Showing preview only (270K chars total). Download the full file or copy to clipboard to get everything.
Repository: xu-song/bert-as-language-model
Branch: master
Commit: fa449de0fac7
Files: 24
Total size: 259.0 KB
Directory structure:
gitextract_lmgwb86k/
├── .gitignore
├── CONTRIBUTING.md
├── LICENSE
├── README.md
├── __init__.py
├── cases/
│ └── test.zh.md
├── create_pretraining_data.py
├── data/
│ └── lm/
│ ├── result.zh.json
│ ├── test.en.tsv
│ └── test.zh.tsv
├── extract_features.py
├── modeling.py
├── modeling_test.py
├── multilingual.md
├── optimization.py
├── optimization_test.py
├── requirements.txt
├── run_classifier.py
├── run_lm_predict.py
├── run_pretraining.py
├── run_squad.py
├── sample_text.txt
├── tokenization.py
└── tokenization_test.py
================================================
FILE CONTENTS
================================================
================================================
FILE: .gitignore
================================================
# Initially taken from Github's Python gitignore file
.idea
# 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/
*.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
.hypothesis/
.pytest_cache/
# Translations
*.mo
*.pot
# Django stuff:
*.log
local_settings.py
db.sqlite3
# Flask stuff:
instance/
.webassets-cache
# Scrapy stuff:
.scrapy
# Sphinx documentation
docs/_build/
# PyBuilder
target/
# Jupyter Notebook
.ipynb_checkpoints
# IPython
profile_default/
ipython_config.py
# pyenv
.python-version
# celery beat schedule file
celerybeat-schedule
# 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/
================================================
FILE: CONTRIBUTING.md
================================================
# How to Contribute
BERT needs to maintain permanent compatibility with the pre-trained model files,
so we do not plan to make any major changes to this library (other than what was
promised in the README). However, we can accept small patches related to
re-factoring and documentation. To submit contributes, there are just a few
small guidelines you need to follow.
## Contributor License Agreement
Contributions to this project must be accompanied by a Contributor License
Agreement. You (or your employer) retain the copyright to your contribution;
this simply gives us permission to use and redistribute your contributions as
part of the project. Head over to <https://cla.developers.google.com/> to see
your current agreements on file or to sign a new one.
You generally only need to submit a CLA once, so if you've already submitted one
(even if it was for a different project), you probably don't need to do it
again.
## Code reviews
All submissions, including submissions by project members, require review. We
use GitHub pull requests for this purpose. Consult
[GitHub Help](https://help.github.com/articles/about-pull-requests/) for more
information on using pull requests.
## Community Guidelines
This project follows
[Google's Open Source Community Guidelines](https://opensource.google.com/conduct/).
================================================
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 [yyyy] [name of copyright owner]
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
================================================
**[🤗Demo](#demo)** |
**[📖cases-en](#test-case)** |
**[📖cases-zh](cases/test.zh.md)** |
## BERT as Language Model
For a sentence <img src="https://www.zhihu.com/equation?tex=S%20=%20w_1,%20w_2,...,%20w_k" alt="S = w_1, w_2,..., w_k" eeimg="1"> , we have
<img src="https://www.zhihu.com/equation?tex=p(S)%20=%20\prod_{i=1}^{k}%20p(w_i%20|%20context)" alt="p(S) = \prod_{i=1}^{k} p(w_i | context)" eeimg="1">
In traditional language model, such as RNN, <img src="https://www.zhihu.com/equation?tex=context%20=%20w_1,%20...,%20w_{i-1}" alt="context = w_1, ..., w_{i-1}" eeimg="1"> ,
<img src="https://www.zhihu.com/equation?tex=p(S)%20=%20\prod_{i=1}^{k}%20p(w_i%20|%20w_1,%20...,%20w_{i-1})" alt="p(S) = \prod_{i=1}^{k} p(w_i | w_1, ..., w_{i-1})" eeimg="1">
In bidirectional language model, it has larger context, <img src="https://www.zhihu.com/equation?tex=context+%3d+w_1%2c+...%2c+w_%7bi-1%7d%2cw_%7bi%2b1%7d%2c...%2cw_k" alt="context = w_1, ..., w_{i-1},w_{i+1},...,w_k" eeimg="1">.
In this implementation, we simply adopt the following approximation,
<img src="https://www.zhihu.com/equation?tex=p(S)+%5capprox+%5cprod_%7bi%3d1%7d%5e%7bk%7d+p(w_i+%7c+w_1%2c+...%2c+w_%7bi-1%7d%2cw_%7bi%2b1%7d%2c+...%2cw_k)" alt="p(S) \approx \prod_{i=1}^{k} p(w_i | w_1, ..., w_{i-1},w_{i+1}, ...,w_k)" eeimg="1">.
<!--
1. 近似相等
2. 句子越长，单个word预测的概率越大，ppl越大？传统的RNN也有这个问题
-->
<!-- n-gram
n-gram models construct tables of conditional probabilities for the next word,
Under Markov assumption, the context is the all the
-->
### Demo
Try out the [Web Demo](https://huggingface.co/spaces/eson/bert-perplexity) at [](https://huggingface.co/spaces/eson/bert-perplexity)
### test-case
> [more cases: 中文](cases/test.zh.md)
```bash
export BERT_BASE_DIR=model/uncased_L-12_H-768_A-12
export INPUT_FILE=data/lm/test.en.tsv
python run_lm_predict.py \
--input_file=$INPUT_FILE \
--vocab_file=$BERT_BASE_DIR/vocab.txt \
--bert_config_file=$BERT_BASE_DIR/bert_config.json \
--init_checkpoint=$BERT_BASE_DIR/bert_model.ckpt \
--max_seq_length=128 \
--output_dir=/tmp/lm_output/
```
for the following test case
```bash
$ cat data/lm/test.en.tsv
there is a book on the desk
there is a plane on the desk
there is a book in the desk
$ cat /tmp/lm/output/test_result.json
```
output:
```yml
# prob: probability
# ppl: perplexity
[
{
"tokens": [
{
"token": "there",
"prob": 0.9988962411880493
},
{
"token": "is",
"prob": 0.013578361831605434
},
{
"token": "a",
"prob": 0.9420605897903442
},
{
"token": "book",
"prob": 0.07452250272035599
},
{
"token": "on",
"prob": 0.9607976675033569
},
{
"token": "the",
"prob": 0.4983428418636322
},
{
"token": "desk",
"prob": 4.040586190967588e-06
}
],
"ppl": 17.69329728285426
},
{
"tokens": [
{
"token": "there",
"prob": 0.996775209903717
},
{
"token": "is",
"prob": 0.03194097802042961
},
{
"token": "a",
"prob": 0.8877727389335632
},
{
"token": "plane",
"prob": 3.4907534427475184e-05 # low probability
},
{
"token": "on",
"prob": 0.1902322769165039
},
{
"token": "the",
"prob": 0.5981084704399109
},
{
"token": "desk",
"prob": 3.3164762953674654e-06
}
],
"ppl": 59.646456254851806
},
{
"tokens": [
{
"token": "there",
"prob": 0.9969795942306519
},
{
"token": "is",
"prob": 0.03379646688699722
},
{
"token": "a",
"prob": 0.9095568060874939
},
{
"token": "book",
"prob": 0.013939591124653816
},
{
"token": "in",
"prob": 0.000823647016659379 # low probability
},
{
"token": "the",
"prob": 0.5844194293022156
},
{
"token": "desk",
"prob": 3.3361218356731115e-06
}
],
"ppl": 54.65941516205144
}
]
```
================================================
FILE: __init__.py
================================================
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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: cases/test.zh.md
================================================
# 总体评价
**优势**
1. bert给句子打分，摆脱了传统auto regressive的局限，可并行。
1. 得益于双向语言模型的全局感受野，bert给word打分准确度较高
**缺陷**
1. 给每个word打分，都要跑一遍inference，计算量较大，且冗余。有优化的空间
1. 该实现中采用的句子概率是近似概率，不够严谨
另外
1. char-level的语言模型，由于词组内的高概率，会使整个句子ppl普遍偏高。
1. 句子间的相对ppl还靠谱。
> **建议**:
用分词后的中文重新pretrain，然后进行word-level language model predict。
# 中文测试
```bash
export BERT_BASE_DIR=model/chinese_L-12_H-768_A-12
export INPUT_FILE=data/lm/test.zh.tsv
python run_lm_predict.py \
--input_file=$INPUT_FILE \
--vocab_file=$BERT_BASE_DIR/vocab.txt \
--bert_config_file=$BERT_BASE_DIR/bert_config.json \
--init_checkpoint=$BERT_BASE_DIR/bert_model.ckpt \
--max_seq_length=128 \
--output_dir=/tmp/lm_output/
```
以下是部分结果，更多见[result.zh.json](/data/lm/result.zh.json)
```yml
[
{
"tokens": [
{
"token": "2016",
"prob": 0.06563900411128998
},
{
"token": "全",
"prob": 0.4981258511543274
},
{
"token": "国",
"prob": 0.9088247418403625
},
{
"token": "低",
"prob": 1.6259804397122934e-05 # 低概率
},
{
"token": "考",
"prob": 0.4023572504520416
},
...
],
"ppl": 13.400421357093588
},
{
"tokens": [
{
"token": "落",
"prob": 0.1483132392168045
},
{
"token": "霞",
"prob": 0.42232587933540344
},
{
"token": "与",
"prob": 0.8615185022354126
},
{
"token": "孤",
"prob": 0.9975666999816895
},
{
"token": "鹜",
"prob": 0.5613960027694702
},
{
"token": "齐",
"prob": 0.18012434244155884
},
{
"token": "跑",
"prob": 1.3388593288254924e-05 # 低概率
},
...
],
"ppl": 11.983086642867598
},
```
中文测试样例来源于[百度云dnnlm](https://cloud.baidu.com/product/nlp/dnnlm_cn)
<!--
英文model跑中文 - UNK的影响
export BERT_BASE_DIR=model/uncased_L-12_H-768_A-12
export INPUT_FILE=data/lm/test.zh.tsv
python run_lm_predict.py \
--input_file=$INPUT_FILE \
--vocab_file=$BERT_BASE_DIR/vocab.txt \
--bert_config_file=$BERT_BASE_DIR/bert_config.json \
--init_checkpoint=$BERT_BASE_DIR/bert_model.ckpt \
--max_seq_length=128 \
--output_dir=/tmp/lm_output/
UNK太多，没有多大意义。
-->
================================================
FILE: create_pretraining_data.py
================================================
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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.
"""Create masked LM/next sentence masked_lm TF examples for BERT."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import random
import tokenization
import tensorflow as tf
flags = tf.flags
FLAGS = flags.FLAGS
flags.DEFINE_string("input_file", None,
"Input raw text file (or comma-separated list of files).")
flags.DEFINE_string(
"output_file", None,
"Output TF example file (or comma-separated list of files).")
flags.DEFINE_string("vocab_file", None,
"The vocabulary file that the BERT model was trained on.")
flags.DEFINE_bool(
"do_lower_case", True,
"Whether to lower case the input text. Should be True for uncased "
"models and False for cased models.")
flags.DEFINE_integer("max_seq_length", 128, "Maximum sequence length.")
flags.DEFINE_integer("max_predictions_per_seq", 20,
"Maximum number of masked LM predictions per sequence.")
flags.DEFINE_integer("random_seed", 12345, "Random seed for data generation.")
flags.DEFINE_integer(
"dupe_factor", 10,
"Number of times to duplicate the input data (with different masks).")
flags.DEFINE_float("masked_lm_prob", 0.15, "Masked LM probability.")
flags.DEFINE_float(
"short_seq_prob", 0.1,
"Probability of creating sequences which are shorter than the "
"maximum length.")
class TrainingInstance(object):
"""A single training instance (sentence pair)."""
def __init__(self, tokens, segment_ids, masked_lm_positions, masked_lm_labels,
is_random_next):
self.tokens = tokens
self.segment_ids = segment_ids
self.is_random_next = is_random_next
self.masked_lm_positions = masked_lm_positions
self.masked_lm_labels = masked_lm_labels
def __str__(self):
s = ""
s += "tokens: %s\n" % (" ".join(
[tokenization.printable_text(x) for x in self.tokens]))
s += "segment_ids: %s\n" % (" ".join([str(x) for x in self.segment_ids]))
s += "is_random_next: %s\n" % self.is_random_next
s += "masked_lm_positions: %s\n" % (" ".join(
[str(x) for x in self.masked_lm_positions]))
s += "masked_lm_labels: %s\n" % (" ".join(
[tokenization.printable_text(x) for x in self.masked_lm_labels]))
s += "\n"
return s
def __repr__(self):
return self.__str__()
def write_instance_to_example_files(instances, tokenizer, max_seq_length,
max_predictions_per_seq, output_files):
"""Create TF example files from `TrainingInstance`s."""
writers = []
for output_file in output_files:
writers.append(tf.python_io.TFRecordWriter(output_file))
writer_index = 0
total_written = 0
for (inst_index, instance) in enumerate(instances):
input_ids = tokenizer.convert_tokens_to_ids(instance.tokens)
input_mask = [1] * len(input_ids)
segment_ids = list(instance.segment_ids)
assert len(input_ids) <= max_seq_length
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
masked_lm_positions = list(instance.masked_lm_positions)
masked_lm_ids = tokenizer.convert_tokens_to_ids(instance.masked_lm_labels)
masked_lm_weights = [1.0] * len(masked_lm_ids)
while len(masked_lm_positions) < max_predictions_per_seq:
masked_lm_positions.append(0)
masked_lm_ids.append(0)
masked_lm_weights.append(0.0)
next_sentence_label = 1 if instance.is_random_next else 0
features = collections.OrderedDict()
features["input_ids"] = create_int_feature(input_ids)
features["input_mask"] = create_int_feature(input_mask)
features["segment_ids"] = create_int_feature(segment_ids)
features["masked_lm_positions"] = create_int_feature(masked_lm_positions)
features["masked_lm_ids"] = create_int_feature(masked_lm_ids)
features["masked_lm_weights"] = create_float_feature(masked_lm_weights)
features["next_sentence_labels"] = create_int_feature([next_sentence_label])
tf_example = tf.train.Example(features=tf.train.Features(feature=features))
writers[writer_index].write(tf_example.SerializeToString())
writer_index = (writer_index + 1) % len(writers)
total_written += 1
if inst_index < 20:
tf.logging.info("*** Example ***")
tf.logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in instance.tokens]))
for feature_name in features.keys():
feature = features[feature_name]
values = []
if feature.int64_list.value:
values = feature.int64_list.value
elif feature.float_list.value:
values = feature.float_list.value
tf.logging.info(
"%s: %s" % (feature_name, " ".join([str(x) for x in values])))
for writer in writers:
writer.close()
tf.logging.info("Wrote %d total instances", total_written)
def create_int_feature(values):
feature = tf.train.Feature(int64_list=tf.train.Int64List(value=list(values)))
return feature
def create_float_feature(values):
feature = tf.train.Feature(float_list=tf.train.FloatList(value=list(values)))
return feature
def create_training_instances(input_files, tokenizer, max_seq_length,
dupe_factor, short_seq_prob, masked_lm_prob,
max_predictions_per_seq, rng):
"""Create `TrainingInstance`s from raw text."""
all_documents = [[]]
# Input file format:
# (1) One sentence per line. These should ideally be actual sentences, not
# entire paragraphs or arbitrary spans of text. (Because we use the
# sentence boundaries for the "next sentence prediction" task).
# (2) Blank lines between documents. Document boundaries are needed so
# that the "next sentence prediction" task doesn't span between documents.
for input_file in input_files:
with tf.gfile.GFile(input_file, "r") as reader:
while True:
line = tokenization.convert_to_unicode(reader.readline())
if not line:
break
line = line.strip()
# Empty lines are used as document delimiters
if not line:
all_documents.append([])
tokens = tokenizer.tokenize(line)
if tokens:
all_documents[-1].append(tokens)
# Remove empty documents
all_documents = [x for x in all_documents if x]
rng.shuffle(all_documents)
vocab_words = list(tokenizer.vocab.keys())
instances = []
for _ in range(dupe_factor):
for document_index in range(len(all_documents)):
instances.extend(
create_instances_from_document(
all_documents, document_index, max_seq_length, short_seq_prob,
masked_lm_prob, max_predictions_per_seq, vocab_words, rng))
rng.shuffle(instances)
return instances
def create_instances_from_document(
all_documents, document_index, max_seq_length, short_seq_prob,
masked_lm_prob, max_predictions_per_seq, vocab_words, rng):
"""Creates `TrainingInstance`s for a single document."""
document = all_documents[document_index]
# Account for [CLS], [SEP], [SEP]
max_num_tokens = max_seq_length - 3
# We *usually* want to fill up the entire sequence since we are padding
# to `max_seq_length` anyways, so short sequences are generally wasted
# computation. However, we *sometimes*
# (i.e., short_seq_prob == 0.1 == 10% of the time) want to use shorter
# sequences to minimize the mismatch between pre-training and fine-tuning.
# The `target_seq_length` is just a rough target however, whereas
# `max_seq_length` is a hard limit.
target_seq_length = max_num_tokens
if rng.random() < short_seq_prob:
target_seq_length = rng.randint(2, max_num_tokens)
# We DON'T just concatenate all of the tokens from a document into a long
# sequence and choose an arbitrary split point because this would make the
# next sentence prediction task too easy. Instead, we split the input into
# segments "A" and "B" based on the actual "sentences" provided by the user
# input.
instances = []
current_chunk = []
current_length = 0
i = 0
while i < len(document):
segment = document[i]
current_chunk.append(segment)
current_length += len(segment)
if i == len(document) - 1 or current_length >= target_seq_length:
if current_chunk:
# `a_end` is how many segments from `current_chunk` go into the `A`
# (first) sentence.
a_end = 1
if len(current_chunk) >= 2:
a_end = rng.randint(1, len(current_chunk) - 1)
tokens_a = []
for j in range(a_end):
tokens_a.extend(current_chunk[j])
tokens_b = []
# Random next
is_random_next = False
if len(current_chunk) == 1 or rng.random() < 0.5:
is_random_next = True
target_b_length = target_seq_length - len(tokens_a)
# This should rarely go for more than one iteration for large
# corpora. However, just to be careful, we try to make sure that
# the random document is not the same as the document
# we're processing.
for _ in range(10):
random_document_index = rng.randint(0, len(all_documents) - 1)
if random_document_index != document_index:
break
random_document = all_documents[random_document_index]
random_start = rng.randint(0, len(random_document) - 1)
for j in range(random_start, len(random_document)):
tokens_b.extend(random_document[j])
if len(tokens_b) >= target_b_length:
break
# We didn't actually use these segments so we "put them back" so
# they don't go to waste.
num_unused_segments = len(current_chunk) - a_end
i -= num_unused_segments
# Actual next
else:
is_random_next = False
for j in range(a_end, len(current_chunk)):
tokens_b.extend(current_chunk[j])
truncate_seq_pair(tokens_a, tokens_b, max_num_tokens, rng)
assert len(tokens_a) >= 1
assert len(tokens_b) >= 1
tokens = []
segment_ids = []
tokens.append("[CLS]")
segment_ids.append(0)
for token in tokens_a:
tokens.append(token)
segment_ids.append(0)
tokens.append("[SEP]")
segment_ids.append(0)
for token in tokens_b:
tokens.append(token)
segment_ids.append(1)
tokens.append("[SEP]")
segment_ids.append(1)
(tokens, masked_lm_positions,
masked_lm_labels) = create_masked_lm_predictions(
tokens, masked_lm_prob, max_predictions_per_seq, vocab_words, rng)
instance = TrainingInstance(
tokens=tokens,
segment_ids=segment_ids,
is_random_next=is_random_next,
masked_lm_positions=masked_lm_positions,
masked_lm_labels=masked_lm_labels)
instances.append(instance)
current_chunk = []
current_length = 0
i += 1
return instances
def create_masked_lm_predictions(tokens, masked_lm_prob,
max_predictions_per_seq, vocab_words, rng):
"""Creates the predictions for the masked LM objective."""
cand_indexes = []
for (i, token) in enumerate(tokens):
if token == "[CLS]" or token == "[SEP]":
continue
cand_indexes.append(i)
rng.shuffle(cand_indexes)
output_tokens = list(tokens)
masked_lm = collections.namedtuple("masked_lm", ["index", "label"]) # pylint: disable=invalid-name
num_to_predict = min(max_predictions_per_seq,
max(1, int(round(len(tokens) * masked_lm_prob))))
masked_lms = []
covered_indexes = set()
for index in cand_indexes:
if len(masked_lms) >= num_to_predict:
break
if index in covered_indexes:
continue
covered_indexes.add(index)
masked_token = None
# 80% of the time, replace with [MASK]
if rng.random() < 0.8:
masked_token = "[MASK]"
else:
# 10% of the time, keep original
if rng.random() < 0.5:
masked_token = tokens[index]
# 10% of the time, replace with random word
else:
masked_token = vocab_words[rng.randint(0, len(vocab_words) - 1)]
output_tokens[index] = masked_token
masked_lms.append(masked_lm(index=index, label=tokens[index]))
masked_lms = sorted(masked_lms, key=lambda x: x.index)
masked_lm_positions = []
masked_lm_labels = []
for p in masked_lms:
masked_lm_positions.append(p.index)
masked_lm_labels.append(p.label)
return (output_tokens, masked_lm_positions, masked_lm_labels)
def truncate_seq_pair(tokens_a, tokens_b, max_num_tokens, rng):
"""Truncates a pair of sequences to a maximum sequence length."""
