 的使用和开发代码
+ [basic_demo](basic_demo/README.md): 在这里包含了
+ 使用 transformers 和 vLLM 后端的交互代码
+ OpenAI API 后端交互代码
+ Batch 推理代码
+ [composite_demo](composite_demo/README.md): 在这里包含了
+ GLM-4-9B-Chat 以及 GLM-4V-9B 开源模型的完整功能演示代码，包含了 All Tools 能力、长文档解读和多模态能力的展示。
+ [fintune_demo](finetune_demo/README.md): 在这里包含了
+ PEFT (LORA, P-Tuning) 微调代码
+ SFT 微调代码
+ [intel_device_demo](intel_device_demo/): 在这里包含了
+ 使用 OpenVINO 部署模型代码
+ 使用 Intel® Extension for Transformers 部署模型代码
## 友情链接
+ [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory): 高效开源微调框架，已支持 GLM-4-9B-Chat 语言模型微调。
+ [SWIFT](https://github.com/modelscope/swift): 魔搭社区的大模型/多模态大模型训练框架，已支持 GLM-4-9B-Chat / GLM-4V-9B
模型微调。
+ [Xorbits Inference](https://github.com/xorbitsai/inference): 性能强大且功能全面的分布式推理框架，轻松一键部署你自己的模型或内置的前沿开源模型。
+ [LangChain-ChatChat](https://github.com/chatchat-space/Langchain-Chatchat): 基于 Langchain 与 ChatGLM 等语言模型的 RAG
与 Agent 应用
+ [self-llm](https://github.com/datawhalechina/self-llm/tree/master/models/GLM-4): Datawhale 团队的提供的 GLM-4-9B
系列模型使用教程。
+ [chatglm.cpp](https://github.com/li-plus/chatglm.cpp): 类似 llama.cpp 的量化加速推理方案，实现笔记本上实时对话
+ [OpenVINO](https://github.com/openvinotoolkit):
Intel 开发的高性能 CPU,GPU及NPU 加速推理方案，可以参考此 [步骤](https://github.com/openvinotoolkit/openvino_notebooks/blob/latest/notebooks/llm-chatbot/llm-chatbot-generate-api.ipynb) 部署 glm-4-9b-chat 模型。
## 协议
+ GLM-4 模型的权重的使用则需要遵循 [模型协议](https://huggingface.co/THUDM/glm-4-9b/blob/main/LICENSE)。
+ 本开源仓库的代码则遵循 [Apache 2.0](LICENSE) 协议。
请您严格遵循开源协议。
## 引用
如果你觉得我们的工作有帮助的话，请考虑引用下列论文。
```
@misc{glm2024chatglm,
title={ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools},
author={Team GLM and Aohan Zeng and Bin Xu and Bowen Wang and Chenhui Zhang and Da Yin and Diego Rojas and Guanyu Feng and Hanlin Zhao and Hanyu Lai and Hao Yu and Hongning Wang and Jiadai Sun and Jiajie Zhang and Jiale Cheng and Jiayi Gui and Jie Tang and Jing Zhang and Juanzi Li and Lei Zhao and Lindong Wu and Lucen Zhong and Mingdao Liu and Minlie Huang and Peng Zhang and Qinkai Zheng and Rui Lu and Shuaiqi Duan and Shudan Zhang and Shulin Cao and Shuxun Yang and Weng Lam Tam and Wenyi Zhao and Xiao Liu and Xiao Xia and Xiaohan Zhang and Xiaotao Gu and Xin Lv and Xinghan Liu and Xinyi Liu and Xinyue Yang and Xixuan Song and Xunkai Zhang and Yifan An and Yifan Xu and Yilin Niu and Yuantao Yang and Yueyan Li and Yushi Bai and Yuxiao Dong and Zehan Qi and Zhaoyu Wang and Zhen Yang and Zhengxiao Du and Zhenyu Hou and Zihan Wang},
year={2024},
eprint={2406.12793},
archivePrefix={arXiv},
primaryClass={id='cs.CL' full_name='Computation and Language' is_active=True alt_name='cmp-lg' in_archive='cs' is_general=False description='Covers natural language processing. Roughly includes material in ACM Subject Class I.2.7. Note that work on artificial languages (programming languages, logics, formal systems) that does not explicitly address natural-language issues broadly construed (natural-language processing, computational linguistics, speech, text retrieval, etc.) is not appropriate for this area.'}
}
```
```
@misc{wang2023cogvlm,
title={CogVLM: Visual Expert for Pretrained Language Models},
author={Weihan Wang and Qingsong Lv and Wenmeng Yu and Wenyi Hong and Ji Qi and Yan Wang and Junhui Ji and Zhuoyi Yang and Lei Zhao and Xixuan Song and Jiazheng Xu and Bin Xu and Juanzi Li and Yuxiao Dong and Ming Ding and Jie Tang},
year={2023},
eprint={2311.03079},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
================================================
FILE: demo/composite_demo/.gitignore
================================================
*venv
*.DS_Store
*model
*.idea/
# Created by https://www.toptal.com/developers/gitignore/api/python
# Edit at https://www.toptal.com/developers/gitignore?templates=python
### Python ###
# Byte-compiled / optimized / DLL files
__pycache__/
*.py[cod]
*$py.class
# C extensions
*.so
# Distribution / packaging
.Python
build/
develop-eggs/
dist/
downloads/
eggs/
.eggs/
lib/
lib64/
parts/
sdist/
var/
wheels/
share/python-wheels/
*.egg-info/
.installed.cfg
*.egg
MANIFEST
# PyInstaller
# Usually these files are written by a python script from a template
# before PyInstaller builds the exe, so as to inject date/other infos into it.
*.manifest
*.spec
# Installer logs
pip-log.txt
pip-delete-this-directory.txt
# Unit test / coverage reports
htmlcov/
.tox/
.nox/
.coverage
.coverage.*
.cache
nosetests.xml
coverage.xml
*.cover
*.py,cover
.hypothesis/
.pytest_cache/
cover/
# Translations
*.mo
*.pot
# Django stuff:
*.log
local_settings.py
db.sqlite3
db.sqlite3-journal
# Flask stuff:
instance/
.webassets-cache
# Scrapy stuff:
.scrapy
# Sphinx documentation
docs/_build/
# PyBuilder
.pybuilder/
target/
# Jupyter Notebook
.ipynb_checkpoints
# IPython
profile_default/
ipython_config.py
# pyenv
# For a library or package, you might want to ignore these files since the code is
# intended to run in multiple environments; otherwise, check them in:
# .python-version
# pipenv
# According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
# However, in case of collaboration, if having platform-specific dependencies or dependencies
# having no cross-platform support, pipenv may install dependencies that don't work, or not
# install all needed dependencies.
#Pipfile.lock
# poetry
# Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
# This is especially recommended for binary packages to ensure reproducibility, and is more
# commonly ignored for libraries.
# https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
#poetry.lock
# pdm
# Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
#pdm.lock
# pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
# in version control.
# https://pdm.fming.dev/#use-with-ide
.pdm.toml
# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
__pypackages__/
# Celery stuff
celerybeat-schedule
celerybeat.pid
# SageMath parsed files
*.sage.py
# Environments
.env
.venv
env/
venv/
ENV/
env.bak/
venv.bak/
# Spyder project settings
.spyderproject
.spyproject
# Rope project settings
.ropeproject
# mkdocs documentation
/site
# mypy
.mypy_cache/
.dmypy.json
dmypy.json
# Pyre type checker
.pyre/
# pytype static type analyzer
.pytype/
# Cython debug symbols
cython_debug/
# PyCharm
# JetBrains specific template is maintained in a separate JetBrains.gitignore that can
# be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
# and can be added to the global gitignore or merged into this file. For a more nuclear
# option (not recommended) you can uncomment the following to ignore the entire idea folder.
#.idea/
### Python Patch ###
# Poetry local configuration file - https://python-poetry.org/docs/configuration/#local-configuration
poetry.toml
# ruff
.ruff_cache/
# LSP config files
pyrightconfig.json
# End of https://www.toptal.com/developers/gitignore/api/python
================================================
FILE: demo/composite_demo/README.md
================================================
# GLM-4-9B Web Demo
Read this in [English](README_en.md)
## 安装
我们建议通过 [Conda](https://docs.conda.io/en/latest/) 进行环境管理。
执行以下命令新建一个 conda 环境并安装所需依赖：
```bash
conda create -n glm-4-demo python=3.12
conda activate glm-4-demo
pip install -r requirements.txt
```
请注意，本项目需要 Python 3.10 或更高版本。
此外，使用 Code Interpreter 还需要安装 Jupyter 内核：
```bash
ipython kernel install --name glm-4-demo --user
```
您可以修改 `~/.local/share/jupyter/kernels/glm-4-demo/kernel.json` 来改变 Jupyter 内核的配置，包括内核的启动参数等。例如，若您希望在使用 All Tools 的 Python 代码执行能力时使用 Matplotlib 画图，可以在 `argv` 数组中添加 `"--matplotlib=inline"`。
若要使用浏览器和搜索功能，还需要启动浏览器后端。首先，根据 [Node.js](https://nodejs.org/en/download/package-manager)
官网的指示安装 Node.js，然后安装包管理器 [PNPM](https://pnpm.io) 之后安装浏览器服务的依赖：
```bash
cd browser
npm install -g pnpm
pnpm install
```
## 运行
1. 修改 `browser/src/config.ts` 中的 `BING_SEARCH_API_KEY` 配置浏览器服务需要使用的 Bing 搜索 API Key：
```diff
export default {
BROWSER_TIMEOUT: 10000,
BING_SEARCH_API_URL: 'https://api.bing.microsoft.com/v7.0',
BING_SEARCH_API_KEY: '<PUT_YOUR_BING_SEARCH_KEY_HERE>',
HOST: 'localhost',
PORT: 3000,
};
```
如果您注册的是Bing Customer Search的API，您可以修改您的配置文件为如下，并且填写您的Custom Configuration ID:
```diff
export default {
LOG_LEVEL: 'debug',
BROWSER_TIMEOUT: 10000,
BING_SEARCH_API_URL: 'https://api.bing.microsoft.com/v7.0/custom/',
BING_SEARCH_API_KEY: 'YOUR_BING_SEARCH_API_KEY',
CUSTOM_CONFIG_ID : 'YOUR_CUSTOM_CONFIG_ID', //将您的Custom Configuration ID放在此处
HOST: 'localhost',
PORT: 3000,
};
```
2. 文生图功能需要调用 CogView API。修改 `src/tools/config.py`
，提供文生图功能需要使用的 [智谱 AI 开放平台](https://open.bigmodel.cn) API Key：
```diff
BROWSER_SERVER_URL = 'http://localhost:3000'
IPYKERNEL = 'glm-4-demo'
ZHIPU_AI_KEY = '<PUT_YOUR_ZHIPU_AI_KEY_HERE>'
COGVIEW_MODEL = 'cogview-3'
```
3. 启动浏览器后端，在单独的 shell 中：
```bash
cd browser
pnpm start
```
4. 运行以下命令在本地加载模型并启动 demo：
```bash
streamlit run src/main.py
```
之后即可从命令行中看到 demo 的地址，点击即可访问。初次访问需要下载并加载模型，可能需要花费一定时间。
如果已经在本地下载了模型，可以通过 `export *_MODEL_PATH=/path/to/model` 来指定从本地加载模型。可以指定的模型包括：
- `CHAT_MODEL_PATH`: 用于 All Tools 模式与文档解读模式，默认为 `THUDM/glm-4-9b-chatglm-4-9b-chat`。
- `VLM_MODEL_PATH`: 用于 VLM 模式，默认为 `THUDM/glm-4v-9b`。
Chat 模型支持使用 [vLLM](https://github.com/vllm-project/vllm) 推理。若要使用，请安装 vLLM 并设置环境变量 `USE_VLLM=1`。
Chat 模型支持使用 [OpenAI API](https://platform.openai.com/docs/api-reference/introduction) 推理。若要使用，请启动basic_demo目录下的openai_api_server并设置环境变量 `USE_API=1`。该功能可以解耦推理服务器和demo服务器。
如果需要自定义 Jupyter 内核，可以通过 `export IPYKERNEL=<kernel_name>` 来指定。
## 使用
GLM-4 Demo 拥有三种模式：
- All Tools: 具有完整工具调用能力的对话模式，原生支持网页浏览、代码执行、图片生成，并支持自定义工具。
- 文档解读: 支持上传文档进行文档解读与对话。
- 多模态: 支持上传图像进行图像理解与对话。
### All Tools
本模式兼容 ChatGLM3-6B 的工具注册流程。
+ 代码能力，绘图能力，联网能力已经自动集成，用户只需按照要求配置对应的Key。
+ 本模式下不支持系统提示词，模型会自动构建提示词。
对话模式下，用户可以直接在侧边栏修改 top_p, temperature 等参数来调整模型的行为。
与模型对话时，模型将会自主决定进行工具调用。
由于原始结果可能较长，默认情况下工具调用结果被隐藏，可以通过展开折叠框查看原始的工具调用结果。
模型拥有进行网页搜索和 Python 代码执行的能力。同时，模型也可以连续调用多个工具。例如：
此时模型通过调用浏览器工具进行搜索获取到了需要的数据，之后将会调用 Python 工具执行代码，利用 Matplotlib 绘图：
如果提供了智谱开放平台 API Key，模型也可以调用 CogView 进行图像生成：
#### 自定义工具
可以通过在 `tool_registry.py` 中注册新的工具来增强模型的能力。只需要使用 `@register_tool`
装饰函数即可完成注册。对于工具声明，函数名称即为工具的名称，函数 docstring
即为工具的说明；对于工具的参数，使用 `Annotated[typ: type, description: str, required: bool]` 标注参数的类型、描述和是否必须。
例如，`get_weather` 工具的注册如下：
```python
@register_tool
def get_weather(
city_name: Annotated[str, 'The name of the city to be queried', True],
) -> str:
"""
Get the weather for `city_name` in the following week
"""
...
```
### 文档解读
用户可以上传文档，使用 GLM-4-9B的长文本能力，对文本进行理解。可以解析 pptx，docx，pdf等文件。
+ 本模式下不支持工具调用和系统提示词。
+ 如果文本很长，可能导致模型需要的显存较高，请确认你的硬件配置。
### 多模态
多模态模式下，用户可以利用 GLM-4V 的多模态理解能力，上传图像并与 GLM-4V 进行多轮对话：
用户可以上传图片，使用 GLM-4-9B的图像理解能力，对图片进行理解。
+ 本模式必须使用 glm-4v-9b 模型。
+ 本模式下不支持工具调用和系统提示词。
+ 模型仅能对一张图片进行理解和联系对话，如需更换图片，需要开启一个新的对话。
+ 图像支持的分辨率为 1120 x 1120
================================================
FILE: demo/composite_demo/README_en.md
================================================
# GLM-4-9B Web Demo
## Installation
We recommend using [Conda](https://docs.conda.io/en/latest/) for environment management.
Execute the following commands to create a conda environment and install the required dependencies:
```bash
conda create -n glm-4-demo python=3.12
conda activate glm-4-demo
pip install -r requirements.txt
```
Please note that this project requires Python 3.10 or higher.
In addition, you need to install the Jupyter kernel to use the Code Interpreter:
```bash
ipython kernel install --name glm-4-demo --user
```
You can modify `~/.local/share/jupyter/kernels/glm-4-demo/kernel.json` to change the configuration of the Jupyter
kernel, including the kernel startup parameters. For example, if you want to use Matplotlib to draw when using the
Python code execution capability of All Tools, you can add `"--matplotlib=inline"` to the `argv` array.
