Handle f([]) vs. f() in fake tensor caching (#162284)

Fixes https://github.com/pytorch/pytorch/issues/162279
Pull Request resolved: https://github.com/pytorch/pytorch/pull/162284
Approved by: https://github.com/manuelcandales, https://github.com/aorenste

test/test_fake_tensor.py

@@ -1943,6 +1943,16 @@ class FakeTensorDispatchCache(TestCase):
             self._test_cache_key(fm, 1.0, 1.0, 1)
             self._test_cache_key(fm, 0.0, 0.0, 0)
 
+    def test_empty_list(self):
+        with FakeTensorMode() as fm:
+            func = aten.any.dims
+            state = _CacheKeyState()
+            x = torch.ones((2, 3))
+            key_x = fm._cache_key(state, func, [x, []], {})
+            key_y = fm._cache_key(state, func, [x], {})
+
+        self.assertNotEqual(key_x, key_y)
+
     def assertHitsMisses(self, hits, misses):
         """
         Helper to assert on the number of recorded hits and misses.

torch/_subclasses/fake_tensor.py

@@ -1677,6 +1677,10 @@ class FakeTensorMode(TorchDispatchMode):
         )
         from torch._higher_order_ops.utils import FunctionalizeCtxWrapper
 
+        if isinstance(args, (list, tuple, dict)):
+            result.append(type(args))
+            result.append(f"length_{len(args)}")
+
         if isinstance(args, dict):
             self._prep_args_for_hash(result, args.keys(), state, id_hashed_objects)
             self._prep_args_for_hash(result, args.values(), state, id_hashed_objects)