[Qwen3_5]Remove unnecessary masked_fill_ in torch_chunk_gated_delta_rule attention computation: "attn = (q_i @ k_i.transpose(-1, -2) * decay_mask[:, :, i]).masked_fill_(mask, 0)"#45215
Merged
Rocketknight1 merged 4 commits intohuggingface:mainfrom Apr 9, 2026
Conversation
Signed-off-by: zj <2716634506@qq.com>
…ule attention computation Signed-off-by: zj <2716634506@qq.com>
Contributor
|
[For maintainers] Suggested jobs to run (before merge) run-slow: olmo_hybrid, qwen3_5, qwen3_5_moe, qwen3_next |
Rocketknight1
approved these changes
Apr 8, 2026
Member
Rocketknight1
left a comment
Rocketknight1
left a comment
There was a problem hiding this comment.
This looks right to me! There is a minor danger of NaNs, I guess, but I think it's fine - multiplication of float32 Q/K values in any remotely reasonable range should not yield NaN
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
github-merge-queue
Bot
removed this pull request from the merge queue due to failed status checks
sirzechs66
pushed a commit
to sirzechs66/transformers
that referenced
this pull request
Apr 18, 2026
…ule attention computation: "attn = (q_i @ k_i.transpose(-1, -2) * decay_mask[:, :, i]).masked_fill_(mask, 0)" (huggingface#45215) * [Qwen3_5]Remove excess mask Signed-off-by: zj <2716634506@qq.com> * [Qwen3_5]Remove unnecessary masked_fill_ in torch_chunk_gated_delta_rule attention computation Signed-off-by: zj <2716634506@qq.com> * Fix: remove brackets to match generated code format --------- Signed-off-by: zj <2716634506@qq.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
3 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.
What does this PR do?
Remove unnecessary masked_fill_(mask, 0) call in torch_chunk_gated_delta_rule.
The decay_mask computed earlier already encodes the causal/lower-triangular structure (upper-triangle values are zero), so masking the attention scores again with masked_fill_(mask, 0) is redundant and adds unnecessary overhead.
Before:
attn = (q_i @ k_i.transpose(-1, -2) * decay_mask[:, :, i]).masked_fill_(mask, 0)
After:
attn = (q_i @ k_i.transpose(-1, -2) * decay_mask[:, :, i])
Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
@ArthurZucker @Cyrilvallez