To use the browser and search functions, you also need to start the browser backend. First, install Node.js according to
the instructions on the [Node.js](https://nodejs.org/en/download/package-manager)
official website, then install the package manager [PNPM](https://pnpm.io) and then install the browser service
dependencies:
```bash
cd browser
npm install -g pnpm
pnpm install
```
## Run
1. Modify `BING_SEARCH_API_KEY` in `browser/src/config.ts` to configure the Bing Search API Key that the browser service
needs to use:
```diff
export default {
BROWSER_TIMEOUT: 10000,
BING_SEARCH_API_URL: 'https://api.bing.microsoft.com/v7.0',
BING_SEARCH_API_KEY: '<PUT_YOUR_BING_SEARCH_KEY_HERE>',
HOST: 'localhost',
PORT: 3000,
};
```
2. The Wenshengtu function needs to call the CogView API. Modify `src/tools/config.py`
, provide the [Zhipu AI Open Platform](https://open.bigmodel.cn) API Key required for the Wenshengtu function:
```diff
BROWSER_SERVER_URL = 'http://localhost:3000'
IPYKERNEL = 'glm4-demo'
ZHIPU_AI_KEY = '<PUT_YOUR_ZHIPU_AI_KEY_HERE>'
COGVIEW_MODEL = 'cogview-3'
```
3. Start the browser backend in a separate shell:
```bash
cd browser
pnpm start
```
4. Run the following commands to load the model locally and start the demo:
```bash
streamlit run src/main.py
```
Then you can see the demo address from the command line and click it to access it. The first access requires downloading
and loading the model, which may take some time.
If you have downloaded the model locally, you can specify to load the model from the local
by `export *_MODEL_PATH=/path/to/model`. The models that can be specified include:
- `CHAT_MODEL_PATH`: used for All Tools mode and document interpretation mode, the default is `THUDM/glm-4-9b-chat`.
- `VLM_MODEL_PATH`: used for VLM mode, the default is `THUDM/glm-4v-9b`.
The Chat model supports reasoning using [vLLM](https://github.com/vllm-project/vllm). To use it, please install vLLM and
set the environment variable `USE_VLLM=1`.
The Chat model also supports reasoning using [OpenAI API](https://platform.openai.com/docs/api-reference/introduction). To use it, please run `openai_api_server.py` in `inference` and set the environment variable `USE_API=1`. This function is used to deploy inference server and demo server in different machine.
If you need to customize the Jupyter kernel, you can specify it by `export IPYKERNEL=<kernel_name>`.
## Usage
GLM4 Demo has three modes:
- All Tools mode
- VLM mode
- Text interpretation mode
### All Tools mode
You can enhance the model's capabilities by registering new tools in `tool_registry.py`. Just use `@register_tool`
decorated function to complete the registration. For tool declarations, the function name is the name of the tool, and
the function docstring
is the description of the tool; for tool parameters, use `Annotated[typ: type, description: str, required: bool]` to
annotate the parameter type, description, and whether it is required.
For example, the registration of the `get_weather` tool is as follows:
```python
@register_tool
def get_weather(
city_name: Annotated[str, 'The name of the city to be queried', True],
) -> str:
"""
Get the weather for `city_name` in the following week
"""
...
```
This mode is compatible with the tool registration process of ChatGLM3-6B.
+ Code capability, drawing capability, and networking capability have been automatically integrated. Users only need to
configure the corresponding Key as required.
+ System prompt words are not supported in this mode. The model will automatically build prompt words.
## Text interpretation mode
Users can upload documents and use the long text capability of GLM-4-9B to understand the text. It can parse pptx, docx,
pdf and other files.
+ Tool calls and system prompt words are not supported in this mode.
+ If the text is very long, the model may require a high amount of GPU memory. Please confirm your hardware
configuration.
## Image Understanding Mode
Users can upload images and use the image understanding capabilities of GLM-4-9B to understand the images.
+ This mode must use the glm-4v-9b model.
+ Tool calls and system prompts are not supported in this mode.
+ The model can only understand and communicate with one image. If you need to change the image, you need to open a new
conversation.
+ The supported image resolution is 1120 x 1120
================================================
FILE: demo/composite_demo/browser/.gitignore
================================================
# Created by https://www.toptal.com/developers/gitignore/api/node
# Edit at https://www.toptal.com/developers/gitignore?templates=node
### Node ###
# Logs
logs
*.log
npm-debug.log*
yarn-debug.log*
yarn-error.log*
lerna-debug.log*
.pnpm-debug.log*
# Diagnostic reports (https://nodejs.org/api/report.html)
report.[0-9]*.[0-9]*.[0-9]*.[0-9]*.json
# Runtime data
pids
*.pid
*.seed
*.pid.lock
# Directory for instrumented libs generated by jscoverage/JSCover
lib-cov
# Coverage directory used by tools like istanbul
coverage
*.lcov
# nyc test coverage
.nyc_output
# Grunt intermediate storage (https://gruntjs.com/creating-plugins#storing-task-files)
.grunt
# Bower dependency directory (https://bower.io/)
bower_components
# node-waf configuration
.lock-wscript
# Compiled binary addons (https://nodejs.org/api/addons.html)
build/Release
# Dependency directories
node_modules/
jspm_packages/
# Snowpack dependency directory (https://snowpack.dev/)
web_modules/
# TypeScript cache
*.tsbuildinfo
# Optional npm cache directory
.npm
# Optional eslint cache
.eslintcache
# Optional stylelint cache
.stylelintcache
# Microbundle cache
.rpt2_cache/
.rts2_cache_cjs/
.rts2_cache_es/
.rts2_cache_umd/
# Optional REPL history
.node_repl_history
# Output of 'npm pack'
*.tgz
# Yarn Integrity file
.yarn-integrity
# dotenv environment variable files
.env
.env.development.local
.env.test.local
.env.production.local
.env.local
# parcel-bundler cache (https://parceljs.org/)
.cache
.parcel-cache
# Next.js build output
.next
out
# Nuxt.js build / generate output
.nuxt
dist
# Gatsby files
.cache/
# Comment in the public line in if your project uses Gatsby and not Next.js
# https://nextjs.org/blog/next-9-1#public-directory-support
# public
# vuepress build output
.vuepress/dist
# vuepress v2.x temp and cache directory
.temp
# Docusaurus cache and generated files
.docusaurus
# Serverless directories
.serverless/
# FuseBox cache
.fusebox/
# DynamoDB Local files
.dynamodb/
# TernJS port file
.tern-port
# Stores VSCode versions used for testing VSCode extensions
.vscode-test
# yarn v2
.yarn/cache
.yarn/unplugged
.yarn/build-state.yml
.yarn/install-state.gz
.pnp.*
### Node Patch ###
# Serverless Webpack directories
.webpack/
# Optional stylelint cache
# SvelteKit build / generate output
.svelte-kit
# End of https://www.toptal.com/developers/gitignore/api/node
================================================
FILE: demo/composite_demo/browser/package.json
================================================
{
"name": "glm4-browser",
"version": "1.0.0",
"description": "Browser system for GLM-4",
"main": "src/server.ts",
"scripts": {
"dev": "npx nodemon src/server",
"start": "npx ts-node src/server.ts"
},
"license": "MIT",
"dependencies": {
"express": "^4.18.3",
"jsdom": "^24.0.0",
"pnpm": "^9.1.2",
"turndown": "^7.1.2",
"winston": "^3.11.0"
},
"devDependencies": {
"@types/express": "^4.17.21",
"@types/jsdom": "^21.1.6",
"@types/node": "^20.11.20",
"@types/turndown": "^5.0.4",
"nodemon": "^3.1.0",
"ts-node": "^10.9.2"
}
}
================================================
FILE: demo/composite_demo/browser/src/browser.ts
================================================
import { JSDOM } from 'jsdom';
import TurndownService from 'turndown';
import config from './config';
import { Message, ToolObservation } from './types';
import { logger, withTimeout } from './utils';
// represent a quote from a display
interface Quote {
text: string;
metadata: Metadata[];
}
interface ActionResult {
contentType: string;
metadataList?: TetherQuoteMetadata[];
metadata?: any;
roleMetadata: string;
message: string;
}
// represent a piece of metadata to be marked in the final answer
interface Metadata {
type: string;
title: string;
url: string;
lines: string[];
}
interface TetherQuoteExtra {
cited_message_idx: number;
evidence_text: string;
}
interface TetherQuoteMetadata {
type: string;
title: string;
url: string;
text: string;
pub_date?: string;
extra?: TetherQuoteExtra;
}
interface Citation {
citation_format_type: string;
start_ix: number;
end_ix: number;
metadata?: TetherQuoteMetadata;
invalid_reason?: string;
}
interface PageState {
aCounter: number;
imgCounter: number;
url: URL;
url_string: string;
hostname: string;
links: string[];
links_meta: TetherQuoteMetadata[];
lines: string[];
line_source: Record<string, Metadata>; // string representation of number interval
title?: string;
}
interface BrowserState {
pageStack: PageState[];
quoteCounter: number;
quotes: Record<string, Quote>;
}
function removeDenseLinks(document: Document, ratioThreshold: number = 0.5) {
// Remove nav elements
const navs = document.querySelectorAll('nav');
navs.forEach(nav => {
if (nav.parentNode) {
nav.parentNode.removeChild(nav);
}
});
// Query for lists, divs, spans, tables, and paragraphs
const elements = document.querySelectorAll('ul, ol, div, span, nav, table, p');
elements.forEach(element => {
if (element === null) return;
const children = Array.from(element.childNodes);
const links = element.querySelectorAll('a');
if (children.length <= 1) return;
const allText = element.textContent ? element.textContent.trim().replace(/\s+/g, '') : '';
const linksText = Array.from(links)
.map(link => (link.textContent ? link.textContent.trim() : ''))
.join('')
.replace(/\s+/g, '');
if (allText.length === 0 || linksText.length === 0) return;
let ratio = linksText.length / allText.length;
if (ratio > ratioThreshold && element.parentNode) {
element.parentNode.removeChild(element);
}
});
}
abstract class BaseBrowser {
public static toolName = 'browser' as const;
public description = 'BaseBrowser';
private turndownService = new TurndownService({
headingStyle: 'atx',
});
private state: BrowserState;
private transform(dom: JSDOM): string {
let state = this.lastPageState();
state.aCounter = 0;
state.imgCounter = 0;
state.links = [];
return this.turndownService.turndown(dom.window.document);
}
private formatPage(state: PageState): string {
let formatted_lines = state.lines.join('\n');
let formatted_title = state.title ? `TITLE: ${state.title}\n\n` : '';
let formatted_range = `\nVisible: 0% - 100%`;
let formatted_message = formatted_title + formatted_lines + formatted_range;
return formatted_message;
}
private newPageState(): PageState {
return {
aCounter: 0,
imgCounter: 0,
url: new URL('about:blank'),
url_string: 'about:blank',
hostname: '',
title: '',
links: [],
links_meta: [],
lines: [],
line_source: {},
};
}
private pushPageState(): PageState {
let state = this.newPageState();
this.state.pageStack.push(state);
return state;
}
private lastPageState(): PageState {
if (this.state.pageStack.length === 0) {
throw new Error('No page state');
}
return this.state.pageStack[this.state.pageStack.length - 1];
}
private formatErrorUrl(url: string): string {
let TRUNCATION_LIMIT = 80;
if (url.length <= TRUNCATION_LIMIT) {
return url;
}
return url.slice(0, TRUNCATION_LIMIT) + `... (URL truncated at ${TRUNCATION_LIMIT} chars)`;
}
protected functions = {
search: async (query: string, recency_days: number = -1) => {
logger.debug(`Searching for: ${query}`);
const search = new URLSearchParams({ q: query });
recency_days > 0 && search.append('recency_days', recency_days.toString());
if (config.CUSTOM_CONFIG_ID) {
search.append('customconfig', config.CUSTOM_CONFIG_ID.toString());
}
const url = `${config.BING_SEARCH_API_URL}/search?${search.toString()}`;
console.log('Full URL:', url); // 输出完整的 URL查看是否正确
return withTimeout(
config.BROWSER_TIMEOUT,
fetch(url, {
headers: {
'Ocp-Apim-Subscription-Key': config.BING_SEARCH_API_KEY,
}
})
.then(
res =>
res.json() as Promise<{
queryContext: {
originalQuery: string;
};
webPages: {
webSearchUrl: string;
totalEstimatedMatches: number;
value: {
id: string;
name: string;
url: string;
datePublished: string; // 2018-05-18T08:00:00.0000000
datePublishedDisplayText: string;
isFamilyFriendly: boolean;
displayUrl: string;
snippet: string;
dateLastCrawled: string;
cachedPageUrl: string;
language: string;
isNavigational: boolean;
}[];
};
rankingResponse: {
mainline: {
items: {
answerType: 'WebPages';
resultIndex: number;
value: {
id: string;
};
}[];
};
};
}>,
),
)
.then(async ({ value: res }) => {
try {
let state = this.pushPageState();
let metadataList: TetherQuoteMetadata[] = [];
for (const [i, entry] of res.webPages.value.entries()) {
const url = new URL(entry.url);
const hostname = url.hostname;
state.lines.push(` # 【${i}†${entry.name}†${hostname}】`);
state.lines.push(entry.snippet);
const quoteMetadata: Metadata = {
type: 'webpage',
title: entry.name,
url: entry.url,
lines: state.lines.slice(2 * i, 2 * i + 2),
};
state.line_source[`${2 * i}-${2 * i + 1}`] = quoteMetadata;
state.links[i] = entry.url;
const returnMetadata: TetherQuoteMetadata = {
type: quoteMetadata.type,
title: quoteMetadata.title,
url: quoteMetadata.url,
text: state.lines[2 * i + 1], // only content, not link
pub_date: entry.datePublished,
};
metadataList.push(returnMetadata);
}
const returnContentType = 'browser_result';
return {
contentType: returnContentType,
roleMetadata: returnContentType,
message: this.formatPage(state),
metadataList,
};
} catch (err) {
throw new Error(`parse error: ${err}`);
}
})
.catch(err => {
logger.error(`搜索请求失败：${query}，错误信息：${err.message}`);
if (err.code === 'ECONNABORTED') {
throw new Error(`Timeout while executing search for: ${query}`);
}
throw new Error(`网络或服务器发生错误，请检查URL: ${url}`);
});
},
open_url: (url: string) => {
logger.debug(`Opening ${url}`);
return withTimeout(
config.BROWSER_TIMEOUT,
fetch(url).then(res => res.text()),
)
.then(async ({ value: res, time }) => {
try {
const state = this.pushPageState();
state.url = new URL(url);
state.url_string = url;
state.hostname = state.url.hostname;
const html = res;
const dom = new JSDOM(html);
const title = dom.window.document.title;
const markdown = this.transform(dom);
state.title = title;
// Remove first line, because it will be served as the title
const lines = markdown.split('\n');
lines.shift();
// Remove consequent empty lines
let i = 0;
while (i < lines.length - 1) {
if (lines[i].trim() === '' && lines[i + 1].trim() === '') {
lines.splice(i, 1);
} else {
i++;
}
}
let page = lines.join('\n');
// The first line feed is not a typo
let text_result = `\nURL: ${url}\n${page}`;
state.lines = text_result.split('\n');
// all lines has only one source
state.line_source = {};
state.line_source[`0-${state.lines.length - 1}`] = {
type: 'webpage',
title: title,
url: url,
lines: state.lines,
};
let message = this.formatPage(state);
const returnContentType = 'browser_result';
return {
contentType: returnContentType,
roleMetadata: returnContentType,
message,
metadataList: state.links_meta,
};
} catch (err) {
throw new Error(`parse error: ${err}`);
}
})
.catch(err => {
logger.error(err.message);
if (err.code === 'ECONNABORTED') {
