All pull requests

Pull requests

• NXP backend: Adding the build diagram for MCUXpresso example

  CLA Signedmodule: nxprelease notes: nxp
  #23741·jirioc opened Oct 13, 2026Collaborator·

  ·
  2

• [executorch][cuda] Trace a tuned quantized GEMM with only its picked config, timed in CUDA graphs

  CLA Signedmeta-exported
  #23740·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

• [executorch][cuda] Reuse PyTorch's access analysis of a repeated user Triton kernel call

  CLA Signedmeta-exported
  #23739·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

• [executorch][cuda] Write a weight storage the methods share only once

  CLA Signedmeta-exported
  #23738·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

• [executorch][cuda] Shorter autotune benchmark window and one compile pool for every split candidate

  CLA Signedmeta-exported
  #23737·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

• [executorch][cuda] Compile an autotuned quantized GEMM's configs in parallel before benchmarking them

  CLA Signedmeta-exported
  #23736·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

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

  CLA Signedmeta-exported
  #23735·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

• [executorch][muse] Benchmark MG batching through the serving runtime

  CLA Signedmeta-exported
  #23734·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

• [executorch][muse] Serve MG batching on CUDA from the launcher

  CLA Signedmeta-exported
  #23733·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

• [executorch][cuda] Test prefix-cache reuse on CudaExecutor

  CLA Signedmeta-exported
  #23732·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

• [executorch][muse] Run the MG batching runtime on CudaExecutor

  CLA Signedmeta-exported
  #23731·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

• [executorch][muse] Hash prepared image identity without CommonCrypto

  CLA Signedmeta-exported
  #23730·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

• [executorch][muse] Export the batching contract for CUDA with --use-offgraph-kv-cache

  CLA Signedmeta-exported
  #23729·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

• [executorch][cuda] Retain emptied KV backing across JIT autotuning

  CLA Signedmeta-exported
  #23728·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

• [executorch][cuda] Autotune SDPA on a visible mask and tune pack GQA with the tile

  CLA Signedmeta-exported
  #23727·Gasoonjia opened Oct 13, 2026Contributor·

  ·
  2

• Bump PyTorch to 2.15.0 everywhere and support the CUDA and ROCm versions it ships for

  ciflow/cudaciflow/rocmCLA Signedmodule: rocm
  #23712·Gasoonjia opened Oct 12, 2026Contributor·

  ·
  2

• [Muse Glimmer] Enable DFlash in the off-graph MLX batching executor

  CLA Signed
  #23711·metascroy opened Oct 12, 2026Contributor·

  ·
  1

• [MLX] Add a batched off-graph cache with DFlash support

  CLA Signed
  #23710·metascroy opened Oct 12, 2026Contributor·

  ·
  4

• [ET-VK][CI] Add end-to-end quantized LLM coverage

  CLA Signed
  #23708·SS-JIA opened Oct 12, 2026Contributor·

  ·
  8

• Arm backend: Fix Llama tests

  ciflow/trunkCLA Signedmodule: armrelease notes: none
  #23705·tom-arm opened Oct 12, 2026Contributor·

  ·
  3

• NXP backend: Update MCUXpresso version in the documentation

  CLA Signedmodule: nxprelease notes: nxp
  #23704·jirioc opened Oct 12, 2026Collaborator·

  ·
  3

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

  CLA Signedpython
  #23703·MartinPavella opened Oct 12, 2026Collaborator·

  ·
  3

  

• Arm backend: CWE-427 · Uncontrolled Search Path Element

  ciflow/trunkCLA Signedmodule: armpartner: armrelease notes: arm
  #23701·wwwind opened Oct 12, 2026Collaborator·

  ·
  2

• Bump PyTorch and TorchAO pins for dev20261011

  ci/pytorch-pin-bumpci-no-tdciflow/cudaciflow/dockerCLA Signed
  #23698·pytorchupdatebot opened Oct 12, 2026Collaborator·

  ·
  2

• transforms: handle rank changes during propagation

  ciflow/trunkCLA Signedhelp wantedmodule: armpartner: armrelease notes: none
  #23697·AdrianLundell opened Oct 12, 2026Collaborator·

  ·
  1