Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,34 @@
|
|
1 |
---
|
2 |
license: llama3
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: llama3
|
3 |
+
datasets:
|
4 |
+
- LooksJuicy/ruozhiba
|
5 |
+
language:
|
6 |
+
- zh
|
7 |
---
|
8 |
+
|
9 |
+
## 基于ruozhiba对Llama3-Instruct进行微调。</br>
|
10 |
+
### 模型:</br>
|
11 |
+
- https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct
|
12 |
+
|
13 |
+
### 数据集:
|
14 |
+
- https://huggingface.co/datasets/LooksJuicy/ruozhiba
|
15 |
+
|
16 |
+
### 训练工具
|
17 |
+
https://github.com/hiyouga/LLaMA-Factory
|
18 |
+
|
19 |
+
### 测评方式:
|
20 |
+
使用opencompass(https://github.com/open-compass/OpenCompass/ ), 测试工具基于CEval和MMLU对微调之后的模型和原始模型进行测试。</br>
|
21 |
+
测试模型分别为:
|
22 |
+
- Llama-3-8B
|
23 |
+
- Llama-3-8B-Instruct
|
24 |
+
- LLama3-Instruct-sft-ruozhiba,使用ruozhiba数据对Llama-3-8B-Instruct使用sft方式lora微调
|
25 |
+
|
26 |
+
### 测试机器
|
27 |
+
8*A800
|
28 |
+
|
29 |
+
### 结果
|
30 |
+
| 模型名称 | CEVAL | MMLU |
|
31 |
+
|--------------------------|-------|------|
|
32 |
+
| LLama3 | 49.91 | 66.62|
|
33 |
+
| LLama3-Instruct | 50.55 | 67.15|
|
34 |
+
| LLama3-Instruct-sft-ruozhiba-3epoch | 50.87 | 67.51|
|