Repository navigation

CUDA: fast, batch-invariant small-batch PTQ1_0 mat-vec (1.5x decode on Ampere) - #218

Open

sudoingX wants to merge 11 commits into

PrismML-Eng:prismfrom

sudoingX:pr-ptq1-mmv

Open

CUDA: fast, batch-invariant small-batch PTQ1_0 mat-vec (1.5x decode on Ampere)#218
sudoingX wants to merge 11 commits into
PrismML-Eng:prismfrom
sudoingX:pr-ptq1-mmv

Commits on Sep 22, 2026
Commits on Oct 10, 2026
- Fix: wait on the PDL dependency in the PTQ1_0 mat-vec kernel before reading vy
  
  LamplighterPaul
  authored and
  sudoingX
  committed
  efe1a06