mct-nightly 1.11.0.20240303.post423__tar.gz → 1.11.0.20240305.post352__tar.gz

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mct-nightly
-Version: 1.11.0.20240303.post423
+Version: 1.11.0.20240305.post352
 Summary: A Model Compression Toolkit for neural networks
 Home-page: UNKNOWN
 License: UNKNOWN
@@ -152,11 +152,11 @@ Description: # Model Compression Toolkit (MCT)
         taking into account the target platform's Single Instruction, Multiple Data (SIMD) capabilities. 
         By pruning groups of channels (SIMD groups), our approach not only reduces model size 
         and complexity, but ensures that better utilization of channels is in line with the SIMD architecture 
-        for a target KPI of weights memory footprint.
+        for a target KPI of weights memory footprint. 
+        [Keras API](https://sony.github.io/model_optimization/docs/api/experimental_api_docs/methods/keras_pruning_experimental.html)
+        [Pytorch API](https://sony.github.io/model_optimization/docs/api/experimental_api_docs/methods/pytorch_pruning_experimental.html)
         
         
-        <u>_Note: Currently, only Keras models pruning is supported._</u>
-        
         #### Results
         
         Results for applying pruning to reduce the parameters of the following models by 50%:
@@ -164,7 +164,7 @@ Description: # Model Compression Toolkit (MCT)
         | Model           | Dense Model Accuracy | Pruned Model Accuracy |
         |-----------------|----------------------|-----------------------|
         | ResNet50 [2]    | 75.1                 | 72.4                  |
-        | DenseNet121 [2] | 75.0                 | 71.15                 |
+        | DenseNet121 [3] | 74.44                | 71.71                 |
         
         
         

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/README.md

@@ -146,11 +146,11 @@ This pruning technique is designed to compress models for specific hardware arch
 taking into account the target platform's Single Instruction, Multiple Data (SIMD) capabilities. 
 By pruning groups of channels (SIMD groups), our approach not only reduces model size 
 and complexity, but ensures that better utilization of channels is in line with the SIMD architecture 
-for a target KPI of weights memory footprint.
+for a target KPI of weights memory footprint. 
+[Keras API](https://sony.github.io/model_optimization/docs/api/experimental_api_docs/methods/keras_pruning_experimental.html)
+[Pytorch API](https://sony.github.io/model_optimization/docs/api/experimental_api_docs/methods/pytorch_pruning_experimental.html)
 
 
-<u>_Note: Currently, only Keras models pruning is supported._</u>
-
 #### Results
 
 Results for applying pruning to reduce the parameters of the following models by 50%:
@@ -158,7 +158,7 @@ Results for applying pruning to reduce the parameters of the following models by
 | Model           | Dense Model Accuracy | Pruned Model Accuracy |
 |-----------------|----------------------|-----------------------|
 | ResNet50 [2]    | 75.1                 | 72.4                  |
-| DenseNet121 [2] | 75.0                 | 71.15                 |
+| DenseNet121 [3] | 74.44                | 71.71                 |
 
 
 

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/mct_nightly.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mct-nightly
-Version: 1.11.0.20240303.post423
+Version: 1.11.0.20240305.post352
 Summary: A Model Compression Toolkit for neural networks
 Home-page: UNKNOWN
 License: UNKNOWN
@@ -152,11 +152,11 @@ Description: # Model Compression Toolkit (MCT)
         taking into account the target platform's Single Instruction, Multiple Data (SIMD) capabilities. 
         By pruning groups of channels (SIMD groups), our approach not only reduces model size 
         and complexity, but ensures that better utilization of channels is in line with the SIMD architecture 
-        for a target KPI of weights memory footprint.
+        for a target KPI of weights memory footprint. 
+        [Keras API](https://sony.github.io/model_optimization/docs/api/experimental_api_docs/methods/keras_pruning_experimental.html)
+        [Pytorch API](https://sony.github.io/model_optimization/docs/api/experimental_api_docs/methods/pytorch_pruning_experimental.html)
         
