feat(scalarization): Implement GradNormScalarizer#733
Open
powerofaisinstudy-debug wants to merge 5 commits into
Open
feat(scalarization): Implement GradNormScalarizer#733powerofaisinstudy-debug wants to merge 5 commits into
powerofaisinstudy-debug wants to merge 5 commits into
Conversation
PierreQuinton
added
cc: feat
github-actions
Bot
changed the title
Contributor
|
Thanks for the PR. Please comply to the CONTRIBUTING.md file to install and run pre-commit. Also please review the code you submit from your agent before submitting it, it contains comments that are directly addressed to you, we are not supposed to resolve these. |
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
This PR adds the GradNormScalarizer to the scalarization package, enabling dynamic loss balancing based on gradient norms as proposed by the GradNorm algorithm.
Changes
Added GradNormScalarizer class inheriting from the Scalarizer base class.
Integrated the weighting logic directly into the forward pass to ensure compatibility with the existing TorchJD architecture.
Implemented _compute_gradient_norms to facilitate gradient tracking during the forward pass.
Testing
Verified the implementation against the library's Scalarizer interface.
Confirmed that the forward method correctly handles the required input tensors.
Note
This implementation addresses the feedback provided in the previous issue regarding the integration of the scalarizer logic.