[BUG/Help] <title>多卡p-tuning v2 运行Segmentation fault

[licy02]

Is there an existing issue for this?

• [X] I have searched the existing issues

Current Behavior

加载模型和数据后突然停止，并报Segmentation fault

Expected Behavior

通过调整transformers版本无效，定位发现是trainer.py中1820行的错误。self.control = self.callback_handler.on_train_begin(args, self.state, self.control)

Steps To Reproduce

单卡运行正常，多卡报错

Environment

- OS:Linux
- Python:3.11
- Transformers:4.27.1
- PyTorch:2.0.1
- CUDA Support (`python -c "import torch; print(torch.cuda.is_available())"`) :True

Anything else?

No response