RelayModelInfo: support non-tensor outputs

Thanks to @jokap11

mlonmcu/flow/tvm/backend/model_info.py

@@ -143,7 +143,7 @@ def parse_relay_main(line):
     output_tensors_str = re.compile(r"-> (.+) {").findall(line)
     # The following depends on InferType annocations
     if len(output_tensors_str) > 0:
-        output_tensor_strs = re.compile(r"Tensor\[\([\di]+(?:, [\di]+)*\), [a-zA-Z0-9_]+\]").findall(
+        output_tensor_strs = re.compile(r"Tensor\[\([\di]+(?:, [\di]+)*\), [a-zA-Z0-9_]+\]|(?:u?int\d+)").findall(
             output_tensors_str[0]
         )
 
@@ -156,10 +156,15 @@ def parse_relay_main(line):
 
         for i, output_name in enumerate(output_tensor_names):
             res = re.compile(r"Tensor\[\(([\di]+(?:, [\di]+)*)\), ([a-zA-Z0-9_]+)\]").match(output_tensor_strs[i])
+            if res is None:
+                res = re.compile(r"(u?int\d+)").match(output_tensor_strs[i])
             assert res is not None
             groups = res.groups()
-            assert len(groups) == 2
-            output_shape_str, output_type = groups
+            assert len(groups) in [1, 2]
+            if len(groups) == 2:
+                output_shape_str, output_type = groups
+            elif len(groups) == 1:
+                output_shape_str, output_type = "1, 1", groups[0]
             output_shape = shape_from_str(output_shape_str)
             output_tensor = TensorInfo(output_name, output_shape, output_type)
             output_tensors.append(output_tensor)