大模型微调：让AI更懂你的业务

最流行的参数高效微调方法

# LoRA微调示例
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import LoraConfig, get_peft_model, TaskType
import torch

# 加载基础模型
model_name = "meta-llama/Llama-2-7b-hf"
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# LoRA配置
lora_config = LoraConfig(
    r=16,                      # LoRA秩
    lora_alpha=32,            # LoRA缩放参数
    target_modules=[          # 目标模块
        "q_proj",
        "k_proj", 
        "v_proj",
        "o_proj"
    ],
    lora_dropout=0.1,         # Dropout率
    bias="none",              # 偏置处理
    task_type=TaskType.CAUSAL_LM
)

# 应用LoRA
model = get_peft_model(model, lora_config)

# 打印可训练参数
model.print_trainable_parameters()
# trainable params: 4,194,304 || all params: 6,742,609,920 || trainable%: 0.06%

# 准备训练数据
def prepare_dataset(examples):
    # 构建指令格式
    texts = []
    for instruction, output in zip(examples['instruction'], examples['output']):
        text = f"""### Instruction:
{instruction}

### Response:
{output}"""
        texts.append(text)
    
    # Tokenize
    encodings = tokenizer(
        texts,
        truncation=True,
        padding=True,
        max_length=512,
        return_tensors='pt'
    )
    
    return encodings

# 训练配置
from transformers import TrainingArguments, Trainer

training_args = TrainingArguments(
    output_dir="./lora_model",
    num_train_epochs=3,
    per_device_train_batch_size=4,
    gradient_accumulation_steps=4,
    warmup_steps=100,
    learning_rate=2e-4,
    fp16=True,
    logging_steps=10,
    save_strategy="epoch",
    evaluation_strategy="epoch"
)

# 开始训练
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=train_dataset,
    eval_dataset=eval_dataset,
    tokenizer=tokenizer
)

trainer.train()

# 保存LoRA权重
model.save_pretrained("./lora_weights")

# 推理时加载
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained(model_name)
model = PeftModel.from_pretrained(base_model, "./lora_weights")
model = model.merge_and_unload()  # 合并权重