Repository navigation

[Triton/Gluon] Improving GFX950 Unified Attention for spec. decode and reduce opimizations - #6336

Merged

vgokhale merged 9 commits into

mainfrom

mekaymak/ua_gluon_gfx950_specdecode_splitkv

Oct 12, 2026

Merged

[Triton/Gluon] Improving GFX950 Unified Attention for spec. decode and reduce opimizations#6336
vgokhale merged 9 commits into
mainfrom
mekaymak/ua_gluon_gfx950_specdecode_splitkv

Commits on Oct 8, 2026
Commits on Oct 9, 2026
- Split the KV of gfx950 gluon prefill when the grid is too small to fill the GPU
  cagrikymk
  committed
  07d82b0
- Pick gfx950 gluon fp8 D=256 2d and D=64 prefill split configs that hold on Triton 3.7
  cagrikymk
  committed
  1ca4105