Issues

• [upstream ut] AssertionError: Tensor-likes are not equal! in test/test_sparse.py

  intel/torch-xpu-ops#5669 · Filter by author zxd1997066 opened on Oct 11, 2026

• [XPU] Track complex dtype support for arange

  intel/torch-xpu-ops#5668 · Filter by author xuhancn opened on Oct 10, 2026

• [upstream_ut] test_equal_xpu: torch.equal returns True for tensor containing NaN compared to itself

  intel/torch-xpu-ops#5663 · Filter by author daisyden opened on Oct 10, 2026

• [upstream_ut] test_bmm_matmul_mixed_dtype_error_xpu: mixed fp16/fp32 bmm/matmul does not raise

  intel/torch-xpu-ops#5662 · Filter by author daisyden opened on Oct 10, 2026

• test_split_with_sizes_copy_out_xpu crashes with Fatal Python error: Floating point exception (SIGFPE)

  intel/torch-xpu-ops#5661 · Filter by author daisyden opened on Oct 10, 2026

• [upstream_ut] test_cdist_backward_large_index GPU segfault: int32 overflow in XPU cdist backward buffer offset

  intel/torch-xpu-ops#5660 · Filter by author daisyden opened on Oct 10, 2026

• Correct normalization of Dirichlet distribution

  intel/torch-xpu-ops#5656 · Filter by author mwiktor-intel opened on Oct 9, 2026
  

• [CI] Windows Nightly CI silently doesn't launch almost any basic test.

  Bug

  intel/torch-xpu-ops#5654 · Filter by author BBBela opened on Oct 9, 2026
  

• UT New failures AssertionError: Tensor-likes are not equal in test_reductions_xpu.py

  skipped_bmg
  intel/torch-xpu-ops#5647 · Filter by author kaileiyx opened on Oct 9, 2026

• [v.2.15.0] Release Tracker

  intel/torch-xpu-ops#5645 · Filter by author chuanqi129 opened on Oct 9, 2026 · PT2.15
  2

• [TorchAO][BMG] Llama-3.1-8B MXFP4 GSM8K accuracy lower than MXFP8 and INT4 RTN

  module: ao
  intel/torch-xpu-ops#5644 · Filter by author LifengWang opened on Oct 9, 2026
  

• Fused SDPA (OVERRIDEABLE / oneDNN) writes past its output buffer when q_len == 16 (head_dim 64, bf16, no mask)

  intel/torch-xpu-ops#5643 · Filter by author John-Giskhan opened on Oct 9, 2026
  1

• XPU lacks some memory stats in dict returned by torch.xpu.memory_stats

  test: utut_upstream
  intel/torch-xpu-ops#5642 · Filter by author BBBela opened on Oct 8, 2026
  

• DLPack stream exchange is not implemented on XPU

  intel/torch-xpu-ops#5641 · Filter by author tszulist-hbn opened on Oct 8, 2026

• [XPU] convolution_backward returns wrong gradients when grad_output is contiguous but has channels-last-looking strides

  intel/torch-xpu-ops#5640 · Filter by author tszulist-hbn opened on Oct 8, 2026

• Document and test integrity of test infra setup for non-CI hardware

  intel/torch-xpu-ops#5639 · Filter by author dvrogozh opened on Oct 8, 2026

• [release/2.15] test_profiler.py::TestProfiler::test_source_multithreaded AssertionError

  hw: BMGhw: LNLmodule: profileros: Windowsos: WSL
  intel/torch-xpu-ops#5638 · Filter by author bjarzemb opened on Oct 8, 2026

• [release/2.15] [BMG] [wsl2] test_cpp_thread.py RuntimeError: Error building extension 'profiler_test_cpp_thread_lib'

  hw: BMGmodule: profileros: WSLtest: ut
  Bug

  intel/torch-xpu-ops#5637 · Filter by author bjarzemb opened on Oct 8, 2026

• [release/2.15] e2e perf regressions on windows introduced in PyTorch nightly 20260924

  os: Windowsperformance
  Bug

  intel/torch-xpu-ops#5635 · Filter by author bjarzemb opened on Oct 8, 2026

• [release/2.15] e2e eager perf regressions on windows introduced in PyTorch nightly 20260909

  os: Windowsperformance
  Bug

  intel/torch-xpu-ops#5633 · Filter by author bjarzemb opened on Oct 8, 2026

• [upstream ut] ValueError: trying to initialize the default process group twice! in test/dynamo/test_logging.py

  intel/torch-xpu-ops#5630 · Filter by author zxd1997066 opened on Oct 8, 2026

• [xpu-alignment] [26-10-07] [26-10-07] XPU rrelu_with_noise drops every write into a non-contiguous noise tensor after the CUDA copy-back landed

  ai_generated
  intel/torch-xpu-ops#5628 · Filter by author torchxpubot opened on Oct 8, 2026

• UT failures 2026-10-7

  regressionskipped_bmg
  intel/torch-xpu-ops#5627 · Filter by author kaileiyx opened on Oct 8, 2026
  1

• RuntimeError: Cannot set preferred backend to cuSOLVER if PyTorch has not been compiled with cuSOLVER in test_linalg_xpu.py

  regressionskipped_bmg
  intel/torch-xpu-ops#5626 · Filter by author kaileiyx opened on Oct 8, 2026

• test_manual_ac_xpu - AssertionError: Scalars are not close!

  intel/torch-xpu-ops#5624 · Filter by author mtruszcx opened on Oct 7, 2026