Repository navigation

Default GPTQ to the fused NVFP4 update - #2598

Open

namgyu-yun wants to merge 5 commits into

NVIDIA:mainfrom

namgyu-yun:perf/gptq-fused-default

Open

Default GPTQ to the fused NVFP4 update#2598
namgyu-yun wants to merge 5 commits into
NVIDIA:mainfrom
namgyu-yun:perf/gptq-fused-default

Commits on Oct 31, 2026
- Default GPTQ to the fused NVFP4 update
  namgyu-yun
  committed
  70e3caf
- Handle irregular widths in the fused GPTQ kernel
  namgyu-yun
  committed
  4e202b5
- Assert the irregular-shape GPTQ test takes the fused path
  namgyu-yun
  committed
  1a78ae8
- Fix fused GPTQ scale rounding and dispatch tests
  namgyu-yun
  committed
  f01facd
- Keep non-plain quantizer forwards on unfused GPTQ
  namgyu-yun
  committed
  f228175