Repository navigation

feat(cake_activation): add fused Blackwell SwiGLU MXFP8 quantization - #4638

Merged

yyihuang merged 16 commits into

flashinfer-ai:mainfrom

yyihuang:feat/fused-swiglu-mxfp8-blackwell

Aug 28, 2026

Merged

feat(cake_activation): add fused Blackwell SwiGLU MXFP8 quantization#4638
yyihuang merged 16 commits into
flashinfer-ai:mainfrom
yyihuang:feat/fused-swiglu-mxfp8-blackwell

Commits on Aug 24, 2026
- feat(activation): add fused gated MXFP8 quantization
  averyhNV
  committed
  c33f6d9
- fix(activation): address gated MXFP8 integration checks
  averyhNV
  committed
  ad97424
- style(activation): format gated MXFP8 Python integration
  averyhNV
  committed
  f85acad
- perf(activation): optimize fused gated MXFP8 backward
  averyhNV
  committed
  623d6f4
- chore: preserve generated gated MXFP8 sources
  averyhNV
  committed
  9637e8f
- docs(activation): document gated MXFP8 APIs
  averyhNV
  committed
  16c9437
- style(activation): apply pre-commit formatting
  averyhNV
  committed
  adac071
- perf(activation): defer backward-both alias barrier
  averyhNV
  committed
  ed69cfe
- perf(activation): launch row TMA before scale stores
  averyhNV
  committed
  dfcbc04
- fix(trace): declare gated MXFP8 init axes
  averyhNV
  committed
  f1d7d84
- perf(activation): specialize gated MXFP8 streaming paths
  averyhNV
  committed
  92889b1
- fix(activation): use CUDA tensor map type in frozen kernels
  averyhNV
  committed
  caa94be
- style(activation): format gated MXFP8 dispatch
  averyhNV
  committed
  d1320fb
- fix(activation): use platform uint64 type in frozen kernels
  averyhNV
  committed
  8f18f4e
- Merge remote-tracking branch 'origin/main' into codex/gated-mxfp8-current-main-staging-20260823
  averyhNV
  committed
  0ef32a5
Commits on Aug 27, 2026
- perf(activation): specialize SM103 gated MXFP8 backward-both
  averyhNV
  committed
  373df13