All pull requests

Pull requests

• Arm backend: Fix set of VULKAN_SDK

  ciflow/trunkCLA Signedmodule: armmodule: vulkanpartner: armrelease notes: arm
  #23163·wwwind opened Sep 25, 2026Collaborator·

  ·
  1

• [Vulkan] Fully delegate dynamic transformer blocks

  CLA Signedmodule: vulkanrelease notes: vulkan
  #23162·mergennachin opened Sep 25, 2026Contributor·

  ·
  2

  8

• Qualcomm AI Engine Direct - Floor per-channel weight scales at the weight width's eps

  CLA Signed
  #23161·msluszniak opened Sep 25, 2026Contributor·

  ·
  1

  2

• Move the sdpa_bitwise_mask_gen test onto the OSS-exported path

  CLA Signedmeta-exported
  #23157·mvartani-meta opened Sep 25, 2026Contributor·

  ·
  5

• [examples] Enable Nemotron 3 Diarization on Vulkan

  CLA Signedmodule: vulkan
  #23155·mergennachin opened Sep 25, 2026Contributor·

  ·
  1

• XNNPACK: fold constant subgraphs before ATen decomposition

  ciflow/trunkCLA Signedmodule: xnnpackpartner: armrelease notes: xnnpack
  #23154·mansnils opened Sep 25, 2026Collaborator·

  ·
  1

• [examples] Enable Nemotron 3 Diarization on XNNPACK

  CLA Signedmodule: vulkan
  #23152·mergennachin opened Sep 25, 2026Contributor·

  ·
  6

• Arm backend: Phase 3 - Vulkan backend uses shared vulkan context.

  CLA Signedmodule: armmodule: vulkanpartner: armrelease notes: arm
  #23151·wwwind opened Sep 25, 2026Collaborator·

  ·
  2

• Arm backend: Add Arm FACTO nightly harness

  CLA Signedpartner: armrelease notes: arm
  #23149·bdemirb opened Sep 25, 2026Collaborator·

  ·
  4

• Ignore missing imports in mypy

  ciflow/trunkCLA Signedrelease notes: none
  #23142·perheld opened Sep 25, 2026Collaborator·

  ·
  4

• Add experimental Hugging Face .tok tokenizer path

  CLA Signed
  #23139·JacobSzwejbka opened Sep 24, 2026Contributor·

  ·
  2

• [MLX] Allow lowering EP's with unused constants

  ciflow/mlxCLA Signed
  #23138·metascroy opened Sep 24, 2026Contributor·

  ·
  4

• Add MLPerf Tiny streaming wakeword model

  CLA Signedrelease notes: examples
  #23133·rascani opened Sep 24, 2026Contributor·

  ·
  1

• Hoist data pointers out of sdpa_bitwise_mask_gen packing loops (#23129)

  CLA Signedmeta-exported
  #23129·mvartani-meta opened Sep 24, 2026Contributor·

  ·
  5

• Retire internal Static Docs build wiring

  CLA Signedmeta-exported
  #23128·JakeStevens opened Sep 24, 2026Contributor·

  ·
  3

• Add memory format to const prop primitive types (#23127)

  CLA Signedmeta-exported
  #23127·DrJessop opened Sep 24, 2026Contributor·

  ·
  3

• Add atomic data replacement to DataLoader

  CLA Signed
  #23125·JacobSzwejbka opened Sep 24, 2026Contributor·

  ·
  3

• Add named data replacement support

  CLA Signed
  #23124·JacobSzwejbka opened Sep 24, 2026Contributor·

  ·
  1

• Ship the CUDA runtime in the Windows CUDA wheel

  CLA Signed
  #23123·Gasoonjia opened Sep 24, 2026Contributor·

  ·
  7

• Optimize CPU SDPA for ring attention

  CLA Signed
  #23122·JacobSzwejbka opened Sep 24, 2026Contributor·

  ·
  2

• Ship linkable libraries in the Windows CPU wheel

  CLA Signed
  #23121·Gasoonjia opened Sep 24, 2026Contributor·

  ·
  3

• Add delegate data replacement APIs

  CLA Signed
  #23120·JacobSzwejbka opened Sep 24, 2026Contributor·

  ·
  2

• Support negative padding in constant_pad_nd

  CLA Signedrelease notes: ops & kernels
  #23116·JakeStevens opened Sep 24, 2026Contributor·

  ·
  1

  1

• Arm backend: Normalize TOSA operator support checks

  ciflow/trunkCLA Signedmodule: armrelease notes: none
  #23115·AdrianLundell opened Sep 24, 2026Collaborator·

  ·
  1

• Update xnnpack test buck label

  CLA Signedmeta-exported
  #23114·JakeStevens opened Sep 24, 2026Contributor·

  ·
  3