[fx] fix qualified name for methods of torch.Tensor (#162407)

This fixes an error in the previous PR.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/162407
Approved by: https://github.com/ezyang, https://github.com/XuehaiPan

test/inductor/test_torchinductor.py

@@ -2998,7 +2998,7 @@ class CommonTemplate:
         x = torch.rand(10)
 
         with x.device:
-            out = torch.compile(fn)(x)
+            out = torch.compile(fn, backend=lambda gm, _: gm)(x)
             ref = fn(x)
             for a, b in zip(out, ref):
                 self.assertTrue(torch.allclose(a, b))

torch/fx/node.py

@@ -155,7 +155,7 @@ def _get_qualified_name(func: Callable[..., Any]) -> str:
         isinstance(func, (types.MethodDescriptorType, types.WrapperDescriptorType))
         and func is getattr(torch.Tensor, func.__name__, None)
     ) or (
-        func.__module__ == torch._tensor
+        func.__module__ == torch._tensor.__name__
         and func.__qualname__ == f"Tensor.{func.__name__}"
     ):
         return f"torch.Tensor.{func.__name__}"