Commits

Branch selector

 main

User selector

All users

Datepicker

All time

Commit History

Commits on Nov 7, 2025

• [common] Remove kvpacked and qkvpacked attention functions for every kernel type. (#2287)

  Show description for 3454f84
  

  3 peopleauthored
  3454f84

  Copy full SHA for 3454f84

• Update cudnn frontend to v1.16.0 (#2362)

  Show description for d20311b
  ksivaman

  authored
  d20311b

  Copy full SHA for d20311b

• [JAX] Default to fused attention in JAX DPA (#2363)

  Show description for 5978f1d
  

  KshitijLakhani

  and

  greptile-apps[bot]

  authored
  5978f1d

  Copy full SHA for 5978f1d

• Disable cuDNN attention for known IMA and NaNs (#2344)

  Show description for 26aad6b
  ksivaman

  authored
  26aad6b

  Copy full SHA for 26aad6b

• Fix sharding of segment position to match id in ring attention. (#2349)

  mgoldfarb-nvidia

  authored
  f62cad9

  Copy full SHA for f62cad9

• [JAX] Add test to check jaxpr that amax is reused for nvfp4 recipe (#2348)

  Show description for 4ff3eed
  

  jberchtold-nvidia

  and

  pre-commit-ci[bot]

  authored
  4ff3eed

  Copy full SHA for 4ff3eed

Commits on Nov 6, 2025

• Make FP8 weights compatible with older MCore version (#2342)

  Show description for b14a3b6
  

  3 peopleauthored
  b14a3b6

  Copy full SHA for b14a3b6

• Fix out of bounds access in the FP4 dequantize kernel (#2346)

  Show description for f3b97c2
  ptrendx

  authored
  f3b97c2

  Copy full SHA for f3b97c2

• [JAX] Try to use pre-downloaded dataset artifacts first (#2345)

  Show description for dcaca2a
  jberchtold-nvidia

  authored
  dcaca2a

  Copy full SHA for dcaca2a

Commits on Nov 5, 2025

• [JAX] Fix bug with pre scale bias (#2300)

  Show description for b6020e3
  pggPL

  authored
  b6020e3

  Copy full SHA for b6020e3

• [PyTorch Debug] Add max_blockwise_dynamic_range stats (#2137)

  Show description for 77a0063
  

  pggPL

  and

  pre-commit-ci[bot]

  authored
  77a0063

  Copy full SHA for 77a0063

Commits on Nov 4, 2025

• [PyTorch debug] Fixes to debug tests failures (#2268)

  Show description for 3d76218
  

  pggPL

  and

  pre-commit-ci[bot]

  authored
  3d76218

  Copy full SHA for 3d76218

Commits on Nov 3, 2025

• [JAX] L1_jax_distributed_test suit with individual executions (#2321)

  Show description for c57ffc5
  phu0ngng

  authored
  c57ffc5

  Copy full SHA for c57ffc5

Commits on Oct 31, 2025

• [Common] Deleted unused header (#2324)

  Show description for e7227af
  Oleg-Goncharov

  authored
  e7227af

  Copy full SHA for e7227af

• [JAX] Fix mesh resource requirement when no mesh (#2307)

  Show description for 006670d
  jberchtold-nvidia

  authored
  006670d

  Copy full SHA for 006670d

Commits on Oct 30, 2025

• [JAX] Ensure JAX reference impl uses an accurate backend in our tests (#2322)

  Show description for 1269b2e
  jberchtold-nvidia

  authored
  1269b2e

  Copy full SHA for 1269b2e

• [PyT] Bump the min version expected to supported FP8 current scaling determinism on Blackwell (#2316)

  Show description for 26370b1
  

  KshitijLakhani

  and

  pre-commit-ci[bot]

  authored
  26370b1

  Copy full SHA for 26370b1

• [Common] Split cast/gated kernels by scaling mode (#2248)

