Skip to content

[Fix] Add error message for exceptions caused by large gradients #34

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

Open
yoshikisd wants to merge 1 commit into
base: master
Choose a base branch
from
Open

[Fix] Add error message for exceptions caused by large gradients #34

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
15 changes: 14 additions & 1 deletion src/cdtools/models/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -414,7 +414,20 @@ def closure():
return total_loss

# This takes the step for this minibatch
loss += optimizer.step(closure).detach().cpu().numpy()
# If an exception occurs at this point, the root cause may be related
# to the presence of large gradients that produce `nan` losses.
# We write a custom error message here to help communicate this point.
try:
loss += optimizer.step(closure).detach().cpu().numpy()
except Exception as e:
print('\n\n')
msg = 'An error has occurred during the parameter update and\n'\
'loss calculation step. This problem may be related to excessively\n'\
'large parameter gradients. Please try lowering the learning\n'\
'rates or disabling grad_required for one or more parameters\n'\

raise Exception(msg) from e



loss /= normalization
Expand Down