mct-nightly 2.3.0.20250425.557__py3-none-any.whl → 2.3.0.20250427.609__py3-none-any.whl

{mct_nightly-2.3.0.20250425.557.dist-info → mct_nightly-2.3.0.20250427.609.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mct-nightly
-Version: 2.3.0.20250425.557
+Version: 2.3.0.20250427.609
 Summary: A Model Compression Toolkit for neural networks
 Classifier: Programming Language :: Python :: 3
 Classifier: License :: OSI Approved :: Apache Software License

{mct_nightly-2.3.0.20250425.557.dist-info → mct_nightly-2.3.0.20250427.609.dist-info}/RECORD

@@ -1,5 +1,5 @@
-mct_nightly-2.3.0.20250425.557.dist-info/licenses/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
-model_compression_toolkit/__init__.py,sha256=kXjmCqsExE8DYRx8JTVs4AhVXF1GF94SPL9EcfyLeks,1557
+mct_nightly-2.3.0.20250427.609.dist-info/licenses/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
+model_compression_toolkit/__init__.py,sha256=n5rc5Vyh3AmYQB3AKBl-pAmV5IKGzeqqRm3AOQFlSa4,1557
 model_compression_toolkit/constants.py,sha256=iJ6vfTjC2oFIZWt8wvHoxEw5YJi3yl0Hd4q30_8q0Zc,3958
 model_compression_toolkit/defaultdict.py,sha256=LSc-sbZYXENMCw3U9F4GiXuv67IKpdn0Qm7Fr11jy-4,2277
 model_compression_toolkit/logger.py,sha256=L3q7tn3Uht0i_7phnlOWMR2Te2zvzrt2HOz9vYEInts,4529
@@ -233,7 +233,7 @@ model_compression_toolkit/core/pytorch/back2framework/factory_model_builder.py,s
 model_compression_toolkit/core/pytorch/back2framework/float_model_builder.py,sha256=tLrlUyYhxVKVjkad1ZAtbRra0HedB3iVfIkZ_dYnQ-4,3419
 model_compression_toolkit/core/pytorch/back2framework/instance_builder.py,sha256=BBHBfTqeWm7L3iDyPBpk0jxvj-rBg1QWI23imkjfIl0,1467
 model_compression_toolkit/core/pytorch/back2framework/mixed_precision_model_builder.py,sha256=K4L8FzJFM8_Ge2MHYkSqzCtoZe-ejEhVq8C1RgecyOc,14531
-model_compression_toolkit/core/pytorch/back2framework/pytorch_model_builder.py,sha256=WccaNiHK12IIimYu29E1oJkQHUdhPCBcIRutefTQ3Ag,19903
+model_compression_toolkit/core/pytorch/back2framework/pytorch_model_builder.py,sha256=9l5ZsEd4lwCSzT6VNItj5rw996BI_eH71x_uV-4gIRM,21101
 model_compression_toolkit/core/pytorch/back2framework/quantized_model_builder.py,sha256=qZNNOlNTTV4ZKPG3q5GDXkIVTPUEr8dvxAS_YiMORmg,3456
 model_compression_toolkit/core/pytorch/back2framework/quantization_wrapper/__init__.py,sha256=cco4TmeIDIh32nj9ZZXVkws4dd9F2UDrmjKzTN8G0V0,697
 model_compression_toolkit/core/pytorch/back2framework/quantization_wrapper/quantized_layer_wrapper.py,sha256=q2JDw10NKng50ee2i9faGzWZ-IydnR2aOMGSn9RoZmc,5773
@@ -528,7 +528,7 @@ model_compression_toolkit/xquant/pytorch/model_analyzer.py,sha256=b93o800yVB3Z-i
 model_compression_toolkit/xquant/pytorch/pytorch_report_utils.py,sha256=UVN_S9ULHBEldBpShCOt8-soT8YTQ5oE362y96qF_FA,3950
 model_compression_toolkit/xquant/pytorch/similarity_functions.py,sha256=CERxq5K8rqaiE-DlwhZBTUd9x69dtYJlkHOPLB54vm8,2354
 model_compression_toolkit/xquant/pytorch/tensorboard_utils.py,sha256=mkoEktLFFHtEKzzFRn_jCnxjhJolK12TZ5AQeDHzUO8,9767
-mct_nightly-2.3.0.20250425.557.dist-info/METADATA,sha256=3W1evd1H2JL0wpqb1frzmmX5nz2fqHcMib1y3remRt8,25560
-mct_nightly-2.3.0.20250425.557.dist-info/WHEEL,sha256=SmOxYU7pzNKBqASvQJ7DjX3XGUF92lrGhMb3R6_iiqI,91
-mct_nightly-2.3.0.20250425.557.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
-mct_nightly-2.3.0.20250425.557.dist-info/RECORD,,
+mct_nightly-2.3.0.20250427.609.dist-info/METADATA,sha256=uT9gv4F4QKU-7z2fkoBSftfQMo-eUFG_n2ANL1X9DbA,25560
+mct_nightly-2.3.0.20250427.609.dist-info/WHEEL,sha256=SmOxYU7pzNKBqASvQJ7DjX3XGUF92lrGhMb3R6_iiqI,91
+mct_nightly-2.3.0.20250427.609.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
+mct_nightly-2.3.0.20250427.609.dist-info/RECORD,,

