Properly register parameter for binary folding test (#128356)

This PR properly registers the tensor used in the module compute as a parameter. This bug was hidden previously because all tensors on the nn modules would be considered constant by dynamo, with inlining NN modules, this is no longer the case.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/128356
Approved by: https://github.com/anijain2305
ghstack dependencies: #128355

test/inductor/test_binary_folding.py

@@ -56,7 +56,7 @@ class BinaryFoldingTemplate(TestCase):
                     self.use_scalar = scalar
                     tensor_size = [1 for _ in range(self.conv.weight.ndim)]
                     tensor_size[1] = self.conv.weight.size(0)
-                    self.tensor = (
+                    self.tensor = torch.nn.Parameter(
                         add_tensor
                         if add_tensor is not None
                         else torch.rand(tensor_size).to(device)
@@ -136,7 +136,11 @@ class BinaryFoldingTemplate(TestCase):
                 nn.Conv2d,
                 pytorch_op,
                 False,
-                add_tensor=torch.rand(32, 1, 32).to(self.device),
+                add_tensor=torch.rand(
+                    32,
+                    1,
+                    32,
+                ).to(self.device),
                 expect_success=False,
             )
 
@@ -156,7 +160,7 @@ class BinaryFoldingTemplate(TestCase):
                 nn.Conv2d,
                 pytorch_op,
                 False,
-                add_tensor=torch.tensor([2]).to(torch.int).to(self.device),
+                add_tensor=torch.tensor([2]).to(torch.float64).to(self.device),
                 expect_success=False,
             )