Training slowdown after 3-5 epochs

[thewh1teagle]

The second epoch finish in 38 seconds and then the next epoch slow down and finish only in 3:50 minutes.
Is there a workaround?

Maybe related to https://discuss.pytorch.org/t/training-gets-slow-down-by-each-batch-slowly/4460/9

[Eyalcohenx]

I also encountered it, but I don't think it's an error, I think it's related to the pretraining phase where only the generator is being trained.
pretraining_steps default is 1000 so after 1000 steps I saw a decrease in training speed.