Commits

Branch selector

 main

User selector

All users

All time

Commit History

Commits on Apr 23, 2026

• Add optimised top-k kernel AIR. (#2890)

  Show description for 0c2e7b0
  

  3 peopleauthored
  0c2e7b0

  Copy full SHA for 0c2e7b0

• [PyTorch] Fix FA4 selection when FA3 is unavailable. (#2909)

  Show description for 9e55a25
  bbuschkaemper

  authored
  9e55a25

  Copy full SHA for 9e55a25

• fix: scope get_full_cu_seqlens cache key by device and inference mode (#2728)

  Show description for ab60f4c
  

  3 peopleauthored
  ab60f4c

  Copy full SHA for ab60f4c

• [PyTorch] Fix CP A2A F16 when NVTE_FP8_DPA_BWD=1 (#2917)

  Show description for 424b031
  cyanguwa

  authored
  424b031

  Copy full SHA for 424b031

• [PyTorch] [torch.compile] Remove internal tensor state from Float8CurrentScalingQuantizer (#2816)

  Show description for a5164fe
  

  pggPL

  and

  pre-commit-ci[bot]

  authored
  a5164fe

  Copy full SHA for a5164fe

Commits on Apr 22, 2026

• Make NS coefficients parameter 2D in Python API (#2904)

  Show description for 3c62f42
  vcherepanov-nv

  authored
  3c62f42

  Copy full SHA for 3c62f42

• [PyT] Fix FSDP2 memory leaks for FP8 weight workspaces and transpose caches (#2805)

  Show description for 0a088c1
  

  3 peopleauthored
  0a088c1

  Copy full SHA for 0a088c1

• Fix flash attention version check. (#2910)

  Show description for 4014f7f
  bbuschkaemper

  authored
  4014f7f

  Copy full SHA for 4014f7f

• Add better ordering enforcment to split_overlap_rs gemms. (#2056)

  Show description for f2ed86b
  

  3 peopleauthored
  f2ed86b

  Copy full SHA for f2ed86b

• Bias/Dbias Support for GroupedLinear (#2885)

  Show description for 0be9046
  

  vthumbe1503

  and

  pre-commit-ci[bot]

  authored
  0be9046

  Copy full SHA for 0be9046

Commits on Apr 21, 2026

• Add MXFP8 attention (#2719)

  Show description for ee5dcec
  

  cyanguwa

  and

  pre-commit-ci[bot]

  authored
  ee5dcec

  Copy full SHA for ee5dcec

• fix(readme): update broken links and modernize project description (#2879)

  Show description for 0e8ff35
  

  3 peopleauthored
  0e8ff35

  Copy full SHA for 0e8ff35

• [PyTorch] Fix cuteDSL kernel incorrect numerics when K is 64 aligned (#2905)

  Show description for 2d92aa6
  ksivaman

  authored
  2d92aa6

  Copy full SHA for 2d92aa6

• [Common] Reduced padding kernel compilation time (#2827)

  Show description for 264da2b
  Oleg-Goncharov

  authored
  264da2b

  Copy full SHA for 264da2b

Commits on Apr 20, 2026

• Changed version to 2.16.0.dev0

  Show description for fff2245
  ptrendx

  committed
  fff2245

  Copy full SHA for fff2245

• adds NVFP4 Fused Adam support (#2797)

  Show description for 549f5ba
  

  4 peopleauthored
  549f5ba

  Copy full SHA for 549f5ba

Commits on Apr 17, 2026

• [JAX] Fix grouped quant checkpointing (#2889)

  Show description for 262bc6c
  

  jberchtold-nvidia

  and

  pre-commit-ci[bot]

  authored
  262bc6c

  Copy full SHA for 262bc6c

• [PyTorch] Add FA4 Support (#2432)

  Show description for c5a4fd5
  yaox12

  authored
  c5a4fd5

  Copy full SHA for c5a4fd5

• [Common, pyTorch] Grouped MXFP8 dequantize support (#2722)

  Show description for be593b1
  

  ptrendx

  and

  pre-commit-ci[bot]

  authored
  be593b1

  Copy full SHA for be593b1

Commits on Apr 16, 2026

• fix CUDA architectures cmake logic (#2832)

  Show description for fca261e
  GaetanLepage

  authored
  fca261e

  Copy full SHA for fca261e

• [Common] Fix fused router for large top-K and expert counts (#2821)

  Show description for 1e9e48c
  

  harryzhou2000

  and

  pre-commit-ci[bot]

  authored
  1e9e48c

  Copy full SHA for 1e9e48c

• [PyTorch] Add test to compare single vs multi-param fused GMLP (#2893)

  Show description for 58a008f
  ksivaman

  authored
  58a008f

  Copy full SHA for 58a008f

• [PyTorch] Minor optimizations in fused grouped MLP (#2888)

  Show description for c9035a4
  ksivaman

  authored
  c9035a4

  Copy full SHA for c9035a4

• [PyTorch] Add method for mcore to register wgrad accumulation hook (#2886)

  Show description for 3a78e15
  

  ksivaman

  and

  gdengk

  authored
  3a78e15

  Copy full SHA for 3a78e15

• [PyTorch] [torch.compile] Split linear forward into forward and setup context. (#2811)

  Show description for 51d9eeb
  

  pggPL

  and

  pre-commit-ci[bot]

  authored
  51d9eeb

  Copy full SHA for 51d9eeb

Commits on Apr 15, 2026

• [Pytorch][JAX] Guard against invalid num_out_tokens in permute_with_mask_map (#2876)

  Show description for 92b0370
  

  tdophung

  and

  pre-commit-ci[bot]

  authored
  92b0370

  Copy full SHA for 92b0370

• Add grouped unswizzle functionality for MXFP8 scaling factors (#2837)

  Show description for a347e09
  

  int-smart

  and

  pre-commit-ci[bot]

  authored
  a347e09

  Copy full SHA for a347e09

• [JAX] Tighten Triton autotuning version gate + autotuning enforce env var (#2875)

  Show description for a817b60
  

  tdophung

  and

  pre-commit-ci[bot]

  authored
  a817b60

  Copy full SHA for a817b60

• Newton-Schulz via cuSOLVERMp (#2706)

  Show description for a073ad5
  

  5 peopleauthored
  a073ad5

  Copy full SHA for a073ad5

• [PyTorch] [torch.compile] Remove module reference from autograd function args (#2791)

  Show description for c6853b6
  

  pggPL

  and

  pre-commit-ci[bot]

  authored
  c6853b6

  Copy full SHA for c6853b6

• [PyTorch] [torch.compile] transformer_engine.pytorch.autocast suport inside torch.compile (#2759)

  Show description for 17aa2e4
  

  pggPL

  and

  pre-commit-ci[bot]

  authored
  17aa2e4

  Copy full SHA for 17aa2e4

• Test Fused MOE with padded tokens (#2880)

  Show description for 52d6e8b
  

  vthumbe1503

  and

  timmoon10

  authored
  52d6e8b

  Copy full SHA for 52d6e8b

• [JAX] MXFP8 Grouped Quant+GEMM (#2763)

  Show description for 70af730
  

  jberchtold-nvidia

  and

  greptile-apps[bot]

  authored
  70af730

  Copy full SHA for 70af730

Commits on Apr 14, 2026

• Reduce number of C++ test cases for MXFP8 cast and activation kernels (#2874)

  Show description for 5d5065f
  timmoon10

  authored
  5d5065f

  Copy full SHA for 5d5065f

• Strip local version labels from package version checks (#2858)

  Show description for c7205a7
  pstjohn

  authored
  c7205a7

  Copy full SHA for c7205a7