         
-        <u>_Note: Currently, only Keras models pruning is supported._</u>
-        
         #### Results
         
         Results for applying pruning to reduce the parameters of the following models by 50%:
@@ -164,7 +164,7 @@ Description: # Model Compression Toolkit (MCT)
         | Model           | Dense Model Accuracy | Pruned Model Accuracy |
         |-----------------|----------------------|-----------------------|
         | ResNet50 [2]    | 75.1                 | 72.4                  |
-        | DenseNet121 [2] | 75.0                 | 71.15                 |
+        | DenseNet121 [3] | 74.44                | 71.71                 |
         
         
         

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/mct_nightly.egg-info/SOURCES.txt

@@ -120,7 +120,6 @@ model_compression_toolkit/core/common/quantization/quantize_node.py
 model_compression_toolkit/core/common/quantization/set_node_quantization_config.py
 model_compression_toolkit/core/common/quantization/quantization_params_generation/__init__.py
 model_compression_toolkit/core/common/quantization/quantization_params_generation/error_functions.py
-model_compression_toolkit/core/common/quantization/quantization_params_generation/kmeans_params.py
 model_compression_toolkit/core/common/quantization/quantization_params_generation/lut_kmeans_params.py
 model_compression_toolkit/core/common/quantization/quantization_params_generation/outlier_filter.py
 model_compression_toolkit/core/common/quantization/quantization_params_generation/power_of_two_selection.py
@@ -131,7 +130,6 @@ model_compression_toolkit/core/common/quantization/quantization_params_generatio
 model_compression_toolkit/core/common/quantization/quantization_params_generation/symmetric_selection.py
 model_compression_toolkit/core/common/quantization/quantization_params_generation/uniform_selection.py
 model_compression_toolkit/core/common/quantization/quantizers/__init__.py
-model_compression_toolkit/core/common/quantization/quantizers/kmeans_quantizer.py
 model_compression_toolkit/core/common/quantization/quantizers/lut_kmeans_quantizer.py
 model_compression_toolkit/core/common/quantization/quantizers/quantizers_helpers.py
 model_compression_toolkit/core/common/quantization/quantizers/uniform_quantizers.py
@@ -263,6 +261,8 @@ model_compression_toolkit/core/pytorch/hessian/weights_trace_hessian_calculator_
 model_compression_toolkit/core/pytorch/mixed_precision/__init__.py
 model_compression_toolkit/core/pytorch/mixed_precision/configurable_activation_quantizer.py
 model_compression_toolkit/core/pytorch/mixed_precision/configurable_weights_quantizer.py
+model_compression_toolkit/core/pytorch/pruning/__init__.py
+model_compression_toolkit/core/pytorch/pruning/pruning_pytorch_implementation.py
 model_compression_toolkit/core/pytorch/quantizer/__init__.py
 model_compression_toolkit/core/pytorch/quantizer/fake_quant_builder.py
 model_compression_toolkit/core/pytorch/quantizer/lut_fake_quant.py
@@ -382,6 +382,8 @@ model_compression_toolkit/gptq/pytorch/quantizer/ste_rounding/symmetric_ste.py
 model_compression_toolkit/pruning/__init__.py
 model_compression_toolkit/pruning/keras/__init__.py
 model_compression_toolkit/pruning/keras/pruning_facade.py
+model_compression_toolkit/pruning/pytorch/__init__.py
+model_compression_toolkit/pruning/pytorch/pruning_facade.py
 model_compression_toolkit/ptq/__init__.py
 model_compression_toolkit/ptq/runner.py
 model_compression_toolkit/ptq/keras/__init__.py

