# 使用Hugging Face快速加载模型
from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
    "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
    torch_dtype=torch.bfloat16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

def format_medical_data(sample):
    return f"""【患者信息】
主诉：{sample['chief_complaint']}
现病史：{sample['history']}

【诊断过程】
1. 初步鉴别：{sample['differential_diagnosis']}
2. 关键检查：{sample['exams']}
3. 确诊依据：{sample['diagnosis_evidence']}

【最终诊断】{sample['final_diagnosis']}"""

from unsloth import FastLanguageModel

model, optimizer = FastLanguageModel.from_pretrained(
    model_name = "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
    max_seq_length = 4096,
    dtype = torch.bfloat16,
    load_in_4bit = True,
)

model = FastLanguageModel.get_peft_model(
    model,
    r=32,  # LoRA矩阵秩
    target_modules=["q_proj", "k_proj", "v_proj"],
    lora_alpha=64,
    lora_dropout=0.1,
)

[客户端] → [负载均衡] → [Cloud Run实例组] 
                    ↘ [医学知识图谱缓存]
                    ↘ [合规性审核模块]

# 使用Flash Attention V2优化
with torch.backends.cuda.sdp_kernel(
    enable_flash=True, 
    enable_math=False, 
    enable_mem_efficient=False
):
    outputs = model.generate(
        input_ids,
        max_new_tokens=256,
        temperature=0.7,
        do_sample=True,
    )

pip install transformers torch

from transformers import AutoTokenizer, AutoModelForCausalLM
 
model_name = "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B"
model = AutoModelForCausalLM.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "请告诉我12的阶乘是多少？"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(inputs.input_ids)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)