LoRA与QLoRA微调Qwen2.5-7B：从数据构建到推理效果对比

import torch
from transformers import AutoModelForCausalLM, TrainingArguments, BitsAndBytesConfig
from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training
from trl import SFTTrainer

# ---- QLoRA 4bit配置（LoRA组删掉这一段，模型用bf16加载）----
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_use_double_quant=True,
    bnb_4bit_compute_dtype=torch.bfloat16,
)

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    quantization_config=bnb_config,   # LoRA组去掉这行，加 torch_dtype=torch.bfloat16
    device_map="auto",
    trust_remote_code=True,
)
model = prepare_model_for_kbit_training(model)  # LoRA组可省略

lora_config = LoraConfig(
    r=16,
    lora_alpha=32,
    lora_dropout=0.05,
    bias="none",
    task_type="CAUSAL_LM",
    target_modules=["q_proj","k_proj","v_proj","o_proj",
                    "gate_proj","up_proj","down_proj"],
)
model = get_peft_model(model, lora_config)
model.print_trainable_parameters()
# 输出: trainable params: 40,370,176 || all params: 7,655,986,688 || 0.5273%

args = TrainingArguments(
    output_dir="./qwen2.5-7b-medical-lora",
    num_train_epochs=3,
    per_device_train_batch_size=2,
    gradient_accumulation_steps=8,      # 等效batch=16
    learning_rate=2e-4,
    lr_scheduler_type="cosine",
    warmup_ratio=0.03,
    logging_steps=10,
    save_strategy="epoch",
    eval_strategy="epoch",
    bf16=True,
    gradient_checkpointing=True,
    optim="paged_adamw_8bit",
    max_grad_norm=0.3,
    report_to="none",
)

trainer = SFTTrainer(
    model=model,
    args=args,
    train_dataset=dataset["train"],
    eval_dataset=dataset["test"],
    dataset_text_field="text",
    max_seq_length=1024,
    packing=False,
)
trainer.train()
trainer.model.save_pretrained("./lora_adapter")

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

base = AutoModelForCausalLM.from_pretrained(
    model_id, torch_dtype=torch.bfloat16, device_map="auto"
)
model = PeftModel.from_pretrained(base, "./lora_adapter")
model = model.merge_and_unload()
model.eval()

tokenizer = AutoTokenizer.from_pretrained(model_id)
prompt = "患者男，58岁，2型糖尿病，eGFR 25，能否使用二甲双胍？"
messages = [
    {"role": "system", "content": "你是一名专业的医疗问诊助手，请严格按JSON格式输出。"},
    {"role": "user", "content": prompt},
]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
with torch.no_grad():
    out = model.generate(**inputs, max_new_tokens=256, temperature=0.1, do_sample=False)
print(tokenizer.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

{
  "diagnosis_suggestion": "eGFR 25属于重度肾功能不全，二甲双胍禁用。",
  "drug_interaction": "建议改用胰岛素或格列喹酮等肾排泄少的降糖药。",
  "risk_level": "high"
}