  Show description for 0e80c84
  

  Oleg-Goncharov

  and

  pre-commit-ci[bot]

  authored
  0e80c84

  Copy full SHA for 0e80c84

• [PyTorch] Fix attention backend and tests for `sm120` (#2320)

  Show description for 490a5f4
  ksivaman

  authored
  490a5f4

  Copy full SHA for 490a5f4

• [JAX] Fix: Skip determinism tests for bprop for all sm >=100 (#2315)

  Show description for 5e8a9a9
  

  KshitijLakhani

  and

  pre-commit-ci[bot]

  authored
  5e8a9a9

  Copy full SHA for 5e8a9a9

• CMake to respect MAX_JOBS or NVTE_MAX_JOBS (#2319)

  Show description for f0295f9
  phu0ngng

  authored
  f0295f9

  Copy full SHA for f0295f9

Commits on Oct 29, 2025

• [Pytorch] Integrate GPT OSS Swiglu in TransformerLayer (#2312)

  Show description for c6cbcc8
  

  3 peopleauthored
  c6cbcc8

  Copy full SHA for c6cbcc8

Commits on Oct 28, 2025

• [JAX] Use TE quantization when TE fused norm is disable (#2303)

  Show description for a8e4346
  phu0ngng

  authored
  a8e4346

  Copy full SHA for a8e4346

Commits on Oct 27, 2025

• Change the pyTorch installation to CUDA 13 in Build All GitHub action (#2308)

  Show description for 4cf2f12
  ptrendx

  authored
  4cf2f12

  Copy full SHA for 4cf2f12

• Submodule checkout during setup (#2293)

  Show description for a019c80
  ksivaman

  authored
  a019c80

  Copy full SHA for a019c80

• Remove `nvidia-mathdx` dependency (#2295)

  Show description for d7c9777
  ksivaman

  authored
  d7c9777

  Copy full SHA for d7c9777

• [PyTorch] Use dummy wgrad in GroupedLinear (#2305)

  Show description for d2945c6
  

  Autumn1998

  and

  yaox12

  authored
  d2945c6

  Copy full SHA for d2945c6

Commits on Oct 25, 2025

• [PyTorch] Add max_logit support for MuonClip (#2195)

  Show description for 87cb26c
  

  cyanguwa

  and

  pre-commit-ci[bot]

  authored
  87cb26c

  Copy full SHA for 87cb26c

Commits on Oct 24, 2025

• [Common] Fix checks in quantize_transpose_vector_blockwise_fp4 (#2299)

  Show description for 060811c
  jberchtold-nvidia

  authored
  060811c

  Copy full SHA for 060811c

• [PyTorch] Support delay_wgrad_compute cudagraph (#1948)

  Show description for 6273ced
  

  buptzyb

  and

  ksivaman

  authored
  6273ced

  Copy full SHA for 6273ced

Commits on Oct 23, 2025

• [PyTorch Debug] Fix issue with microbatching + debug value caching (#2108)

  Show description for 021e1e6
  pggPL

  authored
  021e1e6

  Copy full SHA for 021e1e6

• [JAX] Make SR rng state always 2D (num_devices, 4) to fix partitioning issue (#2294)

  Show description for e2f2a0b
  jberchtold-nvidia

  authored
  e2f2a0b

  Copy full SHA for e2f2a0b

• Overhaul the compilation for the arch-specific features (#2279)

  Show description for eb34783
  

  4 peopleauthored
  eb34783

  Copy full SHA for eb34783

• Include TE core headers in final build (#2291)

  Show description for 66acb8e
  ksivaman

  authored
  66acb8e

  Copy full SHA for 66acb8e

Commits on Oct 22, 2025

• [JAX] Defer TE/JAX cublas shape check on fp8 gemms until lowering (#2292)

  Show description for 2ac3c16
  jberchtold-nvidia

  authored
  2ac3c16

  Copy full SHA for 2ac3c16

Pagination

Previous

Next