[Windows][ARC][UT] test_weight_norm.py::TestNNMethod::test_weight_norm_different_type failed with error AssertionError: Tensor-likes are not close!

[huaiyuzh]

[Windows][ARC][UT] tests/gpu/examples/test_weight_norm.py::TestNNMethod::test_weight_norm_different_type failed with error AssertionError: Tensor-likes are not close!

[Error Details]

================= Session test_weight_norm.py =================
test_weight_norm.py::TestNNMethod::test_weight_norm_different_type, FAILED, E AssertionError: Tensor-likes are not close!, 1.980

self = <test_weight_norm.TestNNMethod testMethod=test_weight_norm_different_type>

def test_weight_norm_different_type(self):
    v = torch.randn(8193  8193).requires_grad_(True)
    g = torch.randn(8193).to(torch.float).requires_grad_(True)
    gw = torch.randn(8193  8193)
    w  n = torch._weight_norm_interface(v  g  dim=0)
    w.backward(gw)
    v_xpu = v.detach().clone().to("xpu").requires_grad_(True)
    g_xpu = g.detach().clone().to("xpu").requires_grad_(True)
    w_xpu  n_xpu = torch._weight_norm_interface(v_xpu  g_xpu  dim=0)
    w_xpu.backward(gw.to("xpu"))
    self.assertEqual(w  w_xpu.cpu()  atol=1e-3  rtol=1e-5)
    self.assertEqual(n  n_xpu.cpu()  atol=1e-1  rtol=1e-5)
    self.assertEqual(v.grad  v_xpu.grad.cpu()  atol=1e-3  rtol=1e-5)

  self.assertEqual(g.grad  g_xpu.grad.cpu()  atol=1e-3  rtol=1e-5)

E AssertionError: Tensor-likes are not close!
E
E Mismatched elements: 1 / 8193 (0.0%)
E Greatest absolute difference: 0.5634427070617676 at index (3813 ) (up to 0.001 allowed)
E Greatest relative difference: 2.646775960922241 at index (3813 ) (up to 1e-05 allowed)
E
E To execute this test run the following from the base repo dir:
E python test_weight_norm.py TestNNMethod.test_weight_norm_different_type
E
E This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0

test_weight_norm.py:204: AssertionError