qwen2 lora微调报错：RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn

[ArlanCooper]

环境按照文档进行安装:https://github.com/datawhalechina/self-llm/blob/master/Qwen2/05-Qwen2-7B-Instruct%20Lora%20%E5%BE%AE%E8%B0%83.md

python版本:3.10.12 cuda： 12.1 os: ubuntu12 运行代码:

# -*- coding: utf-8 -*-

from datasets import Dataset
import pandas as pd
from transformers import (
    AutoTokenizer,
    AutoModelForCausalLM,
    Trainer,
    DataCollatorForSeq2Seq,
)
from peft import get_peft_model
from app.configs.config_llm import ConfigQwen, config_lora, training_args

class Qwen2trainer:
    """
    qwen2微调
    """

    def __init__(self):
        self.train_path = ConfigQwen.train_data_path
        self.model_path = ConfigQwen.model_path
        df = pd.read_json(self.train_path)
        ds = Dataset.from_pandas(df)
        self.max_length = ConfigQwen.MAX_LENGTH
        self.tokenizer = AutoTokenizer.from_pretrained(
            self.model_path, use_fast=False, trust_remote_code=True
        )
        self.tokenized_id = ds.map(
            self.process_func, remove_columns=ds.column_names
        )
        base_model = AutoModelForCausalLM.from_pretrained(
            self.model_path, device_map="auto", trust_remote_code=True
        )
        self.full_model = get_peft_model(base_model, config_lora)

    def process_func(self, example):
        """
        数据处理
        """
        max_length = self.max_length  # 限制最大长度
        instruction = self.tokenizer(
            f"<|im_start|>system\n你是一个有用的助手<|im_end|>\n<|im_start|>user\n{example['instruction'] + example['input']}<|im_end|>\n<|im_start|>assistant\n",
            add_special_tokens=False,
        )  # add_special_tokens 不在开头加 special_tokens
        response = self.tokenizer(f"{example['output']}", add_special_tokens=False)
        input_ids = (
            instruction["input_ids"]
            + response["input_ids"]
            + [self.tokenizer.pad_token_id]
        )
        attention_mask = (
            instruction["attention_mask"] + response["attention_mask"] + [1]
        )  # 因为eos token咱们也是要关注的所以 补充为1
        labels = (
            [-100] * len(instruction["input_ids"])
            + response["input_ids"]
            + [self.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,
        }

    def train(self):
        '''
        训练
        '''
        trainer = Trainer(
            model=self.full_model,
            args=training_args,
            train_dataset=self.tokenized_id,
            data_collator=DataCollatorForSeq2Seq(
                tokenizer=self.tokenizer, padding=True
            ),
        )
        trainer.train()

if __name__ == "__main__":
    qwen2 = Qwen2trainer()
    qwen2.train()

报错信息:

Detected kernel version 4.19.118, which is below the recommended minimum of 5.5.0; this can cause the process to hang. It is recommended to upgrade the kernel to the minimum version or higher.
  0%|                                                                                   | 0/53850 [00:00<?, ?it/s]`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...
/home/powerop/work/conda/envs/qwen2/lib/python3.10/site-packages/torch/utils/checkpoint.py:464: UserWarning: torch.utils.checkpoint: the use_reentrant parameter should be passed explicitly. In version 2.4 we will raise an exception if use_reentrant is not passed. use_reentrant=False is recommended, but if you need to preserve the current default behavior, you can pass use_reentrant=True. Refer to docs for more details on the differences between the two variants.
  warnings.warn(
/home/powerop/work/conda/envs/qwen2/lib/python3.10/site-packages/torch/utils/checkpoint.py:91: UserWarning: None of the inputs have requires_grad=True. Gradients will be None
  warnings.warn(
Traceback (most recent call last):
  File "/home/powerop/work/rwq/llm_finetune/app/finetune_qwen2.py", line 92, in <module>
    qwen2.train()
  File "/home/powerop/work/rwq/llm_finetune/app/finetune_qwen2.py", line 87, in train
    trainer.train()
  File "/home/powerop/work/conda/envs/qwen2/lib/python3.10/site-packages/transformers/trainer.py", line 1885, in train
    return inner_training_loop(
  File "/home/powerop/work/conda/envs/qwen2/lib/python3.10/site-packages/transformers/trainer.py", line 2216, in _inner_training_loop
    tr_loss_step = self.training_step(model, inputs)
  File "/home/powerop/work/conda/envs/qwen2/lib/python3.10/site-packages/transformers/trainer.py", line 3250, in training_step
    self.accelerator.backward(loss)
  File "/home/powerop/work/conda/envs/qwen2/lib/python3.10/site-packages/accelerate/accelerator.py", line 1966, in backward
    loss.backward(**kwargs)
  File "/home/powerop/work/conda/envs/qwen2/lib/python3.10/site-packages/torch/_tensor.py", line 525, in backward
    torch.autograd.backward(
  File "/home/powerop/work/conda/envs/qwen2/lib/python3.10/site-packages/torch/autograd/__init__.py", line 267, in backward
    _engine_run_backward(
  File "/home/powerop/work/conda/envs/qwen2/lib/python3.10/site-packages/torch/autograd/graph.py", line 744, in _engine_run_backward
    return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn

[KMnO4-zx]

看样子应该是少了c++的什么环境，在教程制作过程中未曾遇到你的问题

[ArlanCooper]

看样子应该是少了c++的什么环境，在教程制作过程中未曾遇到你的问题 请问一下有什么解决方案吗？

[ArlanCooper]

我这边已经解决了哈，少敲了一行命令: model.enable_input_require_grads() # 开启梯度检查点时，要执行该方法