Skip to content

Move grad_accum logging to every step #1187

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
merged 3 commits into from
Jun 17, 2022
Merged

Move grad_accum logging to every step #1187

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
8bc7f64
Move grad_accum logging to every step
coryMosaicML Jun 17, 2022
3b7281e
Merge branch 'dev' into grad-accum-logs
coryMosaicML Jun 17, 2022
736ff83
Fix lint
coryMosaicML Jun 17, 2022
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
4 changes: 2 additions & 2 deletions composer/trainer/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1672,7 +1672,6 @@ def _train_batch(self, use_grad_scaling: bool):
else:
original_grad_accum = self.state.grad_accum
self.state.grad_accum = min(2 * self.state.grad_accum, device_batch_size)
self.logger.data_batch({'trainer/grad_accum': self.state.grad_accum})
log.info(('CUDA out of memory detected. Gradient Accumulation '
f'increased from {original_grad_accum} -> {self.state.grad_accum}, '
'and the batch will be retrained.'))
Expand All @@ -1681,7 +1680,8 @@ def _train_batch(self, use_grad_scaling: bool):
# back only to this newly raised error.
raise caught_timeout_error
else:
# Otherwise, return calculated loss
# Otherwise, log grad_accum and return calculated loss
self.logger.data_batch({'trainer/grad_accum': self.state.grad_accum})
return total_loss

def _train_microbatches(self, microbatches: Sequence[Batch], ddp_sync: bool = True):
Expand Down