大模型微调的核心 处理代码

一、 核心代码

# input-01
PROMPT_TEMPLATE = (
        "Below is an instruction that describes a task. "
        "Write a response that appropriately completes the request.\n\n"
        "### Instruction:\n{instruction}\n\n### Response: "
    )

def build_instruction_dataset(data_path: Union[List[str],str],
                tokenizer: transformers.PreTrainedTokenizer,
                max_seq_length: int, data_cache_dir = None,
                preprocessing_num_workers = None,
                ):

    def tokenization(examples):
        sources = []
        targets = []
        prompt = PROMPT_TEMPLATE
        for instruction, input, output in zip(examples['instruction'],examples['input'],examples['output']):
            # input-02
            if input is not None and input !="":
                instruction = instruction+'\n'+input
            source = prompt.format_map({'instruction':instruction})
            target = f"{output}{tokenizer.eos_token}"

            sources.append(source)
            targets.append(target)

        tokenized_sources = tokenizer(sources,return_attention_mask=False)
        tokenized_targets = tokenizer(targets,return_attention_mask=False,add_special_tokens=False)

        all_input_ids = []
        all_labels = []
        for s,t in zip(tokenized_sources['input_ids'],tokenized_targets['input_ids']):
            # input-03，模型的输入是包括input与label的
            input_ids = torch.LongTensor(s + t)[:max_seq_length]
            # output-01, label中输入位全置为-100
            labels = torch.LongTensor([IGNORE_INDEX] * len(s) + t)[:max_seq_length]
            assert len(input_ids) == len(labels)
            all_input_ids.append(input_ids)
            all_labels.append(labels)

        results = {'input_ids':all_input_ids, 'labels': all_labels}
        return results

二、结论：

{'instruction': '创建一个完成这句话的句子。', 
 'input': '她收拾好了行李，然后', 
  'output': '踏上了她的旅程。'}
 
{'instruction': '创作一张对50岁以上的人有趣的模因（meme）。',
  'input': '',
  'output': '一张模因图片，上面是一位拿着拐杖或行走杖的老人望向远方，配有字幕：“当二十多岁的人们谈论‘好旧时光’时”。'}

1、human_input：PROMPT_TEMPLATE + instruction + input
2、model_response: output

微调模型输入：
human_input + model_response

微调模型输出：
[- 100] * len(human_input ) + model_response