while True:
total_length = len(tokens_a) + len(tokens_b)
if total_length <= max_num_tokens:
break
trunc_tokens = tokens_a if len(tokens_a) > len(tokens_b) else tokens_b
assert len(trunc_tokens) >= 1
# We want to sometimes truncate from the front and sometimes from the
# back to add more randomness and avoid biases.
if rng.random() < 0.5:
del trunc_tokens[0]
else:
trunc_tokens.pop()
def main(_):
tf.logging.set_verbosity(tf.logging.INFO)
tokenizer = tokenization.FullTokenizer(
vocab_file=FLAGS.vocab_file, do_lower_case=FLAGS.do_lower_case)
input_files = []
for input_pattern in FLAGS.input_file.split(","):
input_files.extend(tf.gfile.Glob(input_pattern))
tf.logging.info("*** Reading from input files ***")
for input_file in input_files:
tf.logging.info(" %s", input_file)
rng = random.Random(FLAGS.random_seed)
instances = create_training_instances(
input_files, tokenizer, FLAGS.max_seq_length, FLAGS.dupe_factor,
FLAGS.short_seq_prob, FLAGS.masked_lm_prob, FLAGS.max_predictions_per_seq,
rng)
output_files = FLAGS.output_file.split(",")
tf.logging.info("*** Writing to output files ***")
for output_file in output_files:
tf.logging.info(" %s", output_file)
write_instance_to_example_files(instances, tokenizer, FLAGS.max_seq_length,
FLAGS.max_predictions_per_seq, output_files)
if __name__ == "__main__":
flags.mark_flag_as_required("input_file")
flags.mark_flag_as_required("output_file")
flags.mark_flag_as_required("vocab_file")
tf.app.run()
================================================
FILE: data/lm/result.zh.json
================================================
[
{
"tokens": [
{
"token": "2016",
"prob": 0.10201895982027054
},
{
"token": "全",
"prob": 0.8663827180862427
},
{
"token": "国",
"prob": 0.9864780902862549
},
{
"token": "高",
"prob": 0.34686800837516785
},
{
"token": "考",
"prob": 0.9513261914253235
},
{
"token": "卷",
"prob": 0.002143713878467679
},
{
"token": "答",
"prob": 0.04595280811190605
},
{
"token": "题",
"prob": 0.052179522812366486
},
{
"token": "模",
"prob": 0.9097260236740112
},
{
"token": "板",
"prob": 0.5003664493560791
}
],
"ppl": 5.214168292180062
},
{
"tokens": [
{
"token": "2016",
"prob": 0.07322586327791214
},
{
"token": "全",
"prob": 0.4594285190105438
},
{
"token": "国",
"prob": 0.9693033695220947
},
{
"token": "大",
"prob": 0.002398423384875059
},
{
"token": "考",
"prob": 0.35024359822273254
},
{
"token": "卷",
"prob": 0.005357644520699978
},
{
"token": "答",
"prob": 0.05084587633609772
},
{
"token": "题",
"prob": 0.08499259501695633
},
{
"token": "模",
"prob": 0.8091200590133667
},
{
"token": "板",
"prob": 0.4416927099227905
}
],
"ppl": 9.214288062044073
},
{
"tokens": [
{
"token": "2016",
"prob": 0.06563900411128998
},
{
"token": "全",
"prob": 0.4981258511543274
},
{
"token": "国",
"prob": 0.9088247418403625
},
{
"token": "低",
"prob": 1.6259804397122934e-05
},
{
"token": "考",
"prob": 0.4023572504520416
},
{
"token": "卷",
"prob": 0.014088313095271587
},
{
"token": "答",
"prob": 0.0639762282371521
},
{
"token": "题",
"prob": 0.09024914354085922
},
{
"token": "模",
"prob": 0.8052080273628235
},
{
"token": "板",
"prob": 0.4206086993217468
}
],
"ppl": 13.400421357093588
},
{
"tokens": [
{
"token": "床",
"prob": 0.06141529977321625
},
{
"token": "前",
"prob": 0.6264788508415222
},
{
"token": "明",
"prob": 0.19100141525268555
},
{
"token": "月",
"prob": 0.8327909111976624
},
{
"token": "光",
"prob": 0.0027572319377213717
},
{
"token": "，",
"prob": 0.9875602126121521
},
{
"token": "疑",
"prob": 0.26570719480514526
},
{
"token": "是",
"prob": 0.20073072612285614
},
{
"token": "地",
"prob": 0.0013851922703906894
},
{
"token": "上",
"prob": 0.08247583359479904
},
{
"token": "霜",
"prob": 1.1996443838313553e-09
}
],
"ppl": 52.41817724710573
},
{
"tokens": [
{
"token": "床",
"prob": 0.007852567359805107
},
{
"token": "前",
"prob": 0.023406021296977997
},
{
"token": "星",
"prob": 0.0016379707958549261
},
{
"token": "星",
"prob": 0.08062248677015305
},
{
"token": "光",
"prob": 0.003109232522547245
},
{
"token": "，",
"prob": 0.7677633762359619
},
{
"token": "疑",
"prob": 0.02753453515470028
},
{
"token": "是",
"prob": 0.18585893511772156
},
{
"token": "地",
"prob": 0.006168654654175043
},
{
"token": "上",
"prob": 0.09490019828081131
},
{
"token": "霜",
"prob": 5.2765480873517845e-09
}
],
"ppl": 153.20970629080068
},
{
"tokens": [
{
"token": "床",
"prob": 0.05734571814537048
},
{
"token": "前",
"prob": 0.4436746835708618
},
{
"token": "白",
"prob": 0.0020346769597381353
},
{
"token": "月",
"prob": 0.23601755499839783
},
{
"token": "光",
"prob": 0.05537288263440132
},
{
"token": "，",
"prob": 0.9799799919128418
},
{
"token": "疑",
"prob": 0.22007052600383759
},
{
"token": "是",
"prob": 0.21303308010101318
},
{
"token": "地",
"prob": 0.0022917157039046288
},
{
"token": "上",
"prob": 0.07564827054738998
},
{
"token": "霜",
"prob": 6.6843566237650975e-09
}
],
"ppl": 58.55804131999474
},
{
"tokens": [
{
"token": "落",
"prob": 0.14436914026737213
},
{
"token": "霞",
"prob": 0.4585109353065491
},
{
"token": "与",
"prob": 0.8038789629936218
},
{
"token": "孤",
"prob": 0.9961937665939331
},
{
"token": "鹜",
"prob": 0.7430610060691833
},
{
"token": "齐",
"prob": 0.5758452415466309
},
{
"token": "飞",
"prob": 0.8611886501312256
},
{
"token": "，",
"prob": 0.8926917314529419
},
{
"token": "秋",
"prob": 0.26929906010627747
},
{
"token": "水",
"prob": 0.29808318614959717
},
{
"token": "共",
"prob": 0.16475750505924225
},
{
"token": "长",
"prob": 0.04301689937710762
},
{
"token": "天",
"prob": 0.8850820660591125
},
{
"token": "一",
"prob": 0.675239622592926
},
{
"token": "色",
"prob": 5.158061867405195e-06
}
],
"ppl": 5.023060982907458
},
{
"tokens": [
{
"token": "落",
"prob": 0.1483132392168045
},
{
"token": "霞",
"prob": 0.42232587933540344
},
{
"token": "与",
"prob": 0.8615185022354126
},
{
"token": "孤",
"prob": 0.9975666999816895
},
{
"token": "鹜",
"prob": 0.5613960027694702
},
{
"token": "齐",
"prob": 0.18012434244155884
},
{
"token": "跑",
"prob": 1.3388593288254924e-05
},
{
"token": "，",
"prob": 0.8621458411216736
},
{
"token": "秋",
"prob": 0.24820539355278015
},
{
"token": "水",
"prob": 0.32748720049858093
},
{
"token": "共",
"prob": 0.2348739057779312
},
{
"token": "长",
"prob": 0.040592435747385025
},
{
"token": "天",
"prob": 0.9231186509132385
},
{
"token": "一",
"prob": 0.5295999646186829
},
{
"token": "色",
"prob": 2.7847559067595284e-06
}
],
"ppl": 11.983086642867598
},
{
"tokens": [
{
"token": "落",
"prob": 0.14112578332424164
},
{
"token": "霞",
"prob": 0.5331157445907593
},
{
"token": "与",
"prob": 0.41120263934135437
},
{
"token": "孤",
"prob": 0.9943874478340149
},
{
"token": "鹜",
"prob": 0.6526917219161987
},
{
"token": "双",
"prob": 0.03388489782810211
},
{
"token": "飞",
"prob": 0.4864092767238617
},
{
"token": "，",
"prob": 0.8635039925575256
},
{
"token": "秋",
"prob": 0.27505722641944885
},
{
"token": "水",
"prob": 0.2975866198539734
},
{
"token": "共",
"prob": 0.1109926775097847
},
{
"token": "长",
"prob": 0.05403125286102295
},
{
"token": "天",
"prob": 0.8708866238594055
},
{
"token": "一",
"prob": 0.6170234680175781
},
{
"token": "色",
"prob": 1.1969897968810983e-05
}
],
"ppl": 6.352968507499718
},
{
"tokens": [
{
"token": "众",
"prob": 0.9997541308403015
},
{
"token": "里",
"prob": 0.9997285008430481
},
{
"token": "寻",
"prob": 0.9988683462142944
},
{
"token": "他",
"prob": 0.7197673916816711
},
{
"token": "千",
"prob": 0.9998348951339722
},
{
"token": "百",
"prob": 0.9883707761764526
},
{
"token": "度",
"prob": 0.9988011121749878
},
{
"token": "，",
"prob": 0.9991918206214905
},
{
"token": "蓦",
"prob": 0.9992905855178833
},
{
"token": "然",
"prob": 0.999387264251709
},
{
"token": "回",
"prob": 0.9999867677688599
},
{
"token": "首",
"prob": 0.9995602965354919
},
{
"token": "，",
"prob": 0.9979262351989746
},
{
"token": "那",
"prob": 0.957136869430542
},
{
"token": "人",
"prob": 0.9965313076972961
},
{
"token": "却",
"prob": 0.9842029809951782
},
{
"token": "在",
"prob": 0.9936916828155518
},
{
"token": "，",
"prob": 0.9937483072280884
},
{
"token": "灯",
"prob": 0.9998267292976379
},
{
"token": "火",
"prob": 0.9999244213104248
},
{
"token": "阑",
"prob": 0.9999798536300659
},
{
"token": "珊",
"prob": 0.9999972581863403
},
{
"token": "处",
"prob": 1.918866399108765e-09
}
],
"ppl": 2.4378636592211476
},
{
"tokens": [
{
"token": "众",
"prob": 0.9995400905609131
},
{
"token": "里",
"prob": 0.9989792108535767
},
{
"token": "寻",
"prob": 0.9996756315231323
},
{
"token": "她",
"prob": 0.26533517241477966
},
{
"token": "千",
"prob": 0.9998635053634644
},
{
"token": "百",
"prob": 0.9769723415374756
},
{
"token": "度",
"prob": 0.997245192527771
},
{
"token": "，",
"prob": 0.9979918003082275
},
{
"token": "蓦",
"prob": 0.9991554021835327
},
{
"token": "然",
"prob": 0.9993096590042114
},
{
"token": "回",
"prob": 0.9999856948852539
},
{
"token": "首",
"prob": 0.9995132684707642
},
{
"token": "，",
"prob": 0.9980834722518921
},
{
"token": "那",
"prob": 0.969828724861145
},
{
"token": "人",
"prob": 0.9686450958251953
},
{
"token": "却",
"prob": 0.9804452061653137
},
{
"token": "在",
"prob": 0.9953302145004272
},
{
"token": "，",
"prob": 0.9902727007865906
},
{
"token": "灯",
"prob": 0.999835729598999
},
{
"token": "火",
"prob": 0.9999294281005859
},
{
"token": "阑",
"prob": 0.9999833106994629
},
{
"token": "珊",
"prob": 0.9999969005584717
},
{
"token": "处",
"prob": 2.7301996219364355e-09
}
],
"ppl": 2.511098534866279
},
{
"tokens": [
{
"token": "众",
"prob": 0.9973291158676147
},
{
"token": "里",
"prob": 0.9993185997009277
},
{
"token": "寻",
"prob": 0.9946662187576294
},
{
"token": "ta",
"prob": 9.90712862858345e-08
},
{
"token": "千",
"prob": 0.999840259552002
},
{
"token": "百",
"prob": 0.9573060870170593
},
{
"token": "度",
"prob": 0.9984676241874695
},
{
"token": "，",
"prob": 0.998295247554779
},
{
"token": "蓦",
"prob": 0.9991011619567871
},
{
"token": "然",
"prob": 0.9991719722747803
},
{
"token": "回",
"prob": 0.9999833106994629
},
{
"token": "首",
"prob": 0.999624490737915
},
{
"token": "，",
"prob": 0.9986977577209473
},
{
"token": "那",
"prob": 0.9325712323188782
},
{
"token": "人",
"prob": 0.9955593347549438
},
{
"token": "却",
"prob": 0.9584773182868958
},
{
"token": "在",
"prob": 0.9958037734031677
},
{
"token": "，",
"prob": 0.9910857081413269
},
{
"token": "灯",
"prob": 0.9998859167098999
},
{
"token": "火",
"prob": 0.9999510049819946
},
{
"token": "阑",
"prob": 0.9999864101409912
},
{
"token": "珊",
"prob": 0.9999982118606567
},
{
"token": "处",
"prob": 1.1214358330846608e-09
}
],
"ppl": 4.980078099549084
}
================================================
FILE: data/lm/test.en.tsv
================================================
there is a book on the desk
there is a plane on the desk
there is a book in the desk
================================================
FILE: data/lm/test.zh.tsv
================================================
2016全国高考卷答题模板
2016全国大考卷答题模板
2016全国低考卷答题模板
床前明月光，疑是地上霜
床前星星光，疑是地上霜
床前白月光，疑是地上霜
落霞与孤鹜齐飞，秋水共长天一色
落霞与孤鹜齐跑，秋水共长天一色
落霞与孤鹜双飞，秋水共长天一色
众里寻他千百度，蓦然回首，那人却在，灯火阑珊处
众里寻她千百度，蓦然回首，那人却在，灯火阑珊处
众里寻ta千百度，蓦然回首，那人却在，灯火阑珊处
================================================
FILE: extract_features.py
================================================
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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.
"""Extract pre-computed feature vectors from BERT."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import codecs
import collections
import json
import re
import modeling
import tokenization
import tensorflow as tf
flags = tf.flags
FLAGS = flags.FLAGS
flags.DEFINE_string("input_file", None, "")
flags.DEFINE_string("output_file", None, "")
flags.DEFINE_string("layers", "-1,-2,-3,-4", "")
flags.DEFINE_string(
"bert_config_file", None,
"The config json file corresponding to the pre-trained BERT model. "
"This specifies the model architecture.")
flags.DEFINE_integer(
"max_seq_length", 128,
"The maximum total input sequence length after WordPiece tokenization. "
"Sequences longer than this will be truncated, and sequences shorter "
"than this will be padded.")
flags.DEFINE_string(
"init_checkpoint", None,
"Initial checkpoint (usually from a pre-trained BERT model).")
flags.DEFINE_string("vocab_file", None,
"The vocabulary file that the BERT model was trained on.")
flags.DEFINE_bool(
"do_lower_case", True,
"Whether to lower case the input text. Should be True for uncased "
"models and False for cased models.")
flags.DEFINE_integer("batch_size", 32, "Batch size for predictions.")
flags.DEFINE_bool("use_tpu", False, "Whether to use TPU or GPU/CPU.")
flags.DEFINE_string("master", None,
"If using a TPU, the address of the master.")
flags.DEFINE_integer(
"num_tpu_cores", 8,
"Only used if `use_tpu` is True. Total number of TPU cores to use.")
flags.DEFINE_bool(
"use_one_hot_embeddings", False,
"If True, tf.one_hot will be used for embedding lookups, otherwise "
"tf.nn.embedding_lookup will be used. On TPUs, this should be True "
"since it is much faster.")
class InputExample(object):
def __init__(self, unique_id, text_a, text_b):
self.unique_id = unique_id
self.text_a = text_a
self.text_b = text_b
class InputFeatures(object):
"""A single set of features of data."""
def __init__(self, unique_id, tokens, input_ids, input_mask, input_type_ids):
self.unique_id = unique_id
self.tokens = tokens
self.input_ids = input_ids
self.input_mask = input_mask
self.input_type_ids = input_type_ids
def input_fn_builder(features, seq_length):
"""Creates an `input_fn` closure to be passed to TPUEstimator."""
all_unique_ids = []
all_input_ids = []
all_input_mask = []
all_input_type_ids = []
for feature in features:
all_unique_ids.append(feature.unique_id)
all_input_ids.append(feature.input_ids)
all_input_mask.append(feature.input_mask)
all_input_type_ids.append(feature.input_type_ids)
def input_fn(params):
"""The actual input function."""
batch_size = params["batch_size"]
num_examples = len(features)
# This is for demo purposes and does NOT scale to large data sets. We do
# not use Dataset.from_generator() because that uses tf.py_func which is
# not TPU compatible. The right way to load data is with TFRecordReader.
d = tf.data.Dataset.from_tensor_slices({
"unique_ids":
tf.constant(all_unique_ids, shape=[num_examples], dtype=tf.int32),
"input_ids":
tf.constant(
all_input_ids, shape=[num_examples, seq_length],
dtype=tf.int32),
"input_mask":
tf.constant(
all_input_mask,
shape=[num_examples, seq_length],
dtype=tf.int32),
"input_type_ids":
tf.constant(
all_input_type_ids,
shape=[num_examples, seq_length],
dtype=tf.int32),
})
d = d.batch(batch_size=batch_size, drop_remainder=False)
return d
return input_fn
def model_fn_builder(bert_config, init_checkpoint, layer_indexes, use_tpu,
use_one_hot_embeddings):
"""Returns `model_fn` closure for TPUEstimator."""
def model_fn(features, labels, mode, params): # pylint: disable=unused-argument
"""The `model_fn` for TPUEstimator."""
unique_ids = features["unique_ids"]
input_ids = features["input_ids"]
input_mask = features["input_mask"]
input_type_ids = features["input_type_ids"]
model = modeling.BertModel(
config=bert_config,
is_training=False,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=input_type_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
if mode != tf.estimator.ModeKeys.PREDICT:
raise ValueError("Only PREDICT modes are supported: %s" % (mode))
tvars = tf.trainable_variables()
scaffold_fn = None
(assignment_map,
initialized_variable_names) = modeling.get_assignment_map_from_checkpoint(
tvars, init_checkpoint)
if use_tpu:
def tpu_scaffold():
tf.train.init_from_checkpoint(init_checkpoint, assignment_map)
return tf.train.Scaffold()
scaffold_fn = tpu_scaffold
else:
tf.train.init_from_checkpoint(init_checkpoint, assignment_map)
tf.logging.info("**** Trainable Variables ****")
for var in tvars:
init_string = ""
if var.name in initialized_variable_names:
init_string = ", *INIT_FROM_CKPT*"
tf.logging.info(" name = %s, shape = %s%s", var.name, var.shape,
init_string)
all_layers = model.get_all_encoder_layers()
predictions = {
"unique_id": unique_ids,
}
for (i, layer_index) in enumerate(layer_indexes):
predictions["layer_output_%d" % i] = all_layers[layer_index]
output_spec = tf.contrib.tpu.TPUEstimatorSpec(
mode=mode, predictions=predictions, scaffold_fn=scaffold_fn)
return output_spec
return model_fn
def convert_examples_to_features(examples, seq_length, tokenizer):
"""Loads a data file into a list of `InputBatch`s."""
features = []
for (ex_index, example) in enumerate(examples):
tokens_a = tokenizer.tokenize(example.text_a)
tokens_b = None
if example.text_b:
tokens_b = tokenizer.tokenize(example.text_b)
if tokens_b:
# Modifies `tokens_a` and `tokens_b` in place so that the total
# length is less than the specified length.
# Account for [CLS], [SEP], [SEP] with "- 3"
_truncate_seq_pair(tokens_a, tokens_b, seq_length - 3)
else:
# Account for [CLS] and [SEP] with "- 2"
if len(tokens_a) > seq_length - 2:
tokens_a = tokens_a[0:(seq_length - 2)]
# The convention in BERT is:
# (a) For sequence pairs:
# tokens: [CLS] is this jack ##son ##ville ? [SEP] no it is not . [SEP]
# type_ids: 0 0 0 0 0 0 0 0 1 1 1 1 1 1
# (b) For single sequences:
# tokens: [CLS] the dog is hairy . [SEP]
# type_ids: 0 0 0 0 0 0 0
#
# Where "type_ids" are used to indicate whether this is the first
# sequence or the second sequence. The embedding vectors for `type=0` and
# `type=1` were learned during pre-training and are added to the wordpiece
# embedding vector (and position vector). This is not *strictly* necessary
# since the [SEP] token unambiguously separates the sequences, but it makes
# it easier for the model to learn the concept of sequences.
#
# For classification tasks, the first vector (corresponding to [CLS]) is
# used as as the "sentence vector". Note that this only makes sense because
# the entire model is fine-tuned.
tokens = []
input_type_ids = []
tokens.append("[CLS]")
input_type_ids.append(0)
for token in tokens_a:
tokens.append(token)
input_type_ids.append(0)
tokens.append("[SEP]")
input_type_ids.append(0)
if tokens_b:
for token in tokens_b:
tokens.append(token)
input_type_ids.append(1)
tokens.append("[SEP]")
input_type_ids.append(1)
input_ids = tokenizer.convert_tokens_to_ids(tokens)
# The mask has 1 for real tokens and 0 for padding tokens. Only real
# tokens are attended to.
input_mask = [1] * len(input_ids)
# Zero-pad up to the sequence length.
while len(input_ids) < seq_length:
input_ids.append(0)
input_mask.append(0)
input_type_ids.append(0)
assert len(input_ids) == seq_length
assert len(input_mask) == seq_length
assert len(input_type_ids) == seq_length
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("unique_id: %s" % (example.unique_id))
tf.logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
tf.logging.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
tf.logging.info(
"input_type_ids: %s" % " ".join([str(x) for x in input_type_ids]))
features.append(
InputFeatures(
unique_id=example.unique_id,
tokens=tokens,
input_ids=input_ids,
input_mask=input_mask,
input_type_ids=input_type_ids))
return features
def _truncate_seq_pair(tokens_a, tokens_b, max_length):
"""Truncates a sequence pair in place to the maximum length."""
