Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,44 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
+
datasets:
|
4 |
+
- Azure99/blossom-chat-v1
|
5 |
+
- Azure99/blossom-math-v2
|
6 |
+
- Azure99/blossom-wizard-v1
|
7 |
+
- Azure99/blossom-orca-v1
|
8 |
+
language:
|
9 |
+
- zh
|
10 |
+
- en
|
11 |
---
|
12 |
+
# **BLOSSOM-v3.1-mistral-7b**
|
13 |
+
|
14 |
+
### 介绍
|
15 |
+
|
16 |
+
Blossom是一个对话式语言模型,基于Mistral-7B-v0.1预训练模型,在Blossom Orca/Wizard/Chat/Math混合数据集上进行指令精调得来。Blossom拥有强大的通用能力及上下文理解能力,此外,训练使用的高质量中英文数据集也进行了开源。
|
17 |
+
|
18 |
+
训练分为两阶段,第一阶段使用100K Wizard、100K Orca单轮指令数据集,训练1个epoch;第二阶段使用2K Blossom math数学推理数据集、50K Blossom chat多轮对话数据集、以及上一阶段中随机采样1%的数据,训练3个epoch。
|
19 |
+
|
20 |
+
注意:Mistral-7B-v0.1预训练模型的中文知识较为欠缺,因此对于中文场景,更推荐使用[blossom-v3-baichuan2-7b](https://huggingface.co/Azure99/blossom-v3-baichuan2-7b)
|
21 |
+
|
22 |
+
### 推理
|
23 |
+
|
24 |
+
推理采用对话续写的形式。
|
25 |
+
|
26 |
+
单轮对话
|
27 |
+
|
28 |
+
```
|
29 |
+
A chat between a human and an artificial intelligence bot. The bot gives helpful, detailed, and polite answers to the human's questions.
|
30 |
+
|Human|: 你好
|
31 |
+
|Bot|:
|
32 |
+
```
|
33 |
+
|
34 |
+
多轮对话
|
35 |
+
|
36 |
+
```
|
37 |
+
A chat between a human and an artificial intelligence bot. The bot gives helpful, detailed, and polite answers to the human's questions.
|
38 |
+
|Human|: 你好
|
39 |
+
|Bot|: 你好,有什么我能帮助你的?</s>
|
40 |
+
|Human|: 介绍下中国的首都吧
|
41 |
+
|Bot|:
|
42 |
+
```
|
43 |
+
|
44 |
+
注意:在历史对话的Bot输出结尾,拼接一个</s>
|