Repository navigation

[PyTorch][common] Support row-scaled NVFP4 in GroupedLinear - #3776

Draft

cael-ling wants to merge 30 commits into

NVIDIA:mainfrom

cael-ling:pr/nvfp4-row-scaled-grouped-linear

Draft

[PyTorch][common] Support row-scaled NVFP4 in GroupedLinear#3776
cael-ling wants to merge 30 commits into
NVIDIA:mainfrom
cael-ling:pr/nvfp4-row-scaled-grouped-linear

Commits on Oct 8, 2026
- [Common][PyTorch] Add NVFP4 row-scaled grouped cast with graph-safe path
  cael-ling
  committed
  148f681
- Address review: guard grouped row-scaled cast dispatch and inputs
  cael-ling
  committed
  b58bbe1
Commits on Oct 9, 2026
- Guard row-scaled grouped cast against non-2D, partial, and strided routing
  cael-ling
  committed
  0af4dd9
- [pre-commit.ci] auto fixes from pre-commit.com hooks
  pre-commit-ci[bot]
  committed
  07de769
- Fall back to per-expert cast for mixed NVFP4 row/tensor-scaled splits
  cael-ling
  committed
  5536e0c
- Add row-scaled NVFP4 support to grouped-tensor GEMM
  cael-ling
  committed
  91e5585
- [pre-commit.ci] auto fixes from pre-commit.com hooks
  
  pre-commit-ci[bot]
  authored and
  cael-ling
  committed
  26ab746
- Harden row-scaled grouped GEMM post-scale and operand checks
  cael-ling
  committed
  2ce6e70
- Apply row-scaled grouped GEMM column scales over actual group rows
  cael-ling
  committed
  3259847
- Make grouped row-scaled GEMM column post-scale graph-safe
  cael-ling
  committed
  07ceda5
- Fix grouped row-scaled GEMM post-scale gating and cuBLAS test skip
  cael-ling
  committed
  e2517b9
Commits on Oct 22, 2026
Commits on Oct 26, 2026
- Fold row-scaled NVFP4 amax into nvte_group_amax
  
  timmoon10
  and
  claude
  committed
  0432c8e
- [pre-commit.ci] auto fixes from pre-commit.com hooks
  pre-commit-ci[bot]
  committed
  55f69ca
- Fix missing include
  timmoon10
  committed
  32c57f2
Commits on Oct 27, 2026
Commits on Oct 29, 2026
- Support row-scaled NVFP4 in GroupedLinear
  cael-ling
  committed
  aaa2f09
Commits on Oct 30, 2026
- Merge upstream main into pr/nvfp4-row-scaled-grouped-linear
  cael-ling
  committed
  9ee1140
- Fix row-scaled NVFP4 GroupedLinear edge cases
  cael-ling
  committed
  5ee2148
- [pre-commit.ci] auto fixes from pre-commit.com hooks
  pre-commit-ci[bot]
  committed
  151f7ff
Commits on Nov 4, 2026
- Support row-scaled NVFP4 weights
  cael-ling
  committed
  a4710f5

Commits

Commits on Oct 8, 2026

Commits on Oct 9, 2026

Commits on Oct 22, 2026

Commits on Oct 26, 2026

Commits on Oct 27, 2026

Commits on Oct 29, 2026

Commits on Oct 30, 2026

Commits on Nov 4, 2026