mct-nightly 2.4.0.20250618.606__py3-none-any.whl → 2.4.0.20250619.621__py3-none-any.whl

{mct_nightly-2.4.0.20250618.606.dist-info → mct_nightly-2.4.0.20250619.621.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mct-nightly
-Version: 2.4.0.20250618.606
+Version: 2.4.0.20250619.621
 Summary: A Model Compression Toolkit for neural networks
 Author-email: ssi-dnn-dev@sony.com
 Classifier: Programming Language :: Python :: 3

{mct_nightly-2.4.0.20250618.606.dist-info → mct_nightly-2.4.0.20250619.621.dist-info}/RECORD

@@ -1,5 +1,5 @@
-mct_nightly-2.4.0.20250618.606.dist-info/licenses/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
-model_compression_toolkit/__init__.py,sha256=wrRgW-DB8mcY8PhRhrQ3I5qhLKYqFra-UMKgJRufNBs,1557
+mct_nightly-2.4.0.20250619.621.dist-info/licenses/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
+model_compression_toolkit/__init__.py,sha256=IwA7G4npLqjEevZwVWOsQQsli7NOIlB3fVVBKUgnNZU,1557
 model_compression_toolkit/constants.py,sha256=KNgiNLpsMgSYyXMNEbHXd4bFNerQc1D6HH3vpbUq_Gs,4086
 model_compression_toolkit/defaultdict.py,sha256=LSc-sbZYXENMCw3U9F4GiXuv67IKpdn0Qm7Fr11jy-4,2277
 model_compression_toolkit/logger.py,sha256=L3q7tn3Uht0i_7phnlOWMR2Te2zvzrt2HOz9vYEInts,4529
@@ -162,7 +162,7 @@ model_compression_toolkit/core/keras/default_framework_info.py,sha256=n0fkMlQ0Cg
 model_compression_toolkit/core/keras/keras_implementation.py,sha256=x5EOYBrg2chC9-OUlrd0laLpnnHCFhYYAFNKRhVh6aQ,28526
 model_compression_toolkit/core/keras/keras_model_validation.py,sha256=dMS9cqaYmliyzVu2-MrKx4AIubqz3HW3RY4if2TV6U8,1581
 model_compression_toolkit/core/keras/keras_node_prior_info.py,sha256=k9cwu3S-OUGFaOHxH6cyYS2JjxAYHfBddz0laf6Quds,3311
-model_compression_toolkit/core/keras/resource_utilization_data_facade.py,sha256=i1VsqqPsVb98YCepCjypIjnBXY-m_r7pUnPpTSETNWc,5532
+model_compression_toolkit/core/keras/resource_utilization_data_facade.py,sha256=xxZlHyruhLuP2iEgMrZhq_AyAGORTqzweVLARFfpaRw,5643
 model_compression_toolkit/core/keras/tf_tensor_numpy.py,sha256=jzD8FGEEa8ZD7w8IpTRdp-Udf1MwOTgjg2XTS1Givic,2696
 model_compression_toolkit/core/keras/back2framework/__init__.py,sha256=rhIiXg_nBgUZ-baE3M6SzCuQbcnq4iebY1jtJBvKHOM,808
 model_compression_toolkit/core/keras/back2framework/factory_model_builder.py,sha256=UIQgOOdexycrSKombTMJVvTthR7MlrCihoqM8Kg-rnE,2293
@@ -331,14 +331,14 @@ model_compression_toolkit/exporter/model_exporter/keras/export_serialization_for
 model_compression_toolkit/exporter/model_exporter/keras/fakely_quant_keras_exporter.py,sha256=n_iXPwMomMVJTZH9M1WV7OJo11ppXOWkANu41fIlsjY,11702
 model_compression_toolkit/exporter/model_exporter/keras/fakely_quant_tflite_exporter.py,sha256=XoFGkVBikKh1BuxObrMLjfVLDIgy3X0rhmEl08CdJls,3727
 model_compression_toolkit/exporter/model_exporter/keras/int8_tflite_exporter.py,sha256=iTUXaia8XLJmmWdk4iiCah9sxeIyBJy42s9_EpuPhnw,8261
-model_compression_toolkit/exporter/model_exporter/keras/keras_export_facade.py,sha256=NzcX7rxLk__Kpuim_VXaOHS4tyiRtRBoERPE00GbdfA,5862
