[DOC]EmoLLM_Scientist微调指南
This commit is contained in:
parent
0c72c31b4f
commit
8f63772651
@ -1,33 +1,14 @@
|
|||||||
# 微调指南
|
# EmoLLM_Scientist微调指南
|
||||||
|
|
||||||
|
## 数据
|
||||||
|
微调数据共包含3800段对话,借助LLM自动生成,后续进行人工校验。数据路径:'datasets\scientist.json'
|
||||||
|
|
||||||
##数据
|
## 基座
|
||||||
|
基座模型采用InternLM2-Chat-7B,模型介绍请见[InternLM](https://github.com/InternLM/InternLM)
|
||||||
|
|
||||||
##基座
|
## 训练方式
|
||||||
|
基于xtuner的微调,使用xtuner的train命令行工具,使用命令如下:
|
||||||
##训练方式
|
### 安装依赖
|
||||||
|
|
||||||
##配置文件
|
|
||||||
|
|
||||||
## 一、基于xtuner的微调🎉🎉🎉🎉🎉
|
|
||||||
|
|
||||||
### 环境准备
|
|
||||||
|
|
||||||
```markdown
|
|
||||||
datasets==2.16.1
|
|
||||||
deepspeed==0.13.1
|
|
||||||
einops==0.7.0
|
|
||||||
flash_attn==2.5.0
|
|
||||||
mmengine==0.10.2
|
|
||||||
openxlab==0.0.34
|
|
||||||
peft==0.7.1
|
|
||||||
sentencepiece==0.1.99
|
|
||||||
torch==2.1.2
|
|
||||||
transformers==4.36.2
|
|
||||||
xtuner==0.1.11
|
|
||||||
```
|
|
||||||
|
|
||||||
也可以一键安装
|
|
||||||
|
|
||||||
```bash
|
```bash
|
||||||
cd xtuner_config/
|
cd xtuner_config/
|
||||||
@ -35,26 +16,23 @@ pip3 install -r requirements.txt
|
|||||||
```
|
```
|
||||||
|
|
||||||
---
|
---
|
||||||
|
### 运行微调脚本
|
||||||
### 微调
|
|
||||||
|
|
||||||
```bash
|
```bash
|
||||||
cd xtuner_config/
|
cd xtuner_config/
|
||||||
xtuner train internlm2_7b_chat_qlora_e3.py --deepspeed deepspeed_zero2
|
xtuner train internlm2_7b_chat_qlora_e3_scienctist.py --deepspeed deepspeed_zero2
|
||||||
```
|
```
|
||||||
|
|
||||||
---
|
---
|
||||||
|
### 模型转换
|
||||||
|
|
||||||
### 将得到的 PTH 模型转换为 HuggingFace 模型
|
将得到的 PTH 模型转换为 HuggingFace 模型,生成 Adapter 文件夹
|
||||||
|
|
||||||
**即:生成 Adapter 文件夹**
|
|
||||||
|
|
||||||
```bash
|
```bash
|
||||||
cd xtuner_config/
|
cd xtuner_config/
|
||||||
mkdir hf
|
mkdir hf
|
||||||
export MKL_SERVICE_FORCE_INTEL=1
|
export MKL_SERVICE_FORCE_INTEL=1
|
||||||
|
#这里假设训练了3个epoch
|
||||||
xtuner convert pth_to_hf internlm2_7b_chat_qlora_e3.py ./work_dirs/internlm_chat_7b_qlora_oasst1_e3_copy/epoch_3.pth ./hf
|
xtuner convert pth_to_hf internlm2_7b_chat_qlora_e3_scienctist.py ./work_dirs/internlm2_7b_chat_qlora_e3_scienctist/epoch_3.pth ./hf
|
||||||
```
|
```
|
||||||
|
|
||||||
---
|
---
|
||||||
@ -76,16 +54,19 @@ xtuner convert merge ./internlm2-chat-7b ./hf ./merged --max-shard-size 2GB
|
|||||||
|
|
||||||
```
|
```
|
||||||
cd demo/
|
cd demo/
|
||||||
python cli_internlm2.py
|
python cli_internlm2_scientist.py
|
||||||
```
|
```
|
||||||
|
|
||||||
---
|
---
|
||||||
|
|
||||||
## 二、基于Transformers的微调🎉🎉🎉🎉🎉
|
## 模型上传
|
||||||
|
完成测试后可将模型上传到ModelScope和Openxlab平台
|
||||||
|
### ModelScope
|
||||||
|
脚本:'scripts/upload_modelscope.py'
|
||||||
|
[Openxlab模型上传](https://openxlab.org.cn/docs/models/%E4%B8%8A%E4%BC%A0%E6%A8%A1%E5%9E%8B.html)
|
||||||
|
|
||||||
- 请查看[ChatGLM3-6b lora微调指南](ChatGLM3-6b-ft.md)
|
### Openxlab
|
||||||
|
[ModelScope模型上传](https://modelscope.cn/docs/%E6%A8%A1%E5%9E%8B%E7%9A%84%E5%88%9B%E5%BB%BA%E4%B8%8E%E6%96%87%E4%BB%B6%E4%B8%8A%E4%BC%A0)
|
||||||
---
|
|
||||||
|
|
||||||
## 其他
|
## 其他
|
||||||
|
|
||||||
|
Loading…
Reference in New Issue
Block a user