All pull requests

Pull requests

• [xpu] Re-enable the stft complex128 gradient UTs on BMG

  disable_distributeddisable_e2e
  #6158·hoshibara opened Nov 11, 2026Contributor·

  ·
  1

• Fix broken commands and stale facts in .claude skills

  disable_all
  #6157·EikanWang opened Nov 11, 2026Contributor·

  ·
  6

• [xpu] Use 64-bit offsets in nll_loss2d backward when 32-bit indexing is unsafe

  disable_distributeddisable_e2e
  #6154·hoshibara opened Nov 11, 2026Contributor·

  ·
  1

• Register hash_tensor.out for XPU backend

  disable_distributeddisable_e2e
  #6153·msnliu opened Nov 11, 2026Contributor·

  ·
  1

  10

• CI: drive P0 UT from a unified YAML config

  #6147·chuanqi129 opened Nov 11, 2026Contributor·

  ·

• Fix adaptive average pooling backward validation on XPU

  disable_distributeddisable_e2e
  #6146·lchen2331 opened Nov 11, 2026Contributor·

  ·
  1

  1

• Fix SYCL FFT kernel performance scaling with problem size

  disable_distributeddisable_e2e
  #6139·dbollapr opened Nov 10, 2026Contributor·

  ·

• Synchronize test_sparse_xpu.py with upstream

  disable_autodisable_distributeddisable_e2edisable_win
  #6138·pbielak opened Nov 10, 2026Contributor·

  ·

• Refactor test launching mechanism for the extended test suite to match the standard suite.

  disable_autodisable_distributeddisable_e2ewindows_ciwindows_ut
  #6135·BBBela opened Nov 10, 2026Contributor·

  ·
  7

  

• [Test] Add skipped inductor files to upstream_inductor action

  #6134·dominikjastrzebskix opened Nov 10, 2026Contributor·

  ·

• test_layer_norm_fast_path_forward_float32: Avoid inaccurate AVX2 CPU layer_norm as reference on Windows

  disable_builddisable_distributeddisable_e2edisable_win
  #6133·BartoszKokoszko opened Nov 10, 2026Contributor·

  ·
  1

• Add XPU machine acceptance checks; keep check_aot.py AOT-only

  disable_alldisable_auto
  #6132·mengfei25 opened Nov 10, 2026Contributor·

  ·

• [XPU] Reuse horizontal interpolation across bicubic output rows

  disable_distributeddisable_e2e
  #6130·yucai-pro opened Nov 10, 2026Contributor·

  ·

• Fix ProcessGroup hooks never firing for XPU collectives

  disable_e2edisable_win
  #6129·jemitche1 opened Nov 10, 2026·

  ·
  1

• Fix xpu_profiling UT check, update llama GEMM count, and run it in PR CI

  disable_distributeddisable_e2e
  #6126·RUIJIEZHONG66166 opened Nov 10, 2026Contributor·

  ·
  1

  1

• [xccl] Make imported peer memory resident in XPUSymmetricMemory rendezvous

  disable_e2edisable_win
  #6124·pkourdis opened Nov 9, 2026Contributor·

  ·
  6

• Fix feature stride in sparse coalesce_values_kernel

  disable_distributeddisable_e2e
  #6120·jkosnox opened Nov 9, 2026Contributor·

  ·
  1

  5

• Materialize intermediate f16 division result before truncation.

  disable_distributeddisable_e2e
  #6119·AKloniecki opened Nov 9, 2026Contributor·

  ·
  1

• Optimize conflicting-layout binary ops with SLM-assisted transposition

  #6115·lchen2331 opened Nov 9, 2026Contributor·

  ·

• Optimize heuristic of sort

  disable_distributed
  #6112·orrangetabby17 opened Nov 9, 2026Contributor·

  ·
  2

• [WIP] [XPU] Optimize cat op performance for sub-dword aligned shapes

  disable_distributeddisable_e2e
  #6111·KepingYan opened Nov 9, 2026Contributor·

  ·

• [CI] Pin torch-xpu-ops to one commit for nightly build and test

  disable_distributeddisable_e2e
  #6107·RUIJIEZHONG66166 opened Nov 9, 2026Contributor·

  ·
  1

• Fix dtype promotion in XPU max/min elementwise kernel (#5900)

  disable_distributeddisable_e2e
  #6099·xuhancn opened Nov 8, 2026Contributor·

  ·
  1

• Make test_cuda_memory_leak_detection's expected message device-aware (#4405)

  disable_builddisable_distributeddisable_e2edisable_win
  #6098·xuhancn opened Nov 8, 2026Contributor·

  ·
  1

• Vectorize mixed-dtype elementwise ops with compile-time typed loads

  #6097·lchen2331 opened Nov 8, 2026Contributor·

  ·