Soundstream loss doesn't decrease after 1167 steps - version 0.7.1

lucidrains / audiolm-pytorch

Implementation of AudioLM, a SOTA Language Modeling Approach to Audio Generation out of Google Research, in Pytorch

MIT License

2.45k stars 266 forks source link

Soundstream loss doesn't decrease after 1167 steps - version 0.7.1 #61

Closed yigityu closed 1 year ago

yigityu commented 1 year ago

Hi,

First of all, thank you for this project and all the other open source projects you're doing. I'm a big fan of your work.

I was training with the latest version on LibriSpeech dataset and looks like recon_loss shoots up and training goes to nowhere afterwards. I didn't seem to have this with previous releases, but I will roll back, try again and report results here, but this might be a regression with latest changes? I wanted to post it if it would help anyone.

ckwdani commented 1 year ago

Im around 55000 steps and it seems stable too after reducing the learning rate. :) I can also remotely hear a voice already :)