[Window][LNL][Nightly]UT failures AssertionError: Scalars are not equal!

[kaileiyx]

🐛 Describe the bug

Some failed cases about AssertionError: Tensor-likes are not equal!
'''
test_reductions_xpu.py::TestReductionsXPU::test_reduction_vectorize_along_output_xpu_bfloat16
test_reductions_xpu.py::TestReductionsXPU::test_reduction_vectorize_along_output_xpu_float16
test_reductions_xpu.py::TestReductionsXPU::test_reduction_vectorize_along_output_xpu_float32
test_reductions_xpu.py::TestReductionsXPU::test_reduction_vectorize_along_output_xpu_float64
'''

'''
PYTORCH_TEST_WITH_SLOW=1 python ....\test\test_reductions.py TestReductionsXPU.test_reduction_vectorize_along_output_xpu_bfloat16
'''

'''
____ TestReductionsXPU.test_reduction_vectorize_along_output_xpu_bfloat16 _____
[gw0] win32 -- Python 3.10.14 C:\Users\dcai01\miniforge3\envs\xpu-test\python.exe
Traceback (most recent call last):
File "C:\pytorch\third_party\torch-xpu-ops\test\xpu../../../../test\test_reductions.py", line 2465, in test_reduction_vectorize_along_output
run_test(torch.zeros(64, 64, dtype=dtype, device=device))
File "C:\pytorch\third_party\torch-xpu-ops\test\xpu../../../../test\test_reductions.py", line 2462, in run_test
self.assertEqual(output1[i], i)
File "C:\Users\dcai01\miniforge3\envs\xpu-test\lib\site-packages\torch\testing_internal\common_utils.py", line 4255, in assertEqual
raise error_metas.pop()[0].to_error( # type: ignore[index]
AssertionError: Scalars are not equal!

'''

Versions

Environment:
PyTorch Commit: 2.10.0.dev20251101+xpu / f291f529fdd7fe9b617ece2fa373ca684a11fd7f
XPU Commit: 8d373ba
OS: Microsoft Windows 11 Pro
CPU: Intel(R) Core(TM) Ultra 7 258V
GPU and Driver: Intel(R) Arc(TM) 140V GPU (16GB) 32.0.101.7026
oneAPI: intel.oneapi.win.deep-learning-essentials.product=2025.0.1+20