Skip to content

perf(kernel): avoid the activation layout conversion in fp8 moe decode and read mxfp4 gemv weights as column pairs - #653

Merged
jason-fxz merged 2 commits into
mainfrom
perf/decode-moe-gemv
Oct 9, 2026
Merged

perf(kernel): avoid the activation layout conversion in fp8 moe decode and read mxfp4 gemv weights as column pairs#653
jason-fxz merged 2 commits into
mainfrom
perf/decode-moe-gemv

Commits

  1. Commits on Oct 8, 2026