Qwen1.5-7B Lora微调报错：RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn

[feifeifei-hue]

您好，我的代码感觉跟您几乎一模一样，但不知道为什么会报这样的错误，麻烦您给出指导意见，谢谢您~
代码如下：

from datasets import Dataset
import pandas as pd
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, DataCollatorForSeq2Seq, TrainingArguments, Trainer, GenerationConfig
from peft import LoraConfig, TaskType, get_peft_model

df = pd.read_json('test.json', encoding='utf-8')
ds = Dataset.from_pandas(df)

# print(ds[:3])

def process_func(example):
    # Llama分词器会将一个中文字切分为多个token，因此需要放开一些最大长度，保证数据的完整性
    MAX_LENGTH = 384
    input_ids, attention_mask, labels = [], [], []
    # add_special_tokens 不在开头加 special_tokens
    instruction = tokenizer(f"<|im_start|>system\nYou are a friendly and helpful assistant, please strictly follow the prompt I give you to generate content.<|im_end|>\n<|im_start|>user\n{example['instruction'] + example['input']}<|im_end|>\n<|im_start|>assistant\n", add_special_tokens=False)  
    response = tokenizer(f"{example['output']}", add_special_tokens=False)
    input_ids = instruction["input_ids"] + response["input_ids"] + [tokenizer.pad_token_id]
    # 因为eos token咱们也是要关注的所以 补充为1
    attention_mask = instruction["attention_mask"] + response["attention_mask"] + [1]  
    labels = [-100] * len(instruction["input_ids"]) + response["input_ids"] + [tokenizer.pad_token_id]  
    if len(input_ids) > MAX_LENGTH:  # 做一个截断
        input_ids = input_ids[:MAX_LENGTH]
        attention_mask = attention_mask[:MAX_LENGTH]
        labels = labels[:MAX_LENGTH]

    return {
        "input_ids": input_ids,
        "attention_mask": attention_mask,
        "labels": labels
    }

tokenized_id = ds.map(process_func, remove_columns=ds.column_names)
# print(tokenized_id)

# print(tokenizer.decode(tokenized_id[0]['input_ids']))

# print(tokenizer.decode(list(filter(lambda x: x != -100, tokenized_id[1]["labels"]))))

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen1.5-7B-Chat", use_fast=False, trust_remote_code=True)

model = AutoModelForCausalLM.from_pretrained(
    'Qwen/Qwen1.5-7B-Chat', 
    device_map="auto",
    torch_dtype=torch.bfloat16
)

# LoRA
config = LoraConfig(
    task_type=TaskType.CAUSAL_LM, 
    target_modules=["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj"],
    inference_mode=False, # 训练模式
    r=8, # Lora 秩
    lora_alpha=32, # Lora alaph，具体作用参见 Lora 原理
    lora_dropout=0.1# Dropout 比例
)

model = get_peft_model(model, config)

print(config)

print(model.print_trainable_parameters())

args = TrainingArguments(
    output_dir="fine_tuning_output",
    per_device_train_batch_size=4,
    gradient_accumulation_steps=4,
    logging_steps=10,
    num_train_epochs=3,
    save_steps=100,
    learning_rate=1e-4,
    save_on_each_node=True,
    gradient_checkpointing=True
)

trainer = Trainer(
    model=model,
    args=args,
    train_dataset=tokenized_id,
    data_collator=DataCollatorForSeq2Seq(tokenizer=tokenizer, padding=True),
)

trainer.train()

[KMnO4-zx]

[feifeifei-hue]

您好，谢谢您的解答，我这里一直设置的True；

我最终是把print(model.print_trainable_parameters()) 改为了 model.print_trainable_parameters() 解决的，即去掉了print...但是我不知道为什么这样会得到解决，您可以讲解一下吗？