fix: ZeroDivisionError in get_lr when lr_decay_iters == warmup_iters#691
Open
Bortlesboat wants to merge 1 commit into
Open
fix: ZeroDivisionError in get_lr when lr_decay_iters == warmup_iters#691Bortlesboat wants to merge 1 commit into
Bortlesboat wants to merge 1 commit into
Conversation
When lr_decay_iters and warmup_iters are set to the same value (e.g. both equal to max_iters for a short run), the cosine decay formula divides by zero. Guard with max(1, ...) so the edge case returns the peak learning rate gracefully instead of crashing.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Bug
get_lr()divides by zero exactly at the warmup/decay boundary whenlr_decay_iters == warmup_iters:This can occur in short experimental configurations that intentionally disable the cosine-decay interval.
Fix
Clamp the decay interval denominator to at least one:
At the zero-length boundary, the numerator is also zero, so the scheduler returns the peak learning rate. Iterations after
lr_decay_itersstill returnmin_lrthrough the existing earlier branch, and normal positive-length schedules are unchanged.