All pull requests

Pull requests

• Honor set_grad_none=False when clearing gradients

  #2034·sylvesterkaczmarek opened Sep 23, 2026·

  ·

• Restore graph lowering after compiler failures

  #2033·sylvesterkaczmarek opened Sep 23, 2026·

  ·

• Prune unstructured weights by magnitude

  #2032·sylvesterkaczmarek opened Sep 23, 2026·

  ·

• Preserve float64 accumulation in RMSNorm fallback

  #2031·sylvesterkaczmarek opened Sep 23, 2026·

  ·

• [pre-commit.ci] pre-commit autoupdate

  #2030·pre-commit-ci[bot] opened Sep 14, 2026Bot·

  ·

• Replace deprecated torch.cuda.amp API with torch.amp equivalents

  #2025·xyf5432 opened Aug 20, 2026·

  ·
  1

• fix(contrib): call torch.cuda.is_available() in FP16_Optimizer guard

  #2021·andrewwhitecdw opened Jul 26, 2026·

  ·
  4

• extend 64bit indexing for other multi tensor ops

  #2014·POI-WX opened Jun 9, 2026·

  ·

• Avoid retaining saved tensors in fused norm custom ops under no_grad

  #2012·LeSingh1 opened May 31, 2026Contributor·

  ·

• Constrain APEX_ASP_CACHE_DIR to a safe base directory

  #2011·LeSingh1 opened May 31, 2026Contributor·

  ·
  1

• fix missing standard headers in group_norm_v2 host template

  #2010·ZhiyuanChen opened May 28, 2026Contributor·

  ·

• Use TORCH_LIBRARY*, Make Python ABI Stable

  #2009·crcrpar opened May 27, 2026Collaborator·

  ·
  2

• Guard CUDA probe when CUDA_HOME is unset

  #2008·fallintoplace opened May 26, 2026·

  ·

• [contrib][test_clip_grad] Increase rtol for fp16 test to 5e-3

  #2002·crcrpar opened May 12, 2026Collaborator·

  ·

• Group norm v2: add missing include

  #1997·RemiLacroix-IDRIS opened Apr 27, 2026·

  ·

• Add 64-bit indexing fallback for large multi_tensor_l2norm kernels

  #1989·SongXiaoXi opened Mar 25, 2026·

  ·
  4

• Fix traceback formatting in permutation_lib.py

  #1987·yujincheng08 opened Mar 17, 2026Member·

  ·
  3

• Fixed Win11 install issues of compile flags and isfnite() crashing.

  #1977·original-doc opened Jan 18, 2026·

  ·
  3

• Fix FMHA bwd_nl kernel issue for batch_size == 3

  #1975·crcrpar opened Jan 14, 2026Collaborator·

  ·

• Fix E731: Convert lambda assignments to def functions

  #1965·crcrpar with CopilotAI opened Nov 29, 2025·

  ·
  1

  1

  

• Add LibTorch Stable ABI infrastructure (#1946)

  #1956·jackulau opened Nov 24, 2025·

  ·
  20

• fix build python,cpp_ext and cuda_ext error for Visual Studio 2022 on…

  #1931·huhai463127310 opened Aug 29, 2025·

  ·

• fix readme.md install for newer pip

  #1904·ciaxj88 opened May 19, 2025·

  ·

• Update bug_report.md

  #1903·crcrpar opened May 8, 2025Collaborator·

  ·

• Fix the typo of FusedRMSNorm doc

  #1869·cqulilujia opened Dec 30, 2024·

  ·
  4