Update README.md
Browse files
README.md
CHANGED
@@ -12,20 +12,24 @@ tags:
|
|
12 |
- llama3
|
13 |
- chat
|
14 |
---
|
15 |
-
# llama3-instruct 中文DPO版
|
16 |
### 模型介绍
|
17 |
像原版instruct一样,喜欢用有趣中文和表情符号回答问题。
|
18 |
Github:https://github.com/CrazyBoyM/llama3-Chinese-chat
|
19 |
-
|
20 |
-
DPO
|
21 |
-
特点:偏好中文和emoji表情,且不损伤原instruct版模型能力。实测中文DPO版问答性能体验超过现在市面上任何llama3中文微调版 (微调会破坏llama3原版能力,导致遗忘)
|
22 |
![Alt text](https://modelscope.cn/api/v1/models/baicai003/Llama3-Chinese-instruct-DPO-beta0.5/repo?Revision=master&FilePath=image.png&View=true)
|
23 |
|
|
|
|
|
|
|
24 |
### 模型部署
|
25 |
-
网页脚本文件:https://github.com/CrazyBoyM/llama3-Chinese-chat/blob/main/deploy/web_streamlit_for_instruct_v2.py
|
|
|
26 |
```
|
27 |
pip install streamlit
|
28 |
-
|
|
|
29 |
```
|
30 |
|
31 |
### 模型下载
|
@@ -37,10 +41,10 @@ pip install modelscope
|
|
37 |
```python
|
38 |
#SDK模型下载
|
39 |
from modelscope import snapshot_download
|
40 |
-
model_dir = snapshot_download('
|
41 |
```
|
42 |
Git下载
|
43 |
```
|
44 |
#Git模型下载
|
45 |
-
git clone https://www.modelscope.cn/
|
46 |
-
```
|
|
|
12 |
- llama3
|
13 |
- chat
|
14 |
---
|
15 |
+
# llama3.1-8b-instruct 中文DPO版
|
16 |
### 模型介绍
|
17 |
像原版instruct一样,喜欢用有趣中文和表情符号回答问题。
|
18 |
Github:https://github.com/CrazyBoyM/llama3-Chinese-chat
|
19 |
+
|
20 |
+
特点:偏好中文和emoji表情,且不损伤原instruct版模型能力。实测中文DPO版问答性能体验超过现在市面上任何llama3.1中文微调版 (微调会大面积破坏llama3.1原版能力,导致遗忘)
|
|
|
21 |
![Alt text](https://modelscope.cn/api/v1/models/baicai003/Llama3-Chinese-instruct-DPO-beta0.5/repo?Revision=master&FilePath=image.png&View=true)
|
22 |
|
23 |
+
### 训练细节
|
24 |
+
DPO(beta 0.5) + lora rank128, alpha256 + 打开"lm_head", "input_layernorm", "post_attention_layernorm", "norm"层训练.
|
25 |
+
|
26 |
### 模型部署
|
27 |
+
网页脚本文件:https://github.com/CrazyBoyM/llama3-Chinese-chat/blob/main/deploy/web_streamlit_for_instruct_v2.py
|
28 |
+
已经在模型文件夹里内置了一份web.py,可以直接使用。
|
29 |
```
|
30 |
pip install streamlit
|
31 |
+
pip install transformers==4.40.1
|
32 |
+
streamlit run web.py ./llama3.1-8b-instruct-dpo-zh
|
33 |
```
|
34 |
|
35 |
### 模型下载
|
|
|
41 |
```python
|
42 |
#SDK模型下载
|
43 |
from modelscope import snapshot_download
|
44 |
+
model_dir = snapshot_download('shareAI/llama3.1-8b-instruct-dpo-zh')
|
45 |
```
|
46 |
Git下载
|
47 |
```
|
48 |
#Git模型下载
|
49 |
+
git clone https://www.modelscope.cn/shareAI/llama3.1-8b-instruct-dpo-zh.git
|
50 |
+
```
|