Create README_zh-CN.md
Browse files- README_zh-CN.md +56 -0
README_zh-CN.md
ADDED
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# NanoLM-70M-Instruct-v1
|
2 |
+
|
3 |
+
[English](README.md) | 简体中文
|
4 |
+
|
5 |
+
|
6 |
+
## Introduction
|
7 |
+
|
8 |
+
为了探究小模型的潜能,我尝试构建一系列小模型,并存放于 [NanoLM Collections](https://huggingface.co/collections/Mxode/nanolm-66d6d75b4a69536bca2705b2)。
|
9 |
+
|
10 |
+
这是 NanoLM-70M-Instruct-v1。该模型目前仅支持**英文**。
|
11 |
+
|
12 |
+
|
13 |
+
## 模型详情
|
14 |
+
|
15 |
+
NanoLM-70M-Instruct-v1 的分词器和模型架构与 [SmolLM-135M](https://huggingface.co/HuggingFaceTB/SmolLM-135M) 相同,但层数从30减少到12。
|
16 |
+
|
17 |
+
因此,NanoLM-70M-Instruct-v1 的参数量只有 70 M。
|
18 |
+
|
19 |
+
尽管如此,NanoLM-70M-Instruct-v1 仍展示了指令跟随能力。
|
20 |
+
|
21 |
+
|
22 |
+
## 如何使用
|
23 |
+
|
24 |
+
```python
|
25 |
+
import torch
|
26 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
27 |
+
|
28 |
+
model_path = 'Mxode/NanoLM-70M-Instruct-v1'
|
29 |
+
|
30 |
+
model = AutoModelForCausalLM.from_pretrained(model_path).to('cuda:0', torch.bfloat16)
|
31 |
+
tokenizer = AutoTokenizer.from_pretrained(model_path)
|
32 |
+
|
33 |
+
|
34 |
+
text = "Why is it important for entrepreneurs to prioritize financial management?"
|
35 |
+
prompt = tokenizer.apply_chat_template(
|
36 |
+
[
|
37 |
+
{'role': 'system', 'content': 'You are a helpful assistant.'},
|
38 |
+
{'role': 'user', 'content': text}
|
39 |
+
],
|
40 |
+
add_generation_prompt=True,
|
41 |
+
tokenize=True,
|
42 |
+
return_tensors='pt'
|
43 |
+
).to('cuda:0')
|
44 |
+
|
45 |
+
|
46 |
+
outputs = model.generate(
|
47 |
+
prompt,
|
48 |
+
max_new_tokens=1024,
|
49 |
+
do_sample=True,
|
50 |
+
temperature=0.7,
|
51 |
+
repetition_penalty=1.1,
|
52 |
+
eos_token_id=tokenizer.eos_token_id,
|
53 |
+
)
|
54 |
+
response = tokenizer.decode(outputs[0])
|
55 |
+
print(response)
|
56 |
+
```
|