throw new Error(`Timeout while loading page w/ URL: ${url}`);
}
throw new Error(`Failed to load page w/ URL: ${url}`);
});
},
mclick: (ids: number[]) => {
logger.info('Entering mclick', ids);
let promises: Promise<ActionResult>[] = [];
let state = this.lastPageState();
for (let id of ids) {
if (isNaN(id) || id >= state.links.length) {
promises.push(
Promise.reject(
new Error(
`recorded='click(${id})' temporary=None permanent=None new_state=None final=None success=False feedback='Error parsing ID ${id}' metadata={}`,
),
),
);
continue;
}
let url: string;
try {
url = new URL(state.links[id], state.url).href;
} catch (err) {
logger.error(`Failed in getting ${state.links[id]}, ${state.url}`);
promises.push(
Promise.reject(
new Error(
`recorded='click(${id})' temporary=None permanent='${err}' new_state=None final=None success=False feedback='Error parsing URL for ID ${id}' metadata={}`,
),
),
);
continue;
}
const quoteIndex = this.state.quoteCounter++; // ascending in final results
promises.push(
withTimeout(
config.BROWSER_TIMEOUT,
fetch(url).then(res => res.text()),
)
.then(({ value: res, time }) => {
let state = this.newPageState();
state.url = new URL(url);
state.hostname = state.url.hostname;
try {
const html = res;
const dom = new JSDOM(html);
const title = dom.window.document.title;
state.title = title;
removeDenseLinks(dom.window.document);
let quoteText = this.transform(dom);
// remove consecutive newline
quoteText = quoteText.replace(/[\r\n]+/g, '\n');
const quoteLines = quoteText.split('\n');
state.lines = quoteLines;
const metadata = {
type: 'webpage',
title: title,
url: url,
lines: quoteLines,
};
const quoteMetadata = {
type: 'webpage',
title: title,
url: url,
text: quoteText,
};
state.line_source = {};
state.line_source[`0-${state.lines.length - 1}`] = metadata;
this.state.quotes[quoteIndex.toString()] = {
text: quoteText,
metadata: [metadata],
};
const returnContentType = 'quote_result';
return {
contentType: returnContentType,
roleMetadata: `${returnContentType} [${quoteIndex}†source]`,
message: quoteText,
metadataList: [quoteMetadata],
metadata: {
url,
},
};
} catch (err) {
throw new Error(`parse error: ${err}`);
}
})
.catch(err => {
logger.error(err.message);
if (err.code === 'ECONNABORTED') {
throw new Error(`Timeout while loading page w/ URL: ${this.formatErrorUrl(url)}`);
}
throw new Error(`Failed to load page w/ URL: ${this.formatErrorUrl(url)}`);
})
.catch(err => {
// format error message
const returnContentType = 'system_error';
throw {
contentType: returnContentType,
roleMetadata: returnContentType,
message: `recorded='click(${id})' temporary=None permanent='${
err.message
}' new_state=None final=None success=False feedback='Error fetching url ${this.formatErrorUrl(
url,
)}' metadata={}`,
metadata: {
failedURL: url,
},
} as ActionResult;
}),
);
}
return Promise.allSettled(promises).then(async results => {
const actionResults = results.map(r => {
if (r.status === 'fulfilled') {
return r.value;
} else {
logger.error(r.reason);
return r.reason as ActionResult;
}
});
if (results.filter(r => r.status === 'fulfilled').length === 0) {
// collect errors
const err_text = (results as PromiseRejectedResult[])
.map(r => (r.reason as ActionResult).message)
.join('\n');
throw new Error(err_text);
} else {
return actionResults;
}
});
},
};
constructor() {
this.state = {
pageStack: [],
quotes: {},
quoteCounter: 7,
};
this.turndownService.remove('script');
this.turndownService.remove('style');
// Add rules for turndown
this.turndownService.addRule('reference', {
filter: function (node, options: any): boolean {
return (
options.linkStyle === 'inlined' &&
node.nodeName === 'A' &&
node.getAttribute('href') !== undefined
);
},
replacement: (content, node, options): string => {
let state = this.state.pageStack[this.state.pageStack.length - 1];
if (!content || !('getAttribute' in node)) return '';
let href = undefined;
try {
if ('getAttribute' in node) {
const hostname = new URL(node.getAttribute('href')!).hostname;
// Do not append hostname when in the same domain
if (hostname === state.hostname || !hostname) {
href = '';
} else {
href = '†' + hostname;
}
}
} catch (e) {
// To prevent displaying links like '/foo/bar'
href = '';
}
if (href === undefined) return '';
const url = node.getAttribute('href')!;
let linkId = state.links.findIndex(link => link === url);
if (linkId === -1) {
linkId = state.aCounter++;
// logger.debug(`New link[${linkId}]: ${url}`);
state.links_meta.push({
type: 'webpage',
title: node.textContent!,
url: href,
text: node.textContent!,
});
state.links.push(url);
}
return `【${linkId}†${node.textContent}${href}】`;
},
});
this.turndownService.addRule('img', {
filter: 'img',
replacement: (content, node, options): string => {
let state = this.state.pageStack[this.state.pageStack.length - 1];
return `[Image ${state.imgCounter++}]`;
},
});
// Just to change indentation, wondering why this isn't exposed as an option
this.turndownService.addRule('list', {
filter: 'li',
replacement: function (content, node, options) {
content = content
.replace(/^\n+/, '') // remove leading newlines
.replace(/\n+$/, '\n') // replace trailing newlines with just a single one
.replace(/\n/gm, '\n '); // indent
let prefix = options.bulletListMarker + ' ';
const parent = node.parentNode! as Element;
if (parent.nodeName === 'OL') {
const start = parent.getAttribute('start');
const index = Array.prototype.indexOf.call(parent.children, node);
prefix = (start ? Number(start) + index : index + 1) + '. ';
}
return ' ' + prefix + content + (node.nextSibling && !/\n$/.test(content) ? '\n' : '');
},
});
// Remove bold; remove() doesn't work on this, I don't know why
this.turndownService.addRule('emph', {
filter: ['strong', 'b'],
replacement: function (content, node, options) {
if (!content.trim()) return '';
return content;
},
});
}
abstract actionLine(content: string): Promise<ActionResult | ActionResult[]>;
async action(content: string): Promise<ToolObservation[]> {
const lines = content.split('\n');
let results: ActionResult[] = [];
for (const line of lines) {
logger.info(`Action line: ${line}`)
try {
const lineActionResult = await this.actionLine(line);
logger.debug(`Action line result: ${JSON.stringify(lineActionResult, null, 2)}`);
if (Array.isArray(lineActionResult)) {
results = results.concat(lineActionResult);
} else {
results.push(lineActionResult);
}
} catch (err) {
const returnContentType = 'system_error';
results.push({
contentType: returnContentType,
roleMetadata: returnContentType,
message: `Error when executing command ${line}\n${err}`,
metadata: {
failedCommand: line,
},
});
}
}
const observations: ToolObservation[] = [];
for (const result of results) {
const observation: ToolObservation = {
contentType: result.contentType,
result: result.message,
roleMetadata: result.roleMetadata,
metadata: result.metadata ?? {},
};
if (result.metadataList) {
observation.metadata.metadata_list = result.metadataList;
}
observations.push(observation);
}
return observations;
}
postProcess(message: Message, metadata: any) {
const quotePattern = /【(.+?)†(.*?)】/g;
const content = message.content;
let match;
let citations: Citation[] = [];
const citation_format_type = 'tether_og';
while ((match = quotePattern.exec(content))) {
logger.debug(`Citation match: ${match[0]}`);
const start_ix = match.index;
const end_ix = match.index + match[0].length;
let invalid_reason = undefined;
let metadata: TetherQuoteMetadata;
try {
let cited_message_idx = parseInt(match[1]);
let evidence_text = match[2];
let quote = this.state.quotes[cited_message_idx.toString()];
if (quote === undefined) {
invalid_reason = `'Referenced message ${cited_message_idx} in citation 【${cited_message_idx}†${evidence_text}】 is not a quote or tether browsing display.'`;
logger.error(`Triggered citation error with quote undefined: ${invalid_reason}`);
citations.push({
citation_format_type,
start_ix,
end_ix,
invalid_reason,
});
} else {
let extra: TetherQuoteExtra = {
cited_message_idx,
evidence_text,
};
const quote_metadata = quote.metadata[0];
metadata = {
type: 'webpage',
title: quote_metadata.title,
url: quote_metadata.url,
text: quote_metadata.lines.join('\n'),
extra,
};
citations.push({
citation_format_type,
start_ix,
end_ix,
metadata,
});
}
} catch (err) {
logger.error(`Triggered citation error: ${err}`);
invalid_reason = `Citation Error: ${err}`;
citations.push({
start_ix,
end_ix,
citation_format_type,
invalid_reason,
});
}
}
metadata.citations = citations;
}
getState() {
return this.state;
}
}
export class SimpleBrowser extends BaseBrowser {
public description = 'SimpleBrowser';
constructor() {
super();
}
async actionLine(content: string): Promise<ActionResult | ActionResult[]> {
const regex = /(\w+)\(([^)]*)\)/;
const matches = content.match(regex);
if (matches) {
const functionName = matches[1];
let args_string = matches[2];
if (functionName === 'mclick') {
args_string = args_string.trim().slice(1, -1); // remove '[' and ']'
}
const args = args_string.split(',').map(arg => arg.trim());
let result;
switch (functionName) {
case 'search':
logger.debug(`SimpleBrowser action search ${args[0].slice(1, -1)}`);
const recency_days = /(^|\D)(\d+)($|\D)/.exec(args[1])?.[2] as undefined | `${number}`;
result = await this.functions.search(
args[0].slice(1, -1), // slice quote "query"
recency_days && Number(recency_days),
);
break;
case 'open_url':
logger.debug(`SimpleBrowser action open_url ${args[0].slice(1, -1)}`);
result = await this.functions.open_url(args[0].slice(1, -1));
break;
case 'mclick':
logger.debug(`SimpleBrowser action mclick ${args}`);
result = await this.functions.mclick(args.map(x => parseInt(x)));
break;
default:
throw new Error(`Parse Error: ${content}`);
}
return result;
} else {
throw new Error('Parse Error');
}
}
}
if (require.main === module) {
(async () => {
let browser = new SimpleBrowser();
let demo = async (action: string) => {
logger.info(` ------ Begin of Action: ${action} ------`);
let results = await browser.action(action);
for (const [idx, result] of results.entries()) {
logger.info(`[Result ${idx}] contentType: ${result.contentType}`);
logger.info(`[Result ${idx}] roleMetadata: ${result.roleMetadata}`);
logger.info(`[Result ${idx}] result: ${result.result}`);
logger.info(`[Result ${idx}] metadata: ${JSON.stringify(result.metadata, null, 2)}`);
}
logger.info(` ------ End of Action: ${action} ------\n\n`);
};
await demo("search('Apple Latest News')");
await demo('mclick([0, 1, 5, 6])');
await demo('mclick([1, 999999])');
await demo("open_url('https://chatglm.cn')");
await demo("search('zhipu latest News')");
await demo('mclick([0, 1, 5, 6])');
})();
}
================================================
FILE: demo/composite_demo/browser/src/config.ts
================================================
export default {
LOG_LEVEL: 'debug',
BROWSER_TIMEOUT: 10000,
BING_SEARCH_API_URL: 'https://api.bing.microsoft.com/v7.0/custom/',
BING_SEARCH_API_KEY: 'YOUR_BING_SEARCH_API_KEY',
CUSTOM_CONFIG_ID : 'YOUR_CUSTOM_CONFIG_ID', //将您的Custom Configuration ID放在此处
HOST: 'localhost',
PORT: 3000,
};
================================================
FILE: demo/composite_demo/browser/src/server.ts
================================================
import express, { Express, Request, Response } from 'express';
import { SimpleBrowser } from './browser';
import config from './config';
import { logger } from './utils';
const session_history: Record<string, SimpleBrowser> = {};
const app: Express = express();
app.use(express.json());
app.post('/', async (req: Request, res: Response) => {
const {
session_id,
action,
}: {
session_id: string;
action: string;
} = req.body;
logger.info(`session_id: ${session_id}`);
logger.info(`action: ${action}`);
if (!session_history[session_id]) {
session_history[session_id] = new SimpleBrowser();
}
const browser = session_history[session_id];
try {
res.json(await browser.action(action));
} catch (err) {
logger.error(err);
res.status(400).json(err);
}
})
process.on('SIGINT', () => {
process.exit(0);
});
process.on('uncaughtException', e => {
logger.error(e);
});
const { HOST, PORT } = config;
(async () => {
app.listen(PORT, HOST, () => {
logger.info(`⚡️[server]: Server is running at http://${HOST}:${PORT}`);
try {
(<any>process).send('ready');
} catch (err) {}
});
})();
================================================
FILE: demo/composite_demo/browser/src/types.ts
================================================
export interface File {
id: string;
name: string;
size: number;
}
export interface Metadata {
files?: File[];
reference?: string;
}
export interface Message {
role: 'user' | 'assistant' | 'system' | 'observation';
metadata: string;
content: string;
request_metadata?: Metadata;
}
export interface ToolObservation {
contentType: string;
result: string;
text?: string;
roleMetadata?: string; // metadata for <|observation|>${metadata}
metadata: any; // metadata for response
}
================================================
FILE: demo/composite_demo/browser/src/utils.ts
================================================
import winston from 'winston';
import config from './config';
export class TimeoutError extends Error {}
const logLevel = config.LOG_LEVEL;
export const logger = winston.createLogger({
level: logLevel,
format: winston.format.combine(
winston.format.colorize(),
winston.format.printf(info => {
return `${info.level}: ${info.message}`;
}),
),
transports: [new winston.transports.Console()],
});
console.log('LOG_LEVEL', logLevel);
export const parseHrtimeToMillisecond = (hrtime: [number, number]): number => {
return (hrtime[0] + hrtime[1] / 1e9) * 1000;
};
export const promiseWithTime = <T>(
promise: Promise<T>
): Promise<{
value: T;
time: number;
}> => {
return new Promise((resolve, reject) => {
const startTime = process.hrtime();
promise
.then(value => {
resolve({
value: value,
time: parseHrtimeToMillisecond(process.hrtime(startTime))
});
})
.catch(err => reject(err));
});
};
export const withTimeout = <T>(
millis: number,
promise: Promise<T>
): Promise<{
value: T;
time: number;
}> => {
const timeout = new Promise<{ value: T; time: number }>((_, reject) =>
setTimeout(() => reject(new TimeoutError()), millis)
);
return Promise.race([promiseWithTime(promise), timeout]);
};
================================================
FILE: demo/composite_demo/browser/tsconfig.json
================================================
{
"compilerOptions": {
"target": "es2022",
"lib": ["es2022", "dom"],
"module": "commonjs",
"rootDir": "./",
"outDir": "./dist",
"esModuleInterop": true,
"forceConsistentCasingInFileNames": true,
"strict": true,
},
"ts-node": {
"transpileOnly": true
}
}
================================================
FILE: demo/composite_demo/requirements.txt
================================================
# Please install the requirments.txt in inference first!
ipykernel>=6.26.0
ipython>=8.18.1
jupyter_client>=8.6.0
langchain>=0.2.12
langchain-community>=0.2.11
matplotlib>=3.9.1
pymupdf>=1.24.9
python-docx>=1.1.2
python-pptx>=0.6.23
pyyaml>=6.0.1
requests>=2.31.0
streamlit>=1.37.1
zhipuai>=2.1.4
================================================
FILE: demo/composite_demo/src/client.py
================================================
"""
This is a client part of composite_demo.