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/__init__.py

@@ -22,6 +22,6 @@ from model_compression_toolkit.core.common.mixed_precision import mixed_precisio
 from model_compression_toolkit.core.common.quantization.quantization_config import QuantizationConfig, QuantizationErrorMethod, DEFAULTCONFIG
 from model_compression_toolkit.core.common.quantization.core_config import CoreConfig
 from model_compression_toolkit.core.common.mixed_precision.kpi_tools.kpi import KPI
-from model_compression_toolkit.core.common.mixed_precision.mixed_precision_quantization_config import MixedPrecisionQuantizationConfig, MixedPrecisionQuantizationConfigV2
+from model_compression_toolkit.core.common.mixed_precision.mixed_precision_quantization_config import MixedPrecisionQuantizationConfig
 from model_compression_toolkit.core.keras.kpi_data_facade import keras_kpi_data
 from model_compression_toolkit.core.pytorch.kpi_data_facade import pytorch_kpi_data

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/common/framework_implementation.py

@@ -18,7 +18,7 @@ from typing import Callable, Any, List, Tuple, Dict
 import numpy as np
 
 from model_compression_toolkit.constants import HESSIAN_NUM_ITERATIONS
-from model_compression_toolkit.core import MixedPrecisionQuantizationConfigV2
+from model_compression_toolkit.core import MixedPrecisionQuantizationConfig
 from model_compression_toolkit.core import common
 from model_compression_toolkit.core.common import BaseNode
 from model_compression_toolkit.core.common.collectors.statistics_collector import BaseStatsCollector
@@ -304,7 +304,7 @@ class FrameworkImplementation(ABC):
     @abstractmethod
     def get_sensitivity_evaluator(self,
                                   graph: Graph,
-                                  quant_config: MixedPrecisionQuantizationConfigV2,
+                                  quant_config: MixedPrecisionQuantizationConfig,
                                   representative_data_gen: Callable,
                                   fw_info: FrameworkInfo,
                                   hessian_info_service: HessianInfoService = None,

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/common/mixed_precision/mixed_precision_quantization_config.py

@@ -13,16 +13,13 @@
 # limitations under the License.
 # ==============================================================================
 
-from enum import Enum
-from typing import List, Callable, Tuple
+from typing import List, Callable
 
-from model_compression_toolkit.logger import Logger
 from model_compression_toolkit.core.common.mixed_precision.distance_weighting import get_average_weights
-from model_compression_toolkit.core.common.quantization.quantization_config import QuantizationConfig, DEFAULTCONFIG
 from model_compression_toolkit.core.common.similarity_analyzer import compute_mse
 
 
-class MixedPrecisionQuantizationConfigV2:
+class MixedPrecisionQuantizationConfig:
 
     def __init__(self,
                  compute_distance_fn: Callable = None,
@@ -36,8 +33,6 @@ class MixedPrecisionQuantizationConfigV2:
                  metric_normalization_threshold: float = 1e10):
         """
         Class with mixed precision parameters to quantize the input model.
-        Unlike QuantizationConfig, number of bits for quantization is a list of possible bit widths to
-        support mixed-precision model quantization.
 
         Args:
             compute_distance_fn (Callable): Function to compute a distance between two tensors.
@@ -70,67 +65,6 @@ class MixedPrecisionQuantizationConfigV2:
         self.metric_normalization_threshold = metric_normalization_threshold
 
