Keep AIMET custom modules as a single op in connected graph

Signed-off-by: Kevin Hsieh <quic_klhsieh@quicinc.com>

TrainingExtensions/torch/src/python/aimet_torch/meta/connectedgraph.py

@@ -44,6 +44,7 @@
 
 import copy
 from collections import defaultdict
+import inspect
 from typing import Tuple, Union, List, Dict, Type, Optional
 import torch
 
@@ -137,6 +138,7 @@ class ConnectedGraph(AimetCommonConnectedGraph):
         either module or functional) as producers and consumers of tensors.
         Note that the graph has two kinds of nodes: operations and products."""
 
+    # pylint: disable=too-many-instance-attributes
     def __init__(self, model: torch.nn.Module, model_input: Union[torch.Tensor, Tuple]):
         """
         Init function for connected graph.
@@ -157,7 +159,12 @@ def __init__(self, model: torch.nn.Module, model_input: Union[torch.Tensor, Tupl
 
         self._generate_module_lookup_table(model)
         with in_eval_mode(model), torch.no_grad():
+            self._aimet_defined_modules = \
+                tuple(classtype for _, classtype in inspect.getmembers(aimet_modules,
+                                                                       lambda m: inspect.isclass(m) and issubclass(m,
+                                                                                                                   torch.nn.Module)))
             self._construct_graph(model, model_input)
+            del self._aimet_defined_modules
 
         # List of ops in the order they are traversed using the forward function
         self.ordered_ops = self._get_ordered_ops()
@@ -1242,7 +1249,7 @@ def _is_recursive_parsing_needed(self, module: torch.nn.Module,
         recursive_parsing_needed = True
         if is_torch_nn_leaf_module(module) or \
                 is_custom_leaf_module(module, self._find_aten_nodes_in_forward_pass(trace)) or \
-                isinstance(module, tuple(aimet_torch.utils.modules_to_treat_as_leaf)):
+                isinstance(module, (self._aimet_defined_modules, tuple(aimet_torch.utils.modules_to_treat_as_leaf))):
             recursive_parsing_needed = False
 
         return recursive_parsing_needed

TrainingExtensions/torch/test/python/test_connectedgraph.py

@@ -1015,3 +1015,21 @@ def __init__(self):
         assert len(add_1.inputs) == 2
         assert add_1.inputs == [p3, p4]
         assert p5.name not in mcg._products.keys()
+
+    def test_custom_ops_to_treat_as_leaf_module(self):
+        class RMSNormModel(torch.nn.Module):
+            def __init__(self):
+                super(RMSNormModel, self).__init__()
+                self.rms_norm_0 = aimet_modules.RmsNorm([5, 2, 3], [2], 1e-5)
+
+            def forward(self, inp):
+                x = self.rms_norm_0(inp)
+                return x
+
+        model = RMSNormModel()
+        dummy_input = torch.randn(5, 2, 3)
+
+        cg = ConnectedGraph(model, dummy_input)
+        assert len(cg.get_all_ops()) == 1
+        assert cg.ordered_ops[0].inputs[0].is_model_input
+        assert cg.ordered_ops[0].get_module() == model.rms_norm_0