We provide two clients, HFClient and VLLMClient, which are used to interact with the model.
The HFClient is used to interact with the transformers backend, and the VLLMClient is used to interact with the VLLM model.
"""
import json
from collections.abc import Generator
from copy import deepcopy
from enum import Enum, auto
from typing import Protocol
import streamlit as st
from conversation import Conversation, build_system_prompt
from tools.tool_registry import ALL_TOOLS
class ClientType(Enum):
HF = auto()
VLLM = auto()
API = auto()
class Client(Protocol):
def __init__(self, model_path: str): ...
def generate_stream(
self,
tools: list[dict],
history: list[Conversation],
**parameters,
) -> Generator[tuple[str | dict, list[dict]]]: ...
def process_input(history: list[dict], tools: list[dict], role_name_replace: dict = None) -> list[dict]:
chat_history = []
# if len(tools) > 0:
chat_history.append({"role": "system", "content": build_system_prompt(list(ALL_TOOLS), tools)})
for conversation in history:
role = str(conversation.role).removeprefix("<|").removesuffix("|>")
if role_name_replace:
role = role_name_replace.get(role, role)
item = {
"role": role,
"content": conversation.content,
}
if conversation.metadata:
item["metadata"] = conversation.metadata
# Only append image for user
if role == "user" and conversation.image:
item["image"] = conversation.image
chat_history.append(item)
return chat_history
def process_response(output, history):
content = ""
history = deepcopy(history)
for response in output.split("<|assistant|>"):
if "\n" in response:
metadata, content = response.split("\n", maxsplit=1)
else:
metadata, content = "", response
if not metadata.strip():
content = content.strip()
history.append({"role": "assistant", "metadata": metadata, "content": content})
content = content.replace("[[训练时间]]", "2023年")
else:
history.append({"role": "assistant", "metadata": metadata, "content": content})
if history[0]["role"] == "system" and "tools" in history[0]:
parameters = json.loads(content)
content = {"name": metadata.strip(), "parameters": parameters}
else:
content = {"name": metadata.strip(), "content": content}
return content, history
# glm-4v-9b is not available in vLLM backend, use HFClient instead.
@st.cache_resource(max_entries=1, show_spinner="Loading model...")
def get_client(model_path, typ: ClientType) -> Client:
match typ:
case ClientType.HF:
from clients.hf import HFClient
return HFClient(model_path)
case ClientType.VLLM:
try:
from clients.vllm import VLLMClient
except ImportError as e:
e.msg += "; did you forget to install vLLM?"
raise
return VLLMClient(model_path)
case ClientType.API:
from clients.openai import APIClient
return APIClient(model_path)
raise NotImplementedError(f"Client type {typ} is not supported.")
================================================
FILE: demo/composite_demo/src/clients/hf.py
================================================
"""
HuggingFace client.
"""
from collections.abc import Generator
from threading import Thread
import torch
from client import Client, process_input, process_response
from conversation import Conversation
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
class HFClient(Client):
def __init__(self, model_path: str):
self.tokenizer = AutoTokenizer.from_pretrained(
model_path,
trust_remote_code=True,
)
self.model = AutoModelForCausalLM.from_pretrained(
model_path,
torch_dtype=torch.bfloat16,
device_map="cuda",
).eval()
def generate_stream(
self,
tools: list[dict],
history: list[Conversation],
**parameters,
) -> Generator[tuple[str | dict, list[dict]]]:
chat_history = process_input(history, tools)
model_inputs = self.tokenizer.apply_chat_template(
chat_history,
add_generation_prompt=True,
tokenize=True,
return_tensors="pt",
return_dict=True,
).to(self.model.device)
streamer = TextIteratorStreamer(
tokenizer=self.tokenizer,
timeout=5,
skip_prompt=True,
)
generate_kwargs = {
**model_inputs,
"streamer": streamer,
"eos_token_id": [151329, 151336, 151338],
"do_sample": True,
}
generate_kwargs.update(parameters)
t = Thread(target=self.model.generate, kwargs=generate_kwargs)
t.start()
total_text = ""
for token_text in streamer:
total_text += token_text
yield process_response(total_text, chat_history)
================================================
FILE: demo/composite_demo/src/clients/openai.py
================================================
"""
OpenAI API client.
"""
from collections.abc import Generator
from client import Client, process_input, process_response
from conversation import Conversation
from openai import OpenAI
def format_openai_tool(origin_tools):
openai_tools = []
for tool in origin_tools:
openai_param = {}
for param in tool["params"]:
openai_param[param["name"]] = {}
openai_tool = {
"type": "function",
"function": {
"name": tool["name"],
"description": tool["description"],
"parameters": {
"type": "object",
"properties": {
param["name"]: {"type": param["type"], "description": param["description"]}
for param in tool["params"]
},
"required": [param["name"] for param in tool["params"] if param["required"]],
},
},
}
openai_tools.append(openai_tool)
return openai_tools
class APIClient(Client):
def __init__(self, model_path: str):
base_url = "http://127.0.0.1:8000/v1/"
self.client = OpenAI(api_key="EMPTY", base_url=base_url)
self.use_stream = False
self.role_name_replace = {"observation": "tool"}
def generate_stream(
self,
tools: list[dict],
history: list[Conversation],
**parameters,
) -> Generator[tuple[str | dict, list[dict]]]:
chat_history = process_input(history, "", role_name_replace=self.role_name_replace)
# messages = process_input(history, '', role_name_replace=self.role_name_replace)
openai_tools = format_openai_tool(tools)
response = self.client.chat.completions.create(
model="glm-4",
messages=chat_history,
tools=openai_tools,
stream=self.use_stream,
max_tokens=parameters["max_new_tokens"],
temperature=parameters["temperature"],
presence_penalty=1.2,
top_p=parameters["top_p"],
tool_choice="auto",
)
output = response.choices[0].message
if output.tool_calls:
glm4_output = output.tool_calls[0].function.name + "\n" + output.tool_calls[0].function.arguments
else:
glm4_output = output.content
yield process_response(glm4_output, chat_history)
================================================
FILE: demo/composite_demo/src/clients/vllm.py
================================================
"""
vLLM client.
Please install [vLLM](https://github.com/vllm-project/vllm) according to its
installation guide before running this client.
"""
import time
from collections.abc import Generator
from client import Client, process_input, process_response
from conversation import Conversation
from transformers import AutoTokenizer
from vllm import EngineArgs, LLMEngine, SamplingParams
class VLLMClient(Client):
def __init__(self, model_path: str):
self.tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
self.engine_args = EngineArgs(
model=model_path,
tensor_parallel_size=1,
dtype="bfloat16", # torch.bfloat16 is needed.
gpu_memory_utilization=0.6,
enforce_eager=True,
worker_use_ray=False,
)
self.engine = LLMEngine.from_engine_args(self.engine_args)
def generate_stream(
self, tools: list[dict], history: list[Conversation], **parameters
) -> Generator[tuple[str | dict, list[dict]]]:
chat_history = process_input(history, tools)
model_inputs = self.tokenizer.apply_chat_template(chat_history, add_generation_prompt=True, tokenize=False)
parameters["max_tokens"] = parameters.pop("max_new_tokens")
params_dict = {
"n": 1,
"best_of": 1,
"top_p": 1,
"top_k": -1,
"length_penalty": 1,
"stop_token_ids": [151329, 151336, 151338],
}
params_dict.update(parameters)
sampling_params = SamplingParams(**params_dict)
self.engine.add_request(request_id=str(time.time()), inputs=model_inputs, params=sampling_params)
while self.engine.has_unfinished_requests():
request_outputs = self.engine.step()
for request_output in request_outputs:
yield process_response(request_output.outputs[0].text, chat_history)
================================================
FILE: demo/composite_demo/src/conversation.py
================================================
import json
import re
from dataclasses import dataclass
from datetime import datetime
from enum import Enum, auto
import streamlit as st
from PIL.Image import Image
from streamlit.delta_generator import DeltaGenerator
from tools.browser import Quote, quotes
QUOTE_REGEX = re.compile(r"【(\d+)†(.+?)】")
SELFCOG_PROMPT = "你是一个名为 GLM-4 的人工智能助手。你是基于智谱AI训练的语言模型 GLM-4 模型开发的，你的任务是针对用户的问题和要求提供适当的答复和支持。"
DATE_PROMPT = "当前日期: %Y-%m-%d"
TOOL_SYSTEM_PROMPTS = {
"python": "当你向 `python` 发送包含 Python 代码的消息时，该代码将会在一个有状态的 Jupyter notebook 环境中执行。\n`python` 返回代码执行的输出，或在执行 60 秒后返回超时。\n`/mnt/data` 将会持久化存储你的文件。在此会话中，`python` 无法访问互联网。不要使用 `python` 进行任何网络请求或者在线 API 调用，这些在线内容的访问将不会成功。",
"simple_browser": "你可以使用 `simple_browser` 工具。该工具支持以下函数：\n`search(query: str, recency_days: int)`：使用搜索引擎进行查询并显示结果，可以使用 `recency_days` 参数控制搜索内容的时效性。\n`mclick(ids: list[int])`：获取一系列指定 id 的页面内容。每次调用时，须选择3-10个页面。选择多个角度的页面，同时尽可能选择可信任的信息来源。考虑到部分页面是无法加载的，你也可以多打开一些可能有用的页面而不用担心内容过多。\n`open_url(url: str)`：打开指定的 URL。\n\n使用 `【{引用 id}†{引用文本}】` 来引用内容。\n\n操作步骤：1. 使用 `search` 来获得信息列表; 2. 使用 `mclick` 来获取指定 ID 页面的内容; 3. 根据获得的内容进行回复。在回复中应当引用信息来源。\n 如果用户提供了 URL，也可以用 `open_url` 直接打开页面。\n如果初次搜索结果没有找到合适的信息，也可以再次使用 `search` 进行搜索。",
"cogview": "如果用户的请求中包含了对图像的描述，你可以使用 `cogview` 来生成图像并展示给用户。你需要向 `cogview` 发送图像描述，规则：\n- 发送给 `cogview` 的消息必须使用英语。用户的中文描述必须完全翻译为英语。\n- 应当尽可能详细地描述图像生成的需求，需求描述约 100 英文单词。\n- 保持用户原始描述的意图。不要虚构内容或者没见过的人物。\n- 如无特殊说明，所在地为中国，持有中国立场并遵循中国社会主义价值观。",
}
FILE_TEMPLATE = "[File Name]\n{file_name}\n[File Content]\n{file_content}"
def build_system_prompt(
enabled_tools: list[str],
functions: list[dict],
):
value = SELFCOG_PROMPT
value += "\n\n" + datetime.now().strftime(DATE_PROMPT)
if enabled_tools or functions:
value += "\n\n# 可用工具"
contents = []
for tool in enabled_tools:
contents.append(f"\n\n## {tool}\n\n{TOOL_SYSTEM_PROMPTS[tool]}")
for function in functions:
content = f"\n\n## {function['name']}\n\n{json.dumps(function, ensure_ascii=False, indent=4)}"
content += "\n在调用上述函数时，请使用 Json 格式表示调用的参数。"
contents.append(content)
value += "".join(contents)
return value
def response_to_str(response: str | dict[str, str]) -> str:
"""
Convert response to string.
"""
if isinstance(response, dict):
return response.get("name", "") + response.get("content", "")
return response
class Role(Enum):
SYSTEM = auto()
USER = auto()
ASSISTANT = auto()
TOOL = auto()
OBSERVATION = auto()
def __str__(self):
match self:
case Role.SYSTEM:
return "<|system|>"
case Role.USER:
return "<|user|>"
case Role.ASSISTANT | Role.TOOL:
return "<|assistant|>"
case Role.OBSERVATION:
return "<|observation|>"
# Get the message block for the given role
def get_message(self):
# Compare by value here, because the enum object in the session state
# is not the same as the enum cases here, due to streamlit's rerunning
# behavior.
match self.value:
case Role.SYSTEM.value:
return
case Role.USER.value:
return st.chat_message(name="user", avatar="user")
case Role.ASSISTANT.value:
return st.chat_message(name="assistant", avatar="assistant")
case Role.TOOL.value:
return st.chat_message(name="tool", avatar="assistant")
case Role.OBSERVATION.value:
return st.chat_message(name="observation", avatar="assistant")
case _:
st.error(f"Unexpected role: {self}")
@dataclass
class Conversation:
role: Role
content: str | dict
# Processed content
saved_content: str | None = None
metadata: str | None = None
image: str | Image | None = None
def __str__(self) -> str:
metadata_str = self.metadata if self.metadata else ""
return f"{self.role}{metadata_str}\n{self.content}"
# Human readable format
def get_text(self) -> str:
text = self.saved_content or self.content
match self.role.value:
case Role.TOOL.value:
text = f"Calling tool `{self.metadata}`:\n\n```python\n{text}\n```"
case Role.OBSERVATION.value:
text = f"```python\n{text}\n```"
return text
# Display as a markdown block
def show(self, placeholder: DeltaGenerator | None = None) -> str:
if placeholder:
message = placeholder
else:
message = self.role.get_message()
if self.image:
message.image(self.image, width=512)
if self.role == Role.OBSERVATION:
metadata_str = f"from {self.metadata}" if self.metadata else ""
message = message.expander(f"Observation {metadata_str}")
text = self.get_text()
if self.role != Role.USER:
show_text = text
else:
splitted = text.split("files uploaded.\n")
if len(splitted) == 1:
show_text = text
else:
# Show expander for document content
doc = splitted[0]
show_text = splitted[-1]
expander = message.expander("File Content")
expander.markdown(doc)
message.markdown(show_text)
def postprocess_text(text: str, replace_quote: bool) -> str:
text = text.replace(r"\(", "$")
text = text.replace(r"\)", "$")
text = text.replace(r"\[", "$$")
text = text.replace(r"\]", "$$")
text = text.replace("<|assistant|>", "")
text = text.replace("<|observation|>", "")
text = text.replace("<|system|>", "")
text = text.replace("<|user|>", "")
text = text.replace("<|endoftext|>", "")
# Replace quotes
if replace_quote:
for match in QUOTE_REGEX.finditer(text):
quote_id = match.group(1)
quote = quotes.get(quote_id, Quote("未找到引用内容", ""))
text = text.replace(match.group(0), f" (来源：[{quote.title}]({quote.url})) ")
return text.strip()
================================================
FILE: demo/composite_demo/src/main.py
================================================
"""
This demo show the All tools and Long Context chat Capabilities of GLM-4.
Please follow the Readme.md to run the demo.