 
-class MixedPrecisionQuantizationConfig(QuantizationConfig):
-
-    def __init__(self,
-                 qc: QuantizationConfig = DEFAULTCONFIG,
-                 compute_distance_fn: Callable = compute_mse,
-                 distance_weighting_method: Callable = get_average_weights,
-                 num_of_images: int = 32,
-                 configuration_overwrite: List[int] = None,
-                 num_interest_points_factor: float = 1.0):
-        """
-        Class to wrap all different parameters the library quantize the input model according to.
-        Unlike QuantizationConfig, number of bits for quantization is a list of possible bit widths to
-        support mixed-precision model quantization.
-
-        Args:
-            qc (QuantizationConfig): QuantizationConfig object containing parameters of how the model should be quantized.
-            compute_distance_fn (Callable): Function to compute a distance between two tensors.
-            distance_weighting_method (Callable): Function to use when weighting the distances among different layers when computing the sensitivity metric.
-            num_of_images (int): Number of images to use to evaluate the sensitivity of a mixed-precision model comparing to the float model.
-            configuration_overwrite (List[int]): A list of integers that enables overwrite of mixed precision with a predefined one.
-            num_interest_points_factor: A multiplication factor between zero and one (represents percentage) to reduce the number of interest points used to calculate the distance metric.
-
-        """
-
-        super().__init__(**qc.__dict__)
-        self.compute_distance_fn = compute_distance_fn
-        self.distance_weighting_method = distance_weighting_method
-        self.num_of_images = num_of_images
-        self.configuration_overwrite = configuration_overwrite
-
-        assert 0.0 < num_interest_points_factor <= 1.0, "num_interest_points_factor should represent a percentage of " \
-                                                        "the base set of interest points that are required to be " \
-                                                        "used for mixed-precision metric evaluation, " \
-                                                        "thus, it should be between 0 to 1"
-        self.num_interest_points_factor = num_interest_points_factor
-
-    def separate_configs(self) -> Tuple[QuantizationConfig, MixedPrecisionQuantizationConfigV2]:
-        """
-        A function to separate the old MixedPrecisionQuantizationConfig into QuantizationConfig
-        and MixedPrecisionQuantizationConfigV2
-
-        Returns: QuantizationConfig, MixedPrecisionQuantizationConfigV2
-
-        """
-        _dummy_quant_config = QuantizationConfig()
-        _dummy_mp_config_experimental = MixedPrecisionQuantizationConfigV2()
-        qc_dict = {}
-        mp_dict = {}
-        for k, v in self.__dict__.items():
-            if hasattr(_dummy_quant_config, k):
-                qc_dict.update({k: v})
-            elif hasattr(_dummy_mp_config_experimental, k):
-                mp_dict.update({k: v})
-            else:
-                Logger.error(f'Attribute "{k}" mismatch: exists in MixedPrecisionQuantizationConfig but not in '
-                             f'MixedPrecisionQuantizationConfigV2')  # pragma: no cover
-
-        return QuantizationConfig(**qc_dict), MixedPrecisionQuantizationConfigV2(**mp_dict)
-
-
 # Default quantization configuration the library use.
-DEFAULT_MIXEDPRECISION_CONFIG = MixedPrecisionQuantizationConfig(DEFAULTCONFIG,
-                                                                 compute_mse,
-                                                                 get_average_weights)
+DEFAULT_MIXEDPRECISION_CONFIG = MixedPrecisionQuantizationConfig(compute_distance_fn=compute_mse,
+                                                                 distance_weighting_method=get_average_weights)

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/common/mixed_precision/mixed_precision_search_facade.py

@@ -18,7 +18,7 @@ from enum import Enum
 import numpy as np
 from typing import List, Callable, Dict
 
-from model_compression_toolkit.core import MixedPrecisionQuantizationConfigV2
+from model_compression_toolkit.core import MixedPrecisionQuantizationConfig
 from model_compression_toolkit.core.common import Graph
 from model_compression_toolkit.core.common.hessian import HessianInfoService
 from model_compression_toolkit.core.common.mixed_precision.kpi_tools.kpi import KPI, KPITarget
@@ -48,7 +48,7 @@ def search_bit_width(graph_to_search_cfg: Graph,
                      fw_info: FrameworkInfo,
                      fw_impl: FrameworkImplementation,
                      target_kpi: KPI,
-                     mp_config: MixedPrecisionQuantizationConfigV2,
+                     mp_config: MixedPrecisionQuantizationConfig,
                      representative_data_gen: Callable,
                      search_method: BitWidthSearchMethod = BitWidthSearchMethod.INTEGER_PROGRAMMING,
                      hessian_info_service: HessianInfoService=None) -> List[int]:

