Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,45 @@
|
|
1 |
---
|
2 |
license: cc-by-nc-4.0
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: cc-by-nc-4.0
|
3 |
+
datasets:
|
4 |
+
- yuyijiong/Long-Instruction-Chinese
|
5 |
+
language:
|
6 |
+
- zh
|
7 |
+
- en
|
8 |
+
pipeline_tag: text-generation
|
9 |
---
|
10 |
+
[LongAlpaca](https://huggingface.co/Yukang/LongAlpaca-7B)通过对 llama2-chat 进行少量长文本数据的微调,展现出了优秀的长文本对话能力。\
|
11 |
+
LongAlpaca-7b-chinese 和 LongAlpaca 使用类似的训练方法:先使用线性位置插值,然后通过少量长文本数据的微调,使其获得优秀的长文本对话能力。\
|
12 |
+
此模型由atom-7b-chat经过lora微调得到, 通过线性位置插值,将文本长度从4k扩展到16k,可以完成上万字的多文档检索、论文总结等任务,而短对话能力几乎没有下降。\
|
13 |
+
使用方法:
|
14 |
+
```python
|
15 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
16 |
+
from transformers.generation import GenerationConfig
|
17 |
+
import os
|
18 |
+
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
|
19 |
+
|
20 |
+
model_path="yuyijiong/LongAlpaca-7b-chinese"
|
21 |
+
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
|
22 |
+
|
23 |
+
# use auto mode, automatically select precision based on the device.
|
24 |
+
model = AutoModelForCausalLM.from_pretrained(model_path, device_map="auto", load_in_8bit=True).eval()
|
25 |
+
|
26 |
+
question="中国的首都是什么?"
|
27 |
+
input_text = "<s>Human: " + question + "\n</s><s>Assistant: "
|
28 |
+
input_ids = tokenizer(input_text, return_tensors='pt').input_ids.to(model.device)
|
29 |
+
|
30 |
+
with torch.no_grad():
|
31 |
+
with torch.autocast('cuda'):
|
32 |
+
output = model.generate(input_ids=input_ids,
|
33 |
+
max_new_tokens=max_new_tokens,
|
34 |
+
do_sample=True,
|
35 |
+
temperature=0.85,
|
36 |
+
top_k=None,
|
37 |
+
top_p=0.9,
|
38 |
+
use_cache=True,
|
39 |
+
**kwargs)
|
40 |
+
|
41 |
+
reply = tokenizer.decode(output[0], skip_special_tokens=False)
|
42 |
+
reply_return=reply.split('Assistant:')[-1].replace('</s>', '')
|
43 |
+
|
44 |
+
print('模型回答:', reply_return)
|
45 |
+
```
|