# This is a simple heuristic which will always truncate the longer sequence
# one token at a time. This makes more sense than truncating an equal percent
# of tokens from each, since if one sequence is very short then each token
# that's truncated likely contains more information than a longer sequence.
while True:
total_length = len(tokens_a) + len(tokens_b)
if total_length <= max_length:
break
if len(tokens_a) > len(tokens_b):
tokens_a.pop()
else:
tokens_b.pop()
def read_examples(input_file):
"""Read a list of `InputExample`s from an input file."""
examples = []
unique_id = 0
with tf.gfile.GFile(input_file, "r") as reader:
while True:
line = tokenization.convert_to_unicode(reader.readline())
if not line:
break
line = line.strip()
text_a = None
text_b = None
m = re.match(r"^(.*) \|\|\| (.*)$", line)
if m is None:
text_a = line
else:
text_a = m.group(1)
text_b = m.group(2)
examples.append(
InputExample(unique_id=unique_id, text_a=text_a, text_b=text_b))
unique_id += 1
return examples
def main(_):
tf.logging.set_verbosity(tf.logging.INFO)
layer_indexes = [int(x) for x in FLAGS.layers.split(",")]
bert_config = modeling.BertConfig.from_json_file(FLAGS.bert_config_file)
tokenizer = tokenization.FullTokenizer(
vocab_file=FLAGS.vocab_file, do_lower_case=FLAGS.do_lower_case)
is_per_host = tf.contrib.tpu.InputPipelineConfig.PER_HOST_V2
run_config = tf.contrib.tpu.RunConfig(
master=FLAGS.master,
tpu_config=tf.contrib.tpu.TPUConfig(
num_shards=FLAGS.num_tpu_cores,
per_host_input_for_training=is_per_host))
examples = read_examples(FLAGS.input_file)
features = convert_examples_to_features(
examples=examples, seq_length=FLAGS.max_seq_length, tokenizer=tokenizer)
unique_id_to_feature = {}
for feature in features:
unique_id_to_feature[feature.unique_id] = feature
model_fn = model_fn_builder(
bert_config=bert_config,
init_checkpoint=FLAGS.init_checkpoint,
layer_indexes=layer_indexes,
use_tpu=FLAGS.use_tpu,
use_one_hot_embeddings=FLAGS.use_one_hot_embeddings)
# If TPU is not available, this will fall back to normal Estimator on CPU
# or GPU.
estimator = tf.contrib.tpu.TPUEstimator(
use_tpu=FLAGS.use_tpu,
model_fn=model_fn,
config=run_config,
predict_batch_size=FLAGS.batch_size)
input_fn = input_fn_builder(
features=features, seq_length=FLAGS.max_seq_length)
with codecs.getwriter("utf-8")(tf.gfile.Open(FLAGS.output_file,
"w")) as writer:
for result in estimator.predict(input_fn, yield_single_examples=True):
unique_id = int(result["unique_id"])
feature = unique_id_to_feature[unique_id]
output_json = collections.OrderedDict()
output_json["linex_index"] = unique_id
all_features = []
for (i, token) in enumerate(feature.tokens):
all_layers = []
for (j, layer_index) in enumerate(layer_indexes):
layer_output = result["layer_output_%d" % j]
layers = collections.OrderedDict()
layers["index"] = layer_index
layers["values"] = [
round(float(x), 6) for x in layer_output[i:(i + 1)].flat
]
all_layers.append(layers)
features = collections.OrderedDict()
features["token"] = token
features["layers"] = all_layers
all_features.append(features)
output_json["features"] = all_features
writer.write(json.dumps(output_json) + "\n")
if __name__ == "__main__":
flags.mark_flag_as_required("input_file")
flags.mark_flag_as_required("vocab_file")
flags.mark_flag_as_required("bert_config_file")
flags.mark_flag_as_required("init_checkpoint")
flags.mark_flag_as_required("output_file")
tf.app.run()
================================================
FILE: modeling.py
================================================
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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.
"""The main BERT model and related functions."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import copy
import json
import math
import re
import six
import tensorflow as tf
class BertConfig(object):
"""Configuration for `BertModel`."""
def __init__(self,
vocab_size,
hidden_size=768,
num_hidden_layers=12,
num_attention_heads=12,
intermediate_size=3072,
hidden_act="gelu",
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
max_position_embeddings=512,
type_vocab_size=16,
initializer_range=0.02):
"""Constructs BertConfig.
Args:
vocab_size: Vocabulary size of `inputs_ids` in `BertModel`.
hidden_size: Size of the encoder layers and the pooler layer.
num_hidden_layers: Number of hidden layers in the Transformer encoder.
num_attention_heads: Number of attention heads for each attention layer in
the Transformer encoder.
intermediate_size: The size of the "intermediate" (i.e., feed-forward)
layer in the Transformer encoder.
hidden_act: The non-linear activation function (function or string) in the
encoder and pooler.
hidden_dropout_prob: The dropout probability for all fully connected
layers in the embeddings, encoder, and pooler.
attention_probs_dropout_prob: The dropout ratio for the attention
probabilities.
max_position_embeddings: The maximum sequence length that this model might
ever be used with. Typically set this to something large just in case
(e.g., 512 or 1024 or 2048).
type_vocab_size: The vocabulary size of the `token_type_ids` passed into
`BertModel`.
initializer_range: The stdev of the truncated_normal_initializer for
initializing all weight matrices.
"""
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.hidden_act = hidden_act
self.intermediate_size = intermediate_size
self.hidden_dropout_prob = hidden_dropout_prob
self.attention_probs_dropout_prob = attention_probs_dropout_prob
self.max_position_embeddings = max_position_embeddings
self.type_vocab_size = type_vocab_size
self.initializer_range = initializer_range
@classmethod
def from_dict(cls, json_object):
"""Constructs a `BertConfig` from a Python dictionary of parameters."""
config = BertConfig(vocab_size=None)
for (key, value) in six.iteritems(json_object):
config.__dict__[key] = value
return config
@classmethod
def from_json_file(cls, json_file):
"""Constructs a `BertConfig` from a json file of parameters."""
with tf.gfile.GFile(json_file, "r") as reader:
text = reader.read()
return cls.from_dict(json.loads(text))
def to_dict(self):
"""Serializes this instance to a Python dictionary."""
output = copy.deepcopy(self.__dict__)
return output
def to_json_string(self):
"""Serializes this instance to a JSON string."""
return json.dumps(self.to_dict(), indent=2, sort_keys=True) + "\n"
class BertModel(object):
"""BERT model ("Bidirectional Embedding Representations from a Transformer").
Example usage:
```python
# Already been converted into WordPiece token ids
input_ids = tf.constant([[31, 51, 99], [15, 5, 0]])
input_mask = tf.constant([[1, 1, 1], [1, 1, 0]])
token_type_ids = tf.constant([[0, 0, 1], [0, 2, 0]])
config = modeling.BertConfig(vocab_size=32000, hidden_size=512,
num_hidden_layers=8, num_attention_heads=6, intermediate_size=1024)
model = modeling.BertModel(config=config, is_training=True,
input_ids=input_ids, input_mask=input_mask, token_type_ids=token_type_ids)
label_embeddings = tf.get_variable(...)
pooled_output = model.get_pooled_output()
logits = tf.matmul(pooled_output, label_embeddings)
...
```
"""
def __init__(self,
config,
is_training,
input_ids,
input_mask=None,
token_type_ids=None,
use_one_hot_embeddings=True,
scope=None):
"""Constructor for BertModel.
Args:
config: `BertConfig` instance.
is_training: bool. rue for training model, false for eval model. Controls
whether dropout will be applied.
input_ids: int32 Tensor of shape [batch_size, seq_length].
input_mask: (optional) int32 Tensor of shape [batch_size, seq_length].
token_type_ids: (optional) int32 Tensor of shape [batch_size, seq_length].
use_one_hot_embeddings: (optional) bool. Whether to use one-hot word
embeddings or tf.embedding_lookup() for the word embeddings. On the TPU,
it is must faster if this is True, on the CPU or GPU, it is faster if
this is False.
scope: (optional) variable scope. Defaults to "bert".
Raises:
ValueError: The config is invalid or one of the input tensor shapes
is invalid.
"""
config = copy.deepcopy(config)
if not is_training:
config.hidden_dropout_prob = 0.0
config.attention_probs_dropout_prob = 0.0
input_shape = get_shape_list(input_ids, expected_rank=2)
batch_size = input_shape[0]
seq_length = input_shape[1]
if input_mask is None:
input_mask = tf.ones(shape=[batch_size, seq_length], dtype=tf.int32)
if token_type_ids is None:
token_type_ids = tf.zeros(shape=[batch_size, seq_length], dtype=tf.int32)
with tf.variable_scope(scope, default_name="bert"):
with tf.variable_scope("embeddings"):
# Perform embedding lookup on the word ids.
(self.embedding_output, self.embedding_table) = embedding_lookup(
input_ids=input_ids,
vocab_size=config.vocab_size,
embedding_size=config.hidden_size,
initializer_range=config.initializer_range,
word_embedding_name="word_embeddings",
use_one_hot_embeddings=use_one_hot_embeddings)
# Add positional embeddings and token type embeddings, then layer
# normalize and perform dropout.
self.embedding_output = embedding_postprocessor(
input_tensor=self.embedding_output,
use_token_type=True,
token_type_ids=token_type_ids,
token_type_vocab_size=config.type_vocab_size,
token_type_embedding_name="token_type_embeddings",
use_position_embeddings=True,
position_embedding_name="position_embeddings",
initializer_range=config.initializer_range,
max_position_embeddings=config.max_position_embeddings,
dropout_prob=config.hidden_dropout_prob)
with tf.variable_scope("encoder"):
# This converts a 2D mask of shape [batch_size, seq_length] to a 3D
# mask of shape [batch_size, seq_length, seq_length] which is used
# for the attention scores.
attention_mask = create_attention_mask_from_input_mask(
input_ids, input_mask)
# Run the stacked transformer.
# `sequence_output` shape = [batch_size, seq_length, hidden_size].
self.all_encoder_layers = transformer_model(
input_tensor=self.embedding_output,
attention_mask=attention_mask,
hidden_size=config.hidden_size,
num_hidden_layers=config.num_hidden_layers,
num_attention_heads=config.num_attention_heads,
intermediate_size=config.intermediate_size,
intermediate_act_fn=get_activation(config.hidden_act),
hidden_dropout_prob=config.hidden_dropout_prob,
attention_probs_dropout_prob=config.attention_probs_dropout_prob,
initializer_range=config.initializer_range,
do_return_all_layers=True)
self.sequence_output = self.all_encoder_layers[-1]
# The "pooler" converts the encoded sequence tensor of shape
# [batch_size, seq_length, hidden_size] to a tensor of shape
# [batch_size, hidden_size]. This is necessary for segment-level
# (or segment-pair-level) classification tasks where we need a fixed
# dimensional representation of the segment.
with tf.variable_scope("pooler"):
# We "pool" the model by simply taking the hidden state corresponding
# to the first token. We assume that this has been pre-trained
first_token_tensor = tf.squeeze(self.sequence_output[:, 0:1, :], axis=1)
self.pooled_output = tf.layers.dense(
first_token_tensor,
config.hidden_size,
activation=tf.tanh,
kernel_initializer=create_initializer(config.initializer_range))
def get_pooled_output(self):
return self.pooled_output
def get_sequence_output(self):
"""Gets final hidden layer of encoder.
Returns:
float Tensor of shape [batch_size, seq_length, hidden_size] corresponding
to the final hidden of the transformer encoder.
"""
return self.sequence_output
def get_all_encoder_layers(self):
return self.all_encoder_layers
def get_embedding_output(self):
"""Gets output of the embedding lookup (i.e., input to the transformer).
Returns:
float Tensor of shape [batch_size, seq_length, hidden_size] corresponding
to the output of the embedding layer, after summing the word
embeddings with the positional embeddings and the token type embeddings,
then performing layer normalization. This is the input to the transformer.
"""
return self.embedding_output
def get_embedding_table(self):
return self.embedding_table
def gelu(input_tensor):
"""Gaussian Error Linear Unit.
This is a smoother version of the RELU.
Original paper: https://arxiv.org/abs/1606.08415
Args:
input_tensor: float Tensor to perform activation.
Returns:
`input_tensor` with the GELU activation applied.
"""
cdf = 0.5 * (1.0 + tf.erf(input_tensor / tf.sqrt(2.0)))
return input_tensor * cdf
def get_activation(activation_string):
"""Maps a string to a Python function, e.g., "relu" => `tf.nn.relu`.
Args:
activation_string: String name of the activation function.
Returns:
A Python function corresponding to the activation function. If
`activation_string` is None, empty, or "linear", this will return None.
If `activation_string` is not a string, it will return `activation_string`.
Raises:
ValueError: The `activation_string` does not correspond to a known
activation.
"""
# We assume that anything that"s not a string is already an activation
# function, so we just return it.
if not isinstance(activation_string, six.string_types):
return activation_string
if not activation_string:
return None
act = activation_string.lower()
if act == "linear":
return None
elif act == "relu":
return tf.nn.relu
elif act == "gelu":
return gelu
elif act == "tanh":
return tf.tanh
else:
raise ValueError("Unsupported activation: %s" % act)
def get_assignment_map_from_checkpoint(tvars, init_checkpoint):
"""Compute the union of the current variables and checkpoint variables."""
assignment_map = {}
initialized_variable_names = {}
name_to_variable = collections.OrderedDict()
for var in tvars:
name = var.name
m = re.match("^(.*):\\d+$", name)
if m is not None:
name = m.group(1)
name_to_variable[name] = var
init_vars = tf.train.list_variables(init_checkpoint)
assignment_map = collections.OrderedDict()
for x in init_vars:
(name, var) = (x[0], x[1])
if name not in name_to_variable:
continue
assignment_map[name] = name
initialized_variable_names[name] = 1
initialized_variable_names[name + ":0"] = 1
return (assignment_map, initialized_variable_names)
def dropout(input_tensor, dropout_prob):
"""Perform dropout.
Args:
input_tensor: float Tensor.
dropout_prob: Python float. The probability of dropping out a value (NOT of
*keeping* a dimension as in `tf.nn.dropout`).
Returns:
A version of `input_tensor` with dropout applied.
"""
if dropout_prob is None or dropout_prob == 0.0:
return input_tensor
output = tf.nn.dropout(input_tensor, 1.0 - dropout_prob)
return output
def layer_norm(input_tensor, name=None):
"""Run layer normalization on the last dimension of the tensor."""
return tf.contrib.layers.layer_norm(
inputs=input_tensor, begin_norm_axis=-1, begin_params_axis=-1, scope=name)
def layer_norm_and_dropout(input_tensor, dropout_prob, name=None):
"""Runs layer normalization followed by dropout."""
output_tensor = layer_norm(input_tensor, name)
output_tensor = dropout(output_tensor, dropout_prob)
return output_tensor
def create_initializer(initializer_range=0.02):
"""Creates a `truncated_normal_initializer` with the given range."""
return tf.truncated_normal_initializer(stddev=initializer_range)
def embedding_lookup(input_ids,
vocab_size,
embedding_size=128,
initializer_range=0.02,
word_embedding_name="word_embeddings",
use_one_hot_embeddings=False):
"""Looks up words embeddings for id tensor.
Args:
input_ids: int32 Tensor of shape [batch_size, seq_length] containing word
ids.
vocab_size: int. Size of the embedding vocabulary.
embedding_size: int. Width of the word embeddings.
initializer_range: float. Embedding initialization range.
word_embedding_name: string. Name of the embedding table.
use_one_hot_embeddings: bool. If True, use one-hot method for word
embeddings. If False, use `tf.nn.embedding_lookup()`. One hot is better
for TPUs.
Returns:
float Tensor of shape [batch_size, seq_length, embedding_size].
"""
# This function assumes that the input is of shape [batch_size, seq_length,
# num_inputs].
#
# If the input is a 2D tensor of shape [batch_size, seq_length], we
# reshape to [batch_size, seq_length, 1].
if input_ids.shape.ndims == 2:
input_ids = tf.expand_dims(input_ids, axis=[-1])
embedding_table = tf.get_variable(
name=word_embedding_name,
shape=[vocab_size, embedding_size],
initializer=create_initializer(initializer_range))
if use_one_hot_embeddings:
flat_input_ids = tf.reshape(input_ids, [-1])
one_hot_input_ids = tf.one_hot(flat_input_ids, depth=vocab_size)
output = tf.matmul(one_hot_input_ids, embedding_table)
else:
output = tf.nn.embedding_lookup(embedding_table, input_ids)
input_shape = get_shape_list(input_ids)
output = tf.reshape(output,
input_shape[0:-1] + [input_shape[-1] * embedding_size])
return (output, embedding_table)
def embedding_postprocessor(input_tensor,
use_token_type=False,
token_type_ids=None,
token_type_vocab_size=16,
token_type_embedding_name="token_type_embeddings",
use_position_embeddings=True,
position_embedding_name="position_embeddings",
initializer_range=0.02,
max_position_embeddings=512,
dropout_prob=0.1):
"""Performs various post-processing on a word embedding tensor.
Args:
input_tensor: float Tensor of shape [batch_size, seq_length,
embedding_size].
use_token_type: bool. Whether to add embeddings for `token_type_ids`.
token_type_ids: (optional) int32 Tensor of shape [batch_size, seq_length].
Must be specified if `use_token_type` is True.
token_type_vocab_size: int. The vocabulary size of `token_type_ids`.
token_type_embedding_name: string. The name of the embedding table variable
for token type ids.
use_position_embeddings: bool. Whether to add position embeddings for the
position of each token in the sequence.
position_embedding_name: string. The name of the embedding table variable
for positional embeddings.
initializer_range: float. Range of the weight initialization.
max_position_embeddings: int. Maximum sequence length that might ever be
used with this model. This can be longer than the sequence length of
input_tensor, but cannot be shorter.
dropout_prob: float. Dropout probability applied to the final output tensor.
Returns:
float tensor with same shape as `input_tensor`.
Raises:
ValueError: One of the tensor shapes or input values is invalid.
"""
input_shape = get_shape_list(input_tensor, expected_rank=3)
batch_size = input_shape[0]
seq_length = input_shape[1]
width = input_shape[2]
output = input_tensor
if use_token_type:
if token_type_ids is None:
raise ValueError("`token_type_ids` must be specified if"
"`use_token_type` is True.")
token_type_table = tf.get_variable(
name=token_type_embedding_name,
shape=[token_type_vocab_size, width],
initializer=create_initializer(initializer_range))
# This vocab will be small so we always do one-hot here, since it is always
# faster for a small vocabulary.
flat_token_type_ids = tf.reshape(token_type_ids, [-1])
one_hot_ids = tf.one_hot(flat_token_type_ids, depth=token_type_vocab_size)
token_type_embeddings = tf.matmul(one_hot_ids, token_type_table)
token_type_embeddings = tf.reshape(token_type_embeddings,
[batch_size, seq_length, width])
output += token_type_embeddings
if use_position_embeddings:
assert_op = tf.assert_less_equal(seq_length, max_position_embeddings)
with tf.control_dependencies([assert_op]):
full_position_embeddings = tf.get_variable(
name=position_embedding_name,
shape=[max_position_embeddings, width],
initializer=create_initializer(initializer_range))
# Since the position embedding table is a learned variable, we create it
# using a (long) sequence length `max_position_embeddings`. The actual
# sequence length might be shorter than this, for faster training of
# tasks that do not have long sequences.
#
# So `full_position_embeddings` is effectively an embedding table
# for position [0, 1, 2, ..., max_position_embeddings-1], and the current
# sequence has positions [0, 1, 2, ... seq_length-1], so we can just
# perform a slice.
position_embeddings = tf.slice(full_position_embeddings, [0, 0],
[seq_length, -1])
num_dims = len(output.shape.as_list())
# Only the last two dimensions are relevant (`seq_length` and `width`), so
# we broadcast among the first dimensions, which is typically just
# the batch size.
position_broadcast_shape = []
for _ in range(num_dims - 2):
position_broadcast_shape.append(1)
position_broadcast_shape.extend([seq_length, width])
position_embeddings = tf.reshape(position_embeddings,
position_broadcast_shape)
output += position_embeddings
output = layer_norm_and_dropout(output, dropout_prob)
return output
def create_attention_mask_from_input_mask(from_tensor, to_mask):
"""Create 3D attention mask from a 2D tensor mask.
Args:
from_tensor: 2D or 3D Tensor of shape [batch_size, from_seq_length, ...].
to_mask: int32 Tensor of shape [batch_size, to_seq_length].
Returns:
float Tensor of shape [batch_size, from_seq_length, to_seq_length].
"""
from_shape = get_shape_list(from_tensor, expected_rank=[2, 3])
batch_size = from_shape[0]
from_seq_length = from_shape[1]
to_shape = get_shape_list(to_mask, expected_rank=2)
to_seq_length = to_shape[1]
to_mask = tf.cast(
tf.reshape(to_mask, [batch_size, 1, to_seq_length]), tf.float32)
# We don't assume that `from_tensor` is a mask (although it could be). We
# don't actually care if we attend *from* padding tokens (only *to* padding)
# tokens so we create a tensor of all ones.
#
# `broadcast_ones` = [batch_size, from_seq_length, 1]
broadcast_ones = tf.ones(
shape=[batch_size, from_seq_length, 1], dtype=tf.float32)
# Here we broadcast along two dimensions to create the mask.
mask = broadcast_ones * to_mask
return mask
def attention_layer(from_tensor,
to_tensor,
attention_mask=None,
num_attention_heads=1,
size_per_head=512,
query_act=None,
key_act=None,
value_act=None,
attention_probs_dropout_prob=0.0,
initializer_range=0.02,
do_return_2d_tensor=False,
batch_size=None,
from_seq_length=None,
to_seq_length=None):
"""Performs multi-headed attention from `from_tensor` to `to_tensor`.
This is an implementation of multi-headed attention based on "Attention
is all you Need". If `from_tensor` and `to_tensor` are the same, then
this is self-attention. Each timestep in `from_tensor` attends to the
corresponding sequence in `to_tensor`, and returns a fixed-with vector.
This function first projects `from_tensor` into a "query" tensor and
`to_tensor` into "key" and "value" tensors. These are (effectively) a list
of tensors of length `num_attention_heads`, where each tensor is of shape
[batch_size, seq_length, size_per_head].
Then, the query and key tensors are dot-producted and scaled. These are
softmaxed to obtain attention probabilities. The value tensors are then
interpolated by these probabilities, then concatenated back to a single
tensor and returned.
In practice, the multi-headed attention are done with transposes and
reshapes rather than actual separate tensors.
Args:
from_tensor: float Tensor of shape [batch_size, from_seq_length,
from_width].
to_tensor: float Tensor of shape [batch_size, to_seq_length, to_width].
attention_mask: (optional) int32 Tensor of shape [batch_size,
from_seq_length, to_seq_length]. The values should be 1 or 0. The
attention scores will effectively be set to -infinity for any positions in
the mask that are 0, and will be unchanged for positions that are 1.
num_attention_heads: int. Number of attention heads.
size_per_head: int. Size of each attention head.
query_act: (optional) Activation function for the query transform.
key_act: (optional) Activation function for the key transform.
value_act: (optional) Activation function for the value transform.
attention_probs_dropout_prob: (optional) float. Dropout probability of the
attention probabilities.
initializer_range: float. Range of the weight initializer.
do_return_2d_tensor: bool. If True, the output will be of shape [batch_size
* from_seq_length, num_attention_heads * size_per_head]. If False, the
output will be of shape [batch_size, from_seq_length, num_attention_heads
* size_per_head].
batch_size: (Optional) int. If the input is 2D, this might be the batch size
of the 3D version of the `from_tensor` and `to_tensor`.
from_seq_length: (Optional) If the input is 2D, this might be the seq length
of the 3D version of the `from_tensor`.
to_seq_length: (Optional) If the input is 2D, this might be the seq length
of the 3D version of the `to_tensor`.
Returns:
float Tensor of shape [batch_size, from_seq_length,
num_attention_heads * size_per_head]. (If `do_return_2d_tensor` is
true, this will be of shape [batch_size * from_seq_length,
num_attention_heads * size_per_head]).
Raises:
ValueError: Any of the arguments or tensor shapes are invalid.
