mct-nightly 2.0.0.20240508.145608__tar.gz → 2.0.0.20240510.421__tar.gz

{mct-nightly-2.0.0.20240508.145608 → mct-nightly-2.0.0.20240510.421}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mct-nightly
-Version: 2.0.0.20240508.145608
+Version: 2.0.0.20240510.421
 Summary: A Model Compression Toolkit for neural networks
 Home-page: UNKNOWN
 License: UNKNOWN
@@ -47,13 +47,10 @@ Description: # Model Compression Toolkit (MCT)
         Explore the Model Compression Toolkit (MCT) through our tutorials, 
         covering compression techniques for Keras and PyTorch models. Access interactive [notebooks](tutorials/README.md) 
         for hands-on learning. For example:
-        * [Keras MobileNetV2 post training quantization](tutorials/notebooks/keras/ptq/example_keras_imagenet.ipynb)
-        * [Post training quantization with PyTorch](tutorials/notebooks/pytorch/ptq/example_pytorch_quantization_mnist.ipynb)
-        * [Data Generation for ResNet18 with PyTorch](tutorials/notebooks/pytorch/data_generation/example_pytorch_data_generation.ipynb).
+        * [Keras MobileNetV2 post training quantization](tutorials/notebooks/imx500_notebooks/keras/example_keras_mobilenetv2_for_imx500.ipynb)
+        * [Post training quantization with PyTorch](tutorials/notebooks/mct_features_notebooks/pytorch/example_pytorch_ptq_mnist.ipynb)
+        * [Data Generation for ResNet18 with PyTorch](tutorials/notebooks/mct_features_notebooks/pytorch/example_pytorch_data_generation.ipynb).
         
-        Additionally, for quick quantization of a variety of models from well-known collections,
-        visit the [quick-start page](tutorials/quick_start/README.md) and the
-        [results CSV](tutorials/quick_start/results/model_quantization_results.csv).
         
         ### Supported Versions
         
@@ -155,8 +152,6 @@ Description: # Model Compression Toolkit (MCT)
         | ResNet-18 [3]             | 69.86           | 69.63           |69.53|
         | SqueezeNet 1.1 [3]        | 58.128          | 57.678          ||
         
-        For more results, please refer to [quick start](https://github.com/sony/model_optimization/tree/main/tutorials/quick_start).
-        
         
         #### Pruning Results
         

{mct-nightly-2.0.0.20240508.145608 → mct-nightly-2.0.0.20240510.421}/README.md

@@ -41,13 +41,10 @@ For installing the nightly version or installing from source, refer to the [inst
 Explore the Model Compression Toolkit (MCT) through our tutorials, 
 covering compression techniques for Keras and PyTorch models. Access interactive [notebooks](tutorials/README.md) 
 for hands-on learning. For example:
-* [Keras MobileNetV2 post training quantization](tutorials/notebooks/keras/ptq/example_keras_imagenet.ipynb)
-* [Post training quantization with PyTorch](tutorials/notebooks/pytorch/ptq/example_pytorch_quantization_mnist.ipynb)
-* [Data Generation for ResNet18 with PyTorch](tutorials/notebooks/pytorch/data_generation/example_pytorch_data_generation.ipynb).
+* [Keras MobileNetV2 post training quantization](tutorials/notebooks/imx500_notebooks/keras/example_keras_mobilenetv2_for_imx500.ipynb)
+* [Post training quantization with PyTorch](tutorials/notebooks/mct_features_notebooks/pytorch/example_pytorch_ptq_mnist.ipynb)
+* [Data Generation for ResNet18 with PyTorch](tutorials/notebooks/mct_features_notebooks/pytorch/example_pytorch_data_generation.ipynb).
 
-Additionally, for quick quantization of a variety of models from well-known collections,
-visit the [quick-start page](tutorials/quick_start/README.md) and the
-[results CSV](tutorials/quick_start/results/model_quantization_results.csv).
 
