Fix BN layers with parameters as Constant nodes

Signed-off-by: Michael Tuttle <quic_mtuttle@quicinc.com>

TrainingExtensions/onnx/src/python/aimet_onnx/batch_norm_fold.py

@@ -442,24 +442,20 @@ def _update_standalone_batchnorm_ops(model: ModelProto):
     Update weight and bias of standalone batchnorm ops in the model.
     :param model: onnx Model for which batchnorm parameters are to be updated.
     """
-    initalizer_dict = {initializer.name: idx for idx, initializer in enumerate(model.graph.initializer)}
 
     for node in model.graph.node:
         if node.op_type in BatchNormType:
 
             # get parameter names and indices
             weight_name, bias_name, running_mean_name, running_var_name = node.input[1:]
-            idx_w, idx_b = initalizer_dict[weight_name], initalizer_dict[bias_name]
-            idx_rm, idx_rv = initalizer_dict[running_mean_name], initalizer_dict[running_var_name]
-
-            init_w = model.graph.initializer[idx_w]
-            init_b = model.graph.initializer[idx_b]
-            init_rm = model.graph.initializer[idx_rm]
-            init_rv = model.graph.initializer[idx_rv]
-            attr = get_node_attribute(node, "epsilon")
-            if attr is None:
+            init_w, init_b, init_rm, init_rv = [ParamUtils.get_param(model, node, idx) for idx in range(1, 5)]
+
+            attr = [item for item in node.attribute if item.name == "epsilon"]
+            if not attr:
                 attr = onnx.helper.make_attribute("epsilon", 1e-5) # Default epsilon value
                 node.attribute.append(attr)
+            else:
+                attr = attr[0]
 
             epsilon = attr.f
             tensor_w = numpy_helper.to_array(init_w)
@@ -475,13 +471,13 @@ def _update_standalone_batchnorm_ops(model: ModelProto):
             tensor_rv = np.ones(tensor_w.shape, tensor_w.dtype)
             attr.f = 0.
 
-            init_w = numpy_helper.from_array(tensor_w, weight_name)
-            init_b = numpy_helper.from_array(tensor_b, bias_name)
-            init_rm = numpy_helper.from_array(tensor_rm, running_mean_name)
-            init_rv = numpy_helper.from_array(tensor_rv, running_var_name)
+            init_w_ = numpy_helper.from_array(tensor_w, weight_name)
+            init_b_ = numpy_helper.from_array(tensor_b, bias_name)
+            init_rm_ = numpy_helper.from_array(tensor_rm, running_mean_name)
+            init_rv_ = numpy_helper.from_array(tensor_rv, running_var_name)
 
             # update initializers
-            model.graph.initializer[idx_w].CopyFrom(init_w)
-            model.graph.initializer[idx_b].CopyFrom(init_b)
-            model.graph.initializer[idx_rm].CopyFrom(init_rm)
-            model.graph.initializer[idx_rv].CopyFrom(init_rv)
+            init_w.CopyFrom(init_w_)
+            init_b.CopyFrom(init_b_)
+            init_rm.CopyFrom(init_rm_)
+            init_rv.CopyFrom(init_rv_)

TrainingExtensions/onnx/test/python/models/models_for_tests.py

@@ -2515,3 +2515,31 @@ def batchnorm_model():
     onnx.checker.check_model(model, True)
     return model
 
+def batchnorm_model_constants():
+    model = helper.make_model(
+        graph=helper.make_graph(
+            name='BatchnormModel',
+            inputs=[helper.make_tensor_value_info('model_input', TensorProto.FLOAT, shape=[10, 10, 8, 8])],
+            outputs=[helper.make_tensor_value_info('model_output', TensorProto.FLOAT, shape=[10, 10, 8, 8])],
+            initializer=[],
+            nodes=[
+                helper.make_node('Constant', inputs=[], outputs=["batchnorm.weight"],
+                                 value=numpy_helper.from_array(np.abs(np.random.randn(10, )).astype('float32')), name='weight'),
+                helper.make_node('Constant', inputs=[], outputs=["batchnorm.bias"],
+                                 value=numpy_helper.from_array(np.random.randn(10, ).astype('float32')), name='bias'),
+                helper.make_node('Constant', inputs=[], outputs=["batchnorm.input_mean"],
+                                 value=numpy_helper.from_array(np.random.randn(10, ).astype('float32')), name="input_mean"),
+                helper.make_node('Constant', inputs=[], outputs=["batchnorm.input_var"],
+                                 value=numpy_helper.from_array(np.abs(np.random.randn(10, )).astype('float32')), name='input_var'),
+                helper.make_node(
+                    'BatchNormalization',
+                    inputs=['model_input', 'batchnorm.weight', 'batchnorm.bias', 'batchnorm.input_mean', 'batchnorm.input_var'],
+                    outputs=['model_output'],
+                    name='batchnorm'
+                ),
+            ]
+        )
+    )
+    onnx.checker.check_model(model, True)
+    return model
+

TrainingExtensions/onnx/test/python/test_bn_fold.py

@@ -534,3 +534,19 @@ def test_single_batchnorm_layer(self):
             if tensor.name == "batchnorm.input_mean":
                 np_tensor = onnx.numpy_helper.to_array(tensor)
                 assert np.all(np_tensor == np.zeros_like(np_tensor))
+
+    def test_single_bn_layer_with_constants(self):
+        np.random.seed(0)
+        model = models_for_tests.batchnorm_model_constants()
+        dummy_input = make_dummy_input(model)
+        output = rt.InferenceSession(model.SerializeToString(), providers=providers).run(None, dummy_input)[0]
+        _update_standalone_batchnorm_ops(model)
+        output_after_update = rt.InferenceSession(model.SerializeToString(), providers=providers).run(None, dummy_input)[0]
+        assert np.allclose(output, output_after_update, atol=1e-4)
+        for node in model.graph.node:
+            if node.name == "input_var":
+                np_tensor = onnx.numpy_helper.to_array(node.attribute[0].t)
+                assert np.all(np_tensor == np.ones_like(np_tensor))
+            if node.name == "input_mean":
+                np_tensor = onnx.numpy_helper.to_array(node.attribute[0].t)
+                assert np.all(np_tensor == np.zeros_like(np_tensor))