model_compression_toolkit/__init__.py

@@ -27,4 +27,4 @@ from model_compression_toolkit import data_generation
 from model_compression_toolkit import pruning
 from model_compression_toolkit.trainable_infrastructure.keras.load_model import keras_load_quantized_model
 
-__version__ = "2.3.0.20250425.000557"
+__version__ = "2.3.0.20250427.000609"

model_compression_toolkit/core/pytorch/back2framework/pytorch_model_builder.py

@@ -234,7 +234,9 @@ class PytorchModel(torch.nn.Module):
         self.wrapper = wrapper
         self.get_activation_quantizer_holder = get_activation_quantizer_holder_fn
         self.reuse_groups = {}
-        self._add_modules()
+        self._reused_nodes = []
+
+        self._add_all_modules()
 
     # todo: Move to parent class BaseModelBuilder
     @property
@@ -286,17 +288,37 @@ class PytorchModel(torch.nn.Module):
                 node_op = self.wrapper(node, node_builder(node))
         return node_op
 
-    def _add_modules(self):
+    def _add_all_modules(self):
+        """
+        Build and add the modules and functional nodes from node_sort list as attributes to PytorchModel.
+        To assure all required nodes for the reused nodes are already initialized, adds none-reused nodes first,
+        then adds the reused nodes.
+        """
+        self._add_modules(reused_nodes_only=False)  # add none-reused nodes
+        self._add_modules(reused_nodes_only=True)  # add reused nodes
+
+    def _add_modules(self, reused_nodes_only=False):
         """
         Build and add the modules and functional nodes from node_sort list as attributes to PytorchModel
+        Args:
+            reused_nodes_only: whether to go over the reuse nodes list or not.
+                   In case reuse_nodes_only is False - will go over all nodes, and add reused nodes to self._reused_nodes
+                   In case reuse_nodes_only is True - will go over self._reused_nodes only.
+
         """
-        for node in self.node_sort:
-            if node.reuse:
-                # If the node is reused, retrieve the original module
+        nodes = self._reused_nodes if reused_nodes_only else self.node_sort
+        for node in nodes:
+            if node.reuse and reused_nodes_only:
                 if node.reuse_group not in self.reuse_groups:
-                    Logger.critical(f"Reuse group {node.reuse_group} not found for node {node.name}")
+                    raise Exception(f"Reuse group {node.reuse_group} not found for node {node.name}. "
+                                    f"Make sure you first call the method with reused_nodes_only=False")
+                else:
+                    node_op = self.reuse_groups[node.reuse_group]  # retrieve the original module
+
+            elif node.reuse:  # add node to reused list, and go over the list after all other nodes were created
+                self._reused_nodes.append(node)
+                continue
 
-                node_op = self.reuse_groups[node.reuse_group]
             else:
                 # If it's not reused, create a new module
                 node_op = self.wrap(node)