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/common/mixed_precision/sensitivity_evaluation.py

@@ -18,7 +18,7 @@ import numpy as np
 from typing import Callable, Any, List, Tuple
 
 from model_compression_toolkit.constants import AXIS, HESSIAN_OUTPUT_ALPHA
-from model_compression_toolkit.core import FrameworkInfo, MixedPrecisionQuantizationConfigV2
+from model_compression_toolkit.core import FrameworkInfo, MixedPrecisionQuantizationConfig
 from model_compression_toolkit.core.common import Graph, BaseNode
 from model_compression_toolkit.core.common.graph.functional_node import FunctionalNode
 
@@ -37,7 +37,7 @@ class SensitivityEvaluation:
 
     def __init__(self,
                  graph: Graph,
-                 quant_config: MixedPrecisionQuantizationConfigV2,
+                 quant_config: MixedPrecisionQuantizationConfig,
                  representative_data_gen: Callable,
                  fw_info: FrameworkInfo,
                  fw_impl: Any,

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/common/pruning/memory_calculator.py

@@ -307,7 +307,25 @@ class MemoryCalculator:
             total_params += w.size
 
         # Adjust the total parameter count if padded channels are to be included.
-        num_oc = np.sum(output_mask) if output_mask is not None else node.output_shape[-1]
+        if output_mask is not None:
+            num_oc = np.sum(output_mask)
+        else:
+            # Get the node channel axis from framework info
+            channel_axis = self.fw_info.out_channel_axis_mapping.get(node.type)
+            if channel_axis is None:
+                Logger.error("Channel axis is not defined")
+
+            # Check if node.output_shape is a list of lists.
+            # In this case make sure all the out channels are the same value
+            if all(isinstance(sublist, list) for sublist in node.output_shape):
+                compare_value = node.output_shape[0][channel_axis]
+                if all(len(sublist) > channel_axis and sublist[channel_axis] == compare_value for sublist in node.output_shape):
+                    num_oc = compare_value
+                else:
+                    Logger.error("Number of out channels are not the same for all outputs of the node")
+            else:
+                num_oc = node.output_shape[channel_axis]
+
         if include_padded_channels:
             total_params = self.get_node_nparams_with_padded_channels(node, total_params, num_oc, node.get_simd())
 

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/common/quantization/core_config.py

@@ -14,7 +14,7 @@
 # ==============================================================================
 from model_compression_toolkit.core.common.quantization.quantization_config import QuantizationConfig
 from model_compression_toolkit.core.common.quantization.debug_config import DebugConfig
-from model_compression_toolkit.core.common.mixed_precision.mixed_precision_quantization_config import MixedPrecisionQuantizationConfigV2
+from model_compression_toolkit.core.common.mixed_precision.mixed_precision_quantization_config import MixedPrecisionQuantizationConfig
 
 
 class CoreConfig:
@@ -23,14 +23,14 @@ class CoreConfig:
     """
     def __init__(self,
                  quantization_config: QuantizationConfig = QuantizationConfig(),
-                 mixed_precision_config: MixedPrecisionQuantizationConfigV2 = None,
+                 mixed_precision_config: MixedPrecisionQuantizationConfig = None,
                  debug_config: DebugConfig = DebugConfig()
                  ):
         """
 
         Args:
             quantization_config (QuantizationConfig): Config for quantization.
-            mixed_precision_config (MixedPrecisionQuantizationConfigV2): Config for mixed precision quantization (optional, default=None).
+            mixed_precision_config (MixedPrecisionQuantizationConfig): Config for mixed precision quantization (optional, default=None).
             debug_config (DebugConfig): Config for debugging and editing the network quantization process.
         """
         self.quantization_config = quantization_config

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/common/quantization/quantization_fn_selection.py