 ### Supported Versions
 
@@ -149,8 +146,6 @@ In the following table we present the ImageNet validation results for these mode
 | ResNet-18 [3]             | 69.86           | 69.63           |69.53|
 | SqueezeNet 1.1 [3]        | 58.128          | 57.678          ||
 
-For more results, please refer to [quick start](https://github.com/sony/model_optimization/tree/main/tutorials/quick_start).
-
 
 #### Pruning Results
 

{mct-nightly-2.0.0.20240508.145608 → mct-nightly-2.0.0.20240510.421}/mct_nightly.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mct-nightly
-Version: 2.0.0.20240508.145608
+Version: 2.0.0.20240510.421
 Summary: A Model Compression Toolkit for neural networks
 Home-page: UNKNOWN
 License: UNKNOWN
@@ -47,13 +47,10 @@ Description: # Model Compression Toolkit (MCT)
         Explore the Model Compression Toolkit (MCT) through our tutorials, 
         covering compression techniques for Keras and PyTorch models. Access interactive [notebooks](tutorials/README.md) 
         for hands-on learning. For example:
-        * [Keras MobileNetV2 post training quantization](tutorials/notebooks/keras/ptq/example_keras_imagenet.ipynb)
-        * [Post training quantization with PyTorch](tutorials/notebooks/pytorch/ptq/example_pytorch_quantization_mnist.ipynb)
-        * [Data Generation for ResNet18 with PyTorch](tutorials/notebooks/pytorch/data_generation/example_pytorch_data_generation.ipynb).
+        * [Keras MobileNetV2 post training quantization](tutorials/notebooks/imx500_notebooks/keras/example_keras_mobilenetv2_for_imx500.ipynb)
+        * [Post training quantization with PyTorch](tutorials/notebooks/mct_features_notebooks/pytorch/example_pytorch_ptq_mnist.ipynb)
+        * [Data Generation for ResNet18 with PyTorch](tutorials/notebooks/mct_features_notebooks/pytorch/example_pytorch_data_generation.ipynb).
         
-        Additionally, for quick quantization of a variety of models from well-known collections,
-        visit the [quick-start page](tutorials/quick_start/README.md) and the
-        [results CSV](tutorials/quick_start/results/model_quantization_results.csv).
         
         ### Supported Versions
         
@@ -155,8 +152,6 @@ Description: # Model Compression Toolkit (MCT)
         | ResNet-18 [3]             | 69.86           | 69.63           |69.53|
         | SqueezeNet 1.1 [3]        | 58.128          | 57.678          ||
         
-        For more results, please refer to [quick start](https://github.com/sony/model_optimization/tree/main/tutorials/quick_start).
-        
         
         #### Pruning Results
         

{mct-nightly-2.0.0.20240508.145608 → mct-nightly-2.0.0.20240510.421}/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__ = "2.0.0.20240508.145608"
+__version__ = "2.0.0.20240510.000421"

{mct-nightly-2.0.0.20240508.145608 → mct-nightly-2.0.0.20240510.421}/model_compression_toolkit/core/common/visualization/tensorboard_writer.py

@@ -72,19 +72,10 @@ def get_node_properties(node_dict_to_log: dict,
     # Create protobuf for the node's output shapes
     if output_shapes is not None:
         tshape_protos = []
-        is_tf_combined_non_max_suppression = len(output_shapes) == 1 and 'function' in node_dict_to_log and node_dict_to_log['function'] == 'image.combined_non_max_suppression'
-
-        if is_tf_combined_non_max_suppression:
-            combined_nms_output = output_shapes[0]
-            output_shapes = [combined_nms_output.nmsed_boxes,
-                             combined_nms_output.nmsed_scores,
-                             combined_nms_output.nmsed_classes,
-                             combined_nms_output.valid_detections]
-
         for output_shape in output_shapes:  # create protobuf for each output shape
             proto_dims_list = []
             for dim in output_shape:
-                proto_dims_list.append(TensorShapeProto.Dim(size=dim)) # dim shold ne an integer
+                proto_dims_list.append(TensorShapeProto.Dim(size=dim))  # dim is expected to be an integer
             tshape_proto = TensorShapeProto(dim=proto_dims_list)
             tshape_protos.append(tshape_proto)
         node_properties['_output_shapes'] = AttrValue(list=AttrValue.ListValue(shape=tshape_protos))
@@ -272,7 +263,13 @@ class TensorboardWriter(object):
 
