Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,39 @@
|
|
1 |
-
---
|
2 |
-
license: agpl-3.0
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: agpl-3.0
|
3 |
+
language:
|
4 |
+
- zh
|
5 |
+
base_model:
|
6 |
+
- Qwen/Qwen1.5-32B
|
7 |
+
tags:
|
8 |
+
- reflection
|
9 |
+
---
|
10 |
+
|
11 |
+
# Reflection-Chinese-32B · Reflection-中文-32B
|
12 |
+
|
13 |
+
本模型使用[Reflection-Chinese-Dataset](https://huggingface.co/datasets/stvlynn/Reflection-Chinese-Dataset)微调,底模为Qwen1.5-32B
|
14 |
+
|
15 |
+
通过Reflection格式(think-reflect-output)的数据集引导模型形成特定的思维方式,提高正确率
|
16 |
+
|
17 |
+
## Demo
|
18 |
+
|
19 |
+
1. 3.11和3.8哪个大
|
20 |
+
|
21 |
+
![](https://cdn.statically.io/gh/stvlynn/cloudimg@master/blog/2310/截屏2024-09-15-13.22.23.33upadngk6m0.webp)
|
22 |
+
|
23 |
+
2. 鲁迅为什么打周树人
|
24 |
+
|
25 |
+
![](https://cdn.statically.io/gh/stvlynn/cloudimg@master/blog/2310/截屏2024-09-12-13.18.02.3eowy8bgbma0.webp)
|
26 |
+
|
27 |
+
3. 树上几只鸟
|
28 |
+
|
29 |
+
![](https://cdn.statically.io/gh/stvlynn/cloudimg@master/blog/2310/截屏2024-09-12-10.17.59.6c0dbu9ls880.webp)
|
30 |
+
|
31 |
+
4. strawberry(未通过,因为复现成功率低)
|
32 |
+
|
33 |
+
![](https://cdn.statically.io/gh/stvlynn/cloudimg@master/blog/2310/IMG_2685.6gunge0hf5s0.webp)
|
34 |
+
|
35 |
+
## 存在的问题
|
36 |
+
|
37 |
+
1. [Reflection-llama3.1-70B](https://huggingface.co/mattshumer/Reflection-Llama-3.1-70B)在真实性上存在很多疑问,本项目使用的数据集是基于该项目的,所以本项目不保证可用性
|
38 |
+
|
39 |
+
2. 虽然本项目的数据集严格使用<think><reflection><output>标签用来分割内容,但是实际输出并没有这样的效果
|