Repository navigation

[Fix] [AMD] quark: Keep MXFP8 shared experts out of the MXFP4 fused slot on dense-first models - #46800

Merged

BBuf merged 2 commits into

sgl-project:mainfrom

ZhengGong-amd:fix/quark-shared-expert-gate-per-layer

Oct 30, 2026

Merged

[Fix] [AMD] quark: Keep MXFP8 shared experts out of the MXFP4 fused slot on dense-first models#46800
BBuf merged 2 commits into
sgl-project:mainfrom
ZhengGong-amd:fix/quark-shared-expert-gate-per-layer

Commits on Oct 29, 2026
- [Fix] [AMD] quark: Probe every named MoE layer in the shared-experts fusion gate
  ZhengGong-amd
  committed
  3022960
- [Fix] [AMD] minimax: Ask the quantization config whether the shared expert can fuse
  ZhengGong-amd
  committed
  99eccc4