All pull requests

Pull requests

• Validate layout copies on qparams, not fake-tensor dtype (#24071)

  CLA Signedmeta-exportedrelease notes: none
  #24071·pssrawat opened Oct 21, 2026Contributor·

  ·
  2

• Add Python inference bindings to the Native wheel

  CLA Signedmeta-exported
  #24070·JakeStevens opened Oct 21, 2026Contributor·

  ·
  3

• Package Native, Vulkan, and GGML as an executorch-native wheel

  CLA Signedmeta-exportedmodule: vulkan
  #24069·JakeStevens opened Oct 21, 2026Contributor·

  ·
  3

• Fold evaluation BatchNorm during Native AOT export

  CLA Signedmeta-exported
  #24068·JakeStevens opened Oct 21, 2026Contributor·

  ·
  3

• Don't drop a permute that an earlier region already bypassed

  CLA Signed
  #24067·Gasoonjia opened Oct 21, 2026Contributor·

  ·
  2

• [executorch][muse] Accept images in CUDA DFlash batching

  CLA Signedmeta-exported
  #24066·Gasoonjia opened Oct 21, 2026Contributor·

  ·
  2

• [executorch][muse] Serve images on CUDA batching

  CLA Signedmeta-exported
  #24065·Gasoonjia opened Oct 21, 2026Contributor·

  ·
  2

• [executorch][muse] Export vision with the CUDA batching contract

  CLA Signedmeta-exported
  #24064·Gasoonjia opened Oct 21, 2026Contributor·

  ·
  2

• [executorch][cuda] Run CudaExecutor steps from prepared embeddings

  CLA Signedmeta-exported
  #24063·Gasoonjia opened Oct 21, 2026Contributor·

  ·
  2

• [executorch][muse] Share the Muse Glimmer embeddings engine across backends

  CLA Signedmeta-exported
  #24062·Gasoonjia opened Oct 21, 2026Contributor·

  ·
  2

• Qualcomm AI Engine Direct - Lower LPBQ linear as conv2d on HTP to avoid FullyConnected block-scale over-read bug

  CLA Signedrelease notes: qualcomm
  #24061·qti-horodnic opened Oct 21, 2026Contributor·

  ·
  2

• [runtime] Enforce topological EValue ordering for TensorList/OptionalTensorList and null-check CompileSpec fields

  CLA Signedcommunity: contributionmodule: runtimerelease notes: runtime
  #24059·professor-moody opened Oct 21, 2026·

  ·
  6

  

• Arm backend: Don't quantize comparison ops with integer inputs

  ciflow/trunkCLA Signedmodule: armpartner: armrelease notes: none
  #24058·Erik-Lundell opened Oct 21, 2026Collaborator·

  ·
  1

• Qualcomm AI Engine Direct - Exact integer floor_divide and remainder

  CLA Signedcommunity: contributionmodule: qnnpartner: qualcomm
  #24057·msluszniak opened Oct 21, 2026Contributor·

  ·
  2

  

• XNNPACKQuantizer: apply module filters to ops only, not weight nodes

  CLA Signedcommunity: contributionmodule: xnnpackrelease notes: xnnpack
  #24056·msluszniak opened Oct 21, 2026Contributor·

  ·
  2

  

• Vulkan: keep a reshape that follows a dynamically quantized linear

  CLA Signedcommunity: contributionmodule: vulkanrelease notes: vulkan
  #24055·msluszniak opened Oct 21, 2026Contributor·

  ·
  1

  

• Qualcomm AI Engine Direct - Cast TopK int64 indices to int32

  CLA Signedcommunity: contributionmodule: qnnpartner: qualcomm
  #24054·msluszniak opened Oct 21, 2026Contributor·

  ·
  2

  

• Qualcomm AI Engine Direct - Leave a constant op with a runtime value on CPU

  CLA Signedcommunity: contributionmodule: qnnpartner: qualcomm
  #24053·msluszniak opened Oct 21, 2026Contributor·

  ·
  2

  

• [executorch][cuda] Check FQN weight coverage without AOTI's per-buffer warnings

  CLA Signed
  #24052·Gasoonjia opened Oct 21, 2026Contributor·

  ·
  2

• Qualcomm AI Engine Direct - Leave constant-only partitions on CPU

  CLA Signedcommunity: contributionmodule: qnnpartner: qualcomm
  #24051·msluszniak opened Oct 21, 2026Contributor·

  ·
  2

  

• [executorch][muse] Fix the DFlash final-block draft counts test

  CLA Signed
  #24050·Gasoonjia opened Oct 21, 2026Contributor·

  ·
  2

• [executorch][ci] Split unittest-cuda into kernel and system shards

  CLA Signed
  #24047·Gasoonjia opened Oct 21, 2026Contributor·

  ·
  3

• Qualcomm AI Engine Direct - Cast an int64 get_attr right before its first user

  CLA Signedcommunity: contributionmodule: qnnpartner: qualcomm
  #24046·msluszniak opened Oct 21, 2026Contributor·

  ·
  2

  

• Arm backend: Remove deprecated AOT compiler wrapper

  ciflow/trunkCLA Signedmodule: armpartner: armrelease notes: none
  #24045·Sebastian-Larsson opened Oct 21, 2026Collaborator·

  ·
  1

• Preserve eager numeric parity (#24035)

  CLA Signedmeta-exported
  #24035·YIWENX14 opened Oct 21, 2026Contributor·

  ·
  3