Skip to content

[release/2.13] Expected error message is different than actual #2817

New issue
New issue
Open
Bug
pytorch/pytorch
#197247
Open
[release/2.13] Expected error message is different than actual#2817
Bug
pytorch/pytorch
#197247
Assignees
kdrozd-dev
Labels
functionalityFunctional defect: error/assert/hang/timeout (non-numeric)module: opsOther ATen/native ops: autograd, optimizer, fx, exportskippedUsed for temp UT failure to parallel fixtest: utUnit-test failure surfacewait_upstreamNeeds upstream tests adjustment
Milestone
UT Backlog

Description

@Silv3S
Silv3S
opened on Feb 2, 2026

🐛 Describe the bug

Cases:
op_ut,third_party.torch-xpu-ops.test.xpu.test_linalg_xpu.TestLinalgXPU,test_linalg_lstsq_input_checks_xpu_complex128
op_ut,third_party.torch-xpu-ops.test.xpu.test_linalg_xpu.TestLinalgXPU,test_linalg_lstsq_input_checks_xpu_complex64
op_ut,third_party.torch-xpu-ops.test.xpu.test_linalg_xpu.TestLinalgXPU,test_linalg_lstsq_input_checks_xpu_float32
op_ut,third_party.torch-xpu-ops.test.xpu.test_linalg_xpu.TestLinalgXPU,test_linalg_lstsq_input_checks_xpu_float64

E           AssertionError: "`driver` other than `gels` is not supported on CUDA" does not match "torch.linalg.lstsq: parameter `driver` should be one of (gels, gelsy, gelsd, gelss)"
E           
E           To execute this test, run the following from the base repo dir:
E               python test/test_linalg.py TestLinalgXPU.test_linalg_lstsq_input_checks_xpu_complex128
E           
E           This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0

Versions

PyTorch2.10
7e412e0

Activity

  1. Silv3S
    added a parent issue on Feb 2, 2026
  2. Silv3S
    added
    bug_fix_stage5
    on Feb 2, 2026
  3. kdrozd-dev
    self-assigned this
    on Feb 3, 2026
  4. Silv3S
    added
    skippedUsed for temp UT failure to parallel fix
    on Mar 20, 2026
  5. github-actions
    added this to the UT Backlog milestone on Mar 20, 2026
  6. Silv3S
    removed a parent issue on Mar 20, 2026
  7. Silv3S
    marked RuntimeError: torch.linalg.lstsq: parameter driver should be one of (gels, gelsy, gelsd, gelss) #3696 as a duplicate of this issue on May 18, 2026

  8. bjarzemb commented on Jun 17, 2026

    @bjarzemb
    bjarzemb
    on Jun 17, 2026
    Contributor

    I have this issue also on Windows with Intel XPU (Arc B580) in release 2.13 tests.

  9. bjarzemb
    changed the title [-]Expected error message is different than actual[/-] [+][release/2.13] Expected error message is different than actual[/+] on Jun 17, 2026

  10. kdrozd-dev commented on Jun 17, 2026

    @kdrozd-dev
    kdrozd-dev
    on Jun 17, 2026
    Contributor

    I have this issue also on Windows with Intel XPU (Arc B580) in release 2.13 tests.

    Fix is ready we are waiting for community review.

  11. bjarzemb commented on Jun 24, 2026

    @bjarzemb
    bjarzemb
    on Jun 24, 2026
    Contributor

    I have this issue also on LNL (windows) in release 2.13 tests.

  12. daisyden
    removed
    bug_fix_stage5
    on Aug 28, 2026
  13. kdrozd-dev
    added
    wait_upstreamNeeds upstream tests adjustment
    on Sep 30, 2026
  14. kdrozd-dev
    linked a pull request that will close this issue[4/N][Test][XPU] linalg: enable TestLinalgDevice on XPU #197247on Oct 1, 2026
  15. PatrykWilczewski
    marked [PTL][BMG][WSL] test_linalg_lstsq_input_checks_xpu_* expects the CUDA-only driver error, but XPU lstsq runs on CPU fallback and accepts every driver #5838 as a duplicate of this issue on Oct 28, 2026
  16. daisyden
    added
    module: opsOther ATen/native ops: autograd, optimizer, fx, export
    functionalityFunctional defect: error/assert/hang/timeout (non-numeric)
    test: utUnit-test failure surface
    on Nov 11, 2026
  17. daisyden
    added the
    Bug
    issue type on Nov 11, 2026

  18. daisyden commented on Nov 11, 2026

    @daisyden
    daisyden
    on Nov 11, 2026
    Contributor
    label-issue: #2817

    Root cause: the upstream test expects the CUDA-only driver error raised at aten/src/ATen/native/BatchLinearAlgebra.cpp:3538 ("driver other than gels is not supported on CUDA"), but XPU lstsq takes the CPU-fallback path which accepts all drivers and raises the generic "should be one of (gels, gelsy, gelsd, gelss)" message. Error-message expectation mismatch, not incorrect computation; defect owner is the upstream test (wait_upstream).

    Analyzed case: torch-xpu-ops/test/xpu/test_linalg_xpu.py::TestLinalgXPU::test_linalg_lstsq_input_checks_xpu_complex128 (case 1 of 4; the other 3 not analyzed)

    axis value reason
    type Bug Reported UT failure (assertion); issue_type empty, Bug heuristic.
    test test: ut op_ut rows for test/xpu/test_linalg_xpu.py.
    module module: ops torch.linalg.lstsq input-check; ATen native linalg op.
    priority Low Error-message/validation difference vs CUDA without incorrect computation.
    symptom functionality AssertionError on expected vs actual error text.

    Test cases (4):

    • torch-xpu-ops/test/xpu/test_linalg_xpu.py::TestLinalgXPU::test_linalg_lstsq_input_checks_xpu_complex128
    • torch-xpu-ops/test/xpu/test_linalg_xpu.py::TestLinalgXPU::test_linalg_lstsq_input_checks_xpu_complex64
    • torch-xpu-ops/test/xpu/test_linalg_xpu.py::TestLinalgXPU::test_linalg_lstsq_input_checks_xpu_float32
    • torch-xpu-ops/test/xpu/test_linalg_xpu.py::TestLinalgXPU::test_linalg_lstsq_input_checks_xpu_float64

    Applied to this issue. Generated by a local label-issue run.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

functionalityFunctional defect: error/assert/hang/timeout (non-numeric)module: opsOther ATen/native ops: autograd, optimizer, fx, exportskippedUsed for temp UT failure to parallel fixtest: utUnit-test failure surfacewait_upstreamNeeds upstream tests adjustment

Type

Bug

Projects

No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions