108 lines
3.5 KiB
Markdown
108 lines
3.5 KiB
Markdown
|
---
|
|||
|
|
frameworks:
|
|||
|
|
- Pytorch
|
|||
|
|
license: Apache License 2.0
|
|||
|
|
tasks:
|
|||
|
|
- text-generation
|
|||
|
|
|
|||
|
|
#model-type:
|
|||
|
|
##如 gpt、phi、llama、chatglm、baichuan 等
|
|||
|
|
#- gpt
|
|||
|
|
|
|||
|
|
#domain:
|
|||
|
|
##如 nlp、cv、audio、multi-modal
|
|||
|
|
#- nlp
|
|||
|
|
|
|||
|
|
#language:
|
|||
|
|
##语言代码列表 https://help.aliyun.com/document_detail/215387.html?spm=a2c4g.11186623.0.0.9f8d7467kni6Aa
|
|||
|
|
#- cn
|
|||
|
|
|
|||
|
|
#metrics:
|
|||
|
|
##如 CIDEr、Blue、ROUGE 等
|
|||
|
|
#- CIDEr
|
|||
|
|
|
|||
|
|
#tags:
|
|||
|
|
##各种自定义，包括 pretrained、fine-tuned、instruction-tuned、RL-tuned 等训练方法和其他
|
|||
|
|
#- pretrained
|
|||
|
|
|
|||
|
|
#tools:
|
|||
|
|
##如 vllm、fastchat、llamacpp、AdaSeq 等
|
|||
|
|
#- vllm
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
[Github](https://github.com/Zeyi-Lin/Qwen3-Medical-SFT)
|
|||
|
|
|
|||
|
|
- **基础模型**：[Qwen3-1.7B](https://modelscope.cn/models/Qwen/Qwen3-1.7B/summary)
|
|||
|
|
- **微调后模型**：[Qwen3-1.7b-Medical-R1-sft](https://modelscope.cn/models/testUser/Qwen3-1.7b-Medical-R1-sft/summary)
|
|||
|
|
- **数据集**：[delicate_medical_r1_data](https://modelscope.cn/datasets/krisfu/delicate_medical_r1_data)
|
|||
|
|
- **SwanLab**：[qwen3-sft-medical](https://swanlab.cn/@ZeyiLin/qwen3-sft-medical/runs/agps0dkifth5l1xytcdyk/chart)
|
|||
|
|
- **微调方式**：全参数微调、LoRA微调
|
|||
|
|
- **推理风格**：R1推理风格
|
|||
|
|
- **算力要求**：
|
|||
|
|
- **全参数微调**：32GB显存
|
|||
|
|
- **LoRA微调**：28GB显存
|
|||
|
|
- **图文教程**：[Qwen3大模型微调入门实战（完整代码）](https://zhuanlan.zhihu.com/p/1903848838214705484)
|
|||
|
|
|
|||
|
|
## 模型下载
|
|||
|
|
|
|||
|
|
```bash
|
|||
|
|
#安装ModelScope
|
|||
|
|
pip install modelscope
|
|||
|
|
```
|
|||
|
|
```python
|
|||
|
|
#SDK模型下载
|
|||
|
|
from modelscope import snapshot_download
|
|||
|
|
model_dir = snapshot_download('testUser/Qwen3-1.7b-Medical-R1-sft')
|
|||
|
|
```
|
|||
|
|
Git下载
|
|||
|
|
```
|
|||
|
|
#Git模型下载
|
|||
|
|
git clone https://www.modelscope.cn/testUser/Qwen3-1.7b-Medical-R1-sft.git
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
<p style="color: lightgrey;">如果您是本模型的贡献者，我们邀请您根据<a href="https://modelscope.cn/docs/ModelScope%E6%A8%A1%E5%9E%8B%E6%8E%A5%E5%85%A5%E6%B5%81%E7%A8%8B%E6%A6%82%E8%A7%88" style="color: lightgrey; text-decoration: underline;">模型贡献文档</a>，及时完善模型卡片内容。</p>
|
|||
|
|
|
|||
|
|
## 模型推理
|
|||
|
|
|
|||
|
|
```bash
|
|||
|
|
import torch
|
|||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|||
|
|
|
|||
|
|
def predict(messages, model, tokenizer):
|
|||
|
|
if torch.backends.mps.is_available():
|
|||
|
|
device = "mps"
|
|||
|
|
elif torch.cuda.is_available():
|
|||
|
|
device = "cuda"
|
|||
|
|
else:
|
|||
|
|
device = "cpu"
|
|||
|
|
|
|||
|
|
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
|||
|
|
model_inputs = tokenizer([text], return_tensors="pt").to(device)
|
|||
|
|
|
|||
|
|
generated_ids = model.generate(model_inputs.input_ids, max_new_tokens=2048)
|
|||
|
|
generated_ids = [output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)]
|
|||
|
|
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
|||
|
|
|
|||
|
|
return response
|
|||
|
|
|
|||
|
|
|
|||
|
|
# 加载原下载路径的tokenizer和model
|
|||
|
|
tokenizer = AutoTokenizer.from_pretrained("./Qwen3-1.7b-Medical-R1-sft", use_fast=False, trust_remote_code=True)
|
|||
|
|
model = AutoModelForCausalLM.from_pretrained("./Qwen3-1.7b-Medical-R1-sft", device_map="auto", torch_dtype=torch.bfloat16)
|
|||
|
|
|
|||
|
|
test_texts = {
|
|||
|
|
'instruction': "你是一个医学专家，你需要根据用户的问题，给出带有思考的回答。",
|
|||
|
|
'input': "医生，我最近被诊断为糖尿病，听说碳水化合物的选择很重要，我应该选择什么样的碳水化合物呢？"
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
instruction = test_texts['instruction']
|
|||
|
|
input_value = test_texts['input']
|
|||
|
|
|
|||
|
|
messages = [
|
|||
|
|
{"role": "system", "content": f"{instruction}"},
|
|||
|
|
{"role": "user", "content": f"{input_value}"}
|
|||
|
|
]
|
|||
|
|
|
|||
|
|
response = predict(messages, model, tokenizer)
|
|||
|
|
print(response)
|
|||
|
|
```
|