Skip to content

Fix nan eval loss in clip training #293

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

Merged
libinta merged 1 commit into from
Jul 12, 2024
Merged

Fix nan eval loss in clip training #293

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 2 additions & 6 deletions optimum/habana/transformers/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1861,12 +1861,8 @@ def evaluation_loop(
if loss is not None:
# Handle NaN loss
if args.logging_nan_inf_filter and (torch.isnan(loss) or torch.isinf(loss)):
# If loss is NaN or Inf, use the average of previous losses or a small constant
if losses_host is not None and len(losses_host) > 0:
avg_loss = torch.mean(losses_host)
loss = avg_loss
else:
loss = torch.tensor(1e-8, device=loss.device)
# If loss is NaN or Inf, use a small constant
loss = torch.tensor(1e-8, device=loss.device)

losses = self.gather_function((loss.repeat(batch_size)))
all_losses.add(losses)
Expand Down