Conversation
While playing around with some older Gemma2 models, @jackculpepper noticed a bug in how attention is computed. In https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/models/base.py#L24-L42 the mask is created as a (B, 1, N, S) array, and https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/models/gemma2.py#L85-L104 reshapes queries to a 5D array, (B, n_kv_heads, repeats, L, S), so implicit broadcasting will try to make the mask (1, B, 1, L, S) . This usually raises a ValueError but can also fail silently if batch size equals n_kv_heads. This reshapes the mask correctly, and adds a test that fails before, passes after the fix. The issue was verified and reproduced (including the regression test) with Anthropic Claude. Co-authored-by: Jack Culpepper <jackculpepper@gmail.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
While playing around with some older Gemma2 models, @jackculpepper noticed a bug in how attention is computed. In https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/models/base.py#L24-L42 the mask is created as a (B, 1, N, S) array, and https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/models/gemma2.py#L85-L104 reshapes queries to a 5D array, (B, n_kv_heads, repeats, L, S), so implicit broadcasting will try to make the mask (1, B, 1, L, S) . This usually raises a ValueError but can also fail silently if batch size equals n_kv_heads.
This reshapes the mask correctly, and adds a test that fails before, passes after the fix.