初始化项目,由ModelHub XC社区提供模型
Model: Ndlcwx/qwen_1.8B-SFT Source: Original Platform
This commit is contained in:
61
README.md
Normal file
61
README.md
Normal file
@@ -0,0 +1,61 @@
|
||||
# qwen_1.8B-SFT
|
||||
|
||||
此模型是基于qwen_1.8B-base模型进行SFT训练的一个chat模型,由于SFT训练语料较少,约为172万条,并且max_seq_len为512,因此具有一定的局限性,目的主要是测试此模型SFT的可行性。
|
||||
其中要求与依赖项与qwen官网相同,这里不做过多的详细描述。
|
||||
|
||||
## 要求(Requirements)
|
||||
|
||||
* python 3.8及以上版本
|
||||
* pytorch 1.12及以上版本,推荐2.0及以上版本
|
||||
* 建议使用CUDA 11.4及以上(GPU用户、flash-attention用户等需考虑此选项)
|
||||
* python 3.8 and above
|
||||
* pytorch 1.12 and above, 2.0 and above are recommended
|
||||
* CUDA 11.4 and above are recommended (this is for GPU users, flash-attention users, etc.)
|
||||
|
||||
## 依赖项(Dependency)
|
||||
|
||||
运行qwen_1.8B-SFT,请确保满足上述要求,再执行以下pip命令安装依赖库
|
||||
|
||||
```bash
|
||||
pip install transformers==4.32.0 accelerate tiktoken einops scipy transformers_stream_generator==0.0.4 peft deepspeed
|
||||
```
|
||||
另外,推荐安装`flash-attention`库(**当前已支持flash attention 2**),以实现更高的效率和更低的显存占用。
|
||||
|
||||
```bash
|
||||
git clone https://github.com/Dao-AILab/flash-attention
|
||||
cd flash-attention && pip install .
|
||||
# 下方安装可选,安装可能比较缓慢。
|
||||
# pip install csrc/layer_norm
|
||||
# pip install csrc/rotary
|
||||
```
|
||||
<br>
|
||||
|
||||
## 快速使用(Quickstart)
|
||||
|
||||
```python
|
||||
from modelscope import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
|
||||
|
||||
tokenizer = AutoTokenizer.from_pretrained("Ndlcwx/qwen_1.8B-SFT", revision='master', trust_remote_code=True)
|
||||
|
||||
model = AutoModelForCausalLM.from_pretrained("Ndlcwx/qwen_1.8B-SFT", revision='master', device_map="auto", trust_remote_code=True).eval()
|
||||
|
||||
response, history = model.chat(tokenizer, "你好", history=None)
|
||||
print(response)
|
||||
|
||||
response, history = model.chat(tokenizer, "给我讲一个年轻人奋斗创业最终取得成功的故事。", history=history)
|
||||
print(response)
|
||||
|
||||
response, history = model.chat(tokenizer, "给这个故事起一个标题", history=history)
|
||||
print(response)
|
||||
|
||||
response, history = model.chat(tokenizer, "请写一段Python代码", history=history)
|
||||
print(response)
|
||||
```
|
||||
这里仅展示部分问题的回答效果:
|
||||

|
||||

|
||||

|
||||

|
||||

|
||||

|
||||
**可以看出Qwen-1.8B-SFT的效果还是不错的,毕竟预训练模型得到了充分的训练,拥有一个足够“聪明”的大脑**
|
||||
Reference in New Issue
Block a user