Repository navigation
Fix: Correct cross-attention query routing in Post-LN TransformerDecoderLayer - #3382
Merged
angeloskath merged 2 commits intoApr 7, 2026
Conversation
…nd adding comprehensive regression tests
angeloskath
approved these changes
Apr 6, 2026
angeloskath
left a comment
angeloskath
left a comment
Member
There was a problem hiding this comment.
Thanks for catching that!
I removed the last test because it is basically a re-implementation of the layer so it doesn't really help anything. If there is a bug then it is in both and if it is in another component then it will be present in both.
Contributor
Author
Perfect, thanks for the review and taking care of the test! |
1 task done
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Proposed changes
This PR fixes a critical data flow bug in TransformerDecoderLayer when initialized with norm_first=False.
Previously, the cross-attention module was receiving y (the raw, dropped-out output of the self-attention layer) as its query. This inadvertently bypassed the first LayerNorm (self.ln1) and severed the residual connection (x + y). I updated the query input to x to correctly route the output of the first Add & Norm block into the cross-attention block, aligning with standard Post-LN Transformer architecture.
Checklist
pre-commit run --all-filesto format my code / installed pre-commit prior to committing changes