Fix batch>1 causal masks in GLM4 and Qwen3 model variants#3611
Open
NahButch wants to merge 1 commit into
Open
Conversation
Five models (glm4_new, quantized_glm4, quantized_qwen3, qwen3_moe, quantized_qwen3_moe) carried byte-identical copies of the causal_mask helper from qwen3, sharing its bug: the mask data is built independent of the batch size but declared with shape (b, 1, tgt, tgt + offset), so only batch row 0 is valid and rows >= 1 attend over a corrupt pattern. Replace the copies with a shared utils::additive_causal_mask that builds the data once and expands it across the batch (zero-copy view), with unit tests covering batching, KV offset, and sliding window. Same bug as huggingface#3582 (fixed for qwen3 in huggingface#3610). Co-Authored-By: Claude Fable 5 <noreply@anthropic.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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Five models (glm4_new, quantized_glm4, quantized_qwen3, qwen3_moe, quantized_qwen3_moe) carried byte-identical copies of qwen3's
causal_maskhelper, sharing its bug: the mask data is built independent of the batch size but declared with shape(b, 1, tgt, tgt + offset), so only batch row 0 is valid and rows >= 1 attend over a corrupt pattern. Same bug as #3582, fixed for qwen3 in #3610.Replaces the copies with a shared
utils::additive_causal_maskthat builds the data once and expands it across the batch (zero-copy view), with unit tests covering batching, KV offset, and sliding window. Net -26 lines.🤖 Generated with Claude Code