Skip to content

Avoid extra normalization#71

Merged
karimnosseir merged 1 commit into
apple:mainfrom
karimnosseir:user/karimnosseir/mixtral_expert_select_opt
Jul 1, 2026
Merged

Avoid extra normalization#71
karimnosseir merged 1 commit into
apple:mainfrom
karimnosseir:user/karimnosseir/mixtral_expert_select_opt

Conversation

@karimnosseir

Copy link
Copy Markdown
Contributor

Avoid softmax on all experts and extra normalization.

Perf impact is tiny ~2% improvement on decode

Testing: presubmit

Avoid softmax on all experts and extra normalization.

Perf impact is tiny ~2% improvement on decode

Testing: presubmit
@karimnosseir karimnosseir force-pushed the user/karimnosseir/mixtral_expert_select_opt branch from 8818c3f to 83e80bb Compare July 1, 2026 01:17
@karimnosseir karimnosseir merged commit 7f9db7c into apple:main Jul 1, 2026
3 checks passed
@karimnosseir karimnosseir deleted the user/karimnosseir/mixtral_expert_select_opt branch July 1, 2026 17:05
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants