All pull requests

Pull requests

• [ET-VK][benchmark] Add JSON operator and shader benchmark cases

  CLA Signed
  #24391·pytorchbot opened Oct 31, 2026Collaborator·

  ·
  1

• [ET-VK][sdpa] Expose shader override

  CLA Signed
  #24390·pytorchbot opened Oct 31, 2026Collaborator·

  ·
  1

• [ET-VK][runtime] Expose raw shader timestamps and dispatch arguments

  CLA Signed
  #24389·pytorchbot opened Oct 31, 2026Collaborator·

  ·
  1

• [ET-VK][runtime] Support arbitrary explicit local workgroup sizes

  CLA Signed
  #24388·pytorchbot opened Oct 31, 2026Collaborator·

  ·
  1

• [ET-VK][runtime] Derive SPIR-V requirements and enable Vulkan shader features

  CLA Signedmodule: vulkan
  #24387·pytorchbot opened Oct 31, 2026Collaborator·

  ·
  1

• [ET-VK][q8ta] Honor alpha in quantized addition

  CLA Signed
  #24386·pytorchbot opened Oct 31, 2026Collaborator·

  ·
  2

• Support 3-D convolution in the portable convolution kernel

  CLA Signedrelease notes: ops & kernels
  #24385·shoumikhin opened Oct 31, 2026Contributor·

  ·
  4

• Qualcomm AI Engine Direct - Fold views of parameters before annotation

  CLA Signedcommunity: contribution
  #24384·msluszniak opened Oct 31, 2026Contributor·

  ·
  2

• Qualcomm AI Engine Direct - Drop single-chunk split_with_sizes

  CLA Signedcommunity: contribution
  #24383·msluszniak opened Oct 31, 2026Contributor·

  ·
  2

• Qualcomm AI Engine Direct - Decompose min/max.dim without source partitions

  CLA Signedcommunity: contribution
  #24382·msluszniak opened Oct 31, 2026Contributor·

  ·
  2

• Fix mixed-dtype cat and stack in to_edge, the edge check and portable cat

  ciflow/trunkCLA Signedmodule: armrelease notes: exir
  #24381·shoumikhin opened Oct 31, 2026Contributor·

  ·
  7

• Keep ops with more than 6 dimensions out of the XNNPACK delegate

  CLA Signedrelease notes: xnnpack
  #24380·shoumikhin opened Oct 31, 2026Contributor·

  ·
  4

• Read reduce keepdim when it is serialized as Bool (#24379)

  CLA Signedmeta-exported
  #24379·psiddh opened Oct 31, 2026Contributor·

  ·
  4

• Add a resize hook to native_layer_norm (#24378)

  CLA Signedmeta-exported
  #24378·psiddh opened Oct 31, 2026Contributor·

  ·
  4

• Fix 0-d tensors, dynamic output sizes and integer reductions, and default to float32 on CPU, in the OpenVINO backend

  ciflow/trunkCLA Signedmodule: arm
  #24377·pytorchbot opened Oct 31, 2026Collaborator·

  ·
  3

• Add FLUX.1 export and runtime support

  CLA Signed
  #24376·kiymetakdemir opened Oct 31, 2026Contributor·

  ·
  2

• [CoreAI] Add a coreai_compute_unit backend option

  CLA Signed
  #24375·metascroy opened Oct 31, 2026Contributor·

  ·
  2

• [CoreAI] Add multithread stress testing and shared op/model suites

  CLA Signed
  #24374·metascroy opened Oct 31, 2026Contributor·

  ·
  2

• [executorch][native] Run the ggml CPU act_quant sample test in CI

  CLA Signedmeta-exported
  #24373·digantdesai opened Oct 30, 2026Contributor·

  ·
  2

• [executorch][vulkan] Round f16 image stores to nearest even

  CLA Signedmeta-exportedmodule: vulkan
  #24372·digantdesai opened Oct 30, 2026Contributor·

  ·
  2

• [executorch][native-vk] Note the int4 weight-only N % 4 and rank-4 linear gaps

  CLA Signedmeta-exported
  #24371·digantdesai opened Oct 30, 2026Contributor·

  ·
  2

• [executorch][vulkan] Fix dq8ca q4gsw weight sums stride when N % 8 is 1 to 4

  CLA Signedmeta-exportedmodule: vulkan
  #24370·digantdesai opened Oct 30, 2026Contributor·

  ·
  2

• [executorch][vulkan] Accept 1-D activations in 8da4w linear

  CLA Signedmeta-exportedmodule: vulkan
  #24369·digantdesai opened Oct 30, 2026Contributor·

  ·
  2

• [executorch][native] Cover zero, negative, rounded and subnormal GGUF super-block scales in the op samples

  CLA Signed
  #24368·digantdesai opened Oct 30, 2026Contributor·

  ·
  2

• [executorch][native] Decode quantized weights in float32 for precise samples

  CLA Signedmeta-exported
  #24367·digantdesai opened Oct 30, 2026Contributor·

  ·
  2