@@ -17,7 +17,6 @@ from collections.abc import Callable
 from functools import partial
 
 from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
-from model_compression_toolkit.core.common.quantization.quantizers.kmeans_quantizer import kmeans_quantizer
 from model_compression_toolkit.core.common.quantization.quantizers.lut_kmeans_quantizer import lut_kmeans_quantizer
 from model_compression_toolkit.core.common.quantization.quantizers.uniform_quantizers import power_of_two_quantizer, \
     symmetric_quantizer, uniform_quantizer
@@ -40,8 +39,6 @@ def get_weights_quantization_fn(weights_quantization_method: QuantizationMethod)
         quantizer_fn = symmetric_quantizer
     elif weights_quantization_method == QuantizationMethod.UNIFORM:
         quantizer_fn = uniform_quantizer
-    elif weights_quantization_method == QuantizationMethod.KMEANS:
-        quantizer_fn = kmeans_quantizer
     elif weights_quantization_method in [QuantizationMethod.LUT_POT_QUANTIZER, QuantizationMethod.LUT_SYM_QUANTIZER]:
         quantizer_fn = lut_kmeans_quantizer
     else:

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/common/quantization/quantization_params_fn_selection.py

@@ -18,7 +18,6 @@ from functools import partial
 
 from model_compression_toolkit.logger import Logger
 from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
-from model_compression_toolkit.core.common.quantization.quantization_params_generation.kmeans_params import kmeans_tensor
 from model_compression_toolkit.core.common.quantization.quantization_params_generation.lut_kmeans_params import \
     lut_kmeans_tensor, lut_kmeans_histogram
 from model_compression_toolkit.core.common.quantization.quantization_params_generation.symmetric_selection import \
@@ -70,8 +69,6 @@ def get_weights_quantization_params_fn(weights_quantization_method: Quantization
         params_fn = symmetric_selection_tensor
     elif weights_quantization_method == QuantizationMethod.UNIFORM:
         params_fn = uniform_selection_tensor
-    elif weights_quantization_method == QuantizationMethod.KMEANS:
-        params_fn = kmeans_tensor
     elif weights_quantization_method == QuantizationMethod.LUT_POT_QUANTIZER:
         params_fn = partial(lut_kmeans_tensor, is_symmetric=False)
     elif weights_quantization_method == QuantizationMethod.LUT_SYM_QUANTIZER:

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/common/quantization/quantization_params_generation/__init__.py

@@ -14,7 +14,6 @@
 # ==============================================================================
 from model_compression_toolkit.core.common.quantization.quantization_params_generation.power_of_two_selection import power_of_two_no_clipping_selection_min_max, \
     power_of_two_selection_histogram, power_of_two_selection_tensor
-from model_compression_toolkit.core.common.quantization.quantization_params_generation.kmeans_params import kmeans_tensor
 from model_compression_toolkit.core.common.quantization.quantization_params_generation.lut_kmeans_params import lut_kmeans_tensor
 from model_compression_toolkit.core.common.quantization.quantization_params_generation.symmetric_selection import symmetric_no_clipping_selection_min_max
 from model_compression_toolkit.core.common.quantization.quantization_params_generation.uniform_selection import uniform_no_clipping_selection_min_max

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/keras/keras_implementation.py

@@ -57,7 +57,7 @@ else:
         Concatenate, Add
     from keras.layers.core import TFOpLambda
 
-from model_compression_toolkit.core import QuantizationConfig, FrameworkInfo, CoreConfig, MixedPrecisionQuantizationConfigV2
+from model_compression_toolkit.core import QuantizationConfig, FrameworkInfo, CoreConfig, MixedPrecisionQuantizationConfig
 from model_compression_toolkit.core import common
 from model_compression_toolkit.core.common import Graph, BaseNode
 from model_compression_toolkit.core.common.framework_implementation import FrameworkImplementation
@@ -355,7 +355,7 @@ class KerasImplementation(FrameworkImplementation):
 
     def get_sensitivity_evaluator(self,
                                   graph: Graph,
-                                  quant_config: MixedPrecisionQuantizationConfigV2,
+                                  quant_config: MixedPrecisionQuantizationConfig,
                                   representative_data_gen: Callable,
                                   fw_info: FrameworkInfo,
                                   disable_activation_for_metric: bool = False,

