ZhangQiao123/medgemma-27b-it-chinese-medical-qa-lora

pip install transformers peft torch bitsandbytes accelerate huggingface_hub

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
from peft import PeftModel

# 量化配置（推荐用于GPU内存优化）
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_use_double_quant=True,
)

# 加载基础模型
base_model = AutoModelForCausalLM.from_pretrained(
    "google/medgemma-27b-it",
    quantization_config=bnb_config,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True
)

# 加载tokenizer
tokenizer = AutoTokenizer.from_pretrained("google/medgemma-27b-it")
if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token

# 加载LoRA适配器
model = PeftModel.from_pretrained(
    base_model,
    "ZhangQiao123/medgemma-27b-it-chinese-medical-qa-lora"
)

# 🚀 快速测试脚本 - 可直接在Colab中运行
import subprocess
import sys

# 安装依赖
packages = ["transformers", "peft", "bitsandbytes", "accelerate", "huggingface_hub"]
for pkg in packages:
    subprocess.run([sys.executable, "-m", "pip", "install", "-q", pkg])

# 设置认证（需要申请MedGemma访问权限）
from huggingface_hub import login
login()  # 输入您的Hugging Face Token

# 运行测试
exec("""
import torch
import time
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
from peft import PeftModel

# 配置和加载模型
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True, bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16, bnb_4bit_use_double_quant=True
)

tokenizer = AutoTokenizer.from_pretrained("google/medgemma-27b-it")
if tokenizer.pad_token is None: tokenizer.pad_token = tokenizer.eos_token

base_model = AutoModelForCausalLM.from_pretrained(
    "google/medgemma-27b-it", quantization_config=bnb_config,
    device_map="auto", torch_dtype=torch.bfloat16, trust_remote_code=True
)

model = PeftModel.from_pretrained(base_model, "ZhangQiao123/medgemma-27b-it-chinese-medical-qa-lora")

# 测试问题
question = "什么是高血压？它的主要症状有哪些？"
prompt = f"你是一位专业的医学助手，请根据医学知识回答以下问题。请提供准确、专业且易懂的回答。\\n\\n问题：{question}\\n\\n回答："

inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=2048).to(model.device)
with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=300, temperature=0.7, top_p=0.9, do_sample=True, pad_token_id=tokenizer.eos_token_id, repetition_penalty=1.1)

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
answer = response.split("回答：")[1].strip() if "回答：" in response else response[len(prompt):].strip()

print(f"问题: {question}")
print(f"回答: {answer}")
""")

def generate_medical_answer(question):
    prompt = f"作为一名专业的医生，请回答以下医学问题：\n\n问题：{question}\n\n回答："
    
    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=2048)
    inputs = {k: v.to(model.device) for k, v in inputs.items()}
    
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=200,
            temperature=0.7,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id
        )
    
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    answer = response.split("回答：")[-1].strip()
    return answer

# 使用示例
questions = [
    "什么是高血压？",
    "感冒和流感有什么区别？",
    "糖尿病患者应该注意什么饮食？"
]

for question in questions:
    answer = generate_medical_answer(question)
    print(f"问题: {question}")
    print(f"回答: {answer}\n")

@misc{medgemma-chinese-qa-lora-2025,
  title={MedGemma Chinese Medical QA LoRA},
  author={MedGemma Team},
  year={2025},
  howpublished={\url{https://huggingface.co/ZhangQiao123/medgemma-27b-it-chinese-medical-qa-lora}}
}

@article{sellergren2025medgemma,
  title={MedGemma Technical Report},
  author={Sellergren, Andrew and Kazemzadeh, Sahar and Jaroensri, Tiam and Kiraly, Atilla and Traverse, Madeleine and Kohlberger, Timo and Xu, Shawn and Jamil, Fayaz and Hughes, Cían and Lau, Charles and others},
  journal={arXiv preprint arXiv:2507.05201},
  year={2025}
}