[BUG] <title>TypeError: CPMTrainer.training_step() takes 3 positional arguments but 4 were given

[FrancisFan98]

是否已有关于该错误的issue或讨论？ | Is there an existing issue / discussion for this?

• 我已经搜索过已有的issues和讨论 | I have searched the existing issues / discussions

该问题是否在FAQ中有解答？ | Is there an existing answer for this in FAQ?

• 我已经搜索过FAQ | I have searched FAQ

当前行为 | Current Behavior

[rank0]: Traceback (most recent call last):
[rank0]: File "/opt/MiniCPM-V-main/finetune/finetune.py", line 299, in
[rank0]: train()
[rank0]: File "/opt/MiniCPM-V-main/finetune/finetune.py", line 289, in train
[rank0]: trainer.train()
[rank0]: File "/root/miniconda3/envs/transformers/lib/python3.10/site-packages/transformers/trainer.py", line 2164, in train
[rank0]: return inner_training_loop(
[rank0]: File "/root/miniconda3/envs/transformers/lib/python3.10/site-packages/transformers/trainer.py", line 2524, in _inner_training_loop
[rank0]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
[rank0]: TypeError: CPMTrainer.training_step() takes 3 positional arguments but 4 were given

I got this error when I tried to finetune the model, anyone knows how to solve it?

期望行为 | Expected Behavior

No response

复现方法 | Steps To Reproduce

No response

运行环境 | Environment

- OS:
- Python:
- Transformers:
- PyTorch:
- CUDA (`python -c 'import torch; print(torch.version.cuda)'`):

备注 | Anything else?

No response

[Bravo5542]

Same problem, recently appeared

[BugBugdie]

same, anybody solved?

[owl-10]

Same problem, recently appeared

[PhucNDA]

It is because of transformers version:
Transformers:
transformers==4.47.1: vllms (used for inference)
transformers==4.40.0: trainable (used for training)

My current solution is swapping between these two. Looking forward to more feasible approaches...

[yuting89830]

Same problem

[PhucNDA]

Envs little bit messy but this works for me

Training:

pip uninstall vllm-flash-attn
pip uninstall xformers
pip uninstall openai
pip install -r ../requirements.txt

Testing:
pip install vllm==0.5.4

Flipping between these two works well for me