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/keras/kpi_data_facade.py

@@ -14,8 +14,7 @@
 # ==============================================================================
 
 from typing import Callable
-
-from model_compression_toolkit.core import CoreConfig, MixedPrecisionQuantizationConfigV2
+from model_compression_toolkit.core import MixedPrecisionQuantizationConfig, CoreConfig
 from model_compression_toolkit.core.common.mixed_precision.kpi_tools.kpi import KPI
 from model_compression_toolkit.logger import Logger
 from model_compression_toolkit.constants import TENSORFLOW
@@ -36,7 +35,7 @@ if FOUND_TF:
 
     def keras_kpi_data(in_model: Model,
                        representative_data_gen: Callable,
-                       core_config: CoreConfig,
+                       core_config: CoreConfig = CoreConfig(mixed_precision_config=MixedPrecisionQuantizationConfig()),
                        fw_info: FrameworkInfo = DEFAULT_KERAS_INFO,
                        target_platform_capabilities: TargetPlatformCapabilities = KERAS_DEFAULT_TPC) -> KPI:
         """
@@ -73,9 +72,9 @@ if FOUND_TF:
 
         """
 
-        if not isinstance(core_config.mixed_precision_config, MixedPrecisionQuantizationConfigV2):
-            Logger.error("KPI data computation can't be executed without MixedPrecisionQuantizationConfigV2 object."
-                         "Given quant_config is not of type MixedPrecisionQuantizationConfigV2.")
+        if not isinstance(core_config.mixed_precision_config, MixedPrecisionQuantizationConfig):
+            Logger.error("KPI data computation can't be executed without MixedPrecisionQuantizationConfig object."
+                         "Given quant_config is not of type MixedPrecisionQuantizationConfig.")
 
         fw_impl = KerasImplementation()
 

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/keras/pruning/pruning_keras_implementation.py

@@ -44,9 +44,9 @@ class PruningKerasImplementation(KerasImplementation, PruningFrameworkImplementa
         Prunes the entry node of a model in Keras.
 
         Args:
-            node: The entry node to be pruned.
-            output_mask: A numpy array representing the mask to be applied to the output channels.
-            fw_info: Framework-specific information object.
+            node (BaseNode): The entry node to be pruned.
+            output_mask (np.ndarray): A numpy array representing the mask to be applied to the output channels.
+            fw_info (FrameworkInfo): Framework-specific information object.
 
         """
         return _prune_keras_edge_node(node=node,
@@ -63,10 +63,10 @@ class PruningKerasImplementation(KerasImplementation, PruningFrameworkImplementa
         Prunes an intermediate node in a Keras model.
 
         Args:
-            node: The intermediate node to be pruned.
-            input_mask: A numpy array representing the mask to be applied to the input channels.
-            output_mask: A numpy array representing the mask to be applied to the output channels.
-            fw_info: Framework-specific information object.
+            node (BaseNode): The intermediate node to be pruned.
+            input_mask (np.ndarray): A numpy array representing the mask to be applied to the input channels.
+            output_mask (np.ndarray): A numpy array representing the mask to be applied to the output channels.
+            fw_info (FrameworkInfo): Framework-specific information object.
 
         """
         _edit_node_input_shape(input_mask, node)
@@ -85,9 +85,9 @@ class PruningKerasImplementation(KerasImplementation, PruningFrameworkImplementa
         Prunes the exit node of a model in Keras.
 
         Args:
-            node: The exit node to be pruned.
-            input_mask: A numpy array representing the mask to be applied to the input channels.
-            fw_info: Framework-specific information object.
+            node (BaseNode): The exit node to be pruned.
+            input_mask (np.ndarray): A numpy array representing the mask to be applied to the input channels.
+            fw_info (FrameworkInfo): Framework-specific information object.
 