"""
def transpose_for_scores(input_tensor, batch_size, num_attention_heads,
seq_length, width):
output_tensor = tf.reshape(
input_tensor, [batch_size, seq_length, num_attention_heads, width])
output_tensor = tf.transpose(output_tensor, [0, 2, 1, 3])
return output_tensor
from_shape = get_shape_list(from_tensor, expected_rank=[2, 3])
to_shape = get_shape_list(to_tensor, expected_rank=[2, 3])
if len(from_shape) != len(to_shape):
raise ValueError(
"The rank of `from_tensor` must match the rank of `to_tensor`.")
if len(from_shape) == 3:
batch_size = from_shape[0]
from_seq_length = from_shape[1]
to_seq_length = to_shape[1]
elif len(from_shape) == 2:
if (batch_size is None or from_seq_length is None or to_seq_length is None):
raise ValueError(
"When passing in rank 2 tensors to attention_layer, the values "
"for `batch_size`, `from_seq_length`, and `to_seq_length` "
"must all be specified.")
# Scalar dimensions referenced here:
# B = batch size (number of sequences)
# F = `from_tensor` sequence length
# T = `to_tensor` sequence length
# N = `num_attention_heads`
# H = `size_per_head`
from_tensor_2d = reshape_to_matrix(from_tensor)
to_tensor_2d = reshape_to_matrix(to_tensor)
# `query_layer` = [B*F, N*H]
query_layer = tf.layers.dense(
from_tensor_2d,
num_attention_heads * size_per_head,
activation=query_act,
name="query",
kernel_initializer=create_initializer(initializer_range))
# `key_layer` = [B*T, N*H]
key_layer = tf.layers.dense(
to_tensor_2d,
num_attention_heads * size_per_head,
activation=key_act,
name="key",
kernel_initializer=create_initializer(initializer_range))
# `value_layer` = [B*T, N*H]
value_layer = tf.layers.dense(
to_tensor_2d,
num_attention_heads * size_per_head,
activation=value_act,
name="value",
kernel_initializer=create_initializer(initializer_range))
# `query_layer` = [B, N, F, H]
query_layer = transpose_for_scores(query_layer, batch_size,
num_attention_heads, from_seq_length,
size_per_head)
# `key_layer` = [B, N, T, H]
key_layer = transpose_for_scores(key_layer, batch_size, num_attention_heads,
to_seq_length, size_per_head)
# Take the dot product between "query" and "key" to get the raw
# attention scores.
# `attention_scores` = [B, N, F, T]
attention_scores = tf.matmul(query_layer, key_layer, transpose_b=True)
attention_scores = tf.multiply(attention_scores,
1.0 / math.sqrt(float(size_per_head)))
if attention_mask is not None:
# `attention_mask` = [B, 1, F, T]
attention_mask = tf.expand_dims(attention_mask, axis=[1])
# Since attention_mask is 1.0 for positions we want to attend and 0.0 for
# masked positions, this operation will create a tensor which is 0.0 for
# positions we want to attend and -10000.0 for masked positions.
adder = (1.0 - tf.cast(attention_mask, tf.float32)) * -10000.0
# Since we are adding it to the raw scores before the softmax, this is
# effectively the same as removing these entirely.
attention_scores += adder
# Normalize the attention scores to probabilities.
# `attention_probs` = [B, N, F, T]
attention_probs = tf.nn.softmax(attention_scores)
# This is actually dropping out entire tokens to attend to, which might
# seem a bit unusual, but is taken from the original Transformer paper.
attention_probs = dropout(attention_probs, attention_probs_dropout_prob)
# `value_layer` = [B, T, N, H]
value_layer = tf.reshape(
value_layer,
[batch_size, to_seq_length, num_attention_heads, size_per_head])
# `value_layer` = [B, N, T, H]
value_layer = tf.transpose(value_layer, [0, 2, 1, 3])
# `context_layer` = [B, N, F, H]
context_layer = tf.matmul(attention_probs, value_layer)
# `context_layer` = [B, F, N, H]
context_layer = tf.transpose(context_layer, [0, 2, 1, 3])
if do_return_2d_tensor:
# `context_layer` = [B*F, N*V]
context_layer = tf.reshape(
context_layer,
[batch_size * from_seq_length, num_attention_heads * size_per_head])
else:
# `context_layer` = [B, F, N*V]
context_layer = tf.reshape(
context_layer,
[batch_size, from_seq_length, num_attention_heads * size_per_head])
return context_layer
def transformer_model(input_tensor,
attention_mask=None,
hidden_size=768,
num_hidden_layers=12,
num_attention_heads=12,
intermediate_size=3072,
intermediate_act_fn=gelu,
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
initializer_range=0.02,
do_return_all_layers=False):
"""Multi-headed, multi-layer Transformer from "Attention is All You Need".
This is almost an exact implementation of the original Transformer encoder.
See the original paper:
https://arxiv.org/abs/1706.03762
Also see:
https://github.com/tensorflow/tensor2tensor/blob/master/tensor2tensor/models/transformer.py
Args:
input_tensor: float Tensor of shape [batch_size, seq_length, hidden_size].
attention_mask: (optional) int32 Tensor of shape [batch_size, seq_length,
seq_length], with 1 for positions that can be attended to and 0 in
positions that should not be.
hidden_size: int. Hidden size of the Transformer.
num_hidden_layers: int. Number of layers (blocks) in the Transformer.
num_attention_heads: int. Number of attention heads in the Transformer.
intermediate_size: int. The size of the "intermediate" (a.k.a., feed
forward) layer.
intermediate_act_fn: function. The non-linear activation function to apply
to the output of the intermediate/feed-forward layer.
hidden_dropout_prob: float. Dropout probability for the hidden layers.
attention_probs_dropout_prob: float. Dropout probability of the attention
probabilities.
initializer_range: float. Range of the initializer (stddev of truncated
normal).
do_return_all_layers: Whether to also return all layers or just the final
layer.
Returns:
float Tensor of shape [batch_size, seq_length, hidden_size], the final
hidden layer of the Transformer.
Raises:
ValueError: A Tensor shape or parameter is invalid.
"""
if hidden_size % num_attention_heads != 0:
raise ValueError(
"The hidden size (%d) is not a multiple of the number of attention "
"heads (%d)" % (hidden_size, num_attention_heads))
attention_head_size = int(hidden_size / num_attention_heads)
input_shape = get_shape_list(input_tensor, expected_rank=3)
batch_size = input_shape[0]
seq_length = input_shape[1]
input_width = input_shape[2]
# The Transformer performs sum residuals on all layers so the input needs
# to be the same as the hidden size.
if input_width != hidden_size:
raise ValueError("The width of the input tensor (%d) != hidden size (%d)" %
(input_width, hidden_size))
# We keep the representation as a 2D tensor to avoid re-shaping it back and
# forth from a 3D tensor to a 2D tensor. Re-shapes are normally free on
# the GPU/CPU but may not be free on the TPU, so we want to minimize them to
# help the optimizer.
prev_output = reshape_to_matrix(input_tensor)
all_layer_outputs = []
for layer_idx in range(num_hidden_layers):
with tf.variable_scope("layer_%d" % layer_idx):
layer_input = prev_output
with tf.variable_scope("attention"):
attention_heads = []
with tf.variable_scope("self"):
attention_head = attention_layer(
from_tensor=layer_input,
to_tensor=layer_input,
attention_mask=attention_mask,
num_attention_heads=num_attention_heads,
size_per_head=attention_head_size,
attention_probs_dropout_prob=attention_probs_dropout_prob,
initializer_range=initializer_range,
do_return_2d_tensor=True,
batch_size=batch_size,
from_seq_length=seq_length,
to_seq_length=seq_length)
attention_heads.append(attention_head)
attention_output = None
if len(attention_heads) == 1:
attention_output = attention_heads[0]
else:
# In the case where we have other sequences, we just concatenate
# them to the self-attention head before the projection.
attention_output = tf.concat(attention_heads, axis=-1)
# Run a linear projection of `hidden_size` then add a residual
# with `layer_input`.
with tf.variable_scope("output"):
attention_output = tf.layers.dense(
attention_output,
hidden_size,
kernel_initializer=create_initializer(initializer_range))
attention_output = dropout(attention_output, hidden_dropout_prob)
attention_output = layer_norm(attention_output + layer_input)
# The activation is only applied to the "intermediate" hidden layer.
with tf.variable_scope("intermediate"):
intermediate_output = tf.layers.dense(
attention_output,
intermediate_size,
activation=intermediate_act_fn,
kernel_initializer=create_initializer(initializer_range))
# Down-project back to `hidden_size` then add the residual.
with tf.variable_scope("output"):
layer_output = tf.layers.dense(
intermediate_output,
hidden_size,
kernel_initializer=create_initializer(initializer_range))
layer_output = dropout(layer_output, hidden_dropout_prob)
layer_output = layer_norm(layer_output + attention_output)
prev_output = layer_output
all_layer_outputs.append(layer_output)
if do_return_all_layers:
final_outputs = []
for layer_output in all_layer_outputs:
final_output = reshape_from_matrix(layer_output, input_shape)
final_outputs.append(final_output)
return final_outputs
else:
final_output = reshape_from_matrix(prev_output, input_shape)
return final_output
def get_shape_list(tensor, expected_rank=None, name=None):
"""Returns a list of the shape of tensor, preferring static dimensions.
Args:
tensor: A tf.Tensor object to find the shape of.
expected_rank: (optional) int. The expected rank of `tensor`. If this is
specified and the `tensor` has a different rank, and exception will be
thrown.
name: Optional name of the tensor for the error message.
Returns:
A list of dimensions of the shape of tensor. All static dimensions will
be returned as python integers, and dynamic dimensions will be returned
as tf.Tensor scalars.
"""
if name is None:
name = tensor.name
if expected_rank is not None:
assert_rank(tensor, expected_rank, name)
shape = tensor.shape.as_list()
non_static_indexes = []
for (index, dim) in enumerate(shape):
if dim is None:
non_static_indexes.append(index)
if not non_static_indexes:
return shape
dyn_shape = tf.shape(tensor)
for index in non_static_indexes:
shape[index] = dyn_shape[index]
return shape
def reshape_to_matrix(input_tensor):
"""Reshapes a >= rank 2 tensor to a rank 2 tensor (i.e., a matrix)."""
ndims = input_tensor.shape.ndims
if ndims < 2:
raise ValueError("Input tensor must have at least rank 2. Shape = %s" %
(input_tensor.shape))
if ndims == 2:
return input_tensor
width = input_tensor.shape[-1]
output_tensor = tf.reshape(input_tensor, [-1, width])
return output_tensor
def reshape_from_matrix(output_tensor, orig_shape_list):
"""Reshapes a rank 2 tensor back to its original rank >= 2 tensor."""
if len(orig_shape_list) == 2:
return output_tensor
output_shape = get_shape_list(output_tensor)
orig_dims = orig_shape_list[0:-1]
width = output_shape[-1]
return tf.reshape(output_tensor, orig_dims + [width])
def assert_rank(tensor, expected_rank, name=None):
"""Raises an exception if the tensor rank is not of the expected rank.
Args:
tensor: A tf.Tensor to check the rank of.
expected_rank: Python integer or list of integers, expected rank.
name: Optional name of the tensor for the error message.
Raises:
ValueError: If the expected shape doesn't match the actual shape.
"""
if name is None:
name = tensor.name
expected_rank_dict = {}
if isinstance(expected_rank, six.integer_types):
expected_rank_dict[expected_rank] = True
else:
for x in expected_rank:
expected_rank_dict[x] = True
actual_rank = tensor.shape.ndims
if actual_rank not in expected_rank_dict:
scope_name = tf.get_variable_scope().name
raise ValueError(
"For the tensor `%s` in scope `%s`, the actual rank "
"`%d` (shape = %s) is not equal to the expected rank `%s`" %
(name, scope_name, actual_rank, str(tensor.shape), str(expected_rank)))
================================================
FILE: modeling_test.py
================================================
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import json
import random
import re
import modeling
import six
import tensorflow as tf
class BertModelTest(tf.test.TestCase):
class BertModelTester(object):
def __init__(self,
parent,
batch_size=13,
seq_length=7,
is_training=True,
use_input_mask=True,
use_token_type_ids=True,
vocab_size=99,
hidden_size=32,
num_hidden_layers=5,
num_attention_heads=4,
intermediate_size=37,
hidden_act="gelu",
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
max_position_embeddings=512,
type_vocab_size=16,
initializer_range=0.02,
scope=None):
self.parent = parent
self.batch_size = batch_size
self.seq_length = seq_length
self.is_training = is_training
self.use_input_mask = use_input_mask
self.use_token_type_ids = use_token_type_ids
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.intermediate_size = intermediate_size
self.hidden_act = hidden_act
self.hidden_dropout_prob = hidden_dropout_prob
self.attention_probs_dropout_prob = attention_probs_dropout_prob
self.max_position_embeddings = max_position_embeddings
self.type_vocab_size = type_vocab_size
self.initializer_range = initializer_range
self.scope = scope
def create_model(self):
input_ids = BertModelTest.ids_tensor([self.batch_size, self.seq_length],
self.vocab_size)
input_mask = None
if self.use_input_mask:
input_mask = BertModelTest.ids_tensor(
[self.batch_size, self.seq_length], vocab_size=2)
token_type_ids = None
if self.use_token_type_ids:
token_type_ids = BertModelTest.ids_tensor(
[self.batch_size, self.seq_length], self.type_vocab_size)
config = modeling.BertConfig(
vocab_size=self.vocab_size,
hidden_size=self.hidden_size,
num_hidden_layers=self.num_hidden_layers,
num_attention_heads=self.num_attention_heads,
intermediate_size=self.intermediate_size,
hidden_act=self.hidden_act,
hidden_dropout_prob=self.hidden_dropout_prob,
attention_probs_dropout_prob=self.attention_probs_dropout_prob,
max_position_embeddings=self.max_position_embeddings,
type_vocab_size=self.type_vocab_size,
initializer_range=self.initializer_range)
model = modeling.BertModel(
config=config,
is_training=self.is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=token_type_ids,
scope=self.scope)
outputs = {
"embedding_output": model.get_embedding_output(),
"sequence_output": model.get_sequence_output(),
"pooled_output": model.get_pooled_output(),
"all_encoder_layers": model.get_all_encoder_layers(),
}
return outputs
def check_output(self, result):
self.parent.assertAllEqual(
result["embedding_output"].shape,
[self.batch_size, self.seq_length, self.hidden_size])
self.parent.assertAllEqual(
result["sequence_output"].shape,
[self.batch_size, self.seq_length, self.hidden_size])
self.parent.assertAllEqual(result["pooled_output"].shape,
[self.batch_size, self.hidden_size])
def test_default(self):
self.run_tester(BertModelTest.BertModelTester(self))
def test_config_to_json_string(self):
config = modeling.BertConfig(vocab_size=99, hidden_size=37)
obj = json.loads(config.to_json_string())
self.assertEqual(obj["vocab_size"], 99)
self.assertEqual(obj["hidden_size"], 37)
def run_tester(self, tester):
with self.test_session() as sess:
ops = tester.create_model()
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
output_result = sess.run(ops)
tester.check_output(output_result)
self.assert_all_tensors_reachable(sess, [init_op, ops])
@classmethod
def ids_tensor(cls, shape, vocab_size, rng=None, name=None):
"""Creates a random int32 tensor of the shape within the vocab size."""
if rng is None:
rng = random.Random()
total_dims = 1
for dim in shape:
total_dims *= dim
values = []
for _ in range(total_dims):
values.append(rng.randint(0, vocab_size - 1))
return tf.constant(value=values, dtype=tf.int32, shape=shape, name=name)
def assert_all_tensors_reachable(self, sess, outputs):
"""Checks that all the tensors in the graph are reachable from outputs."""
graph = sess.graph
ignore_strings = [
"^.*/assert_less_equal/.*$",
"^.*/dilation_rate$",
"^.*/Tensordot/concat$",
"^.*/Tensordot/concat/axis$",
"^testing/.*$",
]
ignore_regexes = [re.compile(x) for x in ignore_strings]
unreachable = self.get_unreachable_ops(graph, outputs)
filtered_unreachable = []
for x in unreachable:
do_ignore = False
for r in ignore_regexes:
m = r.match(x.name)
if m is not None:
do_ignore = True
if do_ignore:
continue
filtered_unreachable.append(x)
unreachable = filtered_unreachable
self.assertEqual(
len(unreachable), 0, "The following ops are unreachable: %s" %
(" ".join([x.name for x in unreachable])))
@classmethod
def get_unreachable_ops(cls, graph, outputs):
"""Finds all of the tensors in graph that are unreachable from outputs."""
outputs = cls.flatten_recursive(outputs)
output_to_op = collections.defaultdict(list)
op_to_all = collections.defaultdict(list)
assign_out_to_in = collections.defaultdict(list)
for op in graph.get_operations():
for x in op.inputs:
op_to_all[op.name].append(x.name)
for y in op.outputs:
output_to_op[y.name].append(op.name)
op_to_all[op.name].append(y.name)
if str(op.type) == "Assign":
for y in op.outputs:
for x in op.inputs:
assign_out_to_in[y.name].append(x.name)
assign_groups = collections.defaultdict(list)
for out_name in assign_out_to_in.keys():
name_group = assign_out_to_in[out_name]
for n1 in name_group:
assign_groups[n1].append(out_name)
for n2 in name_group:
if n1 != n2:
assign_groups[n1].append(n2)
seen_tensors = {}
stack = [x.name for x in outputs]
while stack:
name = stack.pop()
if name in seen_tensors:
continue
seen_tensors[name] = True
if name in output_to_op:
for op_name in output_to_op[name]:
if op_name in op_to_all:
for input_name in op_to_all[op_name]:
if input_name not in stack:
stack.append(input_name)
expanded_names = []
if name in assign_groups:
for assign_name in assign_groups[name]:
expanded_names.append(assign_name)
for expanded_name in expanded_names:
if expanded_name not in stack:
stack.append(expanded_name)
unreachable_ops = []
for op in graph.get_operations():
is_unreachable = False
all_names = [x.name for x in op.inputs] + [x.name for x in op.outputs]
for name in all_names:
if name not in seen_tensors:
is_unreachable = True
if is_unreachable:
unreachable_ops.append(op)
return unreachable_ops
@classmethod
def flatten_recursive(cls, item):
"""Flattens (potentially nested) a tuple/dictionary/list to a list."""
output = []
if isinstance(item, list):
output.extend(item)
elif isinstance(item, tuple):
output.extend(list(item))
elif isinstance(item, dict):
for (_, v) in six.iteritems(item):
output.append(v)
else:
return [item]
flat_output = []
for x in output:
flat_output.extend(cls.flatten_recursive(x))
return flat_output
if __name__ == "__main__":
tf.test.main()
================================================
FILE: multilingual.md
================================================
## Models
There are two multilingual models currently available. We do not plan to release
more single-language models, but we may release `BERT-Large` versions of these
two in the future:
* **[`BERT-Base, Multilingual Cased (New, recommended)`](https://storage.googleapis.com/bert_models/2018_11_23/multi_cased_L-12_H-768_A-12.zip)**:
104 languages, 12-layer, 768-hidden, 12-heads, 110M parameters
* **[`BERT-Base, Multilingual Uncased (Orig, not recommended)`](https://storage.googleapis.com/bert_models/2018_11_03/multilingual_L-12_H-768_A-12.zip)**:
102 languages, 12-layer, 768-hidden, 12-heads, 110M parameters
* **[`BERT-Base, Chinese`](https://storage.googleapis.com/bert_models/2018_11_03/chinese_L-12_H-768_A-12.zip)**:
Chinese Simplified and Traditional, 12-layer, 768-hidden, 12-heads, 110M
parameters
**The `Multilingual Cased (New)` model also fixes normalization issues in many
languages, so it is recommended in languages with non-Latin alphabets (and is
often better for most languages with Latin alphabets). When using this model,
make sure to pass `--do_lower_case=false` to `run_pretraining.py` and other
scripts.**
See the [list of languages](#list-of-languages) that the Multilingual model
supports. The Multilingual model does include Chinese (and English), but if your
fine-tuning data is Chinese-only, then the Chinese model will likely produce
better results.
## Results
To evaluate these systems, we use the
[XNLI dataset](https://github.com/facebookresearch/XNLI) dataset, which is a
version of [MultiNLI](https://www.nyu.edu/projects/bowman/multinli/) where the
dev and test sets have been translated (by humans) into 15 languages. Note that
the training set was *machine* translated (we used the translations provided by
XNLI, not Google NMT). For clarity, we only report on 6 languages below:
<!-- mdformat off(no table) -->
| System | English | Chinese | Spanish | German | Arabic | Urdu |
| --------------------------------- | -------- | -------- | -------- | -------- | -------- | -------- |
| XNLI Baseline - Translate Train | 73.7 | 67.0 | 68.8 | 66.5 | 65.8 | 56.6 |
| XNLI Baseline - Translate Test | 73.7 | 68.3 | 70.7 | 68.7 | 66.8 | 59.3 |
| BERT - Translate Train Cased | **81.9** | **76.6** | **77.8** | **75.9** | **70.7** | 61.6 |
| BERT - Translate Train Uncased | 81.4 | 74.2 | 77.3 | 75.2 | 70.5 | 61.7 |
| BERT - Translate Test Uncased | 81.4 | 70.1 | 74.9 | 74.4 | 70.4 | **62.1** |
| BERT - Zero Shot Uncased | 81.4 | 63.8 | 74.3 | 70.5 | 62.1 | 58.3 |
<!-- mdformat on -->
The first two rows are baselines from the XNLI paper and the last three rows are
our results with BERT.
**Translate Train** means that the MultiNLI training set was machine translated
from English into the foreign language. So training and evaluation were both
done in the foreign language. Unfortunately, training was done on
machine-translated data, so it is impossible to quantify how much of the lower
accuracy (compared to English) is due to the quality of the machine translation
vs. the quality of the pre-trained model.
**Translate Test** means that the XNLI test set was machine translated from the
foreign language into English. So training and evaluation were both done on
English. However, test evaluation was done on machine-translated English, so the
accuracy depends on the quality of the machine translation system.
**Zero Shot** means that the Multilingual BERT system was fine-tuned on English
MultiNLI, and then evaluated on the foreign language XNLI test. In this case,
machine translation was not involved at all in either the pre-training or
fine-tuning.
Note that the English result is worse than the 84.2 MultiNLI baseline because
this training used Multilingual BERT rather than English-only BERT. This implies
that for high-resource languages, the Multilingual model is somewhat worse than
a single-language model. However, it is not feasible for us to train and
maintain dozens of single-language model. Therefore, if your goal is to maximize
performance with a language other than English or Chinese, you might find it
beneficial to run pre-training for additional steps starting from our
Multilingual model on data from your language of interest.
Here is a comparison of training Chinese models with the Multilingual
`BERT-Base` and Chinese-only `BERT-Base`:
System | Chinese
----------------------- | -------
XNLI Baseline | 67.0
BERT Multilingual Model | 74.2
BERT Chinese-only Model | 77.2
Similar to English, the single-language model does 3% better than the
Multilingual model.