+model_compression_toolkit/exporter/model_exporter/keras/keras_export_facade.py,sha256=2LQ7afCtciq8pDcCfQvwXz-uMlABiZJdRsztIWs6040,5973
 model_compression_toolkit/exporter/model_exporter/keras/mctq_keras_exporter.py,sha256=qXXkv3X_wb7t622EOHwXIxfGLGaDqh0T0y4UxREi4Bo,1976
 model_compression_toolkit/exporter/model_exporter/pytorch/__init__.py,sha256=uZ2RigbY9O2PJ0Il8wPpS_s7frgg9WUGd_SHeKGyl1A,699
 model_compression_toolkit/exporter/model_exporter/pytorch/base_pytorch_exporter.py,sha256=9adOGG1nyviNzuL-1aJXyL0c_VQllSZWiG2gR-puywo,6420
 model_compression_toolkit/exporter/model_exporter/pytorch/export_serialization_format.py,sha256=bPevy6OBqng41PqytBR55e6cBEuyrUS0H8dWX4zgjQ4,967
-model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_onnx_pytorch_exporter.py,sha256=1ix8j7rxc1giPjf2PZKwaaCb5pKo0obUvPmRtklmugY,10056
+model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_onnx_pytorch_exporter.py,sha256=H0vXDcnQhVXAy-WOpnoM8kjTs5gzmYmO2IiDECSUpd0,10239
 model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_torchscript_pytorch_exporter.py,sha256=ksWV2A-Njo-wAxQ_Ye2sLIZXBWJ_WNyjT7-qFFwvV2o,2897
-model_compression_toolkit/exporter/model_exporter/pytorch/pytorch_export_facade.py,sha256=7xuUrHPMiifn23sWfeiqR9wkYhm8EweDRL_vF-JSxMY,6642
+model_compression_toolkit/exporter/model_exporter/pytorch/pytorch_export_facade.py,sha256=pDBAsUSm3Dq5ZKFH1XftvZ5GZn_R63IJpuhUII9Z_k0,6759
 model_compression_toolkit/exporter/model_wrapper/__init__.py,sha256=7CF2zvpTrIEm8qnbuHnLZyTZkwBBxV24V8QA0oxGbh0,1187
 model_compression_toolkit/exporter/model_wrapper/fw_agnostic/__init__.py,sha256=pKAdbTCFM_2BrZXUtTIw0ouKotrWwUDF_hP3rPwCM2k,696
 model_compression_toolkit/exporter/model_wrapper/fw_agnostic/get_inferable_quantizers.py,sha256=Bd3QhAR__YC9Xmobd5qHv9ofh_rPn_eTFV0sXizcBnY,2297
@@ -529,7 +529,7 @@ model_compression_toolkit/xquant/pytorch/model_analyzer.py,sha256=b93o800yVB3Z-i
 model_compression_toolkit/xquant/pytorch/pytorch_report_utils.py,sha256=Y0oBl8qPFsdNrK49XczwmVacInJcOPHslVnFBs-iTCc,3742
 model_compression_toolkit/xquant/pytorch/similarity_functions.py,sha256=CERxq5K8rqaiE-DlwhZBTUd9x69dtYJlkHOPLB54vm8,2354
 model_compression_toolkit/xquant/pytorch/tensorboard_utils.py,sha256=n0HvWBzkBkUJZlS3WeynhpsRTps2qQkjlq7luliBHNU,9627
-mct_nightly-2.4.0.20250618.606.dist-info/METADATA,sha256=4-RG_tJDUZfTOBhT-rkaWLwAhWIoZNB8szWU_njK5iQ,25087
-mct_nightly-2.4.0.20250618.606.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-mct_nightly-2.4.0.20250618.606.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
-mct_nightly-2.4.0.20250618.606.dist-info/RECORD,,
+mct_nightly-2.4.0.20250619.621.dist-info/METADATA,sha256=Kc3thJIr5lW5AUz47ho_FKA41lX6vXg4CUijKaE0elg,25087
+mct_nightly-2.4.0.20250619.621.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+mct_nightly-2.4.0.20250619.621.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
+mct_nightly-2.4.0.20250619.621.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__ = "2.4.0.20250618.000606"
+__version__ = "2.4.0.20250619.000621"

