Feedback about Quickstart

[BhavneetSingh7]

There is the following issue on this page: https://docs.pytorch.org/tutorials/beginner/basics/quickstart_tutorial.html#optimizing-the-model-parameters

System specs: Windows 11, python3.11, pytorch==2.8.0+xpu, Intel oneAPI 2025.2.

Been following this tut, I got this error raising from test function

correct += (pred.argmax(1) == y).type(torch.float).sum().item()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: UR error

Checked all the compatibility of oneAPI, pytorch, and intel_extension_for_pytorch

print(torch.xpu._is_compiled())
print(torch.xpu.is_available())

Both prints True

Really new to ML and NN, but not dev, so tried using torch.FloatTensor
correct += (pred.argmax(1) == y).type(torch.FloatTensor).sum().item()
It works and output almost matches to what's given in tut.

I hope what I did is correct in terms of ML.
If not please suggest where can I lookup to understand this better.

cc @albanD @jbschlosser @gujinghui @EikanWang @fengyuan14 @guangyey

[guangyey]

Hi @BhavneetSingh7 I can run it on Arc770 successfully. May I know what's the output of torch.xpu.get_device_properties() on your machine?
In additional, intel_extension_for_pytorch is unnecessary now. I recommend you to unisntall it and try again.

[BhavneetSingh7]

$ python -c "import torch; print(torch.xpu.get_device_properties())"

_XpuDeviceProperties(name='Intel(R) Iris(R) Xe Graphics', platform_name='Intel(R) oneAPI Unified Runtime over Level-Zero', type='gpu', driver_version='1.6.33511', total_memory=14802MB, max_compute_units=96, gpu_eu_count=96, gpu_subslice_count=12, max_work_group_size=512, max_num_sub_groups=64, sub_group_sizes=[8 16 32], has_fp16=1, has_fp64=0, has_atomic64=1)

[guangyey]

Thanks @BhavneetSingh7 I could reproduce it on the old iGPU. I would try to investigate what the root cause is. Please use the workaround temporarily.