## Fine-tuning Example
The multilingual model does **not** require any special consideration or API
changes. We did update the implementation of `BasicTokenizer` in
`tokenization.py` to support Chinese character tokenization, so please update if
you forked it. However, we did not change the tokenization API.
To test the new models, we did modify `run_classifier.py` to add support for the
[XNLI dataset](https://github.com/facebookresearch/XNLI). This is a 15-language
version of MultiNLI where the dev/test sets have been human-translated, and the
training set has been machine-translated.
To run the fine-tuning code, please download the
[XNLI dev/test set](https://s3.amazonaws.com/xnli/XNLI-1.0.zip) and the
[XNLI machine-translated training set](https://s3.amazonaws.com/xnli/XNLI-MT-1.0.zip)
and then unpack both .zip files into some directory `$XNLI_DIR`.
To run fine-tuning on XNLI. The language is hard-coded into `run_classifier.py`
(Chinese by default), so please modify `XnliProcessor` if you want to run on
another language.
This is a large dataset, so this will training will take a few hours on a GPU
(or about 30 minutes on a Cloud TPU). To run an experiment quickly for
debugging, just set `num_train_epochs` to a small value like `0.1`.
```shell
export BERT_BASE_DIR=/path/to/bert/chinese_L-12_H-768_A-12 # or multilingual_L-12_H-768_A-12
export XNLI_DIR=/path/to/xnli
python run_classifier.py \
--task_name=XNLI \
--do_train=true \
--do_eval=true \
--data_dir=$XNLI_DIR \
--vocab_file=$BERT_BASE_DIR/vocab.txt \
--bert_config_file=$BERT_BASE_DIR/bert_config.json \
--init_checkpoint=$BERT_BASE_DIR/bert_model.ckpt \
--max_seq_length=128 \
--train_batch_size=32 \
--learning_rate=5e-5 \
--num_train_epochs=2.0 \
--output_dir=/tmp/xnli_output/
```
With the Chinese-only model, the results should look something like this:
```
***** Eval results *****
eval_accuracy = 0.774116
eval_loss = 0.83554
global_step = 24543
loss = 0.74603
```
## Details
### Data Source and Sampling
The languages chosen were the
[top 100 languages with the largest Wikipedias](https://meta.wikimedia.org/wiki/List_of_Wikipedias).
The entire Wikipedia dump for each language (excluding user and talk pages) was
taken as the training data for each language
However, the size of the Wikipedia for a given language varies greatly, and
therefore low-resource languages may be "under-represented" in terms of the
neural network model (under the assumption that languages are "competing" for
limited model capacity to some extent).
However, the size of a Wikipedia also correlates with the number of speakers of
a language, and we also don't want to overfit the model by performing thousands
of epochs over a tiny Wikipedia for a particular language.
To balance these two factors, we performed exponentially smoothed weighting of
the data during pre-training data creation (and WordPiece vocab creation). In
other words, let's say that the probability of a language is *P(L)*, e.g.,
*P(English) = 0.21* means that after concatenating all of the Wikipedias
together, 21% of our data is English. We exponentiate each probability by some
factor *S* and then re-normalize, and sample from that distribution. In our case
we use *S=0.7*. So, high-resource languages like English will be under-sampled,
and low-resource languages like Icelandic will be over-sampled. E.g., in the
original distribution English would be sampled 1000x more than Icelandic, but
after smoothing it's only sampled 100x more.
### Tokenization
For tokenization, we use a 110k shared WordPiece vocabulary. The word counts are
weighted the same way as the data, so low-resource languages are upweighted by
some factor. We intentionally do *not* use any marker to denote the input
language (so that zero-shot training can work).
Because Chinese (and Japanese Kanji and Korean Hanja) does not have whitespace
characters, we add spaces around every character in the
[CJK Unicode range](https://en.wikipedia.org/wiki/CJK_Unified_Ideographs_\(Unicode_block\))
before applying WordPiece. This means that Chinese is effectively
character-tokenized. Note that the CJK Unicode block only includes
Chinese-origin characters and does *not* include Hangul Korean or
Katakana/Hiragana Japanese, which are tokenized with whitespace+WordPiece like
all other languages.
For all other languages, we apply the
[same recipe as English](https://github.com/google-research/bert#tokenization):
(a) lower casing+accent removal, (b) punctuation splitting, (c) whitespace
tokenization. We understand that accent markers have substantial meaning in some
languages, but felt that the benefits of reducing the effective vocabulary make
up for this. Generally the strong contextual models of BERT should make up for
any ambiguity introduced by stripping accent markers.
### List of Languages
The multilingual model supports the following languages. These languages were
chosen because they are the top 100 languages with the largest Wikipedias:
* Afrikaans
* Albanian
* Arabic
* Aragonese
* Armenian
* Asturian
* Azerbaijani
* Bashkir
* Basque
* Bavarian
* Belarusian
* Bengali
* Bishnupriya Manipuri
* Bosnian
* Breton
* Bulgarian
* Burmese
* Catalan
* Cebuano
* Chechen
* Chinese (Simplified)
* Chinese (Traditional)
* Chuvash
* Croatian
* Czech
* Danish
* Dutch
* English
* Estonian
* Finnish
* French
* Galician
* Georgian
* German
* Greek
* Gujarati
* Haitian
* Hebrew
* Hindi
* Hungarian
* Icelandic
* Ido
* Indonesian
* Irish
* Italian
* Japanese
* Javanese
* Kannada
* Kazakh
* Kirghiz
* Korean
* Latin
* Latvian
* Lithuanian
* Lombard
* Low Saxon
* Luxembourgish
* Macedonian
* Malagasy
* Malay
* Malayalam
* Marathi
* Minangkabau
* Nepali
* Newar
* Norwegian (Bokmal)
* Norwegian (Nynorsk)
* Occitan
* Persian (Farsi)
* Piedmontese
* Polish
* Portuguese
* Punjabi
* Romanian
* Russian
* Scots
* Serbian
* Serbo-Croatian
* Sicilian
* Slovak
* Slovenian
* South Azerbaijani
* Spanish
* Sundanese
* Swahili
* Swedish
* Tagalog
* Tajik
* Tamil
* Tatar
* Telugu
* Turkish
* Ukrainian
* Urdu
* Uzbek
* Vietnamese
* Volapük
* Waray-Waray
* Welsh
* West
* Western Punjabi
* Yoruba
The **Multilingual Cased (New)** release contains additionally **Thai** and
**Mongolian**, which were not included in the original release.
================================================
FILE: optimization.py
================================================
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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.
"""Functions and classes related to optimization (weight updates)."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import re
import tensorflow as tf
def create_optimizer(loss, init_lr, num_train_steps, num_warmup_steps, use_tpu):
"""Creates an optimizer training op."""
global_step = tf.train.get_or_create_global_step()
learning_rate = tf.constant(value=init_lr, shape=[], dtype=tf.float32)
# Implements linear decay of the learning rate.
learning_rate = tf.train.polynomial_decay(
learning_rate,
global_step,
num_train_steps,
end_learning_rate=0.0,
power=1.0,
cycle=False)
# Implements linear warmup. I.e., if global_step < num_warmup_steps, the
# learning rate will be `global_step/num_warmup_steps * init_lr`.
if num_warmup_steps:
global_steps_int = tf.cast(global_step, tf.int32)
warmup_steps_int = tf.constant(num_warmup_steps, dtype=tf.int32)
global_steps_float = tf.cast(global_steps_int, tf.float32)
warmup_steps_float = tf.cast(warmup_steps_int, tf.float32)
warmup_percent_done = global_steps_float / warmup_steps_float
warmup_learning_rate = init_lr * warmup_percent_done
is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32)
learning_rate = (
(1.0 - is_warmup) * learning_rate + is_warmup * warmup_learning_rate)
# It is recommended that you use this optimizer for fine tuning, since this
# is how the model was trained (note that the Adam m/v variables are NOT
# loaded from init_checkpoint.)
optimizer = AdamWeightDecayOptimizer(
learning_rate=learning_rate,
weight_decay_rate=0.01,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-6,
exclude_from_weight_decay=["LayerNorm", "layer_norm", "bias"])
if use_tpu:
optimizer = tf.contrib.tpu.CrossShardOptimizer(optimizer)
tvars = tf.trainable_variables()
grads = tf.gradients(loss, tvars)
# This is how the model was pre-trained.
(grads, _) = tf.clip_by_global_norm(grads, clip_norm=1.0)
train_op = optimizer.apply_gradients(
zip(grads, tvars), global_step=global_step)
new_global_step = global_step + 1
train_op = tf.group(train_op, [global_step.assign(new_global_step)])
return train_op
class AdamWeightDecayOptimizer(tf.train.Optimizer):
"""A basic Adam optimizer that includes "correct" L2 weight decay."""
def __init__(self,
learning_rate,
weight_decay_rate=0.0,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-6,
exclude_from_weight_decay=None,
name="AdamWeightDecayOptimizer"):
"""Constructs a AdamWeightDecayOptimizer."""
super(AdamWeightDecayOptimizer, self).__init__(False, name)
self.learning_rate = learning_rate
self.weight_decay_rate = weight_decay_rate
self.beta_1 = beta_1
self.beta_2 = beta_2
self.epsilon = epsilon
self.exclude_from_weight_decay = exclude_from_weight_decay
def apply_gradients(self, grads_and_vars, global_step=None, name=None):
"""See base class."""
assignments = []
for (grad, param) in grads_and_vars:
if grad is None or param is None:
continue
param_name = self._get_variable_name(param.name)
m = tf.get_variable(
name=param_name + "/adam_m",
shape=param.shape.as_list(),
dtype=tf.float32,
trainable=False,
initializer=tf.zeros_initializer())
v = tf.get_variable(
name=param_name + "/adam_v",
shape=param.shape.as_list(),
dtype=tf.float32,
trainable=False,
initializer=tf.zeros_initializer())
# Standard Adam update.
next_m = (
tf.multiply(self.beta_1, m) + tf.multiply(1.0 - self.beta_1, grad))
next_v = (
tf.multiply(self.beta_2, v) + tf.multiply(1.0 - self.beta_2,
tf.square(grad)))
update = next_m / (tf.sqrt(next_v) + self.epsilon)
# Just adding the square of the weights to the loss function is *not*
# the correct way of using L2 regularization/weight decay with Adam,
# since that will interact with the m and v parameters in strange ways.
#
# Instead we want ot decay the weights in a manner that doesn't interact
# with the m/v parameters. This is equivalent to adding the square
# of the weights to the loss with plain (non-momentum) SGD.
if self._do_use_weight_decay(param_name):
update += self.weight_decay_rate * param
update_with_lr = self.learning_rate * update
next_param = param - update_with_lr
assignments.extend(
[param.assign(next_param),
m.assign(next_m),
v.assign(next_v)])
return tf.group(*assignments, name=name)
def _do_use_weight_decay(self, param_name):
"""Whether to use L2 weight decay for `param_name`."""
if not self.weight_decay_rate:
return False
if self.exclude_from_weight_decay:
for r in self.exclude_from_weight_decay:
if re.search(r, param_name) is not None:
return False
return True
def _get_variable_name(self, param_name):
"""Get the variable name from the tensor name."""
m = re.match("^(.*):\\d+$", param_name)
if m is not None:
param_name = m.group(1)
return param_name
================================================
FILE: optimization_test.py
================================================
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import optimization
import tensorflow as tf
class OptimizationTest(tf.test.TestCase):
def test_adam(self):
with self.test_session() as sess:
w = tf.get_variable(
"w",
shape=[3],
initializer=tf.constant_initializer([0.1, -0.2, -0.1]))
x = tf.constant([0.4, 0.2, -0.5])
loss = tf.reduce_mean(tf.square(x - w))
tvars = tf.trainable_variables()
grads = tf.gradients(loss, tvars)
global_step = tf.train.get_or_create_global_step()
optimizer = optimization.AdamWeightDecayOptimizer(learning_rate=0.2)
train_op = optimizer.apply_gradients(zip(grads, tvars), global_step)
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
for _ in range(100):
sess.run(train_op)
w_np = sess.run(w)
self.assertAllClose(w_np.flat, [0.4, 0.2, -0.5], rtol=1e-2, atol=1e-2)
if __name__ == "__main__":
tf.test.main()
================================================
FILE: requirements.txt
================================================
tensorflow >= 1.11.0 # CPU Version of TensorFlow.
# tensorflow-gpu >= 1.11.0 # GPU version of TensorFlow.
================================================
FILE: run_classifier.py
================================================
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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.
"""BERT finetuning runner."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import csv
import os
import modeling
import optimization
import tokenization
import tensorflow as tf
flags = tf.flags
FLAGS = flags.FLAGS
## Required parameters
flags.DEFINE_string(
"data_dir", None,
"The input data dir. Should contain the .tsv files (or other data files) "
"for the task.")
flags.DEFINE_string(
"bert_config_file", None,
"The config json file corresponding to the pre-trained BERT model. "
"This specifies the model architecture.")
flags.DEFINE_string("task_name", None, "The name of the task to train.")
flags.DEFINE_string("vocab_file", None,
"The vocabulary file that the BERT model was trained on.")
flags.DEFINE_string(
"output_dir", None,
"The output directory where the model checkpoints will be written.")
## Other parameters
flags.DEFINE_string(
"init_checkpoint", None,
"Initial checkpoint (usually from a pre-trained BERT model).")
flags.DEFINE_bool(
"do_lower_case", True,
"Whether to lower case the input text. Should be True for uncased "
"models and False for cased models.")
flags.DEFINE_integer(
"max_seq_length", 128,
"The maximum total input sequence length after WordPiece tokenization. "
"Sequences longer than this will be truncated, and sequences shorter "
"than this will be padded.")
flags.DEFINE_bool("do_train", False, "Whether to run training.")
flags.DEFINE_bool("do_eval", False, "Whether to run eval on the dev set.")
flags.DEFINE_bool(
"do_predict", False,
"Whether to run the model in inference mode on the test set.")
flags.DEFINE_integer("train_batch_size", 32, "Total batch size for training.")
flags.DEFINE_integer("eval_batch_size", 8, "Total batch size for eval.")
flags.DEFINE_integer("predict_batch_size", 8, "Total batch size for predict.")
flags.DEFINE_float("learning_rate", 5e-5, "The initial learning rate for Adam.")
flags.DEFINE_float("num_train_epochs", 3.0,
"Total number of training epochs to perform.")
flags.DEFINE_float(
"warmup_proportion", 0.1,
"Proportion of training to perform linear learning rate warmup for. "
"E.g., 0.1 = 10% of training.")
flags.DEFINE_integer("save_checkpoints_steps", 1000,
"How often to save the model checkpoint.")
flags.DEFINE_integer("iterations_per_loop", 1000,
"How many steps to make in each estimator call.")
flags.DEFINE_bool("use_tpu", False, "Whether to use TPU or GPU/CPU.")
tf.flags.DEFINE_string(
"tpu_name", None,
"The Cloud TPU to use for training. This should be either the name "
"used when creating the Cloud TPU, or a grpc://ip.address.of.tpu:8470 "
"url.")
tf.flags.DEFINE_string(
"tpu_zone", None,
"[Optional] GCE zone where the Cloud TPU is located in. If not "
"specified, we will attempt to automatically detect the GCE project from "
"metadata.")
tf.flags.DEFINE_string(
"gcp_project", None,
"[Optional] Project name for the Cloud TPU-enabled project. If not "
"specified, we will attempt to automatically detect the GCE project from "
"metadata.")
tf.flags.DEFINE_string("master", None, "[Optional] TensorFlow master URL.")
flags.DEFINE_integer(
"num_tpu_cores", 8,
"Only used if `use_tpu` is True. Total number of TPU cores to use.")
class InputExample(object):
"""A single training/test example for simple sequence classification."""
def __init__(self, guid, text_a, text_b=None, label=None):
"""Constructs a InputExample.
Args:
guid: Unique id for the example.
text_a: string. The untokenized text of the first sequence. For single
sequence tasks, only this sequence must be specified.
text_b: (Optional) string. The untokenized text of the second sequence.
Only must be specified for sequence pair tasks.
label: (Optional) string. The label of the example. This should be
specified for train and dev examples, but not for test examples.
"""
self.guid = guid
self.text_a = text_a
self.text_b = text_b
self.label = label
class InputFeatures(object):
"""A single set of features of data."""
def __init__(self, input_ids, input_mask, segment_ids, label_id):
self.input_ids = input_ids
self.input_mask = input_mask
self.segment_ids = segment_ids
self.label_id = label_id
class DataProcessor(object):
"""Base class for data converters for sequence classification data sets."""
def get_train_examples(self, data_dir):
"""Gets a collection of `InputExample`s for the train set."""
raise NotImplementedError()
def get_dev_examples(self, data_dir):
"""Gets a collection of `InputExample`s for the dev set."""
raise NotImplementedError()
def get_test_examples(self, data_dir):
"""Gets a collection of `InputExample`s for prediction."""
raise NotImplementedError()
def get_labels(self):
"""Gets the list of labels for this data set."""
raise NotImplementedError()
@classmethod
def _read_tsv(cls, input_file, quotechar=None):
"""Reads a tab separated value file."""
with tf.gfile.Open(input_file, "r") as f:
reader = csv.reader(f, delimiter="\t", quotechar=quotechar)
lines = []
for line in reader:
lines.append(line)
return lines
class XnliProcessor(DataProcessor):
"""Processor for the XNLI data set."""
def __init__(self):
self.language = "zh"
def get_train_examples(self, data_dir):
"""See base class."""
lines = self._read_tsv(
os.path.join(data_dir, "multinli",
"multinli.train.%s.tsv" % self.language))
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "train-%d" % (i)
text_a = tokenization.convert_to_unicode(line[0])
text_b = tokenization.convert_to_unicode(line[1])
label = tokenization.convert_to_unicode(line[2])
if label == tokenization.convert_to_unicode("contradictory"):
label = tokenization.convert_to_unicode("contradiction")
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
def get_dev_examples(self, data_dir):
"""See base class."""
lines = self._read_tsv(os.path.join(data_dir, "xnli.dev.tsv"))
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "dev-%d" % (i)
language = tokenization.convert_to_unicode(line[0])
if language != tokenization.convert_to_unicode(self.language):
continue
text_a = tokenization.convert_to_unicode(line[6])
text_b = tokenization.convert_to_unicode(line[7])
label = tokenization.convert_to_unicode(line[1])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
def get_labels(self):
"""See base class."""
return ["contradiction", "entailment", "neutral"]
class MnliProcessor(DataProcessor):
"""Processor for the MultiNLI data set (GLUE version)."""
def get_train_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "train.tsv")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "dev_matched.tsv")),
"dev_matched")
def get_test_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test_matched.tsv")), "test")
def get_labels(self):
"""See base class."""
return ["contradiction", "entailment", "neutral"]
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, tokenization.convert_to_unicode(line[0]))
text_a = tokenization.convert_to_unicode(line[8])
text_b = tokenization.convert_to_unicode(line[9])
if set_type == "test":
label = "contradiction"
else:
label = tokenization.convert_to_unicode(line[-1])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
class MrpcProcessor(DataProcessor):
"""Processor for the MRPC data set (GLUE version)."""
def get_train_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "train.tsv")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "dev.tsv")), "dev")
def get_test_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test.tsv")), "test")
def get_labels(self):
"""See base class."""
return ["0", "1"]
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, i)
text_a = tokenization.convert_to_unicode(line[3])
text_b = tokenization.convert_to_unicode(line[4])
if set_type == "test":
label = "0"
else:
label = tokenization.convert_to_unicode(line[0])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
class ColaProcessor(DataProcessor):
"""Processor for the CoLA data set (GLUE version)."""
def get_train_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "train.tsv")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "dev.tsv")), "dev")
def get_test_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test.tsv")), "test")
def get_labels(self):
"""See base class."""
return ["0", "1"]
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
# Only the test set has a header
if set_type == "test" and i == 0:
continue
guid = "%s-%s" % (set_type, i)
if set_type == "test":
text_a = tokenization.convert_to_unicode(line[1])
label = "0"
else:
text_a = tokenization.convert_to_unicode(line[3])
label = tokenization.convert_to_unicode(line[1])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=None, label=label))
return examples
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer):
"""Converts a single `InputExample` into a single `InputFeatures`."""
label_map = {}
for (i, label) in enumerate(label_list):
label_map[label] = i
tokens_a = tokenizer.tokenize(example.text_a)
tokens_b = None
if example.text_b:
tokens_b = tokenizer.tokenize(example.text_b)
if tokens_b:
# Modifies `tokens_a` and `tokens_b` in place so that the total
# length is less than the specified length.
# Account for [CLS], [SEP], [SEP] with "- 3"
_truncate_seq_pair(tokens_a, tokens_b, max_seq_length - 3)
else:
# Account for [CLS] and [SEP] with "- 2"
if len(tokens_a) > max_seq_length - 2:
tokens_a = tokens_a[0:(max_seq_length - 2)]
# The convention in BERT is:
# (a) For sequence pairs:
# tokens: [CLS] is this jack ##son ##ville ? [SEP] no it is not . [SEP]
# type_ids: 0 0 0 0 0 0 0 0 1 1 1 1 1 1
# (b) For single sequences:
# tokens: [CLS] the dog is hairy . [SEP]
# type_ids: 0 0 0 0 0 0 0
#
# Where "type_ids" are used to indicate whether this is the first
# sequence or the second sequence. The embedding vectors for `type=0` and
# `type=1` were learned during pre-training and are added to the wordpiece
# embedding vector (and position vector). This is not *strictly* necessary
# since the [SEP] token unambiguously separates the sequences, but it makes
# it easier for the model to learn the concept of sequences.
#
# For classification tasks, the first vector (corresponding to [CLS]) is
# used as as the "sentence vector". Note that this only makes sense because
# the entire model is fine-tuned.
tokens = []
segment_ids = []
tokens.append("[CLS]")
segment_ids.append(0)
for token in tokens_a:
tokens.append(token)
segment_ids.append(0)
tokens.append("[SEP]")
segment_ids.append(0)
if tokens_b:
for token in tokens_b:
tokens.append(token)
segment_ids.append(1)
tokens.append("[SEP]")
segment_ids.append(1)
input_ids = tokenizer.convert_tokens_to_ids(tokens)
# The mask has 1 for real tokens and 0 for padding tokens. Only real
# tokens are attended to.
input_mask = [1] * len(input_ids)
# Zero-pad up to the sequence length.
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
label_id = label_map[example.label]
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("guid: %s" % (example.guid))
tf.logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
tf.logging.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" % " ".join([str(x) for x in segment_ids]))
tf.logging.info("label: %s (id = %d)" % (example.label, label_id))
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_id=label_id)
return feature
def file_based_convert_examples_to_features(
examples, label_list, max_seq_length, tokenizer, output_file):
"""Convert a set of `InputExample`s to a TFRecord file."""
writer = tf.python_io.TFRecordWriter(output_file)
for (ex_index, example) in enumerate(examples):
if ex_index % 10000 == 0:
tf.logging.info("Writing example %d of %d" % (ex_index, len(examples)))
feature = convert_single_example(ex_index, example, label_list,
max_seq_length, tokenizer)
def create_int_feature(values):
f = tf.train.Feature(int64_list=tf.train.Int64List(value=list(values)))
return f
features = collections.OrderedDict()
features["input_ids"] = create_int_feature(feature.input_ids)
features["input_mask"] = create_int_feature(feature.input_mask)
features["segment_ids"] = create_int_feature(feature.segment_ids)
features["label_ids"] = create_int_feature([feature.label_id])
tf_example = tf.train.Example(features=tf.train.Features(feature=features))
writer.write(tf_example.SerializeToString())
def file_based_input_fn_builder(input_file, seq_length, is_training,
drop_remainder):
"""Creates an `input_fn` closure to be passed to TPUEstimator."""
name_to_features = {
"input_ids": tf.FixedLenFeature([seq_length], tf.int64),
"input_mask": tf.FixedLenFeature([seq_length], tf.int64),
"segment_ids": tf.FixedLenFeature([seq_length], tf.int64),
"label_ids": tf.FixedLenFeature([], tf.int64),
}
def _decode_record(record, name_to_features):
"""Decodes a record to a TensorFlow example."""
example = tf.parse_single_example(record, name_to_features)
# tf.Example only supports tf.int64, but the TPU only supports tf.int32.
# So cast all int64 to int32.
for name in list(example.keys()):
t = example[name]
if t.dtype == tf.int64:
t = tf.to_int32(t)
example[name] = t
return example
def input_fn(params):
"""The actual input function."""
batch_size = params["batch_size"]
# For training, we want a lot of parallel reading and shuffling.
# For eval, we want no shuffling and parallel reading doesn't matter.
d = tf.data.TFRecordDataset(input_file)
if is_training:
d = d.repeat()
d = d.shuffle(buffer_size=100)
d = d.apply(
tf.contrib.data.map_and_batch(
lambda record: _decode_record(record, name_to_features),
batch_size=batch_size,
drop_remainder=drop_remainder))
return d
return input_fn
def _truncate_seq_pair(tokens_a, tokens_b, max_length):
"""Truncates a sequence pair in place to the maximum length."""
