Update cosine decay to not clamp output, match other implementations #1138
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Proposed changes
This change updates how the 'minimum' value is treated by the cosine decay scheduler. The existing implementation clamps the output to the provided minimum and the new implementation instead smoothly transitions from the initial to final learning rate. This picture clarifies the difference between these two when decaying from 1.0 to 0.5 over 100 steps:
With this change the behavior will also match that of pytorch, optax, and keras
Also renamed 'minimum' to 'end' because this implementation can 'decay' upwards.
EDIT: I'm also not completely sold that removing
minimum
is the right move because it still might be useful to someone. Could be best to keep that and also addend
as a separate parameter. If anyone has opinions about that I'd be happy to change it.Checklist
Put an
x
in the boxes that apply.pre-commit run --all-files
to format my code / installed pre-commit prior to committing changes