Skip to content

Commit a181533

Browse files
committed
v0.5.0
1 parent dbe0b95 commit a181533

File tree

6 files changed

+8
-7
lines changed

6 files changed

+8
-7
lines changed

README.md

Lines changed: 1 addition & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -93,10 +93,9 @@ pip install git+https://github.com/Tongjilibo/bert4torch
9393
### 4.1 版本历史
9494
|更新日期| bert4torch | torch4keras | 版本说明 |
9595
|------| ---------------- | ----------------- |----------- |
96+
|20240418| 0.5.0 | 0.2.2 | 修复chatglm3的bug, 修复save_pretrained时多文件的bug,增加CausalLMLoss, 修改deepspeed的传参逻辑,修改Text2Vec的bug, 完善openai client, 增加get_weight_decay_optim_groups|
9697
|20240317| 0.4.9.post2 | 0.2.1.post2 |增加get_weight_decay_optim_groups函数, attention中允许is_causal,修改repetition_penalty的bug,把baichuan从llama中剥离,修复config_path的bug,允许num_key_value_heads参数,[torch4keras-v0.2.1.post2](https://github.com/Tongjilibo/torch4keras/releases/tag/v0.2.1.post2)更新特性|
9798
|20240221| 0.4.8 | 0.2.0|fastapi发布服务允许闲时offload到cpu, `build_transformer_model`允许从hf下载, 添加`FillMask`的pipeline, 添加`SequenceClassificationTrainer`|
98-
|20240204| 0.4.7 | 0.1.9|修改`save_pretrained`用于保存文件夹, 增加GenerateSpeed用于统计token生成速度,修复t5在use_states=True时候的错误, 修改层次编码的bug, 增加deepseek_moe模型,修复generation并发错误,优化大模型耗时|
99-
|20240116| 0.4.6 | 0.1.8|bug修复,增加`save_pretrained`用于保存`transformer`格式的权重, 增加部分`embedding`模型|
10099

101100
[更多版本](https://github.com/Tongjilibo/bert4torch/blob/master/docs/Update.md)
102101

docs/History.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
## 更新历史
22

3-
- **20240403**:修改Text2Vec的bug
3+
- **20240418**:修改Text2Vec的bug, 完善openai client, 增加get_weight_decay_optim_groups
44
- **20240331**: 修复chatglm3的bug, 修复save_pretrained时多文件的bug,增加CausalLMLoss, 修改deepspeed的传参逻辑
55
- **20240317**: 修复config_path的bug,允许num_key_value_heads参数
66
- **20240316**: 增加get_weight_decay_optim_groups函数, attention中允许is_causal,修改repetition_penalty的bug,把baichuan从llama中剥离,[torch4keras-v0.2.1](https://github.com/Tongjilibo/torch4keras/releases/tag/v0.2.1)更新特性

docs/Update.md

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,8 @@
22

33
|更新日期| bert4torch版本 | torch4keras版本 | 版本说明 |
44
|------| ---------------- | ----------------- |----------- |
5+
|20240418| 0.5.0 | 0.2.2 | 修复chatglm3的bug, 修复save_pretrained时多文件的bug,增加CausalLMLoss, 修改deepspeed的传参逻辑,修改Text2Vec的bug, 完善openai client, 增加get_weight_decay_optim_groups|
6+
|20240317| 0.4.9.post2 | 0.2.1.post2 |增加get_weight_decay_optim_groups函数, attention中允许is_causal,修改repetition_penalty的bug,把baichuan从llama中剥离,修复config_path的bug,允许num_key_value_heads参数,[torch4keras-v0.2.1.post2](https://github.com/Tongjilibo/torch4keras/releases/tag/v0.2.1.post2)更新特性|
57
|20240221| 0.4.8 | 0.2.0|fastapi发布服务允许闲时offload到cpu, `build_transformer_model`允许从hf下载, 添加`FillMask`的pipeline, 添加`SequenceClassificationTrainer`|
68
|20240204| 0.4.7 | 0.1.9|修改`save_pretrained`用于保存文件夹, 增加GenerateSpeed用于统计token生成速度,修复t5在use_states=True时候的错误, 修改层次编码的bug, 增加deepseek_moe模型,修复generation并发错误,优化大模型耗时|
79
|20240116| 0.4.6 | 0.1.8|bug修复,增加`save_pretrained`用于保存`transformer`格式的权重, 增加部分`embedding`模型|

examples/llm/task_chatglm_lora.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -41,7 +41,7 @@
4141
max_target_length = 64
4242
max_seq_length = max_source_length + max_target_length
4343
lr = 5e-4
44-
batch_size = 4 # 根据显存大小调整
44+
batch_size = 16 # 根据显存大小调整
4545
eval_batch_size = 4
4646
grad_accumulation_steps = 1 # 根据显存大小调整
4747
epochs = 1

examples/sequence_labeling/uie/README.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,5 @@
11
# 说明
2-
- convert.py: 下载uie模型并且转成pytorch格式
2+
- convert.py: 下载uie模型并且转成pytorch格式,作者已转换的[权重](https://huggingface.co/Tongjilibo/uie-base)
33
- finetune.sh:数据处理和训练全流程
44
- finetune_step1_dataprocess.py:数据预处理1
55
- finetune_step2_doccano.py:数据预处理2

setup.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -7,13 +7,13 @@
77

88
setup(
99
name='bert4torch',
10-
version='v0.4.9.post2',
10+
version='v0.5.0',
1111
description='an elegant bert4torch',
1212
long_description=long_description,
1313
long_description_content_type="text/markdown",
1414
license='MIT Licence',
1515
url='https://github.com/Tongjilibo/bert4torch',
1616
author='Tongjilibo',
17-
install_requires=['numpy', 'tqdm', 'torch>1.6', 'torch4keras==0.2.1.post2', 'six'],
17+
install_requires=['numpy', 'tqdm', 'torch>1.6', 'torch4keras==0.2.2', 'six'],
1818
packages=find_packages()
1919
)

0 commit comments

Comments
 (0)