All pull requests

Pull requests

• fix(export): get_edge_program_manager returns stale EPM when EDGE_PROGRAM_MANAGER_TRANSFORM ran

  CLA Signedpython
  #23889·MartinPavella opened Oct 15, 2026Collaborator·

  ·
  3

  

• Add Arm Buck2 coverage and declaration checks to GitHub CI

  ciflow/trunkCLA Signedhelp wantedmodule: armpartner: armrelease notes: arm
  #23888·usamahz opened Oct 15, 2026Collaborator·

  ·
  3

  

• NXP backend: refactor dataset_creators to enable better benchmarking

  CLA Signedmodule: nxprelease notes: nxp
  #23886·novak-vaclav opened Oct 15, 2026Collaborator·

  ·
  3

• Quacomm AI Engine Direct - Bump QNN SDK version from 2.37 to 2.50

  CLA Signed
  #23884·chenweng-quic opened Oct 15, 2026Contributor·

  ·
  6

• [ExecuTorch][WebGPU] Add a shared-memory direct kernel for groups == 1 conv2d

  CLA Signed
  #23882·digantdesai opened Oct 15, 2026Contributor·3/3·

  ·
  3

• [ExecuTorch][WebGPU] Run groups == 1 conv_with_clamp on the conv2d GEMM

  CLA Signed
  #23881·digantdesai opened Oct 15, 2026Contributor·2/3·

  ·
  2

• [ExecuTorch][WebGPU] Run stride == kernel conv_transpose2d as a GEMM

  CLA Signed
  #23880·digantdesai opened Oct 15, 2026Contributor·1/3·

  ·
  2

• [MLX] Rename the canonical sequence-cache header

  CLA Signed
  #23879·metascroy opened Oct 15, 2026Contributor·

  ·
  1

• [MLX] Promote the sequence cache and consolidate control tests

  CLA Signed
  #23878·metascroy opened Oct 15, 2026Contributor·

  ·
  1

• Fix Python 3.15 wheel dependency setup

  CLA Signed
  #23876·JacobSzwejbka opened Oct 15, 2026Contributor·

  ·
  2

• CUDA: add prepared-weight Q8 MMQ for large-row prefill

  CLA Signed
  #23873·Gasoonjia opened Oct 14, 2026Contributor·

  ·
  2

• fbcode/executorch/devtools/bundled_program/serialize/__init__.py

  CLA Signedmeta-exported
  #23872·meta-codesync[bot] opened Oct 14, 2026Bot·

  ·
  3

• Implement the portable kernel for _embedding_bag.out

  CLA Signed
  #23871·yenhao-huang opened Oct 14, 2026Contributor·

  ·
  1

  2

• Per-load option to turn the warm-up prediction on or off (#23870)

  CLA Signedmeta-exported
  #23870·derolf opened Oct 14, 2026·

  ·
  4

• FACTO: Add Arm U55 generated-test profile

  CLA Signedhelp wantedpartner: armrelease notes: arm
  #23869·bdemirb opened Oct 14, 2026Collaborator·

  ·
  4

• [ET-VK][profiling] Record dispatch configuration in GPU traces

  CLA Signed
  #23868·SS-JIA opened Oct 14, 2026Contributor·

  ·
  2

• Gate background delegation using forecasts and measured inference rates

  CLA Signed
  #23866·mergennachin opened Oct 14, 2026Contributor·

  ·
  2

• Add a bounded background agent for Terminal-Bench

  CLA Signed
  #23865·mergennachin opened Oct 14, 2026Contributor·

  ·
  3

• Expose request and live engine telemetry from the LLM server

  CLA Signed
  #23864·mergennachin opened Oct 14, 2026Contributor·

  ·
  2

• [executorch][native-ggml] Keep strided permutations as views

  CLA Signed
  #23853·pytorchbot opened Oct 14, 2026Collaborator·

  ·
  1

• Fix Method input layout handling for non-contiguous tensors

  CLA Signedrelease notes: runtime
  #23820·JakeStevens opened Oct 14, 2026Contributor·

  ·
  1

  1

• [executorch][cuda] Test CudaExecutor on every step width up to 65

  CLA Signedmeta-exported
  #23819·Gasoonjia opened Oct 14, 2026Contributor·

  ·
  2

• [executorch][ci] Test Muse Glimmer batching on CUDA in OSS CI

  CLA Signed
  #23818·Gasoonjia opened Oct 14, 2026Contributor·

  ·
  2

• [WIP] Let delegate backends declare memory-planned state buffers

  CLA Signed
  #23817·rascani opened Oct 14, 2026Contributor·

  ·
  2

• Fix FP8 mismatch diagnostics in the test harness

  ciflow/trunkCLA Signedhelp wantedmodule: backendpartner: armrelease notes: none
  #23813·YufengShi-dudu opened Oct 14, 2026Collaborator·

  ·
  1