Fix Qwen3 model conversion perplexity issue#2
Draft
hawkoli1987 wants to merge 1 commit into
Draft
Conversation
Preserve and apply Qwen3's q_norm/k_norm for accurate MLA conversion. Co-authored-by: yuli <yuli@aisingapore.org>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Preserve Qwen3's head-wise QK norms during conversion to fix catastrophic perplexity increase.
Qwen3 models apply per-head RMSNorm (
q_norm/k_norm) after theq_proj/k_projlayers. The original conversion pipeline discarded these modules, leading to unnormalized Q/K activations and incorrect dot product scaling, which caused the observed multi-order-of-magnitude perplexity regression. This PR ensures these norms are preserved and applied correctly throughout the conversion process.