diffusers

[train_custom_diffusion.py] Fix the LR schedulers when `num_train_epochs` is passed in a distributed training env

#9308

Merged