大语言模型-中文chatGLM-LLAMA微调

微调
大语言模型-ChatGLM-Tuning
大语言模型-微调chatglm6b
大语言模型-中文chatGLM-LLAMA微调
大语言模型-alpaca-lora

本地知识库
大语言模型2-document ai解读
大语言模型-DocumentSearch解读
大语言模型-中文Langchain

本文解读代码的地址：
https://github.com/27182812/ChatGLM-LLaMA-chinese-insturct

中文instruct在chatGLM, LLAMA上的表现

数据

json的预处理

相比大语言模型-ChatGLM-Tuning中，是两个函数都放在了dataprocess的一个类中进行，初步看起来需要改变的几乎相同

微调

对于chatGLM和之前文章几乎相同，这里主要关注一下LLAMA

数据

def generate_prompt(data_point):
    # sorry about the formatting disaster gotta move fast
    if data_point["input"]:
        return f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
### Instruction:
{data_point["instruction"]}
### Input:
{data_point["input"]}
### Response:
{data_point["output"]}"""
    else:
        return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.
### Instruction:
{data_point["instruction"]}
### Response:
{data_point["output"]}"""

def tokenize(prompt):
    # there's probably a way to do this with the tokenizer settings
    # but again, gotta move fast
    result = tokenizer(
        prompt,
        truncation=True,
        max_length=CUTOFF_LEN + 1,
        padding="max_length",
    )
    return {
        "input_ids": result["input_ids"][:-1],
        "attention_mask": result["attention_mask"][:-1],
    }

模型

model = LlamaForCausalLM.from_pretrained(
    "decapoda-research/llama-7b-hf",
    load_in_8bit=True,
    device_map="auto",
)
tokenizer = LlamaTokenizer.from_pretrained(
    "decapoda-research/llama-7b-hf", add_eos_token=True
)

model = prepare_model_for_int8_training(model)

config = LoraConfig(
    r=LORA_R,
    lora_alpha=LORA_ALPHA,
    target_modules=["q_proj", "v_proj"],
    lora_dropout=LORA_DROPOUT,
    bias="none",
    task_type="CAUSAL_LM",
)
model = get_peft_model(model, config)
tokenizer.pad_token_id = 0  # unk. we want this to be different from the eos token

微调

data = data.shuffle().map(lambda x: tokenize(generate_prompt(x)))

trainer = transformers.Trainer(
    model=model,
    train_dataset=data["train"],
    args=transformers.TrainingArguments(
        per_device_train_batch_size=MICRO_BATCH_SIZE,
        gradient_accumulation_steps=GRADIENT_ACCUMULATION_STEPS,
        warmup_steps=100,
        num_train_epochs=EPOCHS,
        learning_rate=LEARNING_RATE,
        fp16=True,
        logging_steps=20,
        output_dir="qys-alpaca-chinese",
        save_total_limit=3,
    ),
    data_collator=transformers.DataCollatorForLanguageModeling(tokenizer, mlm=False),
)
model.config.use_cache = False
trainer.train(resume_from_checkpoint=False)
# trainer.train()

model.save_pretrained("qys-alpaca-chinese")

推理

推理代码

tokenizer = LlamaTokenizer.from_pretrained("decapoda-research/llama-7b-hf")

model = LlamaForCausalLM.from_pretrained(
    "decapoda-research/llama-7b-hf",
    load_in_8bit=True,
    torch_dtype=torch.float16,
    device_map="auto",
)

model = PeftModel.from_pretrained( 
    model, "./qys-alpaca-chinese", torch_dtype=torch.float16
)

def generate_prompt(instruction, input=None):
    if input:
        return f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
### Instruction:
{instruction}
### Input:
{input}
### Response:"""
    else:
        return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.
### Instruction:
{instruction}
### Response:"""


instructions = json.load(open("data/zh-data01.json"))

answers = []

with torch.no_grad():
    for idx, item in enumerate(instructions[12:18]):
        feature = format_example(item)
        input_text = feature['context']
        print(input_text)
        inputs = tokenizer(input_text, return_tensors="pt")
        input_ids = inputs["input_ids"].cuda()
        generation_config = GenerationConfig(
            temperature=0.1,
            top_p=0.75,
            top_k=40,
            num_beams=4,
        )
        generation_output = model.generate(
            input_ids=input_ids,
            generation_config=generation_config,
            return_dict_in_generate=True,
            output_scores=True,
            max_new_tokens=256,
        )
        s = generation_output.sequences[0]
        output = tokenizer.decode(s)
        print(output.strip())
        print("--------------------------------------------")