DistributedSampler can't shuffle the dataset

[qmpzzpmq]

almost same things with
pytorch/pytorch#31771
huggingface/transformers#3721

how to fix them in pytorch lighting

[s-rog]

PL already calls set_epoch() for it as part of the training loop.

I'll close this for now, if you have a bugged replicable code sample feel free to re-open.