[dynamo] Graph break when faking named tensors (#120779)

Fixes #120644
Pull Request resolved: https://github.com/pytorch/pytorch/pull/120779
Approved by: https://github.com/zou3519

test/inductor/test_torchinductor.py

@@ -4766,7 +4766,7 @@ class CommonTemplate:
                     dtype=torch.float32,
                     device=a.device,
                 ),
-                torch.zeros(2, 3, names=None),
+                torch.zeros(2, 3),
                 a + torch.ones(8, device=a.device),
                 torch.full((2, 3), 3.1416, device=a.device),
             )

test/test_namedtensor.py

@@ -686,10 +686,12 @@ class TestNamedTensor(TestCase):
 
             self.assertEqual(op(a, a).names, ('N', 'C'))
             self.assertEqual(op(a, c).names, ('N', 'C'))
-
-            with self.assertRaisesRegex(RuntimeError, "do not match"):
+            # TODO: dynamo will throw a slightly different
+            # error message because it's adding fake tensors
+            # `must match the size of` portion is the dynamo error
+            with self.assertRaisesRegex(RuntimeError, "do not match|must match the size of"):
                 op(a, d)
-            with self.assertRaisesRegex(RuntimeError, "do not match"):
+            with self.assertRaisesRegex(RuntimeError, "do not match|must match the size of"):
                 op(a, b)
 
         def test_wildcard(op):

torch/_subclasses/fake_impls.py

@@ -157,6 +157,11 @@ def constructors(fake_mode, func, *args, **kwargs):
     _, new_kwargs = normalize_function(
         func, args=args, kwargs=kwargs, normalize_to_only_use_kwargs=True
     )
+    if "names" in kwargs:
+        raise UnsupportedOperatorException(
+            "torch.compile doesn't support named tensors"
+        )
+
     if func in _like_tensor_constructors:
         default_device = new_kwargs["input"].device
         # TODO: file issue

torch/_subclasses/fake_tensor.py

@@ -448,6 +448,13 @@ class FakeTensor(torch.Tensor):
     # that have dispatch keys which are higher than the "meta" key:
     # https://github.com/pytorch/pytorch/blob/main/c10/core/DispatchKey.h#L189
 
+    # We don't support named tensors; graph break
+    @property
+    def names(self):
+        raise UnsupportedFakeTensorException(
+            "torch.compile doesn't support named tensors"
+        )
+
     @staticmethod
     def __new__(cls, fake_mode, elem, device, constant=None):
         self = torch.Tensor._make_subclass(