大语言模型参数微调过程（附完整代码）

这是一个金融领域大模型微调的具体代码执行过程，具体代码可以详见git仓库。

1.配置参数

model_args,data_args,train_args,finetuning_args,generating_args = get_train_args()

2.加载分词tokenizer

tokenizer = AutoTokenizer.from_pretrained(model_path)

3.读取config文件

config = AutoConfig.from_pretrained(model_path)

4.读取模型文件

model = AutoModelForCausalLM.from_pretrainded(model_path,config)

5.对模型初始化适配器

配置lora的参数，尽可能微调模型较少的参数。

model = init_adapter(model,args)

配置模型为训练模式

model = model.train()

6.开始训练模型

trianer = SftTrainer(model,args,tokenizer)

7.合并原始模型的权重与adapter的权重生成自己的模型

model_name_or_path = "your_LLM_model_path"
    adapter_name_or_path = "your_lora_model_path"
    save_path = "save_model_path"
    
    tokenizer = AutoTokenizer.from_pretrained(
        model_name_or_path,
        trust_remote_code=True
    )
    model = AutoModelForCausalLM.from_pretrained(
        model_name_or_path,
        trust_remote_code=True,
        low_cpu_mem_usage=True,
        torch_dtype=torch.float16,
        device_map="auto"
    )
    model = PeftModel.from_pretrained(model, adapter_name_or_path)
    model = model.merge_and_unload()
 
    tokenizer.save_pretrained(save_path)
    model.save_pretrained(save_path)

全部代码

https://gitee.com/iwuzhichao/sft.git