             # For nodes with an "empty" output shape.
             output_shape = (None,) if n.output_shape == () else n.output_shape
-
+            if 'CombinedNonMaxSuppression' in str(output_shape):
+                # output_shapes is expected to be a list of tuples where each tuple is an output shape.
+                # For NMS layers, we need to align the node's output shapes before creating the node's properties.
+                output_shape = [output_shape.nmsed_boxes,
+                                output_shape.nmsed_scores,
+                                output_shape.nmsed_classes,
+                                output_shape.valid_detections]
             dims = []
             if isinstance(output_shape, list):
                 for o in output_shape:

{mct-nightly-2.0.0.20240508.145608 → mct-nightly-2.0.0.20240510.421}/model_compression_toolkit/data_generation/pytorch/pytorch_data_generation.py

@@ -354,9 +354,9 @@ else:
     # If torch is not installed,
     # we raise an exception when trying to use these functions.
     def get_pytorch_data_generation_config(*args, **kwargs):
-        msg = f"torch and torchvision must be installed to use get_pytorch_data_generation_config. " + ("" if FOUND_TORCH else "'torch' package is missing. ") + ("" if FOUND_TORCHVISION else "'torchvision' package is missing. ") # pragma: no cover
+        msg = f"PyTorch and torchvision must be installed to use get_pytorch_data_generation_config. " + ("" if FOUND_TORCH else "The 'torch' package is missing. ") + ("" if FOUND_TORCHVISION else "The 'torchvision' package is missing. ") # pragma: no cover
         Logger.critical(msg)  # pragma: no cover
 
     def pytorch_data_generation_experimental(*args, **kwargs):
-        msg = f"torch and torchvision must be installed to use pytorch_data_generation_experimental. " + ("" if FOUND_TORCH else "'torch' package is missing. ") + ("" if FOUND_TORCHVISION else "'torchvision' package is missing. ") # pragma: no cover
-        Logger.critical(msg)  # pragma: no cover
+        msg = f"PyTorch and torchvision must be installed to use pytorch_data_generation_experimental. " + ("" if FOUND_TORCH else "The 'torch' package is missing. ") + ("" if FOUND_TORCHVISION else "The 'torchvision' package is missing. ") # pragma: no cover
+        Logger.critical(msg)  # pragma: no cover

{mct-nightly-2.0.0.20240508.145608 → mct-nightly-2.0.0.20240510.421}/model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_onnx_pytorch_exporter.py

@@ -43,7 +43,7 @@ if FOUND_ONNX:
                      save_model_path: str,
                      repr_dataset: Callable,
                      onnx_opset_version: int,
-                     use_onnx_custom_quantizer_ops: bool = False,):
+                     use_onnx_custom_quantizer_ops: bool = False):
             """
 
             Args:
@@ -134,6 +134,5 @@ if FOUND_ONNX:
 
 else:
     def FakelyQuantONNXPyTorchExporter(*args, **kwargs):
-            Logger.critical('Installing onnx is mandatory '
-                            'when using FakelyQuantONNXPyTorchExporter. '
-                            'Could not find onnx package.')  # pragma: no cover
+        Logger.critical("ONNX must be installed to use 'FakelyQuantONNXPyTorchExporter'. "
+                        "The 'onnx' package is missing.")  # pragma: no cover

{mct-nightly-2.0.0.20240508.145608 → mct-nightly-2.0.0.20240510.421}/model_compression_toolkit/exporter/model_exporter/pytorch/pytorch_export_facade.py

@@ -14,7 +14,7 @@
 # ==============================================================================
 from typing import Callable
 
-from model_compression_toolkit.constants import FOUND_TORCH, FOUND_ONNX
+from model_compression_toolkit.constants import FOUND_TORCH
 from model_compression_toolkit.exporter.model_exporter.fw_agonstic.quantization_format import QuantizationFormat
 from model_compression_toolkit.exporter.model_exporter.pytorch.export_serialization_format import \
     PytorchExportSerializationFormat

{mct-nightly-2.0.0.20240508.145608 → mct-nightly-2.0.0.20240510.421}/model_compression_toolkit/target_platform_capabilities/tpc_models/imx500_tpc/v2/tp_model.py

@@ -152,19 +152,6 @@ def generate_tp_model(default_config: OpQuantizationConfig,
     # this configuration will be used for the operation quantization:
     default_configuration_options = tp.QuantizationConfigOptions([default_config])
 
