Enable scheduler and higher starting LR to avoid plateau#53
Draft
michaelmckinsey1 wants to merge 4 commits intoLBANN:mainfrom
Draft
Enable scheduler and higher starting LR to avoid plateau#53michaelmckinsey1 wants to merge 4 commits intoLBANN:mainfrom
michaelmckinsey1 wants to merge 4 commits intoLBANN:mainfrom
Conversation
| dice_score_probs = compute_sharded_dice( | ||
| mask_pred_probs, mask_true_onehot, spatial_mesh | ||
| ) | ||
| dice_loss_curr = 1.0 - dice_score_probs.mean() |
Collaborator
Author
There was a problem hiding this comment.
was background + foreground -> now foreground
michaelmckinsey1
changed the title
| self.scale_reference_starting_learning_rate | ||
| * self.scale_learning_rate_factor | ||
| ** (self.problem_scale - self.scale_reference) | ||
| ) |
Collaborator
Author
There was a problem hiding this comment.
Noticed instability at higher problem scales that is fixed by lowering the LR
Collaborator
Author
There was a problem hiding this comment.
Noticed this is correlated with turning on AMP
Collaborator
Author
There was a problem hiding this comment.
no AMP
vs AMP
Collaborator
Author
There was a problem hiding this comment.
probably should just omit this logic and avoid using AMP
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Use
ExponentialLRscheduler to start at a higher LR to avoid getting stuck atval_dice_score=0.5. Decreases number of epochs required on the order of 2x.mainbranch currently uses a default config with a constant LR.val_dice_scoreis not boosted by background class.main vs PR#53:
1node:
2nodes: