All pull requests

Pull requests

• [XNNPACK] Reject addmm with non-default alpha or beta

  CLA Signedrelease notes: xnnpack
  #24814·JakeStevens opened Nov 9, 2026Contributor·

  ·
  3

• [ET-VK][unary] Compute vec4 leaky_relu with a single mix

  CLA Signedmeta-exported
  #24813·SS-JIA opened Nov 9, 2026Contributor·

  ·
  2

• Add Phonon (Gradium 31M TTS) example with multi-backend export

  CLA Signedmeta-exported
  #24810·psiddh opened Nov 9, 2026Contributor·

  ·
  5

• Swap reverse-broadcast operands once in handle_broadcast_elementwise

  CLA Signed
  #24809·JakeStevens opened Nov 9, 2026Contributor·

  ·
  2

• [ET-VK][conv2d] Fuse residual adds and activations into conv2d during lowering

  CLA Signedmeta-exported
  #24808·SS-JIA opened Nov 9, 2026Contributor·

  ·
  2

• [ET-VK][conv2d] Fuse residual and activation epilogues into conv2d shaders

  CLA Signedmeta-exported
  #24807·SS-JIA opened Nov 9, 2026Contributor·

  ·
  2

• [ET-VK][tools] Write etvk_ops_bench JSON files compactly

  CLA Signedmeta-exported
  #24806·SS-JIA opened Nov 9, 2026Contributor·

  ·
  2

• [MLX] Raise default macOS deployment target to 26.2 for NAX

  CLA Signedrelease notes: build
  #24803·Gasoonjia opened Nov 9, 2026Contributor·

  ·
  1

• Break the exir backend_api <-> program import cycle

  CLA Signed
  #24802·JakeStevens opened Nov 9, 2026Contributor·

  ·
  4

• [runtime] Fix null operator overload logging and named data key dereferences

  CLA Signedcommunity: contributionrelease notes: runtime
  #24801·AhmadHassan-BTed opened Nov 9, 2026·

  ·
  1

  6

• Add an aten.elu kernel and a single-kernel aten.silu

  CLA Signedmeta-exported
  #24800·psiddh opened Nov 9, 2026Contributor·

  ·
  4

• OpenVINO: transpose channels-last tensors at partition edges

  CLA Signed
  #24797·suryasidd opened Nov 9, 2026Collaborator·

  ·
  4

• Arm backend: Add shared device to VGF backend.

  ciflow/trunkCLA Signedmodule: arm
  #24795·wwwind opened Nov 9, 2026Collaborator·

  ·
  2

• Arm backend: Fuse RIFE heads without PixelShuffle

  ciflow/trunkCLA Signedmodule: armpartner: armrelease notes: none
  #24790·oscarandersson8218 opened Nov 9, 2026Collaborator·

  ·
  1

• Add the aoti_torch_cuda_masked_select fallback shim (#24788)

  CLA Signedmeta-exported
  #24788·Gasoonjia opened Nov 9, 2026Contributor·

  ·
  5

• Add the aoti_torch_cuda_nonzero fallback shim

  CLA Signedmeta-exported
  #24787·Gasoonjia opened Nov 9, 2026Contributor·

  ·
  5

• [OpenVINO] Import the partitioner without needing nncf

  CLA Signedcommunity: contribution
  #24785·sohailshk opened Nov 9, 2026·

  ·
  6

• [XNNPACK] Skip linear conversion when the weight transpose was folded

  CLA Signedcommunity: contribution
  #24784·sohailshk opened Nov 9, 2026·

  ·
  1

  5

• CUDA backend: create the temporary kernel library file safely

  CLA Signedmodule: cudarelease notes: backends
  #24783·shoumikhin opened Nov 9, 2026Contributor·

  ·
  17

• Metal backend: free only the destroyed delegate's memory

  CLA Signedcommunity: contribution
  #24782·abdelaziz-mahdy opened Nov 8, 2026Contributor·

  ·
  10

• [ET-WebGPU] Keep fp16 elementwise ops on CPU

  CLA Signedcommunity: contribution
  #24781·abdelaziz-mahdy opened Nov 8, 2026Contributor·

  ·
  3

• Make QuantAbsorption's input-side absorption opt-in (#24780)

  CLA Signedmeta-exported
  #24780·mcremon-meta opened Nov 7, 2026Contributor·

  ·
  3

• Add validate_tensor_layout to check a tensor layout without aborting, and document the raw-bytes overloads

  CLA Signedrelease notes: api
  #24772·shoumikhin opened Nov 7, 2026Contributor·

  ·
  15

• Make nonzero and the Vulkan, OpenVINO and WebGPU delegates follow the declared dim order

  CLA Signedmodule: vulkanrelease notes: backends
  #24771·shoumikhin opened Nov 7, 2026Contributor·

  ·
  1

  24

• Keep a loaded XNNPACK method's thread pool alive, and ship threadpool.h in the wheel

  CLA Signedrelease notes: runtime
  #24770·shoumikhin opened Nov 7, 2026Contributor·

  ·
  21