# This is a simple heuristic which will always truncate the longer sequence
# one token at a time. This makes more sense than truncating an equal percent
# of tokens from each, since if one sequence is very short then each token
# that's truncated likely contains more information than a longer sequence.
while True:
total_length = len(tokens_a) + len(tokens_b)
if total_length <= max_length:
break
if len(tokens_a) > len(tokens_b):
tokens_a.pop()
else:
tokens_b.pop()
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
labels, num_labels, use_one_hot_embeddings):
"""Creates a classification model."""
model = modeling.BertModel(
config=bert_config,
is_training=is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=segment_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
# In the demo, we are doing a simple classification task on the entire
# segment.
#
# If you want to use the token-level output, use model.get_sequence_output()
# instead.
output_layer = model.get_pooled_output()
hidden_size = output_layer.shape[-1].value
output_weights = tf.get_variable(
"output_weights", [num_labels, hidden_size],
initializer=tf.truncated_normal_initializer(stddev=0.02))
output_bias = tf.get_variable(
"output_bias", [num_labels], initializer=tf.zeros_initializer())
with tf.variable_scope("loss"):
if is_training:
# I.e., 0.1 dropout
output_layer = tf.nn.dropout(output_layer, keep_prob=0.9)
logits = tf.matmul(output_layer, output_weights, transpose_b=True)
logits = tf.nn.bias_add(logits, output_bias)
probabilities = tf.nn.softmax(logits, axis=-1)
log_probs = tf.nn.log_softmax(logits, axis=-1)
one_hot_labels = tf.one_hot(labels, depth=num_labels, dtype=tf.float32)
per_example_loss = -tf.reduce_sum(one_hot_labels * log_probs, axis=-1)
loss = tf.reduce_mean(per_example_loss)
return (loss, per_example_loss, logits, probabilities)
def model_fn_builder(bert_config, num_labels, init_checkpoint, learning_rate,
num_train_steps, num_warmup_steps, use_tpu,
use_one_hot_embeddings):
"""Returns `model_fn` closure for TPUEstimator."""
def model_fn(features, labels, mode, params): # pylint: disable=unused-argument
"""The `model_fn` for TPUEstimator."""
tf.logging.info("*** Features ***")
for name in sorted(features.keys()):
tf.logging.info(" name = %s, shape = %s" % (name, features[name].shape))
input_ids = features["input_ids"]
input_mask = features["input_mask"]
segment_ids = features["segment_ids"]
label_ids = features["label_ids"]
is_training = (mode == tf.estimator.ModeKeys.TRAIN)
(total_loss, per_example_loss, logits, probabilities) = create_model(
bert_config, is_training, input_ids, input_mask, segment_ids, label_ids,
num_labels, use_one_hot_embeddings)
tvars = tf.trainable_variables()
initialized_variable_names = {}
scaffold_fn = None
if init_checkpoint:
(assignment_map, initialized_variable_names
) = modeling.get_assignment_map_from_checkpoint(tvars, init_checkpoint)
if use_tpu:
def tpu_scaffold():
tf.train.init_from_checkpoint(init_checkpoint, assignment_map)
return tf.train.Scaffold()
scaffold_fn = tpu_scaffold
else:
tf.train.init_from_checkpoint(init_checkpoint, assignment_map)
tf.logging.info("**** Trainable Variables ****")
for var in tvars:
init_string = ""
if var.name in initialized_variable_names:
init_string = ", *INIT_FROM_CKPT*"
tf.logging.info(" name = %s, shape = %s%s", var.name, var.shape,
init_string)
output_spec = None
if mode == tf.estimator.ModeKeys.TRAIN:
train_op = optimization.create_optimizer(
total_loss, learning_rate, num_train_steps, num_warmup_steps, use_tpu)
output_spec = tf.contrib.tpu.TPUEstimatorSpec(
mode=mode,
loss=total_loss,
train_op=train_op,
scaffold_fn=scaffold_fn)
elif mode == tf.estimator.ModeKeys.EVAL:
def metric_fn(per_example_loss, label_ids, logits):
predictions = tf.argmax(logits, axis=-1, output_type=tf.int32)
accuracy = tf.metrics.accuracy(label_ids, predictions)
loss = tf.metrics.mean(per_example_loss)
return {
"eval_accuracy": accuracy,
"eval_loss": loss,
}
eval_metrics = (metric_fn, [per_example_loss, label_ids, logits])
output_spec = tf.contrib.tpu.TPUEstimatorSpec(
mode=mode,
loss=total_loss,
eval_metrics=eval_metrics,
scaffold_fn=scaffold_fn)
else:
output_spec = tf.contrib.tpu.TPUEstimatorSpec(
mode=mode, predictions=probabilities, scaffold_fn=scaffold_fn)
return output_spec
return model_fn
# This function is not used by this file but is still used by the Colab and
# people who depend on it.
def input_fn_builder(features, seq_length, is_training, drop_remainder):
"""Creates an `input_fn` closure to be passed to TPUEstimator."""
all_input_ids = []
all_input_mask = []
all_segment_ids = []
all_label_ids = []
for feature in features:
all_input_ids.append(feature.input_ids)
all_input_mask.append(feature.input_mask)
all_segment_ids.append(feature.segment_ids)
all_label_ids.append(feature.label_id)
def input_fn(params):
"""The actual input function."""
batch_size = params["batch_size"]
num_examples = len(features)
# This is for demo purposes and does NOT scale to large data sets. We do
# not use Dataset.from_generator() because that uses tf.py_func which is
# not TPU compatible. The right way to load data is with TFRecordReader.
d = tf.data.Dataset.from_tensor_slices({
"input_ids":
tf.constant(
all_input_ids, shape=[num_examples, seq_length],
dtype=tf.int32),
"input_mask":
tf.constant(
all_input_mask,
shape=[num_examples, seq_length],
dtype=tf.int32),
"segment_ids":
tf.constant(
all_segment_ids,
shape=[num_examples, seq_length],
dtype=tf.int32),
"label_ids":
tf.constant(all_label_ids, shape=[num_examples], dtype=tf.int32),
})
if is_training:
d = d.repeat()
d = d.shuffle(buffer_size=100)
d = d.batch(batch_size=batch_size, drop_remainder=drop_remainder)
return d
return input_fn
# This function is not used by this file but is still used by the Colab and
# people who depend on it.
def convert_examples_to_features(examples, label_list, max_seq_length,
tokenizer):
"""Convert a set of `InputExample`s to a list of `InputFeatures`."""
features = []
for (ex_index, example) in enumerate(examples):
if ex_index % 10000 == 0:
tf.logging.info("Writing example %d of %d" % (ex_index, len(examples)))
feature = convert_single_example(ex_index, example, label_list,
max_seq_length, tokenizer)
features.append(feature)
return features
def main(_):
tf.logging.set_verbosity(tf.logging.INFO)
processors = {
"cola": ColaProcessor,
"mnli": MnliProcessor,
"mrpc": MrpcProcessor,
"xnli": XnliProcessor,
}
if not FLAGS.do_train and not FLAGS.do_eval and not FLAGS.do_predict:
raise ValueError(
"At least one of `do_train`, `do_eval` or `do_predict' must be True.")
bert_config = modeling.BertConfig.from_json_file(FLAGS.bert_config_file)
if FLAGS.max_seq_length > bert_config.max_position_embeddings:
raise ValueError(
"Cannot use sequence length %d because the BERT model "
"was only trained up to sequence length %d" %
(FLAGS.max_seq_length, bert_config.max_position_embeddings))
tf.gfile.MakeDirs(FLAGS.output_dir)
task_name = FLAGS.task_name.lower()
if task_name not in processors:
raise ValueError("Task not found: %s" % (task_name))
processor = processors[task_name]()
label_list = processor.get_labels()
tokenizer = tokenization.FullTokenizer(
vocab_file=FLAGS.vocab_file, do_lower_case=FLAGS.do_lower_case)
tpu_cluster_resolver = None
if FLAGS.use_tpu and FLAGS.tpu_name:
tpu_cluster_resolver = tf.contrib.cluster_resolver.TPUClusterResolver(
FLAGS.tpu_name, zone=FLAGS.tpu_zone, project=FLAGS.gcp_project)
is_per_host = tf.contrib.tpu.InputPipelineConfig.PER_HOST_V2
run_config = tf.contrib.tpu.RunConfig(
cluster=tpu_cluster_resolver,
master=FLAGS.master,
model_dir=FLAGS.output_dir,
save_checkpoints_steps=FLAGS.save_checkpoints_steps,
tpu_config=tf.contrib.tpu.TPUConfig(
iterations_per_loop=FLAGS.iterations_per_loop,
num_shards=FLAGS.num_tpu_cores,
per_host_input_for_training=is_per_host))
train_examples = None
num_train_steps = None
num_warmup_steps = None
if FLAGS.do_train:
train_examples = processor.get_train_examples(FLAGS.data_dir)
num_train_steps = int(
len(train_examples) / FLAGS.train_batch_size * FLAGS.num_train_epochs)
num_warmup_steps = int(num_train_steps * FLAGS.warmup_proportion)
model_fn = model_fn_builder(
bert_config=bert_config,
num_labels=len(label_list),
init_checkpoint=FLAGS.init_checkpoint,
learning_rate=FLAGS.learning_rate,
num_train_steps=num_train_steps,
num_warmup_steps=num_warmup_steps,
use_tpu=FLAGS.use_tpu,
use_one_hot_embeddings=FLAGS.use_tpu)
# If TPU is not available, this will fall back to normal Estimator on CPU
# or GPU.
estimator = tf.contrib.tpu.TPUEstimator(
use_tpu=FLAGS.use_tpu,
model_fn=model_fn,
config=run_config,
train_batch_size=FLAGS.train_batch_size,
eval_batch_size=FLAGS.eval_batch_size,
predict_batch_size=FLAGS.predict_batch_size)
if FLAGS.do_train:
train_file = os.path.join(FLAGS.output_dir, "train.tf_record")
file_based_convert_examples_to_features(
train_examples, label_list, FLAGS.max_seq_length, tokenizer, train_file)
tf.logging.info("***** Running training *****")
tf.logging.info(" Num examples = %d", len(train_examples))
tf.logging.info(" Batch size = %d", FLAGS.train_batch_size)
tf.logging.info(" Num steps = %d", num_train_steps)
train_input_fn = file_based_input_fn_builder(
input_file=train_file,
seq_length=FLAGS.max_seq_length,
is_training=True,
drop_remainder=True)
estimator.train(input_fn=train_input_fn, max_steps=num_train_steps)
if FLAGS.do_eval:
eval_examples = processor.get_dev_examples(FLAGS.data_dir)
eval_file = os.path.join(FLAGS.output_dir, "eval.tf_record")
file_based_convert_examples_to_features(
eval_examples, label_list, FLAGS.max_seq_length, tokenizer, eval_file)
tf.logging.info("***** Running evaluation *****")
tf.logging.info(" Num examples = %d", len(eval_examples))
tf.logging.info(" Batch size = %d", FLAGS.eval_batch_size)
# This tells the estimator to run through the entire set.
eval_steps = None
# However, if running eval on the TPU, you will need to specify the
# number of steps.
if FLAGS.use_tpu:
# Eval will be slightly WRONG on the TPU because it will truncate
# the last batch.
eval_steps = int(len(eval_examples) / FLAGS.eval_batch_size)
eval_drop_remainder = True if FLAGS.use_tpu else False
eval_input_fn = file_based_input_fn_builder(
input_file=eval_file,
seq_length=FLAGS.max_seq_length,
is_training=False,
drop_remainder=eval_drop_remainder)
result = estimator.evaluate(input_fn=eval_input_fn, steps=eval_steps)
output_eval_file = os.path.join(FLAGS.output_dir, "eval_results.txt")
with tf.gfile.GFile(output_eval_file, "w") as writer:
tf.logging.info("***** Eval results *****")
for key in sorted(result.keys()):
tf.logging.info(" %s = %s", key, str(result[key]))
writer.write("%s = %s\n" % (key, str(result[key])))
if FLAGS.do_predict:
predict_examples = processor.get_test_examples(FLAGS.data_dir)
predict_file = os.path.join(FLAGS.output_dir, "predict.tf_record")
file_based_convert_examples_to_features(predict_examples, label_list,
FLAGS.max_seq_length, tokenizer,
predict_file)
tf.logging.info("***** Running prediction*****")
tf.logging.info(" Num examples = %d", len(predict_examples))
tf.logging.info(" Batch size = %d", FLAGS.predict_batch_size)
if FLAGS.use_tpu:
# Warning: According to tpu_estimator.py Prediction on TPU is an
# experimental feature and hence not supported here
raise ValueError("Prediction in TPU not supported")
predict_drop_remainder = True if FLAGS.use_tpu else False
predict_input_fn = file_based_input_fn_builder(
input_file=predict_file,
seq_length=FLAGS.max_seq_length,
is_training=False,
drop_remainder=predict_drop_remainder)
result = estimator.predict(input_fn=predict_input_fn)
output_predict_file = os.path.join(FLAGS.output_dir, "test_results.tsv")
with tf.gfile.GFile(output_predict_file, "w") as writer:
tf.logging.info("***** Predict results *****")
for prediction in result:
output_line = "\t".join(
str(class_probability) for class_probability in prediction) + "\n"
writer.write(output_line)
if __name__ == "__main__":
flags.mark_flag_as_required("data_dir")
flags.mark_flag_as_required("task_name")
flags.mark_flag_as_required("vocab_file")
flags.mark_flag_as_required("bert_config_file")
flags.mark_flag_as_required("output_dir")
tf.app.run()
================================================
FILE: run_lm_predict.py
================================================
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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.
"""BERT language model predict."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import json
import modeling
import tokenization
import numpy as np
import tensorflow as tf
flags = tf.flags
FLAGS = flags.FLAGS
flags.DEFINE_integer("max_predictions_per_seq", 20,
"In this task, it also refers to maximum number of masked tokens per word.")
flags.DEFINE_string(
"bert_config_file", None,
"The config json file corresponding to the pre-trained BERT model. "
"This specifies the model architecture.")
flags.DEFINE_string(
"input_file", None,
"The config json file corresponding to the pre-trained BERT model. "
"This specifies the model architecture.")
flags.DEFINE_string(
"output_dir", None,
"The output directory where the model checkpoints will be written.")
flags.DEFINE_string("vocab_file", None,
"The vocabulary file that the BERT model was trained on.")
## Other parameters
flags.DEFINE_string(
"init_checkpoint", None,
"Initial checkpoint (usually from a pre-trained BERT model).")
flags.DEFINE_bool(
"do_lower_case", True,
"Whether to lower case the input text. Should be True for uncased "
"models and False for cased models.")
flags.DEFINE_integer(
"max_seq_length", 128,
"The maximum total input sequence length after WordPiece tokenization. "
"Sequences longer than this will be truncated, and sequences shorter "
"than this will be padded.")
flags.DEFINE_integer("predict_batch_size", 8, "Total batch size for predict.")
flags.DEFINE_bool("use_tpu", False, "Whether to use TPU or GPU/CPU.")
tf.flags.DEFINE_string(
"tpu_name", None,
"The Cloud TPU to use for training. This should be either the name "
"used when creating the Cloud TPU, or a grpc://ip.address.of.tpu:8470 "
"url.")
tf.flags.DEFINE_string(
"tpu_zone", None,
"[Optional] GCE zone where the Cloud TPU is located in. If not "
"specified, we will attempt to automatically detect the GCE project from "
"metadata.")
tf.flags.DEFINE_string(
"gcp_project", None,
"[Optional] Project name for the Cloud TPU-enabled project. If not "
"specified, we will attempt to automatically detect the GCE project from "
"metadata.")
tf.flags.DEFINE_string("master", None, "[Optional] TensorFlow master URL.")
flags.DEFINE_integer(
"num_tpu_cores", 8,
"Only used if `use_tpu` is True. Total number of TPU cores to use.")
class InputExample(object):
def __init__(self, unique_id, text):
self.unique_id = unique_id
self.text = text
def read_examples(input_file):
"""Read a list of `InputExample`s from an input file."""
examples = []
unique_id = 0
with tf.gfile.GFile(input_file, "r") as reader:
while True:
line = tokenization.convert_to_unicode(reader.readline())
if not line:
break
line = line.strip()
unique_id += 1
examples.append(
InputExample(unique_id, line))
unique_id += 1
return examples
def model_fn_builder(bert_config, init_checkpoint, use_tpu,
use_one_hot_embeddings):
"""Returns `model_fn` closure for TPUEstimator."""
def model_fn(features, mode, params): # pylint: disable=unused-argument
"""The `model_fn` for TPUEstimator."""
tf.logging.info("*** Features ***")
for name in sorted(features.keys()):
tf.logging.info(" name = %s, shape = %s" % (name, features[name].shape))
input_ids = features["input_ids"]
input_mask = features["input_mask"]
segment_ids = features["segment_ids"]
masked_lm_positions = features["masked_lm_positions"]
masked_lm_ids = features["masked_lm_ids"]
model = modeling.BertModel(
config=bert_config,
is_training=False,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=segment_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
masked_lm_example_loss = get_masked_lm_output(
bert_config, model.get_sequence_output(), model.get_embedding_table(),
masked_lm_positions, masked_lm_ids)
tvars = tf.trainable_variables()
initialized_variable_names = {}
scaffold_fn = None
if init_checkpoint:
(assignment_map, initialized_variable_names
) = modeling.get_assignment_map_from_checkpoint(tvars, init_checkpoint)
if use_tpu:
def tpu_scaffold():
tf.train.init_from_checkpoint(init_checkpoint, assignment_map)
return tf.train.Scaffold()
scaffold_fn = tpu_scaffold
else:
tf.train.init_from_checkpoint(init_checkpoint, assignment_map)
tf.logging.info("**** Trainable Variables ****")
for var in tvars:
init_string = ""
if var.name in initialized_variable_names:
init_string = ", *INIT_FROM_CKPT*"
tf.logging.info(" name = %s, shape = %s%s", var.name, var.shape,
init_string)
output_spec = None
if mode == tf.estimator.ModeKeys.PREDICT:
output_spec = tf.contrib.tpu.TPUEstimatorSpec(
mode=mode, predictions=masked_lm_example_loss, scaffold_fn=scaffold_fn) # 输出mask_word的score
return output_spec
return model_fn
def get_masked_lm_output(bert_config, input_tensor, output_weights, positions,
label_ids):
"""Get loss and log probs for the masked LM."""
input_tensor = gather_indexes(input_tensor, positions)
with tf.variable_scope("cls/predictions"):
# We apply one more non-linear transformation before the output layer.
# This matrix is not used after pre-training.
with tf.variable_scope("transform"):
input_tensor = tf.layers.dense(
input_tensor,
units=bert_config.hidden_size,
activation=modeling.get_activation(bert_config.hidden_act),
kernel_initializer=modeling.create_initializer(
bert_config.initializer_range))
input_tensor = modeling.layer_norm(input_tensor)
# The output weights are the same as the input embeddings, but there is
# an output-only bias for each token.
output_bias = tf.get_variable(
"output_bias",
shape=[bert_config.vocab_size],
initializer=tf.zeros_initializer())
logits = tf.matmul(input_tensor, output_weights, transpose_b=True)
logits = tf.nn.bias_add(logits, output_bias)
log_probs = tf.nn.log_softmax(logits, axis=-1)
label_ids = tf.reshape(label_ids, [-1])
one_hot_labels = tf.one_hot(
label_ids, depth=bert_config.vocab_size, dtype=tf.float32)
per_example_loss = -tf.reduce_sum(log_probs * one_hot_labels, axis=[-1])
loss = tf.reshape(per_example_loss, [-1, tf.shape(positions)[1]])
# TODO: dynamic gather from per_example_loss
return loss
def gather_indexes(sequence_tensor, positions):
"""Gathers the vectors at the specific positions over a minibatch."""
sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
batch_size = sequence_shape[0]
seq_length = sequence_shape[1]
width = sequence_shape[2]
flat_offsets = tf.reshape(
tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
flat_positions = tf.reshape(positions + flat_offsets, [-1])
flat_sequence_tensor = tf.reshape(sequence_tensor,
[batch_size * seq_length, width])
output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
return output_tensor
def input_fn_builder(features, seq_length, max_predictions_per_seq):
"""Creates an `input_fn` closure to be passed to TPUEstimator."""
all_input_ids = []
all_input_mask = []
all_segment_ids = []
all_masked_lm_positions = []
all_masked_lm_ids = []
for feature in features:
all_input_ids.append(feature.input_ids)
all_input_mask.append(feature.input_mask)
all_segment_ids.append(feature.segment_ids)
all_masked_lm_positions.append(feature.masked_lm_positions)
all_masked_lm_ids.append(feature.masked_lm_ids)
def input_fn(params):
"""The actual input function."""
