-
Notifications
You must be signed in to change notification settings - Fork 517
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.
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
Refactored scheduler callbacks (epoch-based/step-based warmup) #568
Conversation
…iles to respect new config names (Keep the old name for BC as well but emit a warning when used).
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nice work (:
There are some comments inline., mostly discussing some edge cases and mildly (possible) inaccuracies.
Its important IMO we add unit tests for scheduling. Please see the different use cases I had in tests/unit_tests/lr_warmup_test.py.
# Conflicts: # src/super_gradients/training/utils/callbacks/callbacks.py # tests/unit_tests/lr_warmup_test.py
…(It is not that important which method it override here, as they would have same effect)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks awesome!
I just have one final suggestion for the unit test, see my comment there.
Please also update the docs in Trainer.train() (we have there the long list of training params docs), and shortly in default_training_hyperparams.yaml.
…' into feature/SG-525-step-based-warmup
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
Changes:
warmup_mode: linear_step
will emit deprecation warning. But will continue to work without any changes.Two new modes with explicit meaning:
warmup_mode: linear_batch_step
warmup_mode: linear_epoch_step
Per-batch warmup:
Old Per-epoch warmup: