Update README.md
Browse files
README.md
CHANGED
@@ -51,17 +51,30 @@ Anima模型基于QLoRA开源的[33B guanaco](https://huggingface.co/timdettmers/
|
|
51 |
|
52 |
#### 如何训练
|
53 |
|
54 |
-
|
55 |
-
|
56 |
-
```
|
57 |
-
# 1. install dependencies
|
58 |
-
pip install -r requirements.txt
|
59 |
-
# 2.
|
60 |
-
cd training
|
61 |
-
./run_Amina_training.sh
|
62 |
-
```
|
63 |
-
|
64 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
|
66 |
## 📊验证评估
|
67 |
|
|
|
51 |
|
52 |
#### 如何训练
|
53 |
|
54 |
+
1. 重现Anima的模型训练过程:使用以下步骤可以重现Anima 33B模型(单卡80GB H100或双卡 40GB A100均测试过可运行):
|
55 |
+
|
56 |
+
```bash
|
57 |
+
# 1. install dependencies
|
58 |
+
pip install -r requirements.txt
|
59 |
+
# 2.
|
60 |
+
cd training
|
61 |
+
./run_Amina_training.sh
|
62 |
+
```
|
63 |
+
|
64 |
+
2. 基于Anima finetune训练其他model:
|
65 |
+
|
66 |
+
```bash
|
67 |
+
# 1. install dependencies
|
68 |
+
pip install -r requirements.txt
|
69 |
+
# 2.
|
70 |
+
cd training
|
71 |
+
./run_finetune_raining_based_on_Anima.sh
|
72 |
+
```
|
73 |
+
注:可以修改run_finetune_raining_based_on_Anima.sh中的--dataset和--dataset_format参数使用其他训练数据dataset。
|
74 |
+
|
75 |
+
#### 多卡训练
|
76 |
+
由于使用Hugging Face Accelerate,天然支持多卡训练。
|
77 |
+
我们测试过双卡40GB的A100,可以直接运行。
|
78 |
|
79 |
## 📊验证评估
|
80 |
|