batch_size = params["batch_size"]
num_examples = len(features)
# This is for demo purposes and does NOT scale to large data sets. We do
# not use Dataset.from_generator() because that uses tf.py_func which is
# not TPU compatible. The right way to load data is with TFRecordReader.
d = tf.data.Dataset.from_tensor_slices({
"input_ids":
tf.constant(
all_input_ids, shape=[num_examples, seq_length],
dtype=tf.int32),
"input_mask":
tf.constant(
all_input_mask,
shape=[num_examples, seq_length],
dtype=tf.int32),
"segment_ids":
tf.constant(
all_segment_ids,
shape=[num_examples, seq_length],
dtype=tf.int32),
"masked_lm_positions":
tf.constant(
all_masked_lm_positions,
shape=[num_examples, max_predictions_per_seq],
dtype=tf.int32),
"masked_lm_ids":
tf.constant(
all_masked_lm_ids,
shape=[num_examples, max_predictions_per_seq],
dtype=tf.int32)
})
d = d.batch(batch_size=batch_size, drop_remainder=False)
return d
return input_fn
# This function is not used by this file but is still used by the Colab and
# people who depend on it.
def convert_examples_to_features(examples, max_seq_length, tokenizer):
"""Convert a set of `InputExample`s to a list of `InputFeatures`."""
all_features = []
all_tokens = []
for (ex_index, example) in enumerate(examples):
if ex_index % 10000 == 0:
tf.logging.info("Writing example %d of %d" % (ex_index, len(examples)))
features, tokens = convert_single_example(ex_index, example,
max_seq_length, tokenizer)
all_features.extend(features)
all_tokens.extend(tokens)
return all_features, all_tokens
tokenizer = tokenization.FullTokenizer(
vocab_file=FLAGS.vocab_file, do_lower_case=FLAGS.do_lower_case)
MASKED_TOKEN = "[MASK]"
MASKED_ID = tokenizer.convert_tokens_to_ids([MASKED_TOKEN])[0]
def create_masked_lm_prediction(input_ids, mask_position, mask_count=1):
new_input_ids = list(input_ids)
masked_lm_labels = []
masked_lm_positions = list(range(mask_position, mask_position + mask_count))
for i in masked_lm_positions:
new_input_ids[i] = MASKED_ID
masked_lm_labels.append(input_ids[i])
return new_input_ids, masked_lm_positions, masked_lm_labels
class InputFeatures(object):
"""A single set of features of data."""
def __init__(self, input_ids, segment_ids, input_mask, masked_lm_positions,
masked_lm_ids):
self.input_ids = input_ids,
self.segment_ids = segment_ids,
self.input_mask = input_mask,
self.masked_lm_positions = masked_lm_positions,
self.masked_lm_ids = masked_lm_ids,
def convert_single_example(ex_index, example, max_seq_length,
tokenizer):
"""Converts a single `InputExample` into a single `InputFeatures`."""
tokens = tokenizer.tokenize(example.text)
# Account for [CLS] and [SEP] with "- 2"
if len(tokens) > max_seq_length - 2:
tokens = tokens[0:(max_seq_length - 2)]
input_tokens = []
segment_ids = []
input_tokens.append("[CLS]")
segment_ids.append(0)
for token in tokens:
input_tokens.append(token)
segment_ids.append(0)
input_tokens.append("[SEP]")
segment_ids.append(0)
input_ids = tokenizer.convert_tokens_to_ids(input_tokens)
# The mask has 1 for real tokens and 0 for padding tokens. Only real
# tokens are attended to.
input_mask = [1] * len(input_ids)
# Zero-pad up to the sequence length.
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
if ex_index < 5:
tf.logging.info("*** Example ***")
tf.logging.info("id: %s" % (example.unique_id))
tf.logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in input_tokens]))
tf.logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
tf.logging.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
tf.logging.info("segment_ids: %s" % " ".join([str(x) for x in segment_ids]))
features = create_sequential_mask(input_tokens, input_ids, input_mask, segment_ids,
FLAGS.max_predictions_per_seq)
return features, input_tokens
def is_subtoken(x):
return x.startswith("##")
def create_sequential_mask(input_tokens, input_ids, input_mask, segment_ids,
max_predictions_per_seq):
"""Mask each token/word sequentially"""
features = []
i = 1
while i < len(input_tokens) - 1:
mask_count = 1
while is_subtoken(input_tokens[i+mask_count]):
mask_count += 1
input_ids_new, masked_lm_positions, masked_lm_labels = create_masked_lm_prediction(input_ids, i, mask_count)
while len(masked_lm_positions) < max_predictions_per_seq:
masked_lm_positions.append(0)
masked_lm_labels.append(0)
feature = InputFeatures(
input_ids=input_ids_new,
input_mask=input_mask,
segment_ids=segment_ids,
masked_lm_positions=masked_lm_positions,
masked_lm_ids=masked_lm_labels)
features.append(feature)
i += mask_count
return features
def parse_result(result, all_tokens, output_file=None):
with tf.gfile.GFile(output_file, "w") as writer:
tf.logging.info("***** Predict results *****")
i = 0
sentences = []
for word_loss in result:
# start of a sentence
if all_tokens[i] == "[CLS]":
sentence = {}
tokens = []
sentence_loss = 0.0
word_count_per_sent = 0
i += 1
# add token
tokens.append({"token": tokenization.printable_text(all_tokens[i]),
"prob": float(np.exp(-word_loss[0])) })
sentence_loss += word_loss[0]
word_count_per_sent += 1
i += 1
token_count_per_word = 0
while is_subtoken(all_tokens[i]):
token_count_per_word += 1
tokens.append({"token": tokenization.printable_text(all_tokens[i]),
"prob": float(np.exp(-word_loss[token_count_per_word]))})
sentence_loss += word_loss[token_count_per_word]
i += 1
# end of a sentence
if all_tokens[i] == "[SEP]":
sentence["tokens"] = tokens
sentence["ppl"] = float(np.exp(sentence_loss / word_count_per_sent))
sentences.append(sentence)
i += 1
if output_file is not None:
tf.logging.info("Saving results to %s" % output_file)
writer.write(json.dumps(sentences, indent=2, ensure_ascii=False))
def main(_):
tf.logging.set_verbosity(tf.logging.INFO)
bert_config = modeling.BertConfig.from_json_file(FLAGS.bert_config_file)
if FLAGS.max_seq_length > bert_config.max_position_embeddings:
raise ValueError(
"Cannot use sequence length %d because the BERT model "
"was only trained up to sequence length %d" %
(FLAGS.max_seq_length, bert_config.max_position_embeddings))
tf.gfile.MakeDirs(FLAGS.output_dir)
tpu_cluster_resolver = None
if FLAGS.use_tpu and FLAGS.tpu_name:
tpu_cluster_resolver = tf.contrib.cluster_resolver.TPUClusterResolver(
FLAGS.tpu_name, zone=FLAGS.tpu_zone, project=FLAGS.gcp_project)
is_per_host = tf.contrib.tpu.InputPipelineConfig.PER_HOST_V2
run_config = tf.contrib.tpu.RunConfig(
cluster=tpu_cluster_resolver,
master=FLAGS.master,
model_dir=FLAGS.output_dir,
tpu_config=tf.contrib.tpu.TPUConfig(
num_shards=FLAGS.num_tpu_cores,
per_host_input_for_training=is_per_host))
model_fn = model_fn_builder(
bert_config=bert_config,
init_checkpoint=FLAGS.init_checkpoint,
use_tpu=FLAGS.use_tpu,
use_one_hot_embeddings=FLAGS.use_tpu)
# If TPU is not available, this will fall back to normal Estimator on CPU
# or GPU.
estimator = tf.contrib.tpu.TPUEstimator(
use_tpu=FLAGS.use_tpu,
model_fn=model_fn,
config=run_config,
predict_batch_size=FLAGS.predict_batch_size)
predict_examples = read_examples(FLAGS.input_file)
features, all_tokens = convert_examples_to_features(predict_examples,
FLAGS.max_seq_length, tokenizer)
tf.logging.info("***** Running prediction*****")
tf.logging.info(" Num examples = %d", len(predict_examples))
tf.logging.info(" Batch size = %d", FLAGS.predict_batch_size)
if FLAGS.use_tpu:
# Warning: According to tpu_estimator.py Prediction on TPU is an
# experimental feature and hence not supported here
raise ValueError("Prediction in TPU not supported")
predict_input_fn = input_fn_builder(
features=features,
seq_length=FLAGS.max_seq_length,
max_predictions_per_seq=FLAGS.max_predictions_per_seq)
result = estimator.predict(input_fn=predict_input_fn)
output_predict_file = os.path.join(FLAGS.output_dir, "test_results.json")
parse_result(result, all_tokens, output_predict_file)
if __name__ == "__main__":
tf.app.run()
================================================
FILE: run_pretraining.py
================================================
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# 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.
"""Run masked LM/next sentence masked_lm pre-training for BERT."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import modeling
import optimization
import tensorflow as tf
flags = tf.flags
FLAGS = flags.FLAGS
## Required parameters
flags.DEFINE_string(
"bert_config_file", None,
"The config json file corresponding to the pre-trained BERT model. "
"This specifies the model architecture.")
flags.DEFINE_string(
"input_file", None,
"Input TF example files (can be a glob or comma separated).")
flags.DEFINE_string(
"output_dir", None,
"The output directory where the model checkpoints will be written.")
## Other parameters
flags.DEFINE_string(
"init_checkpoint", None,
"Initial checkpoint (usually from a pre-trained BERT model).")
flags.DEFINE_integer(
"max_seq_length", 128,
"The maximum total input sequence length after WordPiece tokenization. "
"Sequences longer than this will be truncated, and sequences shorter "
"than this will be padded. Must match data generation.")
flags.DEFINE_integer(
"max_predictions_per_seq", 20,
"Maximum number of masked LM predictions per sequence. "
"Must match data generation.")
flags.DEFINE_bool("do_train", False, "Whether to run training.")
flags.DEFINE_bool("do_eval", False, "Whether to run eval on the dev set.")
flags.DEFINE_integer("train_batch_size", 32, "Total batch size for training.")
flags.DEFINE_integer("eval_batch_size", 8, "Total batch size for eval.")
flags.DEFINE_float("learning_rate", 5e-5, "The initial learning rate for Adam.")
flags.DEFINE_integer("num_train_steps", 100000, "Number of training steps.")
flags.DEFINE_integer("num_warmup_steps", 10000, "Number of warmup steps.")
flags.DEFINE_integer("save_checkpoints_steps", 1000,
"How often to save the model checkpoint.")
flags.DEFINE_integer("iterations_per_loop", 1000,
"How many steps to make in each estimator call.")
flags.DEFINE_integer("max_eval_steps", 100, "Maximum number of eval steps.")
flags.DEFINE_bool("use_tpu", False, "Whether to use TPU or GPU/CPU.")
tf.flags.DEFINE_string(
"tpu_name", None,
"The Cloud TPU to use for training. This should be either the name "
"used when creating the Cloud TPU, or a grpc://ip.address.of.tpu:8470 "
"url.")
tf.flags.DEFINE_string(
"tpu_zone", None,
"[Optional] GCE zone where the Cloud TPU is located in. If not "
"specified, we will attempt to automatically detect the GCE project from "
"metadata.")
tf.flags.DEFINE_string(
"gcp_project", None,
"[Optional] Project name for the Cloud TPU-enabled project. If not "
"specified, we will attempt to automatically detect the GCE project from "
"metadata.")
tf.flags.DEFINE_string("master", None, "[Optional] TensorFlow master URL.")
flags.DEFINE_integer(
"num_tpu_cores", 8,
"Only used if `use_tpu` is True. Total number of TPU cores to use.")
def model_fn_builder(bert_config, init_checkpoint, learning_rate,
num_train_steps, num_warmup_steps, use_tpu,
use_one_hot_embeddings):
"""Returns `model_fn` closure for TPUEstimator."""
def model_fn(features, labels, mode, params): # pylint: disable=unused-argument
"""The `model_fn` for TPUEstimator."""
tf.logging.info("*** Features ***")
for name in sorted(features.keys()):
tf.logging.info(" name = %s, shape = %s" % (name, features[name].shape))
input_ids = features["input_ids"]
input_mask = features["input_mask"]
segment_ids = features["segment_ids"]
masked_lm_positions = features["masked_lm_positions"]
masked_lm_ids = features["masked_lm_ids"]
masked_lm_weights = features["masked_lm_weights"]
next_sentence_labels = features["next_sentence_labels"]
is_training = (mode == tf.estimator.ModeKeys.TRAIN)
model = modeling.BertModel(
config=bert_config,
is_training=is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=segment_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
(masked_lm_loss,
masked_lm_example_loss, masked_lm_log_probs) = get_masked_lm_output(
bert_config, model.get_sequence_output(), model.get_embedding_table(),
masked_lm_positions, masked_lm_ids, masked_lm_weights)
(next_sentence_loss, next_sentence_example_loss,
next_sentence_log_probs) = get_next_sentence_output(
bert_config, model.get_pooled_output(), next_sentence_labels)
total_loss = masked_lm_loss + next_sentence_loss
tvars = tf.trainable_variables()
initialized_variable_names = {}
scaffold_fn = None
if init_checkpoint:
(assignment_map, initialized_variable_names
) = modeling.get_assignment_map_from_checkpoint(tvars, init_checkpoint)
if use_tpu:
def tpu_scaffold():
tf.train.init_from_checkpoint(init_checkpoint, assignment_map)
return tf.train.Scaffold()
scaffold_fn = tpu_scaffold
else:
tf.train.init_from_checkpoint(init_checkpoint, assignment_map)
tf.logging.info("**** Trainable Variables ****")
for var in tvars:
init_string = ""
if var.name in initialized_variable_names:
init_string = ", *INIT_FROM_CKPT*"
tf.logging.info(" name = %s, shape = %s%s", var.name, var.shape,
init_string)
output_spec = None
if mode == tf.estimator.ModeKeys.TRAIN:
train_op = optimization.create_optimizer(
total_loss, learning_rate, num_train_steps, num_warmup_steps, use_tpu)
output_spec = tf.contrib.tpu.TPUEstimatorSpec(
mode=mode,
loss=total_loss,
train_op=train_op,
scaffold_fn=scaffold_fn)
elif mode == tf.estimator.ModeKeys.EVAL:
def metric_fn(masked_lm_example_loss, masked_lm_log_probs, masked_lm_ids,
masked_lm_weights, next_sentence_example_loss,
next_sentence_log_probs, next_sentence_labels):
"""Computes the loss and accuracy of the model."""
masked_lm_log_probs = tf.reshape(masked_lm_log_probs,
[-1, masked_lm_log_probs.shape[-1]])
masked_lm_predictions = tf.argmax(
masked_lm_log_probs, axis=-1, output_type=tf.int32)
masked_lm_example_loss = tf.reshape(masked_lm_example_loss, [-1])
masked_lm_ids = tf.reshape(masked_lm_ids, [-1])
masked_lm_weights = tf.reshape(masked_lm_weights, [-1])
masked_lm_accuracy = tf.metrics.accuracy(
labels=masked_lm_ids,
predictions=masked_lm_predictions,
weights=masked_lm_weights)
masked_lm_mean_loss = tf.metrics.mean(
values=masked_lm_example_loss, weights=masked_lm_weights)
next_sentence_log_probs = tf.reshape(
next_sentence_log_probs, [-1, next_sentence_log_probs.shape[-1]])
next_sentence_predictions = tf.argmax(
next_sentence_log_probs, axis=-1, output_type=tf.int32)
next_sentence_labels = tf.reshape(next_sentence_labels, [-1])
next_sentence_accuracy = tf.metrics.accuracy(
labels=next_sentence_labels, predictions=next_sentence_predictions)
next_sentence_mean_loss = tf.metrics.mean(
values=next_sentence_example_loss)
return {
"masked_lm_accuracy": masked_lm_accuracy,
"masked_lm_loss": masked_lm_mean_loss,
"next_sentence_accuracy": next_sentence_accuracy,
"next_sentence_loss": next_sentence_mean_loss,
}
eval_metrics = (metric_fn, [
masked_lm_example_loss, masked_lm_log_probs, masked_lm_ids,
masked_lm_weights, next_sentence_example_loss,
next_sentence_log_probs, next_sentence_labels
])
output_spec = tf.contrib.tpu.TPUEstimatorSpec(
mode=mode,
loss=total_loss,
eval_metrics=eval_metrics,
scaffold_fn=scaffold_fn)
else:
raise ValueError("Only TRAIN and EVAL modes are supported: %s" % (mode))
return output_spec
return model_fn
def get_masked_lm_output(bert_config, input_tensor, output_weights, positions,
label_ids, label_weights):
"""Get loss and log probs for the masked LM."""
input_tensor = gather_indexes(input_tensor, positions)
with tf.variable_scope("cls/predictions"):
# We apply one more non-linear transformation before the output layer.
# This matrix is not used after pre-training.
with tf.variable_scope("transform"):
input_tensor = tf.layers.dense(
input_tensor,
units=bert_config.hidden_size,
activation=modeling.get_activation(bert_config.hidden_act),
kernel_initializer=modeling.create_initializer(
bert_config.initializer_range))
input_tensor = modeling.layer_norm(input_tensor)
# The output weights are the same as the input embeddings, but there is
# an output-only bias for each token.
output_bias = tf.get_variable(
"output_bias",
shape=[bert_config.vocab_size],
initializer=tf.zeros_initializer())
logits = tf.matmul(input_tensor, output_weights, transpose_b=True)
logits = tf.nn.bias_add(logits, output_bias)
log_probs = tf.nn.log_softmax(logits, axis=-1)
label_ids = tf.reshape(label_ids, [-1])
label_weights = tf.reshape(label_weights, [-1])
one_hot_labels = tf.one_hot(
label_ids, depth=bert_config.vocab_size, dtype=tf.float32)
# The `positions` tensor might be zero-padded (if the sequence is too
# short to have the maximum number of predictions). The `label_weights`
# tensor has a value of 1.0 for every real prediction and 0.0 for the
# padding predictions.
per_example_loss = -tf.reduce_sum(log_probs * one_hot_labels, axis=[-1])
numerator = tf.reduce_sum(label_weights * per_example_loss)
denominator = tf.reduce_sum(label_weights) + 1e-5
loss = numerator / denominator
return (loss, per_example_loss, log_probs)
def get_next_sentence_output(bert_config, input_tensor, labels):
"""Get loss and log probs for the next sentence prediction."""
# Simple binary classification. Note that 0 is "next sentence" and 1 is
# "random sentence". This weight matrix is not used after pre-training.
with tf.variable_scope("cls/seq_relationship"):
output_weights = tf.get_variable(
"output_weights",
shape=[2, bert_config.hidden_size],
initializer=modeling.create_initializer(bert_config.initializer_range))
output_bias = tf.get_variable(
"output_bias", shape=[2], initializer=tf.zeros_initializer())
logits = tf.matmul(input_tensor, output_weights, transpose_b=True)
logits = tf.nn.bias_add(logits, output_bias)
log_probs = tf.nn.log_softmax(logits, axis=-1)
labels = tf.reshape(labels, [-1])
one_hot_labels = tf.one_hot(labels, depth=2, dtype=tf.float32)
per_example_loss = -tf.reduce_sum(one_hot_labels * log_probs, axis=-1)
loss = tf.reduce_mean(per_example_loss)
return (loss, per_example_loss, log_probs)
def gather_indexes(sequence_tensor, positions):
"""Gathers the vectors at the specific positions over a minibatch."""
sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
batch_size = sequence_shape[0]
seq_length = sequence_shape[1]
width = sequence_shape[2]
flat_offsets = tf.reshape(
tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
flat_positions = tf.reshape(positions + flat_offsets, [-1])
flat_sequence_tensor = tf.reshape(sequence_tensor,
[batch_size * seq_length, width])
output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
return output_tensor
def input_fn_builder(input_files,
max_seq_length,
max_predictions_per_seq,
is_training,
num_cpu_threads=4):
"""Creates an `input_fn` closure to be passed to TPUEstimator."""
def input_fn(params):
"""The actual input function."""
batch_size = params["batch_size"]
name_to_features = {
"input_ids":
tf.FixedLenFeature([max_seq_length], tf.int64),
"input_mask":
tf.FixedLenFeature([max_seq_length], tf.int64),
"segment_ids":
tf.FixedLenFeature([max_seq_length], tf.int64),
"masked_lm_positions":
tf.FixedLenFeature([max_predictions_per_seq], tf.int64),
"masked_lm_ids":
tf.FixedLenFeature([max_predictions_per_seq], tf.int64),
"masked_lm_weights":
tf.FixedLenFeature([max_predictions_per_seq], tf.float32),
"next_sentence_labels":
tf.FixedLenFeature([1], tf.int64),
}
# For training, we want a lot of parallel reading and shuffling.
# For eval, we want no shuffling and parallel reading doesn't matter.
if is_training:
d = tf.data.Dataset.from_tensor_slices(tf.constant(input_files))
d = d.repeat()
d = d.shuffle(buffer_size=len(input_files))
# `cycle_length` is the number of parallel files that get read.
cycle_length = min(num_cpu_threads, len(input_files))
# `sloppy` mode means that the interleaving is not exact. This adds
# even more randomness to the training pipeline.
d = d.apply(
tf.contrib.data.parallel_interleave(
tf.data.TFRecordDataset,
sloppy=is_training,
cycle_length=cycle_length))
d = d.shuffle(buffer_size=100)
else:
d = tf.data.TFRecordDataset(input_files)
# Since we evaluate for a fixed number of steps we don't want to encounter
# out-of-range exceptions.
d = d.repeat()
# We must `drop_remainder` on training because the TPU requires fixed
# size dimensions. For eval, we assume we are evaluating on the CPU or GPU
# and we *don't* want to drop the remainder, otherwise we wont cover
# every sample.
d = d.apply(
tf.contrib.data.map_and_batch(
lambda record: _decode_record(record, name_to_features),
batch_size=batch_size,
num_parallel_batches=num_cpu_threads,
drop_remainder=True))
return d
return input_fn
def _decode_record(record, name_to_features):
"""Decodes a record to a TensorFlow example."""
example = tf.parse_single_example(record, name_to_features)
# tf.Example only supports tf.int64, but the TPU only supports tf.int32.
# So cast all int64 to int32.
for name in list(example.keys()):
t = example[name]
if t.dtype == tf.int64:
t = tf.to_int32(t)
example[name] = t
return example
def main(_):
tf.logging.set_verbosity(tf.logging.INFO)
if not FLAGS.do_train and not FLAGS.do_eval:
raise ValueError("At least one of `do_train` or `do_eval` must be True.")
