Fix LayerNorm Scaling implementation#698
Open
davidgonmar wants to merge 1 commit into
Open
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Issue
The current LayerNorm Scaling implementation has a silent bug when used in the training pipeline.
The scale is implemented as a buffer and initialized in the
__init__method ofLayerNormScaledTransformerBlock. However, model initialization insrc/olmo_core/nn/transformer/model.pycallsto_emptybefore initializing each module’s weights individually. This effectively erases the contents of theln_scalebuffer and causes incorrect behavior during training.Fix
There are two possible fixes. The first is to keep using a buffer and implement a custom initialization function. The second is to implement the scale as a Python scalar, which is simpler and is the approach taken in this PR.