99 lines
3.5 KiB
Markdown
99 lines
3.5 KiB
Markdown
|
|
# Llama3-Chinese-ORPO
|
|||
|
|
基于Llama3,通过进一步CPT,SFT,ORPO得到的中文版Llama3
|
|||
|
|
|
|||
|
|
|
|||
|
|
=log: 2024/04/21 12:00=: 本仓库是出于 *学习目的* 而进行的一个实践项目。
|
|||
|
|
|
|||
|
|
## News
|
|||
|
|
[2024/04/22] We release the first version of Llama3-Chinese-ORPO.
|
|||
|
|
|
|||
|
|
|
|||
|
|
## 模型列表
|
|||
|
|
|
|||
|
|
| 模型名称 | 模型简介 | ModelScope下载链接 | Huggingface下载链接 |
|
|||
|
|
|--------------------------------|---------------------------------------------------------------------------------------------------------------------|---|---|
|
|||
|
|
| Llama3-Chinese-pro-8.4B-sft-1M | 在对llama3-8b的base版本(meta-llama/Meta-Llama-3-8B)进行llama-pro拓展了两层后,进而通过一百万条SFT数据训练而得到。 | [Llama3-Chinese-pro-8.4B-sft-1M](https://modelscope.cn/models/linjh1118/Llama3-Chinese-pro-8.4B-sft-1M) | |
|
|||
|
|
| Llama3-Chinese-pro-8.4B-sft-1M-ORPO | 基于上一行产出的模型继续做ORPO进而得到 | | |
|
|||
|
|
| | | | |
|
|||
|
|
|
|||
|
|
## 测试表现
|
|||
|
|
|
|||
|
|
<details>
|
|||
|
|
<summary>展开查看 测试表现</summary>
|
|||
|
|
|
|||
|
|
|
|||
|
|
1. **随意的五个问题**
|
|||
|
|
|
|||
|
|
### 先插入放两张没有训练完毕的Llama3-Chinese-pro-8.4B-sft-1M的回复,哈哈
|
|||
|
|
|
|||
|
|

|
|||
|
|
|
|||
|
|
|
|||
|
|
### 回到正经测评
|
|||
|
|
|
|||
|
|

|
|||
|
|
|
|||
|
|

|
|||
|
|
|
|||
|
|
</details>
|
|||
|
|
|
|||
|
|
## 主要贡献者 Contributors
|
|||
|
|
**排名不分*先后***
|
|||
|
|
|
|||
|
|
<table>
|
|||
|
|
<tr>
|
|||
|
|
<td align='center'>
|
|||
|
|
<img src="https://avatars.githubusercontent.com/u/67041238?v=4" alt="Contributor 1" height="150">
|
|||
|
|
<br>
|
|||
|
|
<b>linjh1118 @ Northeastern University</b>
|
|||
|
|
<br>
|
|||
|
|
<a href='https://github.com/linjh1118'>Llama3-Chinese-ORPO项目的核心贡献者</a>
|
|||
|
|
</td>
|
|||
|
|
<td align='center'>
|
|||
|
|
<img src="https://avatars.githubusercontent.com/u/59789526?v=4" alt="Contributor 2" height="150">
|
|||
|
|
<br>
|
|||
|
|
<b>Charon-ops @ Tsinghua University</b>
|
|||
|
|
<br>
|
|||
|
|
<a href='https://github.com/AIR-hl'>Llama3-Chinese-ORPO项目的核心贡献者</a>
|
|||
|
|
</td>
|
|||
|
|
<td align='center'>
|
|||
|
|
<img src="https://avatars.githubusercontent.com/u/94281210?v=4" alt="Contributor 3" height="150">
|
|||
|
|
<br>
|
|||
|
|
<b>zzlf @ Northeastern University</b>
|
|||
|
|
<br>
|
|||
|
|
<a href='https://github.com/zzlf'>Llama3-Chinese-ORPO项目的核心贡献者</a>
|
|||
|
|
</td>
|
|||
|
|
</tr>
|
|||
|
|
</table>
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
<table>
|
|||
|
|
<tr>
|
|||
|
|
<td align='center'>
|
|||
|
|
<img src="https://avatars.githubusercontent.com/u/66808901?v=4" alt="Contributor 2" height="150">
|
|||
|
|
<br>
|
|||
|
|
<b>AIR-hl @ Northeastern University</b>
|
|||
|
|
<br>
|
|||
|
|
<a href='https://github.com/AIR-hl'>Llama3-Chinese-ORPO项目的核心贡献者</a>
|
|||
|
|
</td>
|
|||
|
|
<td align='center'>
|
|||
|
|
<img src="https://avatars.githubusercontent.com/u/116535337?v=4" alt="Contributor 3" height="150">
|
|||
|
|
<br>
|
|||
|
|
<b>yanyi74 @ Northeastern University</b>
|
|||
|
|
<br>
|
|||
|
|
<a href='https://github.com/yanyi74'>Llama3-Chinese-ORPO项目的核心贡献者</a>
|
|||
|
|
</td>
|
|||
|
|
<td align='center'>
|
|||
|
|
<img src="https://avatars.githubusercontent.com/u/84018897?v=4" alt="Contributor 1" height="150">
|
|||
|
|
<br>
|
|||
|
|
<b>MrChen-NEU @ Northeastern University</b>
|
|||
|
|
<br>
|
|||
|
|
<a href='https://github.com/MrChen-NEU'>Llama3-Chinese-ORPO项目的核心贡献者</a>
|
|||
|
|
</td>
|
|||
|
|
</tr>
|
|||
|
|
</table>
|
|||
|
|
|
|||
|
|
|
|||
|
|
非常非常感谢前人的优秀工作:https://github.com/CrazyBoyM/llama3-Chinese-chat (全球第一个中文llama3)
|