Extend assert statement to include ListVariable

test/dynamo/test_repros.py

@@ -3168,6 +3168,29 @@ def check_type(obj, types_or_checks):
         res = opt_check_type(torch.randn(4), [torch.Tensor])
         self.assertEqual(ref, res)
 
+    def test_kwargs_out_list_variable(self):
+        class Repro(torch.nn.Module):
+            def __init__(self):
+                super().__init__()
+
+            def forward(self, param):
+                z = torch.frexp(**param)
+                return z
+
+        model = Repro()
+        params = {"input": torch.tensor([[0.0, 1, 2, 4]])}
+        params["out"] = [
+            torch.empty(0, dtype=torch.float32),  # mantissa
+            torch.empty(0, dtype=torch.int32),  # exponent
+        ]
+
+        model = torch.compile(model, backend="eager")
+        mantissa, exponent = model(params)
+        ref_mantissa = torch.tensor([[0.0000, 0.5000, 0.5000, 0.5000]])
+        ref_exponent = torch.tensor([[0, 1, 2, 3]], dtype=torch.int32)
+        self.assertEqual(ref_mantissa, mantissa)
+        self.assertEqual(ref_exponent, exponent)
+
 
 if __name__ == "__main__":
     from torch._dynamo.test_case import run_tests

torch/_dynamo/variables/torch.py

@@ -640,7 +640,7 @@ def check_any_unspec(x):
                 # torch.sigmoid mutate the tensors in the out field. Track such
                 # tensors and rewrite the symbolic locals.
                 if isinstance(tensor_variable, TupleVariable):
-                    assert isinstance(kwargs["out"], TupleVariable)
+                    assert isinstance(kwargs["out"], (TupleVariable, ListVariable))
                     output_tensor_names = [
                         tx.find_symbolic_locals_name(x) for x in kwargs["out"].items
                     ]