All pull requests

Pull requests

• Remove no-op copies so mutated inputs are reinplaced after to_edge_transform_and_lower

  #23583·IlyasMoutawwakil opened Oct 8, 2026·

  ·
  3

• Qualcomm AI Engine Direct - e2e test migration #1

  CLA Signed
  #23582·haowhsu-quic opened Oct 8, 2026Contributor·

  ·
  2

• Fix pass ordering for sharded LLM export in QNN backend

  CLA Signed
  #23581·SabYic opened Oct 8, 2026·

  ·
  1

  10

• Arm backend: Rename Ethos-U driver implementations

  ciflow/trunkCLA Signedhelp wantedmodule: armpartner: armrelease notes: none
  #23579·perheld opened Oct 8, 2026Collaborator·

  ·
  2

• Arm backend: Add --backend all option for convenience.

  ciflow/trunkCLA Signedmodule: armrelease notes: arm
  #23578·wwwind opened Oct 8, 2026Collaborator·

  ·
  1

• Arm backend: Remove redundant compile-spec overrides

  ciflow/trunkCLA Signedmodule: armpartner: armrelease notes: none
  #23577·Sebastian-Larsson opened Oct 8, 2026Collaborator·

  ·
  1

• [WebGPU] Support et_vk.swiglu

  CLA Signed
  #23574·SS-JIA opened Oct 8, 2026Contributor·

  ·
  2

• [ET-VK][q8ta] Support input broadcasting in q8ta_add

  CLA Signed
  #23573·pytorchbot opened Oct 8, 2026Collaborator·

  ·
  2

• Qualcomm AI Engine Direct - Enforce terminal pass ordering

  CLA Signed
  #23572·chenweng-quic opened Oct 8, 2026Contributor·

  ·
  2

• [executorch][cuda] Run INT5 decode linears on autotuned QuantizedGemmFamily Triton kernels and delete the last C shim (int5_plain_mm)

  CLA Signed
  #23571·Gasoonjia opened Oct 8, 2026Contributor·

  ·
  2

• [executorch][cuda] Run INT8 decode linears on autotuned QuantizedGemmFamily Triton kernels and delete the int8_plain_mm C shim

  CLA Signed
  #23570·Gasoonjia opened Oct 8, 2026Contributor·

  ·
  2

• [executorch][cuda] Run INT6 decode linears on autotuned QuantizedGemmFamily Triton kernels and delete the int6_plain_mm C shim

  CLA Signed
  #23569·Gasoonjia opened Oct 8, 2026Contributor·

  ·
  2

• [executorch][cuda] Run INT4 decode linears on autotuned QuantizedGemmFamily Triton kernels and delete the int4_plain_mm C shim

  CLA Signed
  #23568·Gasoonjia opened Oct 8, 2026Contributor·

  ·
  2

• [executorch][cuda] Add QuantizedGemmFamily: per-M quantized GEMM ops with format-owned legality and shared dispatch

  CLA Signed
  #23567·Gasoonjia opened Oct 8, 2026Contributor·

  ·
  2

• [executorch][cuda] Pick launch-time kernel parameters such as split-K by timing them during AOTI compile

  CLA Signedmeta-exported
  #23566·Gasoonjia opened Oct 8, 2026Contributor·

  ·
  2

• [executorch][cuda] Autotune data-dependent kernels on representative inputs declared by the kernel author

  CLA Signed
  #23565·Gasoonjia opened Oct 8, 2026Contributor·

  ·
  2

• [executorch][cuda] Time AOTI autotune candidates with CUDA graphs so a busy CPU does not decide the pick

  CLA Signed
  #23564·Gasoonjia opened Oct 8, 2026Contributor·

  ·
  2

• Preserve in-place lifetimes and buffer allocation

  CLA Signedmeta-exported
  #23563·JakeStevens opened Oct 8, 2026Contributor·

  ·
  3

• Avoid copying the whole file when reading .pte and .ptd headers

  CLA Signedrelease notes: none
  #23561·HiroForYou opened Oct 7, 2026Contributor·

  ·
  5

• Add int16 support to Cadence quantized add reference

  CLA Signedmeta-exported
  #23560·DrJessop opened Oct 7, 2026Contributor·

  ·
  3

• Add missing dtype support to quantized mul per tensor

  CLA Signedmeta-exported
  #23559·DrJessop opened Oct 7, 2026Contributor·

  ·
  3

• Test portable export of MLPerf Tiny streaming wakeword

  CLA Signed
  #23556·rascani opened Oct 7, 2026Contributor·

  ·
  1

  3

• [ET-VK][q4gsw] Fuse sibling q4 linears (q/k/v) into one GEMV with split outputs

  CLA Signedmeta-exported
  #23552·SS-JIA opened Oct 7, 2026Contributor·

  ·
  2

• [ET-VK][q4gsw] Compute dq8ca input qparams in the op; skip them for GEMV

  CLA Signedmeta-exported
  #23551·SS-JIA opened Oct 7, 2026Contributor·

  ·
  2

• [ET-VK][q4gsw] Transposed int4 weight blocks + row-owner M=1 GEMV

  CLA Signedmeta-exported
  #23550·SS-JIA opened Oct 7, 2026Contributor·

  ·
  2