Update emformer tests to avoid 0/1 specialization issue (pytorch#18850)

Summary: The tests break in some dynamo configurations. Set min batch to
2 to be exportable in all.

Reviewed By: abhinaykukkadapu

Differential Revision: D100391669

Author: GregoryComer

backends/test/harness/tester.py

@@ -131,6 +131,9 @@ def generate_random_inputs(self):
                 assert isinstance(self.example_inputs[arg_idx], torch.Tensor)
                 ex_shape = list(self.example_inputs[arg_idx].shape)
                 dynamic_dim_spec = self.dynamic_shapes[arg_idx]
+                if dynamic_dim_spec is None or dynamic_dim_spec == {}:
+                    input_shapes.append(torch.Size(ex_shape))
+                    continue
                 for dim_idx, dim_spec in dynamic_dim_spec.items():
                     assert dim_idx < len(ex_shape)
                     if isinstance(dim_spec, torch.export.dynamic_shapes._DerivedDim):

backends/xnnpack/test/models/emformer_rnnt.py

@@ -50,14 +50,20 @@ def test_fp32_emformer_joiner(self):
 
     def test_fp32_emformer_joiner_dynamic(self):
         joiner = self.Joiner()
+        example_inputs = (
+            torch.rand([2, 128, 1024]),
+            torch.tensor([128]),
+            torch.rand([2, 128, 1024]),
+            torch.tensor([128]),
+        )
         dynamic_shapes = (
             {0: torch.export.Dim("batch", min=1, max=4)},
             None,
             {0: torch.export.Dim("batch", min=1, max=4)},
             None,
         )
         (
-            Tester(joiner, joiner.get_example_inputs(), dynamic_shapes=dynamic_shapes)
+            Tester(joiner, example_inputs, dynamic_shapes=dynamic_shapes)
             .export()
             .to_edge_transform_and_lower()
             .check(["torch.ops.higher_order.executorch_call_delegate"])
@@ -117,14 +123,18 @@ def test_fp32_emformer_transcriber(self):
 
     def test_fp32_emformer_transcriber_dynamic(self):
         transcriber = self.Transcriber()
+        example_inputs = (
+            torch.randn(2, 128, 80),
+            torch.tensor([128]),
+        )
         dynamic_shapes = (
             {0: torch.export.Dim("batch", min=1, max=4)},
             None,
         )
         (
             Tester(
                 transcriber,
-                transcriber.get_example_inputs(),
+                example_inputs,
                 dynamic_shapes=dynamic_shapes,
             )
             .export()