"""
import os
import traceback
from enum import Enum
from io import BytesIO
from uuid import uuid4
import streamlit as st
from client import Client, ClientType, get_client
from conversation import (
FILE_TEMPLATE,
Conversation,
Role,
postprocess_text,
response_to_str,
)
from PIL import Image
from streamlit.delta_generator import DeltaGenerator
from tools.tool_registry import dispatch_tool, get_tools
from utils import extract_docx, extract_pdf, extract_pptx, extract_text
CHAT_MODEL_PATH = os.environ.get("CHAT_MODEL_PATH", "THUDM/glm-4-9b-chat")
VLM_MODEL_PATH = os.environ.get("VLM_MODEL_PATH", "THUDM/glm-4v-9b")
USE_VLLM = os.environ.get("USE_VLLM", "0") == "1"
USE_API = os.environ.get("USE_API", "0") == "1"
class Mode(str, Enum):
ALL_TOOLS = "🛠️ All Tools"
LONG_CTX = "📝 文档解读"
VLM = "🖼️ 多模态"
def append_conversation(
conversation: Conversation,
history: list[Conversation],
placeholder: DeltaGenerator | None = None,
) -> None:
"""
Append a conversation piece into history, meanwhile show it in a new markdown block
"""
history.append(conversation)
conversation.show(placeholder)
st.set_page_config(
page_title="GLM-4 Demo",
page_icon=":robot:",
layout="centered",
initial_sidebar_state="expanded",
)
st.title("GLM-4 Demo")
st.markdown(
"<sub>智谱AI 公开在线技术文档: https://zhipu-ai.feishu.cn/wiki/RuMswanpkiRh3Ok4z5acOABBnjf </sub> \n\n <sub> 更多 GLM-4 开源模型的使用方法请参考文档。</sub>",
unsafe_allow_html=True,
)
with st.sidebar:
top_p = st.slider("top_p", 0.0, 1.0, 0.8, step=0.01)
top_k = st.slider("top_k", 1, 20, 10, step=1, key="top_k")
temperature = st.slider("temperature", 0.0, 1.5, 0.95, step=0.01)
repetition_penalty = st.slider("repetition_penalty", 0.0, 2.0, 1.0, step=0.01)
max_new_tokens = st.slider("max_new_tokens", 1, 4096, 2048, step=1)
cols = st.columns(2)
export_btn = cols[0]
clear_history = cols[1].button("Clear", use_container_width=True)
retry = export_btn.button("Retry", use_container_width=True)
if clear_history:
page = st.session_state.page
client = st.session_state.client
st.session_state.clear()
st.session_state.page = page
st.session_state.client = client
st.session_state.files_uploaded = False
st.session_state.uploaded_texts = ""
st.session_state.uploaded_file_nums = 0
st.session_state.history = []
if "files_uploaded" not in st.session_state:
st.session_state.files_uploaded = False
if "session_id" not in st.session_state:
st.session_state.session_id = uuid4()
if "history" not in st.session_state:
st.session_state.history = []
first_round = len(st.session_state.history) == 0
def build_client(mode: Mode) -> Client:
match mode:
case Mode.ALL_TOOLS:
st.session_state.top_k = 10
typ = ClientType.VLLM if USE_VLLM else ClientType.HF
typ = ClientType.API if USE_API else typ
return get_client(CHAT_MODEL_PATH, typ)
case Mode.LONG_CTX:
st.session_state.top_k = 10
typ = ClientType.VLLM if USE_VLLM else ClientType.HF
return get_client(CHAT_MODEL_PATH, typ)
case Mode.VLM:
st.session_state.top_k = 1
# vLLM is not available for VLM mode
return get_client(VLM_MODEL_PATH, ClientType.HF)
# Callback function for page change
def page_changed() -> None:
global client
new_page: str = st.session_state.page
st.session_state.history.clear()
st.session_state.client = build_client(Mode(new_page))
page = st.radio(
"选择功能",
[mode.value for mode in Mode],
key="page",
horizontal=True,
index=None,
label_visibility="hidden",
on_change=page_changed,
)
HELP = """
### 🎉 欢迎使用 GLM-4!
请在上方选取一个功能。每次切换功能时，将会重新加载模型并清空对话历史。
文档解读模式与 VLM 模式仅支持在第一轮传入文档或图像。
""".strip()
if page is None:
st.markdown(HELP)
exit()
if page == Mode.LONG_CTX:
if first_round:
uploaded_files = st.file_uploader(
"上传文件",
type=["pdf", "txt", "py", "docx", "pptx", "json", "cpp", "md"],
accept_multiple_files=True,
)
if uploaded_files and not st.session_state.files_uploaded:
uploaded_texts = []
for uploaded_file in uploaded_files:
file_name: str = uploaded_file.name
random_file_name = str(uuid4())
file_extension = os.path.splitext(file_name)[1]
file_path = os.path.join("/tmp", random_file_name + file_extension)
with open(file_path, "wb") as f:
f.write(uploaded_file.getbuffer())
if file_name.endswith(".pdf"):
content = extract_pdf(file_path)
elif file_name.endswith(".docx"):
content = extract_docx(file_path)
elif file_name.endswith(".pptx"):
content = extract_pptx(file_path)
else:
content = extract_text(file_path)
uploaded_texts.append(FILE_TEMPLATE.format(file_name=file_name, file_content=content))
os.remove(file_path)
st.session_state.uploaded_texts = "\n\n".join(uploaded_texts)
st.session_state.uploaded_file_nums = len(uploaded_files)
else:
st.session_state.uploaded_texts = ""
st.session_state.uploaded_file_nums = 0
elif page == Mode.VLM:
if first_round:
uploaded_image = st.file_uploader(
"上传图片",
type=["png", "jpg", "jpeg", "bmp", "tiff", "webp"],
accept_multiple_files=False,
)
if uploaded_image:
data: bytes = uploaded_image.read()
image = Image.open(BytesIO(data)).convert("RGB")
st.session_state.uploaded_image = image
else:
st.session_state.uploaded_image = None
prompt_text = st.chat_input("Chat with GLM-4!", key="chat_input")
if prompt_text == "" and retry == False:
print("\n== Clean ==\n")
st.session_state.history = []
exit()
history: list[Conversation] = st.session_state.history
if retry:
print("\n== Retry ==\n")
last_user_conversation_idx = None
for idx, conversation in enumerate(history):
if conversation.role.value == Role.USER.value:
last_user_conversation_idx = idx
if last_user_conversation_idx is not None:
prompt_text = history[last_user_conversation_idx].content
print(f"New prompt: {prompt_text}, idx = {last_user_conversation_idx}")
del history[last_user_conversation_idx:]
for conversation in history:
conversation.show()
tools = get_tools() if page == Mode.ALL_TOOLS else []
client: Client = st.session_state.client
def main(prompt_text: str):
global client
assert client is not None
if prompt_text:
prompt_text = prompt_text.strip()
# Append uploaded files
uploaded_texts = st.session_state.get("uploaded_texts")
if page == Mode.LONG_CTX and uploaded_texts and first_round:
meta_msg = "{} files uploaded.\n".format(st.session_state.uploaded_file_nums)
prompt_text = uploaded_texts + "\n\n\n" + meta_msg + prompt_text
# Clear after first use
st.session_state.files_uploaded = True
st.session_state.uploaded_texts = ""
st.session_state.uploaded_file_nums = 0
image = st.session_state.get("uploaded_image")
if page == Mode.VLM and image and first_round:
st.session_state.uploaded_image = None
role = Role.USER
append_conversation(Conversation(role, prompt_text, image=image), history)
placeholder = st.container()
message_placeholder = placeholder.chat_message(name="assistant", avatar="assistant")
markdown_placeholder = message_placeholder.empty()
def add_new_block():
nonlocal message_placeholder, markdown_placeholder
message_placeholder = placeholder.chat_message(name="assistant", avatar="assistant")
markdown_placeholder = message_placeholder.empty()
def commit_conversation(
role: Role,
text: str,
metadata: str | None = None,
image: str | None = None,
new: bool = False,
):
processed_text = postprocess_text(text, role.value == Role.ASSISTANT.value)
conversation = Conversation(role, text, processed_text, metadata, image)
# Use different placeholder for new block
placeholder = message_placeholder if new else markdown_placeholder
append_conversation(
conversation,
history,
placeholder,
)
response = ""
for _ in range(10):
last_response = None
history_len = None
try:
for response, chat_history in client.generate_stream(
tools=tools,
history=history,
temperature=temperature,
top_p=top_p,
top_k=top_k,
repetition_penalty=repetition_penalty,
max_new_tokens=max_new_tokens,
):
if history_len is None:
history_len = len(chat_history)
elif history_len != len(chat_history):
commit_conversation(Role.ASSISTANT, last_response)
add_new_block()
history_len = len(chat_history)
last_response = response
replace_quote = chat_history[-1]["role"] == "assistant"
markdown_placeholder.markdown(postprocess_text(str(response) + "●", replace_quote=replace_quote))
else:
metadata = page == Mode.ALL_TOOLS and isinstance(response, dict) and response.get("name") or None
role = Role.TOOL if metadata else Role.ASSISTANT
text = response.get("content") if metadata else response_to_str(response)
commit_conversation(role, text, metadata)
if metadata:
add_new_block()
try:
with markdown_placeholder:
with st.spinner(f"Calling tool {metadata}..."):
observations = dispatch_tool(metadata, text, str(st.session_state.session_id))
except Exception as e:
traceback.print_exc()
st.error(f'Uncaught exception in `"{metadata}"`: {e}')
break
for observation in observations:
observation.text = observation.text
commit_conversation(
Role.OBSERVATION,
observation.text,
observation.role_metadata,
observation.image_url,
new=True,
)
add_new_block()
continue
else:
break
except Exception:
traceback.print_exc()
st.error(f"Uncaught exception: {traceback.format_exc()}")
else:
st.error("Too many chaining function calls!")
main(prompt_text)
================================================
FILE: demo/composite_demo/src/tools/browser.py
================================================
"""
Simple browser tool.
# Usage
Please start the backend browser server according to the instructions in the README.
"""
import re
from dataclasses import dataclass
from pprint import pprint
import requests
import streamlit as st
from .config import BROWSER_SERVER_URL
from .interface import ToolObservation
QUOTE_REGEX = re.compile(r"\[(\d+)†(.+?)\]")
@dataclass
class Quote:
title: str
url: str
# Quotes for displaying reference
if "quotes" not in st.session_state:
st.session_state.quotes = {}
quotes: dict[str, Quote] = st.session_state.quotes
def map_response(response: dict) -> ToolObservation:
# Save quotes for reference
print("===BROWSER_RESPONSE===")
pprint(response)
role_metadata = response.get("roleMetadata")
metadata = response.get("metadata")
if role_metadata.split()[0] == "quote_result" and metadata:
quote_id = QUOTE_REGEX.search(role_metadata.split()[1]).group(1)
quote: dict[str, str] = metadata["metadata_list"][0]
quotes[quote_id] = Quote(quote["title"], quote["url"])
elif role_metadata == "browser_result" and metadata:
for i, quote in enumerate(metadata["metadata_list"]):
quotes[str(i)] = Quote(quote["title"], quote["url"])
return ToolObservation(
content_type=response.get("contentType"),
text=response.get("result"),
role_metadata=role_metadata,
metadata=metadata,
)
def tool_call(code: str, session_id: str) -> list[ToolObservation]:
request = {
"session_id": session_id,
"action": code,
}
response = requests.post(BROWSER_SERVER_URL, json=request).json()
return list(map(map_response, response))
================================================
FILE: demo/composite_demo/src/tools/cogview.py
================================================
import streamlit as st
from zhipuai import ZhipuAI
from zhipuai.types.image import GeneratedImage
from .config import COGVIEW_MODEL, ZHIPU_AI_KEY
from .interface import ToolObservation
@st.cache_resource
def get_zhipu_client():
return ZhipuAI(api_key=ZHIPU_AI_KEY)
def map_response(img: GeneratedImage):
return ToolObservation(
content_type="image",
text="CogView 已经生成并向用户展示了生成的图片。",
image_url=img.url,
role_metadata="cogview_result",
)
def tool_call(prompt: str, session_id: str) -> list[ToolObservation]:
client = get_zhipu_client()
response = client.images.generations(model=COGVIEW_MODEL, prompt=prompt).data
return list(map(map_response, response))
================================================
FILE: demo/composite_demo/src/tools/config.py
================================================
BROWSER_SERVER_URL = "http://localhost:3000"
IPYKERNEL = "glm-4-demo"
ZHIPU_AI_KEY = ""
COGVIEW_MODEL = "cogview-3"
================================================
FILE: demo/composite_demo/src/tools/interface.py
================================================
from dataclasses import dataclass
from typing import Any
@dataclass
class ToolObservation:
content_type: str
text: str
image_url: str | None = None
role_metadata: str | None = None
metadata: Any = None
================================================
FILE: demo/composite_demo/src/tools/python.py
================================================
import queue
import re
from pprint import pprint
from subprocess import PIPE
from typing import Literal
import jupyter_client
import streamlit as st
from .config import IPYKERNEL
from .interface import ToolObservation
ANSI_ESCAPE = re.compile(r"(\x9B|\x1B\[|\u001b\[)[0-?]*[ -/]*[@-~]")
CODE = re.compile(r"```([^\n]*)\n(.*?)```")
class CodeKernel:
def __init__(
self,
kernel_name="kernel",
kernel_id=None,
kernel_config_path="",
python_path=None,
ipython_path=None,
init_file_path="./startup.py",
verbose=1,
):
self.kernel_name = kernel_name
self.kernel_id = kernel_id
self.kernel_config_path = kernel_config_path
self.python_path = python_path
self.ipython_path = ipython_path
self.init_file_path = init_file_path
self.verbose = verbose
if python_path is None and ipython_path is None:
env = None
else:
env = {"PATH": self.python_path + ":$PATH", "PYTHONPATH": self.python_path}
# Initialize the backend kernel
self.kernel_manager = jupyter_client.KernelManager(
kernel_name=IPYKERNEL, connection_file=self.kernel_config_path, exec_files=[self.init_file_path], env=env
)
if self.kernel_config_path:
self.kernel_manager.load_connection_file()
self.kernel_manager.start_kernel(stdout=PIPE, stderr=PIPE)
print("Backend kernel started with the configuration: {}".format(self.kernel_config_path))
else:
self.kernel_manager.start_kernel(stdout=PIPE, stderr=PIPE)
print("Backend kernel started with the configuration: {}".format(self.kernel_manager.connection_file))
if verbose:
pprint(self.kernel_manager.get_connection_info())
# Initialize the code kernel
self.kernel = self.kernel_manager.blocking_client()
# self.kernel.load_connection_file()
self.kernel.start_channels()
print("Code kernel started.")
def execute(self, code):
self.kernel.execute(code)
try:
shell_msg = self.kernel.get_shell_msg(timeout=30)
io_msg_content = self.kernel.get_iopub_msg(timeout=30)["content"]
while True:
msg_out = io_msg_content
### Poll the message
try:
io_msg_content = self.kernel.get_iopub_msg(timeout=30)["content"]
if "execution_state" in io_msg_content and io_msg_content["execution_state"] == "idle":
break
except queue.Empty:
break
return shell_msg, msg_out
except Exception as e:
print(e)
return None
def execute_interactive(self, code, verbose=False):
shell_msg = self.kernel.execute_interactive(code)
if shell_msg is queue.Empty:
if verbose:
print("Timeout waiting for shell message.")
self.check_msg(shell_msg, verbose=verbose)
return shell_msg
def inspect(self, code, verbose=False):
msg_id = self.kernel.inspect(code)
shell_msg = self.kernel.get_shell_msg(timeout=30)
if shell_msg is queue.Empty:
if verbose:
print("Timeout waiting for shell message.")
self.check_msg(shell_msg, verbose=verbose)
return shell_msg
def get_error_msg(self, msg, verbose=False) -> str | None:
if msg["content"]["status"] == "error":
try:
error_msg = msg["content"]["traceback"]
except:
try:
error_msg = msg["content"]["traceback"][-1].strip()
except:
error_msg = "Traceback Error"
if verbose:
print("Error: ", error_msg)
return error_msg
return None
def check_msg(self, msg, verbose=False):
status = msg["content"]["status"]
if status == "ok":
if verbose:
print("Execution succeeded.")
elif status == "error":
for line in msg["content"]["traceback"]:
if verbose:
print(line)
def shutdown(self):
# Shutdown the backend kernel
self.kernel_manager.shutdown_kernel()
print("Backend kernel shutdown.")
