Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Figure out where the NaNs in autoencoder training are coming from: #4

Open
robogast opened this issue Nov 2, 2021 · 1 comment
Open

Figure out where the NaNs in autoencoder training are coming from: #4

robogast opened this issue Nov 2, 2021 · 1 comment

Comments

@robogast
Copy link
Member

robogast commented Nov 2, 2021 •
edited
Loading

Source:
2021-11-01/12-14-29/0/lightning_logs/version_8317429
2021-11-01/12-14-29/0/lightning_logs/version_8317429

@robogast robogast changed the title Figure out where the NaNs in training are coming from: Figure out where the NaNs in autoencoder training are coming from: Apr 5, 2022
@robogast
Copy link
Member Author

TL;DR: instability probably happens in the VQ-layer, but still unsure what exactly happens.
Increasing commitment loss, and making sure cdist compute_mode is non-mm seems to at least mitigate the issue.
Forcing 32-bit with torch.cuda.amp.autocast(enabled=False) doesn't solve the issue.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@robogast