微调大模型的tokenize_function编写求教

最近在学习大模型，关于tokenize_function或者preprocess_function,GPT等大模型给出的代码demo中通常会有2个版本：

def preprocess_function(examples):
    inputs = [ex["input"] for ex in examples]
    targets = [ex["output"] for ex in examples]
    model_inputs = tokenizer(inputs, max_length=512, truncation=True, padding="max_length")
    labels = tokenizer(targets, max_length=512, truncation=True, padding="max_length")["input_ids"]
    model_inputs["labels"] = labels
    return model_inputs

def preprocess_function(examples):
    inputs = examples["text"]
    model_inputs = tokenizer(inputs, max_length=512, truncation=True, padding="max_length")
    labels = model_inputs["input_ids"].copy()
    model_inputs["labels"] = labels
    return model_inputs

前者是将模型的输出传给labels，后者是将整个文本传给labels
两种数据结构背后的训练逻辑看起来完全不一样啊，我到底应该用哪一种？还是说选择哪一种逻辑跟代码其他的设置相关？