Skip to content

Commit

Permalink
v0.1.9
Browse files Browse the repository at this point in the history
  • Loading branch information
Tongjilibo committed Aug 8, 2022
1 parent dccb068 commit 42b5c01
Show file tree
Hide file tree
Showing 4 changed files with 6 additions and 4 deletions.
3 changes: 2 additions & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,7 @@ pip install git+https://www.github.com/Tongjilibo/bert4torch.git
- [**实验指标**](https://github.com/Tongjilibo/bert4torch/blob/master/examples/Performance.md)

## 版本说明
- **v0.1.9**:增加mixup/manifold_mixup/temporal_ensembling策略, 修复pgd策略param.grad为空的问题,修改tokenizer支持批量
- **v0.1.8**:修复原来CRF训练中loss陡增的问题,修复xlnet的token_type_ids输入显存占用大的问题
- **v0.1.7**:增加EarlyStop,CRF中自带转bool类型
- **v0.1.6**:增加transformer_xl、xlnet、t5_pegasus模型,prompt、预训练等示例,支持增加embedding输入,EMA策略,修复tokenizer和sinusoid的bug
Expand All @@ -33,7 +34,7 @@ pip install git+https://www.github.com/Tongjilibo/bert4torch.git
- **v0.1.3**:初始版本

## 更新:
- **2022年7月27更新**:增加mixup/manifold_mixup策略, 修复pgd策略param.grad为空的问题,修改tokenizer支持批量,增加uie示例
- **2022年7月27更新**:增加mixup/manifold_mixup/temporal_ensembling策略, 修复pgd策略param.grad为空的问题,修改tokenizer支持批量,增加uie示例
- **2022年7月16更新**:修复原来CRF训练中loss陡增的问题,修复xlnet的token_type_ids输入显存占用大的问题
- **2022年7月10更新**:增加金融中文FAQ示例,天池新闻分类top1案例,增加EarlyStop,CRF中自带转bool类型
- **2022年6月29更新**:增加ner的实验,测试crf不同初始化的效果,bert-whitening中文实验
Expand Down
2 changes: 1 addition & 1 deletion bert4torch/__init__.py
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
#! -*- coding: utf-8 -*-


__version__ = '0.1.8'
__version__ = '0.1.9'
1 change: 1 addition & 0 deletions examples/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -89,6 +89,7 @@
- [task_sentiment_UDA.py](https://github.com/Tongjilibo/bert4torch/blob/master/examples/training_trick/task_sentiment_UDA.py):通过[UDA](https://arxiv.org/abs/1904.12848)半监督学习提升分类效果,在原来Losss上加一致性损失。
- [task_sentiment_mixup.py](https://github.com/Tongjilibo/bert4torch/blob/master/examples/training_trick/task_sentiment_mixup.py):通过[Mixup](https://github.com/vikasverma1077/manifold_mixup)提升模型泛化性能。
- [task_sentiment_exponential_moving_average.py](https://github.com/Tongjilibo/bert4torch/blob/master/examples/training_trick/task_sentiment_exponential_moving_average.py):EMA指数滑动平均
- [task_sentiment_TemporalEnsembling.py](https://github.com/Tongjilibo/bert4torch/blob/master/examples/training_trick/task_sentiment_TemporalEnsembling.py):通过[TemporalEnsembling](https://github.com/s-laine/tempens)[pytorch第三方](https://github.com/ferretj/temporal-ensembling)提升模型泛化性能。
- [task_sentiment_R-Drop.py](https://github.com/Tongjilibo/bert4torch/blob/master/examples/training_trick/task_sentiment_R-Drop.py):通过[R-Drop](https://github.com/dropreg/R-Drop)提升分类效果,可以视为用dropout加噪下的UDA。
- [task_amp.py](https://github.com/Tongjilibo/bert4torch/blob/master/examples/training_trick/task_amp.py):Pytorch的amp混合精度训练
- [task_data_parallel.py](https://github.com/Tongjilibo/bert4torch/blob/master/examples/training_trick/task_data_parallel.py):DataParallel模式的多GPU训练方式
Expand Down
4 changes: 2 additions & 2 deletions setup.py
Original file line number Diff line number Diff line change
Expand Up @@ -4,12 +4,12 @@

setup(
name='bert4torch',
version='0.1.8',
version='0.1.9',
description='an elegant bert4torch',
long_description='bert4torch: https://github.com/Tongjilibo/bert4torch',
license='MIT Licence',
url='https://github.com/Tongjilibo/bert4torch',
author='Tongjilibo',
install_requires=['torch>1.0'],
install_requires=['torch>1.6'],
packages=find_packages()
)

0 comments on commit 42b5c01

Please sign in to comment.