transformers

Fix: Conditionally import `torch.distributed.fsdp` in `trainer_seq2seq.py`

#44507

Merged