[Help] <ptuning时，增大target_length后训练，模型不收敛>

[qdchenxiaoyan]

Is there an existing issue for this?

• [X] I have searched the existing issues

Current Behavior

上次ptuning摘要时，source_length=5000，target_length=120，训练过程正常，推理结果也正常 这次source_length=50，target_length=2000，训练过程loss震荡不收敛，推理结果也出现无序、多次重复的情况

Expected Behavior

No response

Steps To Reproduce

1

Environment

- OS:linux 3.10.0-1160.76.1.el7.x86_64
- Python: 3.10.12
- Transformers: 4.30.2 
- Torch:2.0.1 
- CUDA Support (`python -c "import torch; print(torch.cuda.is_available())"`) :True

Anything else?

No response

[MIhappen]

兄弟，解决了吗？