FENRlR / MB-iSTFT-VITS2

Application of MB-iSTFT-VITS components to vits2_pytorch
MIT License
111 stars 27 forks source link

Does batch size make a difference to end quality? #5

Open DavidNTompkins opened 1 year ago

DavidNTompkins commented 1 year ago

Apologies if this is too general a question - but would using a smaller batch size impact the end-quality of the model? I'm ~5 days into a size 16 batch run on a 3080ti. (using MB-Mini) I've noticed the quality hasn't improved very much past the first 100k iterations. Is this just an issue with the batch size?

I'll post results after another few days.

FENRlR commented 1 year ago

It would not be exactly the same with 50k under 32 batches. But since I haven't tried 16 batch setting, there's not much that I can tell.

nshmyrev commented 1 year ago

Batch size 8 works just fine. You might want to reduce learning rate a bit though.