Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit History

Commits on Dec 2, 2025

• [pre-commit.ci] pre-commit autoupdate (#1966)

  Show description for 5c3b61d
  pre-commit-ci[bot]

  authored
  5c3b61d

  Copy full SHA for 5c3b61d

Commits on Nov 27, 2025

• Add pre-commit and GitHub Actions workflow for it (#1949)

  Show description for 1f20398
  crcrpar

  authored
  1f20398

  Copy full SHA for 1f20398

Commits on Nov 17, 2025

• Use `[[maybe_unused]]` to suppress warning in csrc/fused_dense.cpp (#1954)

  Show description for 9d1358f
  crcrpar

  authored
  9d1358f

  Copy full SHA for 9d1358f

• Fix pr1951: use `.options` for `.type`, not `.scalar_type` (#1953)

  Show description for 1d6a48c
  crcrpar

  authored
  1d6a48c

  Copy full SHA for 1d6a48c

• use `.scalar_type` not `.type` (#1951)

  Show description for 419adeb
  crcrpar

  authored
  419adeb

  Copy full SHA for 419adeb

• give `[[maybe_unused]]` attribute (#1950)

  Show description for 195e8e2
  crcrpar

  authored
  195e8e2

  Copy full SHA for 195e8e2

Commits on Nov 13, 2025

• [layer/rms norm] Pass `Tensor` by pointer, use `optional` if necessary, and replace `NULL` with `nullptr`. (#1948)

  Show description for c1fa083
  

  crcrpar

  and

  Copilot

  authored
  c1fa083

  Copy full SHA for c1fa083

Commits on Nov 8, 2025

• Remove `compat.h` (#1942)

  Show description for e900daa
  

  crcrpar

  and

  Copilot

  authored
  e900daa

  Copy full SHA for e900daa

Commits on Oct 27, 2025

• Finish cleaning up usage of VERSION_GE_1_1/1_3/1_5 macros (#1937)

  Source-Roc

  authored
  4b46e00

  Copy full SHA for 4b46e00

Commits on Aug 21, 2025

• Remove version dependent macros & extension specific arch flags except fmha (#1918)

  crcrpar

  authored
  26bba57

  Copy full SHA for 26bba57

Commits on Nov 1, 2024

• `Tensor.type()` -> `Tensor.scalar_type()` (#1855)

  Show description for e552ad9
  crcrpar

  authored
  e552ad9

  Copy full SHA for e552ad9

Commits on Aug 17, 2024

• Fix illegal memory access with multi_tensor_apply size above INT_MAX (#1825)

  Show description for 79e3dc4
  gdb

  authored
  79e3dc4

  Copy full SHA for 79e3dc4

Commits on Jul 2, 2024

• release gil (#1816)

  Show description for 23c1f86
  crcrpar

  authored
  23c1f86

  Copy full SHA for 23c1f86

Commits on Apr 26, 2024

• Fix reduce_blocks_into_lanes race condition (#1798)

  Show description for a7de60e
  Fuzzkatt

  authored
  a7de60e

  Copy full SHA for a7de60e

Commits on Apr 19, 2024

• Add 2D Fused RoPE (#1784)

  Show description for b5df1cc
  yaox12

  authored
  b5df1cc

  Copy full SHA for b5df1cc

Commits on Jan 12, 2024

• Fused RoPE for `thd` format (#1756)

  Show description for f058162
  yaox12

  authored
  f058162

  Copy full SHA for f058162

Commits on Jan 5, 2024

• 64-bit indexing Adam (#1765)

  Show description for 87c4deb
  eqy

  authored
  87c4deb

  Copy full SHA for 87c4deb

Commits on Nov 29, 2023

• [FusedRoPE] Fuse type conversion and cos/sin (#1752)

  Show description for 37d83fc
  yaox12

  authored
  37d83fc

  Copy full SHA for 37d83fc

Commits on Nov 23, 2023

• Avoid `.contiguous()` in fused RoPE (#1751)

  Show description for dc5fa38
  yaox12

  authored
  dc5fa38

  Copy full SHA for dc5fa38

Commits on Nov 16, 2023

• fix a bug in fused rope (#1750)

  Show description for fd4ae7d
  yaox12

  authored
  fd4ae7d

  Copy full SHA for fd4ae7d

Commits on Nov 14, 2023

• A fused `apply_rotary_pos_emb` implementation for Megatron-Core (#1746)

  Show description for 08f7402
  yaox12

  authored
  08f7402

  Copy full SHA for 08f7402

Commits on Sep 30, 2023

• Add hysteresis support for AMP gradient scale update (#1733)

  Show description for 6a77872
  

  4 peopleauthored
  6a77872

  Copy full SHA for 6a77872

Commits on Sep 28, 2023

• Massively reduce LayerNorm/RMSNorm GPU memory usage in modern networks by tricking torch autograd (#1715)

  Show description for 6ff4548
  RuiWang1998

  authored
  6ff4548

  Copy full SHA for 6ff4548

Commits on Sep 19, 2023

• Add multi_tensor_unscale_l2norm_cuda (#1727)

  Show description for 741bdf5
  minitu

  authored
  741bdf5

  Copy full SHA for 741bdf5

Commits on Aug 8, 2023

• Add BF16 support to FusedMixedPrecisionLamb (#1439)

  Show description for 6309120
  

  nv-joseli

  and

  crcrpar

  authored
  6309120

  Copy full SHA for 6309120

Commits on Jun 1, 2023

• tensor.type() is deprecated, we should use scalar_type() (#1672)

  Show description for ecb5614
  

  ycsos

  and

  yuchuang

  authored
  ecb5614

  Copy full SHA for ecb5614

Commits on Apr 18, 2023

• add bf16 support for FusedDense (#1627)

  Agoniii

  authored
  1ba89ee

  Copy full SHA for 1ba89ee

Commits on Apr 14, 2023

• Add support for FP32 master weights in FusedAdam optimizer (#1625)

  Show description for 6bd01c4
  

  4 peopleauthored
  6bd01c4

  Copy full SHA for 6bd01c4

Commits on Mar 22, 2023

• Revert "Add support for FP32 master weights in FusedAdam optimizer (#1623)" (#1624)

  Show description for 4060aef
  crcrpar

  authored
  4060aef

  Copy full SHA for 4060aef

• Add support for FP32 master weights in FusedAdam optimizer (#1623)

  Show description for 6952004
  

  3 peopleauthored
  6952004

  Copy full SHA for 6952004

Commits on Mar 7, 2023

• s/maximimize/maximize/ (#1604)

  Show description for be819ea
  EIFY

  authored
  be819ea

  Copy full SHA for be819ea

Commits on Feb 25, 2023

• Decouple optimizer state and grad dtypes in distributed Adam optimizer (#1575)

  Show description for 03c9d80
  timmoon10

  authored
  03c9d80

  Copy full SHA for 03c9d80

Commits on Feb 9, 2023

• decrease warnings & remove append_nvcc_threads from setup.py (#1581)

  Show description for 0145d69
  crcrpar

  authored
  0145d69

  Copy full SHA for 0145d69

Commits on Feb 3, 2023

• [fused softmax] integer overflow and non-contiguous memory (#1568)

  Show description for 5c9625c
  cyanguwa

  authored
  5c9625c

  Copy full SHA for 5c9625c

Commits on Jan 31, 2023

• Capturable Adam optimizer (#1560)

  Show description for ad68a2f
  gilshm

  authored
  ad68a2f

  Copy full SHA for ad68a2f