# Shutdown the code kernel
self.kernel.shutdown()
print("Code kernel shutdown.")
def restart(self):
# Restart the backend kernel
self.kernel_manager.restart_kernel()
# print("Backend kernel restarted.")
def interrupt(self):
# Interrupt the backend kernel
self.kernel_manager.interrupt_kernel()
# print("Backend kernel interrupted.")
def is_alive(self):
return self.kernel.is_alive()
def clean_ansi_codes(input_string):
return ANSI_ESCAPE.sub("", input_string)
def extract_code(text: str) -> str:
matches = CODE.findall(text, re.DOTALL)
return matches[-1][1]
def execute(code: str, kernel: CodeKernel) -> tuple[Literal["text", "image"] | None, str]:
res = ""
res_type = None
code = code.replace("<|observation|>", "")
code = code.replace("<|assistant|>python", "")
code = code.replace("<|assistant|>", "")
code = code.replace("<|user|>", "")
code = code.replace("<|system|>", "")
msg, output = kernel.execute(code)
if msg["metadata"]["status"] == "timeout":
return res_type, "Timed out"
elif msg["metadata"]["status"] == "error":
return res_type, clean_ansi_codes("\n".join(kernel.get_error_msg(msg, verbose=True)))
if "text" in output:
res_type = "text"
res = output["text"]
elif "data" in output:
for key in output["data"]:
if "text/plain" in key:
res_type = "text"
res = output["data"][key]
elif "image/png" in key:
res_type = "image"
res = output["data"][key]
break
return res_type, res
@st.cache_resource
def get_kernel() -> CodeKernel:
return CodeKernel()
def tool_call(code: str, session_id: str) -> list[ToolObservation]:
kernel = get_kernel()
res_type, res = execute(code, kernel)
# Convert base64 to data uri
text = "[Image]" if res_type == "image" else res
image = f"data:image/png;base64,{res}" if res_type == "image" else None
return [ToolObservation(res_type, text, image)]
================================================
FILE: demo/composite_demo/src/tools/tool_registry.py
================================================
"""
This code is the tool registration part. By registering the tool, the model can call the tool.
This code provides extended functionality to the model, enabling it to call and interact with a variety of utilities
through defined interfaces.
"""
import copy
import inspect
import json
import subprocess
import traceback
from collections.abc import Callable
from types import GenericAlias
from typing import Annotated, get_origin
from .browser import tool_call as browser
from .cogview import tool_call as cogview
from .interface import ToolObservation
from .python import tool_call as python
ALL_TOOLS = {
"simple_browser": browser,
"python": python,
"cogview": cogview,
}
_TOOL_HOOKS = {}
_TOOL_DESCRIPTIONS = []
def register_tool(func: Callable):
tool_name = func.__name__
tool_description = inspect.getdoc(func).strip()
python_params = inspect.signature(func).parameters
tool_params = []
for name, param in python_params.items():
annotation = param.annotation
if annotation is inspect.Parameter.empty:
raise TypeError(f"Parameter `{name}` missing type annotation")
if get_origin(annotation) != Annotated:
raise TypeError(f"Annotation type for `{name}` must be typing.Annotated")
typ, (description, required) = annotation.__origin__, annotation.__metadata__
typ: str = str(typ) if isinstance(typ, GenericAlias) else typ.__name__
if not isinstance(description, str):
raise TypeError(f"Description for `{name}` must be a string")
if not isinstance(required, bool):
raise TypeError(f"Required for `{name}` must be a bool")
tool_params.append(
{
"name": name,
"description": description,
"type": typ,
"required": required,
}
)
tool_def = {
"name": tool_name,
"description": tool_description,
"params": tool_params,
}
# print("[registered tool] " + pformat(tool_def))
_TOOL_HOOKS[tool_name] = func
_TOOL_DESCRIPTIONS.append(tool_def)
return func
def dispatch_tool(tool_name: str, code: str, session_id: str) -> list[ToolObservation]:
# Dispatch predefined tools
if tool_name in ALL_TOOLS:
return ALL_TOOLS[tool_name](code, session_id)
code = code.strip().rstrip("<|observation|>").strip()
# Dispatch custom tools
try:
tool_params = json.loads(code)
except json.JSONDecodeError as e:
err = f"Error decoding JSON: {e}"
return [ToolObservation("system_error", err)]
if tool_name not in _TOOL_HOOKS:
err = f"Tool `{tool_name}` not found. Please use a provided tool."
return [ToolObservation("system_error", err)]
tool_hook = _TOOL_HOOKS[tool_name]
try:
ret: str = tool_hook(**tool_params)
return [ToolObservation(tool_name, str(ret))]
except:
err = traceback.format_exc()
return [ToolObservation("system_error", err)]
def get_tools() -> list[dict]:
return copy.deepcopy(_TOOL_DESCRIPTIONS)
# Tool Definitions
@register_tool
def random_number_generator(
seed: Annotated[int, "The random seed used by the generator", True],
range: Annotated[tuple[int, int], "The range of the generated numbers", True],
) -> int:
"""
Generates a random number x, s.t. range[0] <= x < range[1]
"""
if not isinstance(seed, int):
raise TypeError("Seed must be an integer")
if not isinstance(range, tuple):
raise TypeError("Range must be a tuple")
if not isinstance(range[0], int) or not isinstance(range[1], int):
raise TypeError("Range must be a tuple of integers")
import random
return random.Random(seed).randint(*range)
@register_tool
def get_weather(
city_name: Annotated[str, "The name of the city to be queried", True],
) -> str:
"""
Get the current weather for `city_name`
"""
if not isinstance(city_name, str):
raise TypeError("City name must be a string")
key_selection = {
"current_condition": [
"temp_C",
"FeelsLikeC",
"humidity",
"weatherDesc",
"observation_time",
],
}
import requests
try:
resp = requests.get(f"https://wttr.in/{city_name}?format=j1")
resp.raise_for_status()
resp = resp.json()
ret = {k: {_v: resp[k][0][_v] for _v in v} for k, v in key_selection.items()}
except:
import traceback
ret = "Error encountered while fetching weather data!\n" + traceback.format_exc()
return str(ret)
@register_tool
def get_shell(
query: Annotated[str, "The command should run in Linux shell", True],
) -> str:
"""
Use shell to run command
"""
if not isinstance(query, str):
raise TypeError("Command must be a string")
try:
result = subprocess.run(
query,
shell=True,
check=True,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE,
text=True,
)
return result.stdout
except subprocess.CalledProcessError as e:
return e.stderr
if __name__ == "__main__":
# print(dispatch_tool("get_shell", {"query": "pwd"}))
print(get_tools())
================================================
FILE: demo/composite_demo/src/utils.py
================================================
import docx
from langchain_community.document_loaders import PyMuPDFLoader
from pptx import Presentation
def extract_text(path):
return open(path, "r").read()
def extract_pdf(path):
loader = PyMuPDFLoader(path)
data = loader.load()
data = [x.page_content for x in data]
content = "\n\n".join(data)
return content
def extract_docx(path):
doc = docx.Document(path)
data = []
for paragraph in doc.paragraphs:
data.append(paragraph.text)
content = "\n\n".join(data)
return content
def extract_pptx(path):
prs = Presentation(path)
text = ""
for slide in prs.slides:
for shape in slide.shapes:
if hasattr(shape, "text"):
text += shape.text + "\n"
return text
================================================
FILE: demo/intel_device_demo/itrex/README.md
================================================
# 使用 Intel® Extension for Transformers 推理 GLM-4-9B-Chat 模型
本示例介绍如何使用 Intel® Extension for Transformers 推理 GLM-4-9B-Chat 模型。
## 设备和依赖检查
### 相关推理测试数据
**本文档的数据均在以下硬件环境测试,实际运行环境需求和运行占用的显存略有不同，请以实际运行环境为准。**
测试硬件信息:
+ OS: Ubuntu 22.04 (本教程一定需要在Linux环境下执行)
+ Memory: 512GB
+ Python: 3.10.12
+ CPU: Intel(R) Xeon(R) Platinum 8358 CPU / 12th Gen Intel i5-12400
## 安装依赖
在开始推理之前，请你先安装`inference`中的依赖，同时您需要安装本目录下的依赖项：
```shell
pip install -r requirements.txt
```
## 运行模型推理
```shell
python itrex_cli_demo.py
```
如果您是第一次推理，会有一次模型转换权重的过程，转换后的模型权重存放在`runtime_outputs`文件夹下，这大概会消耗`60G`的硬盘空间。
转换完成后，文件夹下有两个文件：
+ ne_chatglm2_f32.bin 52G(如果您不使用FP32进行推理，可以删掉这个文件)
+ ne_chatglm2_q_nf4_bestla_cfp32_sym_sfp32_g32.bin 8.1G
如果您不是第一次推理，则会跳过这个步骤，直接开始对话，推理效果如下：
```shell
Welcome to the CLI chat. Type your messages below.
User: 你好
AVX:1 AVX2:1 AVX512F:1 AVX512BW:1 AVX_VNNI:0 AVX512_VNNI:1 AMX_INT8:0 AMX_BF16:0 AVX512_BF16:0 AVX512_FP16:0
beam_size: 1, do_sample: 1, top_k: 40, top_p: 0.900, continuous_batching: 0, max_request_num: 1, early_stopping: 0, scratch_size_ratio: 1.000
model_file_loader: loading model from runtime_outs/ne_chatglm2_q_nf4_bestla_cfp32_sym_sfp32_g32.bin
Loading the bin file with NE format...
load_ne_hparams 0.hparams.n_vocab = 151552
load_ne_hparams 1.hparams.n_embd = 4096
load_ne_hparams 2.hparams.n_mult = 0
load_ne_hparams 3.hparams.n_head = 32
load_ne_hparams 4.hparams.n_head_kv = 0
load_ne_hparams 5.hparams.n_layer = 40
load_ne_hparams 6.hparams.n_rot = 0
load_ne_hparams 7.hparams.ftype = 0
load_ne_hparams 8.hparams.max_seq_len = 131072
load_ne_hparams 9.hparams.alibi_bias_max = 0.000
load_ne_hparams 10.hparams.clip_qkv = 0.000
load_ne_hparams 11.hparams.par_res = 0
load_ne_hparams 12.hparams.word_embed_proj_dim = 0
load_ne_hparams 13.hparams.do_layer_norm_before = 0
load_ne_hparams 14.hparams.multi_query_group_num = 2
load_ne_hparams 15.hparams.ffn_hidden_size = 13696
load_ne_hparams 16.hparams.inner_hidden_size = 0
load_ne_hparams 17.hparams.n_experts = 0
load_ne_hparams 18.hparams.n_experts_used = 0
load_ne_hparams 19.hparams.n_embd_head_k = 0
load_ne_hparams 20.hparams.norm_eps = 0.000000
load_ne_hparams 21.hparams.freq_base = 5000000.000
load_ne_hparams 22.hparams.freq_scale = 1.000
load_ne_hparams 23.hparams.rope_scaling_factor = 0.000
load_ne_hparams 24.hparams.original_max_position_embeddings = 0
load_ne_hparams 25.hparams.use_yarn = 0
load_ne_vocab 26.vocab.bos_token_id = 1
load_ne_vocab 27.vocab.eos_token_id = 151329
load_ne_vocab 28.vocab.pad_token_id = 151329
load_ne_vocab 29.vocab.sep_token_id = -1
init: hparams.n_vocab = 151552
init: hparams.n_embd = 4096
init: hparams.n_mult = 0
init: hparams.n_head = 32
init: hparams.n_layer = 40
init: hparams.n_rot = 0
init: hparams.ffn_hidden_size = 13696
init: n_parts = 1
load: ctx size = 16528.38 MB
load: layers[0].ffn_fusion = 1
load: scratch0 = 4096.00 MB
load: scratch1 = 2048.00 MB
load: scratch2 = 4096.00 MB
load: mem required = 26768.38 MB (+ memory per state)
.............................................................................................
model_init_from_file: support_bestla_kv = 1
kv_cache_init: run_mha_reordered = 1
model_init_from_file: kv self size = 690.00 MB
Assistant:
你好👋！我是人工智能助手，很高兴为你服务。有什么可以帮助你的吗？
```
================================================
FILE: demo/intel_device_demo/itrex/README_en.md
================================================
# Using Intel® Extension for Transformers to Inference the GLM-4-9B-Chat Model
This example introduces how to use Intel® Extension for Transformers to inference the GLM-4-9B-Chat model.
## Device and Dependency Check
### Relevant Inference Test Data
**The data in this document is tested on the following hardware environment. The actual running environment requirements and memory usage may vary slightly. P
gitextract_0zacjmse/
├── .github/
│ ├── ISSUE_TEMPLATE/
│ │ ├── bug_report.yaml
│ │ └── feature-request.yaml
│ └── PULL_REQUEST_TEMPLATE.md
├── .gitignore
├── .pre-commit-config.yaml
├── LICENSE
├── README.md
├── README_20240605.md
├── README_zh.md
├── README_zh_240605.md
├── demo/
│ ├── composite_demo/
│ │ ├── .gitignore
│ │ ├── README.md
│ │ ├── README_en.md
│ │ ├── browser/
│ │ │ ├── .gitignore
│ │ │ ├── package.json
│ │ │ ├── src/
│ │ │ │ ├── browser.ts
│ │ │ │ ├── config.ts
│ │ │ │ ├── server.ts
│ │ │ │ ├── types.ts
│ │ │ │ └── utils.ts
│ │ │ └── tsconfig.json
│ │ ├── requirements.txt
│ │ └── src/
│ │ ├── client.py
│ │ ├── clients/
│ │ │ ├── hf.py
│ │ │ ├── openai.py
│ │ │ └── vllm.py
│ │ ├── conversation.py
│ │ ├── main.py
│ │ ├── tools/
│ │ │ ├── browser.py
│ │ │ ├── cogview.py
│ │ │ ├── config.py
│ │ │ ├── interface.py
│ │ │ ├── python.py
│ │ │ └── tool_registry.py
│ │ └── utils.py
│ └── intel_device_demo/
│ ├── itrex/
│ │ ├── README.md
│ │ ├── README_en.md
│ │ ├── itrex_cli_demo.py
│ │ └── requirements.txt
│ └── openvino/
│ ├── README.md
│ ├── README_en.md
│ ├── convert.py
│ ├── openvino_cli_demo.py
│ └── requirements.txt
├── finetune/
│ ├── .gitignore
│ ├── README.md
│ ├── README_zh.md
│ ├── configs/
│ │ ├── ds_zero_2.json
│ │ ├── ds_zero_3.json
│ │ ├── lora.yaml
│ │ └── sft.yaml
│ ├── finetune.py
│ ├── finetune_vision.py
│ └── requirements.txt
├── inference/
│ ├── README.md
│ ├── README_zh.md
│ ├── glm4v_api_request.py
│ ├── glm4v_server.py
│ ├── requirements.txt
│ ├── trans_batch_demo.py
│ ├── trans_cli_demo.py
│ ├── trans_cli_vision_demo.py
│ ├── trans_stress_test.py
│ ├── trans_web_demo.py
│ ├── trans_web_vision_demo.py
│ ├── vllm_cli_demo.py
│ └── vllm_cli_vision_demo.py
├── pyproject.toml
└── resources/
└── WECHAT.md
SYMBOL INDEX (187 symbols across 29 files)
FILE: demo/composite_demo/browser/src/browser.ts
type Quote (line 9) | interface Quote {
type ActionResult (line 14) | interface ActionResult {
type Metadata (line 23) | interface Metadata {
type TetherQuoteExtra (line 30) | interface TetherQuoteExtra {
type TetherQuoteMetadata (line 35) | interface TetherQuoteMetadata {
type Citation (line 44) | interface Citation {
type PageState (line 52) | interface PageState {
type BrowserState (line 66) | interface BrowserState {
function removeDenseLinks (line 72) | function removeDenseLinks(document: Document, ratioThreshold: number = 0...
method transform (line 116) | private transform(dom: JSDOM): string {
method formatPage (line 125) | private formatPage(state: PageState): string {
method newPageState (line 133) | private newPageState(): PageState {
method pushPageState (line 149) | private pushPageState(): PageState {
method lastPageState (line 155) | private lastPageState(): PageState {
method formatErrorUrl (line 162) | private formatErrorUrl(url: string): string {
method constructor (line 477) | constructor() {
method action (line 574) | async action(content: string): Promise<ToolObservation[]> {
method postProcess (line 616) | postProcess(message: Message, metadata: any) {
method getState (line 676) | getState() {
class SimpleBrowser (line 681) | class SimpleBrowser extends BaseBrowser {
method constructor (line 684) | constructor() {
method actionLine (line 688) | async actionLine(content: string): Promise<ActionResult | ActionResult...
