Add cfg option --scheduler.warmup_min_lr
#542
Merged
Conversation
This file contains 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
For setting the starting LR during the warmup period.
|
Thanks for doing this! |
2015aroras
approved these changes
Apr 11, 2024
olmo/optim.py
Outdated
| @@ -480,7 +481,11 @@ def get_max_grad_norm_ratio( | |||
| return self._get_max_grad_norm_coeff(initial_max_grad_norm_ratio, step, max_steps) | |||
|
|
|||
| def _linear_warmup(self, initial_lr: float, step: int, warmup_steps: int = 2000) -> float: | |||
| return initial_lr * (0.1 + 0.9 * min(step, warmup_steps) / warmup_steps) | |||
| if self.warmup_min_lr is not None: | |||
| assert initial_lr > self.warmup_min_lr | |||
There was a problem hiding this comment.
nit: You could also add a assert self.warmup_min_lr >= 0 for good measure
olmo/optim.py
Outdated
| if self.warmup_min_lr is not None: | ||
| assert initial_lr > self.warmup_min_lr | ||
| return self.warmup_min_lr + (initial_lr - self.warmup_min_lr) * min(step, warmup_steps) / warmup_steps | ||
| else: |
There was a problem hiding this comment.
From a cleaner code perspective, maybe it would be better to have just 1 formula and use self.warmup_min_lr to set the starting lr. Something like:
warmup_min_lr = self.warmup_min_lr if self.warmup_min_lr is not None else 0.1 * initial_lr
...
return warmup_min_lr + (initial_lr - warmup_min_lr) * min(step, warmup_steps) / warmup_steps
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
For setting the starting LR during the warmup period. When no set this defaults to the current behavior of starting at 10% of the target LR.