Incompatible with huggingface transformers?

[hxu105]

Howdy,

I am having one issue when I import torchdrug and transformers at the same time. Importing torchdrug will somehow disable some functionality for hugging face transformers and peft as well as torch.distributed.fsdp.FullySharedDataParallel.

In multiple GPU settings, importing torchdrug will cause AssertionError: Expects a fully sharded module but got FullyShardedDataParallel( (_fsdp_wrapped_module):... I just find these errors quite weird for me, if you can provide any hint or solution, I will be very grateful for your help.

Many thanks,