FILE: demo/composite_demo/browser/src/types.ts
type File (line 1) | interface File {
type Metadata (line 7) | interface Metadata {
type Message (line 12) | interface Message {
type ToolObservation (line 19) | interface ToolObservation {
FILE: demo/composite_demo/browser/src/utils.ts
class TimeoutError (line 5) | class TimeoutError extends Error {}
FILE: demo/composite_demo/src/client.py
class ClientType (line 20) | class ClientType(Enum):
class Client (line 26) | class Client(Protocol):
method __init__ (line 27) | def __init__(self, model_path: str): ...
method generate_stream (line 29) | def generate_stream(
function process_input (line 37) | def process_input(history: list[dict], tools: list[dict], role_name_repl...
function process_response (line 60) | def process_response(output, history):
function get_client (line 84) | def get_client(model_path, typ: ClientType) -> Client:
FILE: demo/composite_demo/src/clients/hf.py
class HFClient (line 14) | class HFClient(Client):
method __init__ (line 15) | def __init__(self, model_path: str):
method generate_stream (line 26) | def generate_stream(
FILE: demo/composite_demo/src/clients/openai.py
function format_openai_tool (line 12) | def format_openai_tool(origin_tools):
class APIClient (line 37) | class APIClient(Client):
method __init__ (line 38) | def __init__(self, model_path: str):
method generate_stream (line 44) | def generate_stream(
FILE: demo/composite_demo/src/clients/vllm.py
class VLLMClient (line 17) | class VLLMClient(Client):
method __init__ (line 18) | def __init__(self, model_path: str):
method generate_stream (line 30) | def generate_stream(
FILE: demo/composite_demo/src/conversation.py
function build_system_prompt (line 26) | def build_system_prompt(
function response_to_str (line 45) | def response_to_str(response: str | dict[str, str]) -> str:
class Role (line 54) | class Role(Enum):
method __str__ (line 61) | def __str__(self):
method get_message (line 73) | def get_message(self):
class Conversation (line 93) | class Conversation:
method __str__ (line 101) | def __str__(self) -> str:
method get_text (line 106) | def get_text(self) -> str:
method show (line 116) | def show(self, placeholder: DeltaGenerator | None = None) -> str:
function postprocess_text (line 145) | def postprocess_text(text: str, replace_quote: bool) -> str:
FILE: demo/composite_demo/src/main.py
class Mode (line 36) | class Mode(str, Enum):
function append_conversation (line 42) | def append_conversation(
function build_client (line 101) | def build_client(mode: Mode) -> Client:
function page_changed (line 119) | def page_changed() -> None:
function main (line 221) | def main(prompt_text: str):
FILE: demo/composite_demo/src/tools/browser.py
class Quote (line 24) | class Quote:
function map_response (line 36) | def map_response(response: dict) -> ToolObservation:
function tool_call (line 59) | def tool_call(code: str, session_id: str) -> list[ToolObservation]:
FILE: demo/composite_demo/src/tools/cogview.py
function get_zhipu_client (line 10) | def get_zhipu_client():
function map_response (line 14) | def map_response(img: GeneratedImage):
function tool_call (line 23) | def tool_call(prompt: str, session_id: str) -> list[ToolObservation]:
FILE: demo/composite_demo/src/tools/interface.py
class ToolObservation (line 6) | class ToolObservation:
FILE: demo/composite_demo/src/tools/python.py
class CodeKernel (line 18) | class CodeKernel:
method __init__ (line 19) | def __init__(
method execute (line 63) | def execute(self, code):
method execute_interactive (line 83) | def execute_interactive(self, code, verbose=False):
method inspect (line 92) | def inspect(self, code, verbose=False):
method get_error_msg (line 102) | def get_error_msg(self, msg, verbose=False) -> str | None:
method check_msg (line 116) | def check_msg(self, msg, verbose=False):
method shutdown (line 126) | def shutdown(self):
method restart (line 134) | def restart(self):
method interrupt (line 139) | def interrupt(self):
method is_alive (line 144) | def is_alive(self):
function clean_ansi_codes (line 148) | def clean_ansi_codes(input_string):
function extract_code (line 152) | def extract_code(text: str) -> str:
function execute (line 157) | def execute(code: str, kernel: CodeKernel) -> tuple[Literal["text", "ima...
function get_kernel (line 189) | def get_kernel() -> CodeKernel:
function tool_call (line 193) | def tool_call(code: str, session_id: str) -> list[ToolObservation]:
FILE: demo/composite_demo/src/tools/tool_registry.py
function register_tool (line 32) | def register_tool(func: Callable):
function dispatch_tool (line 71) | def dispatch_tool(tool_name: str, code: str, session_id: str) -> list[To...
function get_tools (line 98) | def get_tools() -> list[dict]:
function random_number_generator (line 106) | def random_number_generator(
function get_weather (line 126) | def get_weather(
function get_shell (line 161) | def get_shell(
FILE: demo/composite_demo/src/utils.py
function extract_text (line 6) | def extract_text(path):
function extract_pdf (line 10) | def extract_pdf(path):
function extract_docx (line 18) | def extract_docx(path):
function extract_pptx (line 27) | def extract_pptx(path):
FILE: demo/intel_device_demo/itrex/itrex_cli_demo.py
class StopOnTokens (line 17) | class StopOnTokens(StoppingCriteria):
method __call__ (line 18) | def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTen...
function initialize_model_and_tokenizer (line 26) | def initialize_model_and_tokenizer():
function get_user_input (line 37) | def get_user_input():
function main (line 41) | def main():
FILE: demo/intel_device_demo/openvino/openvino_cli_demo.py
class StopOnTokens (line 10) | class StopOnTokens(StoppingCriteria):
method __init__ (line 11) | def __init__(self, token_ids):
method __call__ (line 14) | def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTen...
function convert_history_to_token (line 50) | def convert_history_to_token(history: List[Tuple[str, str]]):
FILE: finetune/finetune.py
class DataCollatorForSeq2Seq (line 38) | class DataCollatorForSeq2Seq(_DataCollatorForSeq2Seq):
method __call__ (line 39) | def __call__(self, features, return_tensors=None):
class Seq2SeqTrainer (line 58) | class Seq2SeqTrainer(_Seq2SeqTrainer):
method prediction_step (line 59) | def prediction_step(
class DataConfig (line 87) | class DataConfig(object):
method data_format (line 94) | def data_format(self) -> str:
method data_files (line 98) | def data_files(self) -> dict[NamedSplit, str]:
class FinetuningConfig (line 110) | class FinetuningConfig(object):
method __post_init__ (line 124) | def __post_init__(self):
method from_dict (line 139) | def from_dict(cls, **kwargs) -> "FinetuningConfig":
method from_file (line 157) | def from_file(cls, path: Union[str, Path]) -> "FinetuningConfig":
function _load_datasets (line 166) | def _load_datasets(
class DataManager (line 184) | class DataManager(object):
method __init__ (line 185) | def __init__(self, data_dir: str, data_config: DataConfig):
method _get_dataset (line 195) | def _get_dataset(self, split: NamedSplit) -> Optional[Dataset]:
method get_dataset (line 198) | def get_dataset(
function process_message (line 221) | def process_message(message):
function process_batch (line 231) | def process_batch(
function process_batch_eval (line 277) | def process_batch_eval(
function load_tokenizer_and_model (line 324) | def load_tokenizer_and_model(
function compute_metrics (line 346) | def compute_metrics(eval_preds: EvalPrediction, tokenizer):
function main (line 371) | def main(
FILE: finetune/finetune_vision.py
class DataCollatorForSeq2Seq (line 40) | class DataCollatorForSeq2Seq(_DataCollatorForSeq2Seq):
method __call__ (line 41) | def __call__(self, features, return_tensors=None):
class Seq2SeqTrainer (line 60) | class Seq2SeqTrainer(_Seq2SeqTrainer):
method prediction_step (line 61) | def prediction_step(
class DataConfig (line 94) | class DataConfig(object):
method data_format (line 101) | def data_format(self) -> str:
method data_files (line 105) | def data_files(self) -> dict[NamedSplit, str]:
class FinetuningConfig (line 117) | class FinetuningConfig(object):
method __post_init__ (line 131) | def __post_init__(self):
method from_dict (line 146) | def from_dict(cls, **kwargs) -> "FinetuningConfig":
method from_file (line 164) | def from_file(cls, path: Union[str, Path]) -> "FinetuningConfig":
function _load_datasets (line 173) | def _load_datasets(
class DataManager (line 191) | class DataManager(object):
method __init__ (line 192) | def __init__(self, data_dir: str, data_config: DataConfig):
method _get_dataset (line 202) | def _get_dataset(self, split: NamedSplit) -> Optional[Dataset]:
method get_dataset (line 205) | def get_dataset(
function process_batch (line 231) | def process_batch(
function process_batch_eval (line 317) | def process_batch_eval(
function load_tokenizer_and_model (line 388) | def load_tokenizer_and_model(
function compute_metrics (line 412) | def compute_metrics(eval_preds: EvalPrediction, tokenizer):
function main (line 437) | def main(
FILE: inference/glm4v_api_request.py
function create_chat_completion (line 18) | def create_chat_completion(messages, use_stream=False):
function encode_image (line 38) | def encode_image(image_path):
function glm4v_simple_image_chat (line 52) | def glm4v_simple_image_chat(use_stream=False, img_path=None):
FILE: inference/glm4v_server.py
function lifespan (line 29) | async def lifespan(app: FastAPI):
class ModelCard (line 51) | class ModelCard(BaseModel):
class ModelList (line 66) | class ModelList(BaseModel):
class ImageUrl (line 71) | class ImageUrl(BaseModel):
class TextContent (line 75) | class TextContent(BaseModel):
class ImageUrlContent (line 80) | class ImageUrlContent(BaseModel):
class ChatMessageInput (line 88) | class ChatMessageInput(BaseModel):
class ChatMessageResponse (line 94) | class ChatMessageResponse(BaseModel):
class DeltaMessage (line 100) | class DeltaMessage(BaseModel):
class ChatCompletionRequest (line 105) | class ChatCompletionRequest(BaseModel):
class ChatCompletionResponseChoice (line 116) | class ChatCompletionResponseChoice(BaseModel):
class ChatCompletionResponseStreamChoice (line 121) | class ChatCompletionResponseStreamChoice(BaseModel):
class UsageInfo (line 126) | class UsageInfo(BaseModel):
class ChatCompletionResponse (line 132) | class ChatCompletionResponse(BaseModel):
function list_models (line 141) | async def list_models():
function create_chat_completion (line 151) | async def create_chat_completion(request: ChatCompletionRequest):
function predict (line 188) | def predict(model_id: str, params: dict):
function generate_glm4v (line 210) | def generate_glm4v(model: AutoModel, tokenizer: AutoTokenizer, params: d...
function process_history_and_images (line 223) | def process_history_and_images(
function generate_stream_glm4v (line 283) | def generate_stream_glm4v(model: AutoModel, tokenizer: AutoTokenizer, pa...
FILE: inference/trans_batch_demo.py
function process_model_outputs (line 20) | def process_model_outputs(inputs, outputs, tokenizer):
function batch (line 28) | def batch(
FILE: inference/trans_cli_demo.py
class StopOnTokens (line 39) | class StopOnTokens(StoppingCriteria):
method __call__ (line 40) | def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTen...
FILE: inference/trans_cli_vision_demo.py
class StopOnTokens (line 48) | class StopOnTokens(StoppingCriteria):
method __call__ (line 49) | def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTen...
FILE: inference/trans_stress_test.py
function stress_test (line 13) | def stress_test(run_name, input_token_len, n, output_token_len, swanlab_...
FILE: inference/trans_web_demo.py
function preprocess_messages (line 12) | def preprocess_messages(history, system_prompt):
function predict (line 30) | def predict(history, system_prompt, max_length, top_p, top_k, temperature):
function main (line 58) | def main():
FILE: inference/trans_web_vision_demo.py
class StopOnTokens (line 32) | class StopOnTokens(StoppingCriteria):
method __call__ (line 33) | def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTen...
function get_image (line 41) | def get_image(image_path=None, image_url=None):
function chatbot (line 50) | def chatbot(image_path=None, image_url=None, assistant_prompt=""):
FILE: inference/vllm_cli_demo.py
function load_model_and_tokenizer (line 26) | def load_model_and_tokenizer(model_dir: str, enable_lora: bool):
function vllm_gen (line 50) | async def vllm_gen(
function chat (line 85) | async def chat():
FILE: inference/vllm_cli_vision_demo.py
function load_model_and_tokenizer (line 24) | def load_model_and_tokenizer(model_dir: str):
function vllm_gen (line 41) | async def vllm_gen(messages: List[Dict[str, str]], top_p: float, tempera...
function chat (line 59) | async def chat():
Condensed preview — 69 files, each showing path, character count, and a content snippet. Download the .json file or copy for the full structured content (353K chars).
[
{
"path": ".github/ISSUE_TEMPLATE/bug_report.yaml",
"chars": 2941,
"preview": "name: \"\\U0001F41B Bug Report\"\ndescription: Submit a bug report to help us improve GLM-4-9B / 提交一个 Bug 问题报告来帮助我们改进 GLM-4-"
},
{
"path": ".github/ISSUE_TEMPLATE/feature-request.yaml",
"chars": 1001,
"preview": "name: \"\\U0001F680 Feature request\"\ndescription: Submit a request for a new GLM-4-9B feature / 提交一个新的 GLM-4-9B 的功能建议\nlabe"
},
{
"path": ".github/PULL_REQUEST_TEMPLATE.md",
"chars": 1370,
"preview": "# Contribution Guide\n\nWe welcome your contributions to this repository. To ensure elegant code style and better code qua"
},
{
"path": ".gitignore",
"chars": 39,
"preview": "*venv\n*.DS_Store\n*.idea/\ndataset\ntest*\n"
},
{
"path": ".pre-commit-config.yaml",
"chars": 525,
"preview": "repos:\n - repo: https://github.com/astral-sh/ruff-pre-commit\n rev: v0.4.5\n hooks:\n - id: ruff\n args: "
},
{
"path": "LICENSE",
"chars": 11338,
"preview": " Apache License\n Version 2.0, January 2004\n "
},
{
"path": "README.md",
"chars": 23767,
"preview": "# GLM-4-0414 Model Series\n\n<p align=\"center\">\n👋 Join our <a href=\"https://discord.gg/8cnQKdAprg\" target=\"_blank\">Discord"
},
{
"path": "README_20240605.md",
"chars": 24202,
"preview": "# GLM-4\n\n<p align=\"center\">\n 📄<a href=\"https://arxiv.org/pdf/2406.12793\" target=\"_blank\"> Report </a> • 🤗 <a href=\"https"
},
{
"path": "README_zh.md",
"chars": 18009,
"preview": "# GLM-4-0414 系列模型\n\n<p align=\"center\">\n👋 加入我们的 <a href=\"https://discord.gg/8cnQKdAprg\" target=\"_blank\">Discord</a>, <a hr"
},
{
"path": "README_zh_240605.md",
"chars": 20757,
"preview": "# GLM-4\n\n<p align=\"center\">\n 📄<a href=\"https://arxiv.org/pdf/2406.12793\" target=\"_blank\"> Report </a> • 🤗 <a href=\"https"
},
{
"path": "demo/composite_demo/.gitignore",
"chars": 3527,
"preview": "*venv\n*.DS_Store\n*model\n*.idea/\n\n# Created by https://www.toptal.com/developers/gitignore/api/python\n# Edit at https://w"
},
{
"path": "demo/composite_demo/README.md",
"chars": 4348,
"preview": "# GLM-4-9B Web Demo\n\nRead this in [English](README_en.md)\n\n\n\n## 安装\n\n我们建议通过 [Conda](https"
},
{
"path": "demo/composite_demo/README_en.md",
"chars": 5336,
"preview": "# GLM-4-9B Web Demo\n\n\n\n## Installation\n\nWe recommend using [Conda](https://docs.conda.io"
},
{
"path": "demo/composite_demo/browser/.gitignore",
"chars": 2392,
"preview": "# Created by https://www.toptal.com/developers/gitignore/api/node\n# Edit at https://www.toptal.com/developers/gitignore?"