-    # Create a QuantizationConfigOptions for quantizing constants in functional ops.
-    # Constant configuration is similar to the default eight bit configuration except for PoT
-    # quantization method for the constant.
-    # Since the constants are not named attributes of the layer, we use the default_weight_attr_config to
-    # define the desired quantization properties for them.
-    const_config = default_config.clone_and_edit(
-        default_weight_attr_config=default_config.default_weight_attr_config.clone_and_edit(
-            enable_weights_quantization=True))
-    if not (const_config.default_weight_attr_config.weights_quantization_method == tp.QuantizationMethod.POWER_OF_TWO and
-            const_config.default_weight_attr_config.weights_per_channel_threshold is False):
-        mct.logger.Logger.error('Constant quantization config should be per-tensor PoT.')
-    const_configuration_options = tp.QuantizationConfigOptions([const_config])
-
     # Create a TargetPlatformModel and set its default quantization config.
     # This default configuration will be used for all operations
     # unless specified otherwise (see OperatorsSet, for example):
@@ -198,10 +185,10 @@ def generate_tp_model(default_config: OpQuantizationConfig,
         # Define operations sets without quantization configuration
         # options (useful for creating fusing patterns, for example):
         any_relu = tp.OperatorsSet("AnyReLU")
-        add = tp.OperatorsSet("Add", const_configuration_options)
-        sub = tp.OperatorsSet("Sub", const_configuration_options)
-        mul = tp.OperatorsSet("Mul", const_configuration_options)
-        div = tp.OperatorsSet("Div", const_configuration_options)
+        add = tp.OperatorsSet("Add")
+        sub = tp.OperatorsSet("Sub")
+        mul = tp.OperatorsSet("Mul")
+        div = tp.OperatorsSet("Div")
         prelu = tp.OperatorsSet("PReLU")
         swish = tp.OperatorsSet("Swish")
         sigmoid = tp.OperatorsSet("Sigmoid")

{mct-nightly-2.0.0.20240508.145608 → mct-nightly-2.0.0.20240510.421}/model_compression_toolkit/target_platform_capabilities/tpc_models/imx500_tpc/v2_lut/tp_model.py

@@ -151,19 +151,6 @@ def generate_tp_model(default_config: OpQuantizationConfig,
     # this configuration will be used for the operation quantization:
     default_configuration_options = tp.QuantizationConfigOptions([default_config])
 
-    # Create a QuantizationConfigOptions for quantizing constants in functional ops.
-    # Constant configuration is similar to the default eight bit configuration except for PoT
-    # quantization method for the constant.
-    # Since the constants are not named attributes of the layer, we use the default_weight_attr_config to
-    # define the desired quantization properties for them.
-    const_config = default_config.clone_and_edit(
-        default_weight_attr_config=default_config.default_weight_attr_config.clone_and_edit(
-            enable_weights_quantization=True))
-    if not (const_config.default_weight_attr_config.weights_quantization_method == tp.QuantizationMethod.POWER_OF_TWO and
-            const_config.default_weight_attr_config.weights_per_channel_threshold is False):
-        mct.logger.Logger.error('Constant quantization config should be per-tensor PoT.')
-    const_configuration_options = tp.QuantizationConfigOptions([const_config])
-
     # Create a TargetPlatformModel and set its default quantization config.
     # This default configuration will be used for all operations
     # unless specified otherwise (see OperatorsSet, for example):
@@ -195,10 +182,10 @@ def generate_tp_model(default_config: OpQuantizationConfig,
         # Define operations sets without quantization configuration
         # options (useful for creating fusing patterns, for example):
         any_relu = tp.OperatorsSet("AnyReLU")
-        add = tp.OperatorsSet("Add", const_configuration_options)
-        sub = tp.OperatorsSet("Sub", const_configuration_options)
-        mul = tp.OperatorsSet("Mul", const_configuration_options)
-        div = tp.OperatorsSet("Div", const_configuration_options)
+        add = tp.OperatorsSet("Add")
+        sub = tp.OperatorsSet("Sub")
+        mul = tp.OperatorsSet("Mul")
+        div = tp.OperatorsSet("Div")
         prelu = tp.OperatorsSet("PReLU")
         swish = tp.OperatorsSet("Swish")
         sigmoid = tp.OperatorsSet("Sigmoid")