All pull requests

Pull requests

• [Common] Make it easier to debug C++ UT failures

  #3783·kainzhong opened Oct 29, 2026Collaborator·8 of 13 tasks·

  ·
  6

• Fix performance of AdamCapturableFunctor kernel

  #3782·ksivaman opened Oct 29, 2026Member·7 of 14 tasks·

  ·
  1

  10

• [PyTorch] Fix FP8 DotProductAttention backward scales stuck at 1.0 without a Linear-family module

  community-contribution
  #3780·sha7doww opened Oct 29, 2026·6 of 13 tasks·

  ·
  1

  3

• [Common] Add explicit CUDA state lifecycle APIs for in-process restarts

  #3779·Oleg-Goncharov opened Oct 29, 2026Collaborator·7 of 13 tasks·

  ·
  16

• [PyTorch][common] Support row-scaled NVFP4 in GroupedLinear

  community-contribution
  #3776·cael-ling opened Oct 29, 2026Contributor·1 of 7 tasks·

  ·
  4

• Swizzle compact padding fusion

  #3768·vthumbe1503 opened Oct 28, 2026Collaborator·13 tasks·

  ·

• Reduce per-call CPU overhead in OperationFuser

  #3767·vthumbe1503 opened Oct 28, 2026Collaborator·5 tasks done·

  ·

• [Common] Add caller-owned workspaces with per-stream resource isolation

  #3766·Oleg-Goncharov opened Oct 28, 2026Collaborator·8 of 13 tasks·

  ·
  1

• fix(fp8): floor tiny amax so scale_inv stays at least FLT_MIN

  community-contribution
  #3764·DylanWanggg opened Oct 28, 2026·5 of 13 tasks·

  ·
  5

• [PyTorch] Add bounded-memory CPU optimizer state export

  community-contribution
  #3763·ZenAlexa opened Oct 28, 2026·2 tasks done·

  ·
  1

  4

• [Common] Look up table based activation calculation in MXFP8 quantization fused kernels

  #3761·tdophung opened Oct 28, 2026Collaborator·13 tasks·

  ·
  2

• [PyTorch] Remove EP handle cache and reconstructing handle on the fly

  2.22
  #3760·phu0ngng opened Oct 27, 2026Collaborator·8 of 13 tasks·

  ·
  10

• Remove dummy weight gradients from fused accumulation

  #3759·vthumbe1503 opened Oct 27, 2026Collaborator·5 tasks done·

  ·
  2

• MxFP8 gradient cross entropy option

  #3758·sraman-rgb opened Oct 27, 2026Collaborator·13 tasks·

  ·
  1

  

• [PyTorch] Fuse BHSD permute and GEMM swizzle of MXFP8 attention scales

  #3757·sraman-rgb opened Oct 27, 2026Collaborator·13 tasks·

  ·
  1

  

• [PyTorch] Fuse NVFP4 global scales and GEMM alpha in the grouped MLP

  #3756·sraman-rgb opened Oct 27, 2026Collaborator·13 tasks·

  ·
  1

• [PyTorch] Optionally derive FC1 dy NVFP4 amaxes from the dGLU epilogue

  #3755·sraman-rgb opened Oct 27, 2026Collaborator·13 tasks·

  ·
  2

• [PyTorch] Cache GEMM-swizzled scales of all-gathered NVFP4 weights

  #3754·sraman-rgb opened Oct 27, 2026Collaborator·13 tasks·

  ·
  3

• [Common][PyTorch] Add NVFP4 EP dispatch and combine backward

  #3753·sraman-rgb opened Oct 27, 2026Collaborator·13 tasks·

  ·
  3

• Reduce CPU overhead in grouped-tensor GroupedLinear and MLP

  #3752·vthumbe1503 opened Oct 27, 2026Collaborator·2 tasks done·

  ·
  2

• [Common, PyTorch] Write both NVFP4 amax outputs in the reduction

  community-contribution
  #3750·kimchi-developer opened Oct 27, 2026·7 of 13 tasks·

  ·
  3

• [Common, PyTorch] Fuse NVFP4 scale swizzling on SM120 and SM121

  community-contribution
  #3749·kimchi-developer opened Oct 27, 2026·6 of 13 tasks·

  ·
  1

• Fix LayerNormLinear dummy gradients for sharded weights

  community-contribution
  #3748·zhongbozhu opened Oct 27, 2026Collaborator·13 tasks·

  ·
  12

• [JAX] Control MeshResource at API call granularity for Dense APIs

  #3747·jberchtold-nvidia opened Oct 26, 2026Collaborator·8 of 13 tasks·

  ·

• [Common] Make CI fail if CuTeDSL kernels with supported configs are not used

  #3746·kainzhong opened Oct 26, 2026Collaborator·9 of 13 tasks·

  ·
  15