Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/bojone/t5_in_bert4keras
整理一下在keras中使用T5模型的要点
https://github.com/bojone/t5_in_bert4keras
Last synced: 3 months ago
JSON representation
整理一下在keras中使用T5模型的要点
- Host: GitHub
- URL: https://github.com/bojone/t5_in_bert4keras
- Owner: bojone
- Created: 2020-10-30T04:26:35.000Z (about 4 years ago)
- Default Branch: main
- Last Pushed: 2022-03-04T08:52:00.000Z (over 2 years ago)
- Last Synced: 2024-06-24T05:36:24.985Z (5 months ago)
- Language: Python
- Size: 1.26 MB
- Stars: 171
- Watchers: 7
- Forks: 28
- Open Issues: 5
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- StarryDivineSky - bojone/t5_in_bert4keras
README
# T5 in bert4keras
整理一下在keras中使用T5模型的要点，尤其是中文场景下的使用要点。以多国语言版mT5为例。
博客链接：https://kexue.fm/archives/7867
本项目实验环境：tensorflow 1.14 + keras 2.3.1 + bert4keras 0.9.1
## 模型下载
首先，要想办法下载Google开放的权重，最简单的方式，是找一台能科学上网的服务器，在上面安装gsutil，然后执行
```shell
gsutil cp -r gs://t5-data/pretrained_models/mt5/small .
```
T5使用sentencepiece作为tokenizer，mT5的tokenizer模型下载地址为
```shell
gsutil cp -r gs://t5-data/vocabs/mc4.250000.100extra/sentencepiece.model .
```
笔者精简好的tokenizer文件：[sentencepiece_cn.model](https://github.com/bojone/t5_in_bert4keras/blob/main/tokenizer/sentencepiece_cn.model)和[sentencepiece_cn_keep_tokens.json](https://github.com/bojone/t5_in_bert4keras/blob/main/tokenizer/sentencepiece_cn_keep_tokens.json)
另外，为了方便国内用户，笔者将small版和base版整理分享到[百度网盘](https://pan.baidu.com/s/1YWaStqB6Epkxfyx6WcOzWw)(mwfc)了。
## Config
T5模型的配置文件是gin格式的，这不符合bert4keras的输入，使用者请根据所给的gin和下述模版构建对应的config.json文件。
下面是mT5 small版的参考config.json：
```python
{
"hidden_dropout_prob": 0.1,
"hidden_size": 512,
"initializer_range": 0.02,
"intermediate_size": 1024,
"num_attention_heads": 6,
"attention_head_size": 64,
"num_hidden_layers": 8,
"vocab_size": 250112,
"hidden_act": ["gelu", "linear"]
}
```
一般要修改的是`hidden_size`、`intermediate_size`、`num_attention_heads`、`attention_head_size`和`num_hidden_layers`这几个参数。
## 基本使用
```python
# 模型路径
config_path = '/root/kg/bert/mt5/mt5_small/t5_config.json'
checkpoint_path = '/root/kg/bert/mt5/mt5_small/model.ckpt-1000000'
spm_path = '/root/kg/bert/mt5/sentencepiece.model'
# 加载分词器
tokenizer = SpTokenizer(spm_path, token_start=None, token_end='</s>')
# 加载模型
t5 = build_transformer_model(
config_path=config_path,
checkpoint_path=checkpoint_path,
model='t5.1.1',
return_keras_model=False,
name='T5',
)
encoder = t5.encoder
decoder = t5.decoder
model = t5.model
```
## 中文优化
```python
# 模型路径
config_path = '/root/kg/bert/mt5/mt5_base/t5_config.json'
checkpoint_path = '/root/kg/bert/mt5/mt5_base/model.ckpt-1000000'
spm_path = '/root/kg/bert/mt5/sentencepiece_cn.model'
keep_tokens_path = '/root/kg/bert/mt5/sentencepiece_cn_keep_tokens.json'
# 加载分词器
tokenizer = SpTokenizer(spm_path, token_start=None, token_end='</s>')
keep_tokens = json.load(open(keep_tokens_path))
# 加载模型
t5 = build_transformer_model(
config_path=config_path,
checkpoint_path=checkpoint_path,
keep_tokens=keep_tokens,
model='t5.1.1',
return_keras_model=False,
name='T5',
)
encoder = t5.encoder
decoder = t5.decoder
model = t5.model
```
细节请参考：[task_autotitle_csl.py](https://github.com/bojone/t5_in_bert4keras/blob/main/task_autotitle_csl.py)。
## 交流联系
QQ交流群：67729435，微信群请加机器人微信号spaces_ac_cn