model_compression_toolkit/core/keras/resource_utilization_data_facade.py

@@ -28,12 +28,14 @@ if FOUND_TF:
         AttachTpcToKeras
     from model_compression_toolkit.target_platform_capabilities.constants import DEFAULT_TP_MODEL
     from model_compression_toolkit.core.keras.keras_implementation import KerasImplementation
+    from model_compression_toolkit.core.keras.default_framework_info import set_keras_info
     from tensorflow.keras.models import Model
 
     from model_compression_toolkit import get_target_platform_capabilities
 
     KERAS_DEFAULT_TPC = get_target_platform_capabilities(TENSORFLOW, DEFAULT_TP_MODEL)
 
+    @set_keras_info
     def keras_resource_utilization_data(in_model: Model,
                                         representative_data_gen: Callable,
                                         core_config: CoreConfig = CoreConfig(

model_compression_toolkit/exporter/model_exporter/keras/keras_export_facade.py

@@ -21,6 +21,7 @@ from model_compression_toolkit.logger import Logger
 
 if FOUND_TF:
     import keras
+    from model_compression_toolkit.core.keras.default_framework_info import set_keras_info
     from model_compression_toolkit.exporter.model_wrapper.keras.validate_layer import is_keras_layer_exportable
     from model_compression_toolkit.exporter.model_exporter.keras.fakely_quant_keras_exporter import \
         FakelyQuantKerasExporter
@@ -36,6 +37,7 @@ if FOUND_TF:
         KerasExportSerializationFormat.TFLITE: [QuantizationFormat.FAKELY_QUANT, QuantizationFormat.INT8]
     }
 
+    @set_keras_info
     def keras_export_model(model: keras.models.Model,
                            save_model_path: str,
                            is_layer_exportable_fn: Callable = is_keras_layer_exportable,

model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_onnx_pytorch_exporter.py

@@ -73,23 +73,25 @@ if FOUND_ONNX:
             Returns:
                 Fake-quant PyTorch model.
             """
-            # List all activation quantization holders with num_bits>8 and replace them with Identity, because
-            # ONNX doesn't support quantization of more than 8 bits for torch.fake_quantize_per_tensor_affine.
-            act_holder_list = [n for n, m in self.model.named_modules()
-                               if isinstance(m, PytorchActivationQuantizationHolder) and
-                               m.activation_holder_quantizer.num_bits > 8]
-            for act_holder in act_holder_list: # pragma: no cover
-                obj = self.model
-                attrs = act_holder.split(".")
-                for a in attrs[:-1]:
-                    obj = getattr(obj, a)
-                if hasattr(obj, attrs[-1]):
-                    delattr(obj, attrs[-1])
-                    setattr(obj, attrs[-1], torch.nn.Identity())
-                else:
-                    Logger.info(f"During removal of activation quantization of a quantizer (with bits > 8) in ONNX FQ "
-                                f"export, deletion of activation holder '{act_holder}' failed — could not locate one or"
-                                f"more intermediate attributes in the path.")
+            # When exporting using Fakely Quant Quantization Format list all activation quantization holders with
+            # num_bits>8 and replace them with Identity, because ONNX doesn't support quantization of more than 8 bits
+            # for torch.fake_quantize_per_tensor_affine.
+            if not self._use_onnx_custom_quantizer_ops:
+                act_holder_list = [n for n, m in self.model.named_modules()
+                                   if isinstance(m, PytorchActivationQuantizationHolder) and
+                                   m.activation_holder_quantizer.num_bits > 8]
+                for act_holder in act_holder_list: # pragma: no cover
+                    obj = self.model
+                    attrs = act_holder.split(".")
+                    for a in attrs[:-1]:
+                        obj = getattr(obj, a)
+                    if hasattr(obj, attrs[-1]):
+                        delattr(obj, attrs[-1])
+                        setattr(obj, attrs[-1], torch.nn.Identity())
+                    else:
+                        Logger.info(f"During removal of activation quantization of a quantizer (with bits > 8) in ONNX"
+                                    f"FQ export, deletion of activation holder '{act_holder}' failed — could not locate"
+                                    f"one or more intermediate attributes in the path.")
 
             for layer in self.model.children():
                 self.is_layer_exportable_fn(layer)

model_compression_toolkit/exporter/model_exporter/pytorch/pytorch_export_facade.py

@@ -27,6 +27,7 @@ DEFAULT_ONNX_OPSET_VERSION = 15
 
 if FOUND_TORCH:
     import torch.nn
+    from model_compression_toolkit.core.pytorch.default_framework_info import set_pytorch_info
     from model_compression_toolkit.exporter.model_exporter.pytorch.fakely_quant_onnx_pytorch_exporter import FakelyQuantONNXPyTorchExporter
     from model_compression_toolkit.exporter.model_exporter.pytorch.fakely_quant_torchscript_pytorch_exporter import FakelyQuantTorchScriptPyTorchExporter
     from model_compression_toolkit.exporter.model_wrapper.pytorch.validate_layer import is_pytorch_layer_exportable
@@ -41,6 +42,7 @@ if FOUND_TORCH:
         PytorchExportSerializationFormat.ONNX: [QuantizationFormat.FAKELY_QUANT, QuantizationFormat.MCTQ]
     }
 
+    @set_pytorch_info
     def pytorch_export_model(model: torch.nn.Module,
                              save_model_path: str,
                              repr_dataset: Callable,