},
{
"path": "demo/composite_demo/browser/package.json",
"chars": 597,
"preview": "{\n \"name\": \"glm4-browser\",\n \"version\": \"1.0.0\",\n \"description\": \"Browser system for GLM-4\",\n \"main\": \"src/server.ts\""
},
{
"path": "demo/composite_demo/browser/src/browser.ts",
"chars": 24124,
"preview": "import { JSDOM } from 'jsdom';\nimport TurndownService from 'turndown';\n\nimport config from './config';\nimport { Message,"
},
{
"path": "demo/composite_demo/browser/src/config.ts",
"chars": 318,
"preview": "export default {\n LOG_LEVEL: 'debug',\n BROWSER_TIMEOUT: 10000,\n BING_SEARCH_API_URL: 'https://api.bing.microsof"
},
{
"path": "demo/composite_demo/browser/src/server.ts",
"chars": 1162,
"preview": "import express, { Express, Request, Response } from 'express';\n\nimport { SimpleBrowser } from './browser';\nimport config"
},
{
"path": "demo/composite_demo/browser/src/types.ts",
"chars": 506,
"preview": "export interface File {\n id: string;\n name: string;\n size: number;\n}\n\nexport interface Metadata {\n files?: File[];\n "
},
{
"path": "demo/composite_demo/browser/src/utils.ts",
"chars": 1372,
"preview": "import winston from 'winston';\n\nimport config from './config';\n\nexport class TimeoutError extends Error {}\n\nconst logLev"
},
{
"path": "demo/composite_demo/browser/tsconfig.json",
"chars": 297,
"preview": "{\n \"compilerOptions\": {\n \"target\": \"es2022\",\n \"lib\": [\"es2022\", \"dom\"],\n \"module\": \"commonjs\",\n \"rootDir\": "
},
{
"path": "demo/composite_demo/requirements.txt",
"chars": 297,
"preview": "# Please install the requirments.txt in inference first!\n\nipykernel>=6.26.0\nipython>=8.18.1\njupyter_client>=8.6.0\nlangch"
},
{
"path": "demo/composite_demo/src/client.py",
"chars": 3394,
"preview": "\"\"\"\n\nThis is a client part of composite_demo.\nWe provide two clients, HFClient and VLLMClient, which are used to interac"
},
{
"path": "demo/composite_demo/src/clients/hf.py",
"chars": 1748,
"preview": "\"\"\"\nHuggingFace client.\n\"\"\"\n\nfrom collections.abc import Generator\nfrom threading import Thread\n\nimport torch\nfrom clien"
},
{
"path": "demo/composite_demo/src/clients/openai.py",
"chars": 2430,
"preview": "\"\"\"\nOpenAI API client.\n\"\"\"\n\nfrom collections.abc import Generator\n\nfrom client import Client, process_input, process_res"
},
{
"path": "demo/composite_demo/src/clients/vllm.py",
"chars": 1943,
"preview": "\"\"\"\nvLLM client.\n\nPlease install [vLLM](https://github.com/vllm-project/vllm) according to its\ninstallation guide before"
},
{
"path": "demo/composite_demo/src/conversation.py",
"chars": 6139,
"preview": "import json\nimport re\nfrom dataclasses import dataclass\nfrom datetime import datetime\nfrom enum import Enum, auto\n\nimpor"
},
{
"path": "demo/composite_demo/src/main.py",
"chars": 11749,
"preview": "\"\"\"\n\nThis demo show the All tools and Long Context chat Capabilities of GLM-4.\nPlease follow the Readme.md to run the de"
},
{
"path": "demo/composite_demo/src/tools/browser.py",
"chars": 1703,
"preview": "\"\"\"\nSimple browser tool.\n\n# Usage\n\nPlease start the backend browser server according to the instructions in the README.\n"
},
{
"path": "demo/composite_demo/src/tools/cogview.py",
"chars": 717,
"preview": "import streamlit as st\nfrom zhipuai import ZhipuAI\nfrom zhipuai.types.image import GeneratedImage\n\nfrom .config import C"
},
{
"path": "demo/composite_demo/src/tools/config.py",
"chars": 118,
"preview": "BROWSER_SERVER_URL = \"http://localhost:3000\"\n\nIPYKERNEL = \"glm-4-demo\"\n\nZHIPU_AI_KEY = \"\"\nCOGVIEW_MODEL = \"cogview-3\"\n"
},
{
"path": "demo/composite_demo/src/tools/interface.py",
"chars": 224,
"preview": "from dataclasses import dataclass\nfrom typing import Any\n\n\n@dataclass\nclass ToolObservation:\n content_type: str\n t"
},
{
"path": "demo/composite_demo/src/tools/python.py",
"chars": 6552,
"preview": "import queue\nimport re\nfrom pprint import pprint\nfrom subprocess import PIPE\nfrom typing import Literal\n\nimport jupyter_"
},
{
"path": "demo/composite_demo/src/tools/tool_registry.py",
"chars": 5338,
"preview": "\"\"\"\nThis code is the tool registration part. By registering the tool, the model can call the tool.\nThis code provides ex"
},
{
"path": "demo/composite_demo/src/utils.py",
"chars": 765,
"preview": "import docx\nfrom langchain_community.document_loaders import PyMuPDFLoader\nfrom pptx import Presentation\n\n\ndef extract_t"
},
{
"path": "demo/intel_device_demo/itrex/README.md",
"chars": 3347,
"preview": "# 使用 Intel® Extension for Transformers 推理 GLM-4-9B-Chat 模型\n\n本示例介绍如何使用 Intel® Extension for Transformers 推理 GLM-4-9B-Chat"
},
{
"path": "demo/intel_device_demo/itrex/README_en.md",
"chars": 4000,
"preview": "\n# Using Intel® Extension for Transformers to Inference the GLM-4-9B-Chat Model\n\nThis example introduces how to use Inte"
},
{
"path": "demo/intel_device_demo/itrex/itrex_cli_demo.py",
"chars": 3003,
"preview": "\"\"\"\nThis script creates a CLI demo with transformers backend for the glm-4-9b model with Intel® Extension for Transforme"
},
{
"path": "demo/intel_device_demo/itrex/requirements.txt",
"chars": 148,
"preview": "cmake>=3.29.5.1\nhuggingface-hub>=0.23.4\ngit+https://github.com/intel/neural-speed.git@main#egg=neural-speed\nintel-extens"
},
{
"path": "demo/intel_device_demo/openvino/README.md",
"chars": 3312,
"preview": "# 使用 OpenVINO 部署 GLM-4-9B-Chat 模型\n\nRead this in [English](README_en.md).\n\n[OpenVINO](https://www.intel.com/content/www/u"
},
{
"path": "demo/intel_device_demo/openvino/README_en.md",
"chars": 3879,
"preview": "# Deploy the GLM-4-9B-Chat model using OpenVINO\n\n[OpenVINO](https://www.intel.com/content/www/us/en/developer/tools/open"
},
{
"path": "demo/intel_device_demo/openvino/convert.py",
"chars": 2378,
"preview": "\"\"\"\nThis script is used to convert the original model to OpenVINO IR format.\nThe Origin Code can check https://github.co"
},
{
"path": "demo/intel_device_demo/openvino/openvino_cli_demo.py",
"chars": 3655,
"preview": "import argparse\nfrom threading import Thread\nfrom typing import List, Tuple\n\nimport torch\nfrom optimum.intel.openvino im"
},
{
"path": "demo/intel_device_demo/openvino/requirements.txt",
"chars": 126,
"preview": "optimum>=1.20.0\noptimum-intel @ git+https://github.com/huggingface/optimum-intel.git@c1ee8ac0864e25e22ea56b5a37a35451531"
},
{
"path": "finetune/.gitignore",
"chars": 8,
"preview": "output/\n"
},
{
"path": "finetune/README.md",
"chars": 13011,
"preview": "# GLM-4-9B Chat Fine-tuning\n\n[中文阅读](README_zh.md)\n\n## Hardware Check\n\nAll fine-tuning tests were performed in the follow"
},
{
"path": "finetune/README_zh.md",
"chars": 9897,
"preview": "# GLM-4-9B Chat 对话模型微调\n\nRead this in [English](README)\n\n## 硬件检查\n\n所有微调测试均在以下环境和硬件下测试:\n\n> OS: Ubuntu 22.04\n>\n> Memory: 512"
},
{
"path": "finetune/configs/ds_zero_2.json",
"chars": 729,
"preview": "{\n \"fp16\": {\n \"enabled\": \"auto\",\n \"loss_scale\": 0,\n \"loss_scale_window\": 1000,\n \"initial_"
},
{
"path": "finetune/configs/ds_zero_3.json",
"chars": 772,
"preview": "{\n \"train_micro_batch_size_per_gpu\": \"auto\",\n \"zero_allow_untested_optimizer\": true,\n \"bf16\": {\n \"enabled\": \"auto\""
},
{
"path": "finetune/configs/lora.yaml",
"chars": 1333,
"preview": "data_config:\n train_file: train.jsonl\n val_file: dev.jsonl\n test_file: dev.jsonl\n num_proc: 1\n\ncombine: True\nfreezeV"
},
{
"path": "finetune/configs/sft.yaml",
"chars": 1107,
"preview": "data_config:\n train_file: train.jsonl\n val_file: dev.jsonl\n test_file: dev.jsonl\n num_proc: 1\n\ncombine: True\nfreezeV"
},
{
"path": "finetune/finetune.py",
"chars": 18224,
"preview": "# -*- coding: utf-8 -*-\nimport dataclasses as dc\nimport functools\nimport os\nfrom collections.abc import Callable, Mappin"
},
{
"path": "finetune/finetune_vision.py",
"chars": 19916,
"preview": "# -*- coding: utf-8 -*-\nimport dataclasses as dc\nimport functools\nimport os\nfrom collections.abc import Callable, Mappin"
},
{
"path": "finetune/requirements.txt",
"chars": 142,
"preview": "jieba>=0.42.1\ndatasets>=2.20.0\npeft>=0.15.1\ndeepspeed>=0.16.5\nnltk==3.8.1\nrouge_chinese==1.0.3\nruamel.yaml>=0.18.6\ntyper"
},
{
"path": "inference/README.md",
"chars": 4971,
"preview": "# Inference\n\n[中文阅读](README_zh.md)\n\nPlease follow the steps in the document strictly to avoid unnecessary errors.\n\n## Dev"
},
{
"path": "inference/README_zh.md",
"chars": 3336,
"preview": "# Inference\n\nRead this in [English](README.md)\n\n请严格按照文档的步骤进行操作，以避免不必要的错误。\n\n## 设备和依赖检查\n\n### 安装依赖\n\n```shell\npip install -r"
},
{
"path": "inference/glm4v_api_request.py",
"chars": 3144,
"preview": "\"\"\"\nThis script creates a OpenAI Request demo for the glm-4v-9b model, just Use OpenAI API to interact with the model.\nF"
},
{
"path": "inference/glm4v_server.py",
"chars": 13199,
"preview": "import base64\nimport gc\nimport sys\nimport threading\nimport time\nfrom contextlib import asynccontextmanager\nfrom io impor"
},
{
"path": "inference/requirements.txt",
"chars": 440,
"preview": "torch>=2.6.0\ntorchvision>=0.21.0\ntransformers>=4.51.3\nsentencepiece>=0.2.0\njinja2>=3.1.4\npydantic>=2.11.1\ntimm>=1.0.15\nt"
},
{
"path": "inference/trans_batch_demo.py",
"chars": 2965,
"preview": "\"\"\"\n\nHere is an example of using batch request GLM-4-0414 Models and glm-4-9b-chat-hf models with the transformers libra"
},
{
"path": "inference/trans_cli_demo.py",
"chars": 3243,
"preview": "\"\"\"\nThis script creates a CLI demo with transformers backend for the glm-4-9b-chat model,\nallowing users to interact wit"
},
{
"path": "inference/trans_cli_vision_demo.py",
"chars": 3721,
"preview": "\"\"\"\nThis script creates a CLI demo with transformers backend for the glm-4v-9b model,\nallowing users to interact with th"
},
{
"path": "inference/trans_stress_test.py",
"chars": 6571,
"preview": "import argparse\nimport datetime\nimport time\nfrom threading import Thread\n\nimport torch\nfrom transformers import AutoMode"
},
{
"path": "inference/trans_web_demo.py",
"chars": 3525,
"preview": "from threading import Thread\n\nimport gradio as gr\nfrom transformers import AutoModelForCausalLM, AutoTokenizer, TextIter"
},
{
"path": "inference/trans_web_vision_demo.py",
"chars": 3543,
"preview": "\"\"\"\nThis script creates a Gradio demo with a Transformers backend for the glm-4v-9b model, allowing users to interact wi"
},
{
"path": "inference/vllm_cli_demo.py",
"chars": 3514,
"preview": "\"\"\"\nThis script creates a CLI demo with vllm backand for the glm-4-9b model,\nallowing users to interact with the model t"
},
{
"path": "inference/vllm_cli_vision_demo.py",
"chars": 3091,
"preview": "\"\"\"\nThis script creates a CLI demo with vllm backand for the glm-4v-9b model,\nallowing users to interact with the model "
},
{
"path": "pyproject.toml",
"chars": 699,
"preview": "[tool.ruff]\nline-length = 119\n\n[tool.ruff.lint]\n# Never enforce `E501` (line length violations).\nignore = [\"C901\", \"E501"
},
{
"path": "resources/WECHAT.md",
"chars": 162,
"preview": "<div align=\"center\">\n<img src=wechat.jpg width=\"60%\"/>\n\n<p> 扫码加入「GLM-4交流群」 </p>\n<p> Scan the QR code to follow to join t"
}
]
About this extraction
This page contains the full source code of the zai-org/GLM-4 GitHub repository, extracted and formatted as plain text for AI agents and large language models (LLMs). The extraction includes 69 files (328.3 KB), approximately 91.5k tokens, and a symbol index with 187 extracted functions, classes, methods, constants, and types. Use this with OpenClaw, Claude, ChatGPT, Cursor, Windsurf, or any other AI tool that accepts text input. You can copy the full output to your clipboard or download it as a .txt file.
Extracted by GitExtract — free GitHub repo to text converter for AI. Built by Nikandr Surkov.