bert_config = modeling.BertConfig.from_json_file(FLAGS.bert_config_file)
tf.gfile.MakeDirs(FLAGS.output_dir)
input_files = []
for input_pattern in FLAGS.input_file.split(","):
input_files.extend(tf.gfile.Glob(input_pattern))
tf.logging.info("*** Input Files ***")
for input_file in input_files:
tf.logging.info(" %s" % input_file)
tpu_cluster_resolver = None
if FLAGS.use_tpu and FLAGS.tpu_name:
tpu_cluster_resolver = tf.contrib.cluster_resolver.TPUClusterResolver(
FLAGS.tpu_name, zone=FLAGS.tpu_zone, project=FLAGS.gcp_project)
is_per_host = tf.contrib.tpu.InputPipelineConfig.PER_HOST_V2
run_config = tf.contrib.tpu.RunConfig(
cluster=tpu_cluster_resolver,
master=FLAGS.master,
model_dir=FLAGS.output_dir,
save_checkpoints_steps=FLAGS.save_checkpoints_steps,
tpu_config=tf.contrib.tpu.TPUConfig(
iterations_per_loop=FLAGS.iterations_per_loop,
num_shards=FLAGS.num_tpu_cores,
per_host_input_for_training=is_per_host))
model_fn = model_fn_builder(
bert_config=bert_config
gitextract_lmgwb86k/ ├── .gitignore ├── CONTRIBUTING.md ├── LICENSE ├── README.md ├── __init__.py ├── cases/ │ └── test.zh.md ├── create_pretraining_data.py ├── data/ │ └── lm/ │ ├── result.zh.json │ ├── test.en.tsv │ └── test.zh.tsv ├── extract_features.py ├── modeling.py ├── modeling_test.py ├── multilingual.md ├── optimization.py ├── optimization_test.py ├── requirements.txt ├── run_classifier.py ├── run_lm_predict.py ├── run_pretraining.py ├── run_squad.py ├── sample_text.txt ├── tokenization.py └── tokenization_test.py
SYMBOL INDEX (195 symbols across 12 files)
FILE: create_pretraining_data.py
class TrainingInstance (line 65) | class TrainingInstance(object):
method __init__ (line 68) | def __init__(self, tokens, segment_ids, masked_lm_positions, masked_lm...
method __str__ (line 76) | def __str__(self):
method __repr__ (line 89) | def __repr__(self):
function write_instance_to_example_files (line 93) | def write_instance_to_example_files(instances, tokenizer, max_seq_length,
function create_int_feature (line 166) | def create_int_feature(values):
function create_float_feature (line 171) | def create_float_feature(values):
function create_training_instances (line 176) | def create_training_instances(input_files, tokenizer, max_seq_length,
function create_instances_from_document (line 220) | def create_instances_from_document(
function create_masked_lm_predictions (line 335) | def create_masked_lm_predictions(tokens, masked_lm_prob,
function truncate_seq_pair (line 390) | def truncate_seq_pair(tokens_a, tokens_b, max_num_tokens, rng):
function main (line 408) | def main(_):
FILE: extract_features.py
class InputExample (line 81) | class InputExample(object):
method __init__ (line 83) | def __init__(self, unique_id, text_a, text_b):
class InputFeatures (line 89) | class InputFeatures(object):
method __init__ (line 92) | def __init__(self, unique_id, tokens, input_ids, input_mask, input_typ...
function input_fn_builder (line 100) | def input_fn_builder(features, seq_length):
function model_fn_builder (line 148) | def model_fn_builder(bert_config, init_checkpoint, layer_indexes, use_tpu,
function convert_examples_to_features (line 210) | def convert_examples_to_features(examples, seq_length, tokenizer):
function _truncate_seq_pair (line 302) | def _truncate_seq_pair(tokens_a, tokens_b, max_length):
function read_examples (line 319) | def read_examples(input_file):
function main (line 343) | def main(_):
FILE: modeling.py
class BertConfig (line 30) | class BertConfig(object):
method __init__ (line 33) | def __init__(self,
method from_dict (line 82) | def from_dict(cls, json_object):
method from_json_file (line 90) | def from_json_file(cls, json_file):
method to_dict (line 96) | def to_dict(self):
method to_json_string (line 101) | def to_json_string(self):
class BertModel (line 106) | class BertModel(object):
method __init__ (line 130) | def __init__(self,
method get_pooled_output (line 235) | def get_pooled_output(self):
method get_sequence_output (line 238) | def get_sequence_output(self):
method get_all_encoder_layers (line 247) | def get_all_encoder_layers(self):
method get_embedding_output (line 250) | def get_embedding_output(self):
method get_embedding_table (line 261) | def get_embedding_table(self):
function gelu (line 265) | def gelu(input_tensor):
function get_activation (line 281) | def get_activation(activation_string):
function get_assignment_map_from_checkpoint (line 318) | def get_assignment_map_from_checkpoint(tvars, init_checkpoint):
function dropout (line 345) | def dropout(input_tensor, dropout_prob):
function layer_norm (line 363) | def layer_norm(input_tensor, name=None):
function layer_norm_and_dropout (line 369) | def layer_norm_and_dropout(input_tensor, dropout_prob, name=None):
function create_initializer (line 376) | def create_initializer(initializer_range=0.02):
function embedding_lookup (line 381) | def embedding_lookup(input_ids,
function embedding_postprocessor (line 430) | def embedding_postprocessor(input_tensor,
function create_attention_mask_from_input_mask (line 526) | def create_attention_mask_from_input_mask(from_tensor, to_mask):
function attention_layer (line 560) | def attention_layer(from_tensor,
function transformer_model (line 756) | def transformer_model(input_tensor,
function get_shape_list (line 897) | def get_shape_list(tensor, expected_rank=None, name=None):
function reshape_to_matrix (line 934) | def reshape_to_matrix(input_tensor):
function reshape_from_matrix (line 948) | def reshape_from_matrix(output_tensor, orig_shape_list):
function assert_rank (line 961) | def assert_rank(tensor, expected_rank, name=None):
FILE: modeling_test.py
class BertModelTest (line 29) | class BertModelTest(tf.test.TestCase):
class BertModelTester (line 31) | class BertModelTester(object):
method __init__ (line 33) | def __init__(self,
method create_model (line 71) | def create_model(self):
method check_output (line 114) | def check_output(self, result):
method test_default (line 126) | def test_default(self):
method test_config_to_json_string (line 129) | def test_config_to_json_string(self):
method run_tester (line 135) | def run_tester(self, tester):
method ids_tensor (line 147) | def ids_tensor(cls, shape, vocab_size, rng=None, name=None):
method assert_all_tensors_reachable (line 162) | def assert_all_tensors_reachable(self, sess, outputs):
method get_unreachable_ops (line 194) | def get_unreachable_ops(cls, graph, outputs):
method flatten_recursive (line 257) | def flatten_recursive(cls, item):
FILE: optimization.py
function create_optimizer (line 25) | def create_optimizer(loss, init_lr, num_train_steps, num_warmup_steps, u...
class AdamWeightDecayOptimizer (line 84) | class AdamWeightDecayOptimizer(tf.train.Optimizer):
method __init__ (line 87) | def __init__(self,
method apply_gradients (line 105) | def apply_gradients(self, grads_and_vars, global_step=None, name=None):
method _do_use_weight_decay (line 156) | def _do_use_weight_decay(self, param_name):
method _get_variable_name (line 166) | def _get_variable_name(self, param_name):
FILE: optimization_test.py
class OptimizationTest (line 23) | class OptimizationTest(tf.test.TestCase):
method test_adam (line 25) | def test_adam(self):
FILE: run_classifier.py
class InputExample (line 127) | class InputExample(object):
method __init__ (line 130) | def __init__(self, guid, text_a, text_b=None, label=None):
class InputFeatures (line 148) | class InputFeatures(object):
method __init__ (line 151) | def __init__(self, input_ids, input_mask, segment_ids, label_id):
class DataProcessor (line 158) | class DataProcessor(object):
method get_train_examples (line 161) | def get_train_examples(self, data_dir):
method get_dev_examples (line 165) | def get_dev_examples(self, data_dir):
method get_test_examples (line 169) | def get_test_examples(self, data_dir):
method get_labels (line 173) | def get_labels(self):
method _read_tsv (line 178) | def _read_tsv(cls, input_file, quotechar=None):
class XnliProcessor (line 188) | class XnliProcessor(DataProcessor):
method __init__ (line 191) | def __init__(self):
method get_train_examples (line 194) | def get_train_examples(self, data_dir):
method get_dev_examples (line 213) | def get_dev_examples(self, data_dir):
method get_labels (line 231) | def get_labels(self):
class MnliProcessor (line 236) | class MnliProcessor(DataProcessor):
method get_train_examples (line 239) | def get_train_examples(self, data_dir):
method get_dev_examples (line 244) | def get_dev_examples(self, data_dir):
method get_test_examples (line 250) | def get_test_examples(self, data_dir):
method get_labels (line 255) | def get_labels(self):
method _create_examples (line 259) | def _create_examples(self, lines, set_type):
class MrpcProcessor (line 277) | class MrpcProcessor(DataProcessor):
method get_train_examples (line 280) | def get_train_examples(self, data_dir):
method get_dev_examples (line 285) | def get_dev_examples(self, data_dir):
method get_test_examples (line 290) | def get_test_examples(self, data_dir):
method get_labels (line 295) | def get_labels(self):
method _create_examples (line 299) | def _create_examples(self, lines, set_type):
class ColaProcessor (line 317) | class ColaProcessor(DataProcessor):
method get_train_examples (line 320) | def get_train_examples(self, data_dir):
method get_dev_examples (line 325) | def get_dev_examples(self, data_dir):
method get_test_examples (line 330) | def get_test_examples(self, data_dir):
method get_labels (line 335) | def get_labels(self):
method _create_examples (line 339) | def _create_examples(self, lines, set_type):
function convert_single_example (line 358) | def convert_single_example(ex_index, example, label_list, max_seq_length,
function file_based_convert_examples_to_features (line 450) | def file_based_convert_examples_to_features(
function file_based_input_fn_builder (line 477) | def file_based_input_fn_builder(input_file, seq_length, is_training,
function _truncate_seq_pair (line 524) | def _truncate_seq_pair(tokens_a, tokens_b, max_length):
function create_model (line 541) | def create_model(bert_config, is_training, input_ids, input_mask, segmen...
function model_fn_builder (line 586) | def model_fn_builder(bert_config, num_labels, init_checkpoint, learning_...
function input_fn_builder (line 671) | def input_fn_builder(features, seq_length, is_training, drop_remainder):
function convert_examples_to_features (line 725) | def convert_examples_to_features(examples, label_list, max_seq_length,
function main (line 741) | def main(_):
FILE: run_lm_predict.py
class InputExample (line 97) | class InputExample(object):
method __init__ (line 98) | def __init__(self, unique_id, text):
function read_examples (line 103) | def read_examples(input_file):
function model_fn_builder (line 120) | def model_fn_builder(bert_config, init_checkpoint, use_tpu,
function get_masked_lm_output (line 185) | def get_masked_lm_output(bert_config, input_tensor, output_weights, posi...
function gather_indexes (line 223) | def gather_indexes(sequence_tensor, positions):
function input_fn_builder (line 239) | def input_fn_builder(features, seq_length, max_predictions_per_seq):
function convert_examples_to_features (line 299) | def convert_examples_to_features(examples, max_seq_length, tokenizer):
function create_masked_lm_prediction (line 322) | def create_masked_lm_prediction(input_ids, mask_position, mask_count=1):
class InputFeatures (line 332) | class InputFeatures(object):
method __init__ (line 335) | def __init__(self, input_ids, segment_ids, input_mask, masked_lm_posit...
function convert_single_example (line 344) | def convert_single_example(ex_index, example, max_seq_length,
function is_subtoken (line 393) | def is_subtoken(x):
function create_sequential_mask (line 396) | def create_sequential_mask(input_tokens, input_ids, input_mask, segment_...
function parse_result (line 422) | def parse_result(result, all_tokens, output_file=None):
function main (line 462) | def main(_):
FILE: run_pretraining.py
function model_fn_builder (line 109) | def model_fn_builder(bert_config, init_checkpoint, learning_rate,
function get_masked_lm_output (line 240) | def get_masked_lm_output(bert_config, input_tensor, output_weights, posi...
function get_next_sentence_output (line 285) | def get_next_sentence_output(bert_config, input_tensor, labels):
function gather_indexes (line 308) | def gather_indexes(sequence_tensor, positions):
function input_fn_builder (line 324) | def input_fn_builder(input_files,
function _decode_record (line 391) | def _decode_record(record, name_to_features):
function main (line 406) | def main(_):
FILE: run_squad.py
class SquadExample (line 157) | class SquadExample(object):
method __init__ (line 163) | def __init__(self,
method __str__ (line 179) | def __str__(self):
method __repr__ (line 182) | def __repr__(self):
class InputFeatures (line 197) | class InputFeatures(object):
method __init__ (line 200) | def __init__(self,
function read_squad_examples (line 227) | def read_squad_examples(input_file, is_training):
function convert_examples_to_features (line 309) | def convert_examples_to_features(examples, tokenizer, max_seq_length,
function _improve_answer_span (line 476) | def _improve_answer_span(doc_tokens, input_start, input_end, tokenizer,
function _check_is_max_context (line 513) | def _check_is_max_context(doc_spans, cur_span_index, position):
function create_model (line 550) | def create_model(bert_config, is_training, input_ids, input_mask, segmen...
function model_fn_builder (line 590) | def model_fn_builder(bert_config, init_checkpoint, learning_rate,
function input_fn_builder (line 687) | def input_fn_builder(input_file, seq_length, is_training, drop_remainder):
function write_predictions (line 741) | def write_predictions(all_examples, all_features, all_results, n_best_size,
function get_final_text (line 927) | def get_final_text(pred_text, orig_text, do_lower_case):
function _get_best_indexes (line 1023) | def _get_best_indexes(logits, n_best_size):
function _compute_softmax (line 1035) | def _compute_softmax(scores):
class FeatureWriter (line 1058) | class FeatureWriter(object):
method __init__ (line 1061) | def __init__(self, filename, is_training):
method process_feature (line 1067) | def process_feature(self, feature):
method close (line 1093) | def close(self):
function validate_flags_or_throw (line 1097) | def validate_flags_or_throw(bert_config):
function main (line 1123) | def main(_):
FILE: tokenization.py
function convert_to_unicode (line 27) | def convert_to_unicode(text):
function printable_text (line 47) | def printable_text(text):
function load_vocab (line 70) | def load_vocab(vocab_file):
function convert_by_vocab (line 85) | def convert_by_vocab(vocab, items):
function convert_tokens_to_ids (line 93) | def convert_tokens_to_ids(vocab, tokens):
function convert_ids_to_tokens (line 97) | def convert_ids_to_tokens(inv_vocab, ids):
function whitespace_tokenize (line 101) | def whitespace_tokenize(text):
class FullTokenizer (line 110) | class FullTokenizer(object):
method __init__ (line 113) | def __init__(self, vocab_file, do_lower_case=True):
method tokenize (line 119) | def tokenize(self, text):
method convert_tokens_to_ids (line 127) | def convert_tokens_to_ids(self, tokens):
method convert_ids_to_tokens (line 130) | def convert_ids_to_tokens(self, ids):
class BasicTokenizer (line 134) | class BasicTokenizer(object):
method __init__ (line 137) | def __init__(self, do_lower_case=True):
method tokenize (line 145) | def tokenize(self, text):
method _run_strip_accents (line 169) | def _run_strip_accents(self, text):
method _run_split_on_punc (line 180) | def _run_split_on_punc(self, text):
method _tokenize_chinese_chars (line 200) | def _tokenize_chinese_chars(self, text):
method _is_chinese_char (line 213) | def _is_chinese_char(self, cp):
method _clean_text (line 235) | def _clean_text(self, text):
class WordpieceTokenizer (line 249) | class WordpieceTokenizer(object):
method __init__ (line 252) | def __init__(self, vocab, unk_token="[UNK]", max_input_chars_per_word=...
method tokenize (line 257) | def tokenize(self, text):
function _is_whitespace (line 311) | def _is_whitespace(char):
function _is_control (line 323) | def _is_control(char):
function _is_punctuation (line 335) | def _is_punctuation(char):
FILE: tokenization_test.py
class TokenizationTest (line 26) | class TokenizationTest(tf.test.TestCase):
method test_full_tokenizer (line 28) | def test_full_tokenizer(self):
method test_chinese (line 47) | def test_chinese(self):
method test_basic_tokenizer_lower (line 54) | def test_basic_tokenizer_lower(self):
method test_basic_tokenizer_no_lower (line 62) | def test_basic_tokenizer_no_lower(self):
method test_wordpiece_tokenizer (line 69) | def test_wordpiece_tokenizer(self):
method test_convert_tokens_to_ids (line 89) | def test_convert_tokens_to_ids(self):
method test_is_whitespace (line 103) | def test_is_whitespace(self):
method test_is_control (line 113) | def test_is_control(self):
method test_is_punctuation (line 121) | def test_is_punctuation(self):
Condensed preview — 24 files, each showing path, character count, and a content snippet. Download the .json file or copy for the full structured content (280K chars).
[
{
"path": ".gitignore",
"chars": 1367,
"preview": "# Initially taken from Github's Python gitignore file\n.idea\n\n# Byte-compiled / optimized / DLL files\n__pycache__/\n*.py[c"
},
{
"path": "CONTRIBUTING.md",
"chars": 1323,
"preview": "# How to Contribute\n\nBERT needs to maintain permanent compatibility with the pre-trained model files,\nso we do not plan "
},
{
"path": "LICENSE",
"chars": 11358,
"preview": "\n Apache License\n Version 2.0, January 2004\n "
},
{
"path": "README.md",
"chars": 4283,
"preview": "**[🤗Demo](#demo)** |\n**[📖cases-en](#test-case)** |\n**[📖cases-zh](cases/test.zh.md)** |\n\n\n\n\n## BERT as Language Model\n\nFo"
},
{
"path": "__init__.py",
"chars": 616,
"preview": "# coding=utf-8\n# Copyright 2018 The Google AI Language Team Authors.\n#\n# Licensed under the Apache License, Version 2.0 "
},
{
"path": "cases/test.zh.md",
"chars": 2277,
"preview": "\n# 总体评价\n\n\n\n\n\n**优势**\n1. bert给句子打分，摆脱了传统auto regressive的局限，可并行。\n1. 得益于双向语言模型的全局感受野，bert给word打分准确度较高\n\n**缺陷**\n1. 给每个word打分，都"
},
{
"path": "create_pretraining_data.py",
"chars": 15208,
"preview": "# coding=utf-8\n# Copyright 2018 The Google AI Language Team Authors.\n#\n# Licensed under the Apache License, Version 2.0 "
},
{
"path": "data/lm/result.zh.json",
"chars": 13941,
"preview": "[\n {\n \"tokens\": [\n {\n \"token\": \"2016\",\n \"prob\": 0.10201895982027054\n },\n {\n \"tok"
},
{
"path": "data/lm/test.en.tsv",
"chars": 86,
"preview": "there is a book on the desk\nthere is a plane on the desk\nthere is a book in the desk\n\n"
},
{
"path": "data/lm/test.zh.tsv",
"chars": 198,
"preview": "2016全国高考卷答题模板\n2016全国大考卷答题模板\n2016全国低考卷答题模板\n床前明月光，疑是地上霜\n床前星星光，疑是地上霜\n床前白月光，疑是地上霜\n落霞与孤鹜齐飞，秋水共长天一色\n落霞与孤鹜齐跑，秋水共长天一色\n落霞与孤鹜双飞，秋水"
},
{
"path": "extract_features.py",
"chars": 13898,
"preview": "# coding=utf-8\n# Copyright 2018 The Google AI Language Team Authors.\n#\n# Licensed under the Apache License, Version 2.0 "
},
{
"path": "modeling.py",
"chars": 38086,
"preview": "# coding=utf-8\n# Copyright 2018 The Google AI Language Team Authors.\n#\n# Licensed under the Apache License, Version 2.0 "
},
{
"path": "modeling_test.py",
"chars": 9191,
"preview": "# coding=utf-8\n# Copyright 2018 The Google AI Language Team Authors.\n#\n# Licensed under the Apache License, Version 2.0 "
},
{
"path": "multilingual.md",
"chars": 11289,
"preview": "## Models\n\nThere are two multilingual models currently available. We do not plan to release\nmore single-language models,"
},
{
"path": "optimization.py",
"chars": 6046,
"preview": "# coding=utf-8\n# Copyright 2018 The Google AI Language Team Authors.\n#\n# Licensed under the Apache License, Version 2.0 "
},
{
"path": "optimization_test.py",
"chars": 1721,
"preview": "# coding=utf-8\n# Copyright 2018 The Google AI Language Team Authors.\n#\n# Licensed under the Apache License, Version 2.0 "
},
{
"path": "requirements.txt",
"chars": 110,
"preview": "tensorflow >= 1.11.0 # CPU Version of TensorFlow.\n# tensorflow-gpu >= 1.11.0 # GPU version of TensorFlow.\n"
},
{
"path": "run_classifier.py",
"chars": 31833,
"preview": "# coding=utf-8\n# Copyright 2018 The Google AI Language Team Authors.\n#\n# Licensed under the Apache License, Version 2.0 "
},
{
"path": "run_lm_predict.py",
"chars": 17990,
"preview": "# coding=utf-8\n# Copyright 2018 The Google AI Language Team Authors.\n#\n# Licensed under the Apache License, Version 2.0 "
},
{
"path": "run_pretraining.py",
"chars": 18667,
"preview": "# coding=utf-8\n# Copyright 2018 The Google AI Language Team Authors.\n#\n# Licensed under the Apache License, Version 2.0 "
},
{
"path": "run_squad.py",
"chars": 46391,
"preview": "# coding=utf-8\n# Copyright 2018 The Google AI Language Team Authors.\n#\n# Licensed under the Apache License, Version 2.0 "
},
{
"path": "sample_text.txt",
"chars": 4364,
"preview": "This text is included to make sure Unicode is handled properly: 力加勝北区ᴵᴺᵀᵃছজটডণত\nText should be one-sentence-per-line, wi"
},
{
"path": "tokenization.py",
"chars": 10559,
"preview": "# coding=utf-8\n# Copyright 2018 The Google AI Language Team Authors.\n#\n# Licensed under the Apache License, Version 2.0 "
},
{
"path": "tokenization_test.py",
"chars": 4386,
"preview": "# coding=utf-8\n# Copyright 2018 The Google AI Language Team Authors.\n#\n# Licensed under the Apache License, Version 2.0 "
}
]
About this extraction
This page contains the full source code of the xu-song/bert-as-language-model GitHub repository, extracted and formatted as plain text for AI agents and large language models (LLMs). The extraction includes 24 files (259.0 KB), approximately 65.1k tokens, and a symbol index with 195 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.