         """
         return _prune_keras_edge_node(node=node,
@@ -100,10 +100,10 @@ class PruningKerasImplementation(KerasImplementation, PruningFrameworkImplementa
         Determines whether a node is an entry node in a Keras model.
 
         Args:
-            node: The node to be checked.
+            node (BaseNode): The node to be checked.
 
         Returns:
-            Boolean indicating if the node is an entry node.
+            bool: Boolean indicating if the node is an entry node.
         """
         return _is_keras_node_pruning_section_edge(node)
 
@@ -115,26 +115,26 @@ class PruningKerasImplementation(KerasImplementation, PruningFrameworkImplementa
         Determines whether a node is an exit node in a Keras model.
 
         Args:
-            node: The node to be checked.
-            corresponding_entry_node: The entry node of the pruning section that is checked.
-            fw_info: Framework-specific information object.
+            node (BaseNode): The node to be checked.
+            corresponding_entry_node (BaseNode): The entry node of the pruning section that is checked.
+            fw_info (FrameworkInfo): Framework-specific information object.
 
         Returns:
-            Boolean indicating if the node is an exit node.
+            bool: Boolean indicating if the node is an exit node.
         """
         return _is_keras_node_pruning_section_edge(node) and PruningSection.has_matching_channel_count(node,
                                                                                                        corresponding_entry_node,
                                                                                                        fw_info)
 
-    def is_node_intermediate_pruning_section(self, node) -> bool:
+    def is_node_intermediate_pruning_section(self, node: BaseNode) -> bool:
         """
         Determines whether a node is part of the intermediate section in the pruning process of a Keras model.
 
         Args:
-            node: The node to be checked.
+            node (BaseNode): The node to be checked.
 
         Returns:
-            Boolean indicating if the node is part of the intermediate pruning section.
+            bool: Boolean indicating if the node is part of the intermediate pruning section.
         """
         # Nodes that are not Conv2D, Conv2DTranspose, DepthwiseConv2D, or Dense are considered intermediate.
         return node.type not in [keras.layers.DepthwiseConv2D,

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/pytorch/constants.py

@@ -62,6 +62,9 @@ DIM = 'dim'
 IN_CHANNELS = 'in_channels'
 OUT_CHANNELS = 'out_channels'
 NUM_FEATURES = 'num_features'
+NUM_PARAMETERS = 'num_parameters'
+IN_FEATURES = 'in_features'
+OUT_FEATURES = 'out_features'
 
 # torch devices
 CUDA = 'cuda'

{mct-nightly-1.11.0.20240303.post423 → mct-nightly-1.11.0.20240305.post352}/model_compression_toolkit/core/pytorch/kpi_data_facade.py

@@ -22,7 +22,7 @@ from model_compression_toolkit.core.common.mixed_precision.kpi_tools.kpi import
 from model_compression_toolkit.core.common.framework_info import FrameworkInfo
 from model_compression_toolkit.core.common.mixed_precision.kpi_tools.kpi_data import compute_kpi_data
 from model_compression_toolkit.core.common.quantization.core_config import CoreConfig
-from model_compression_toolkit.core.common.mixed_precision.mixed_precision_quantization_config import MixedPrecisionQuantizationConfigV2
+from model_compression_toolkit.core.common.mixed_precision.mixed_precision_quantization_config import MixedPrecisionQuantizationConfig
 from model_compression_toolkit.constants import FOUND_TORCH
 
 if FOUND_TORCH:
@@ -38,7 +38,7 @@ if FOUND_TORCH:
 
     def pytorch_kpi_data(in_model: Module,
                          representative_data_gen: Callable,
-                         core_config: CoreConfig = CoreConfig(), # TODO: Why pytorch is initilized and keras not?
+                         core_config: CoreConfig = CoreConfig(mixed_precision_config=MixedPrecisionQuantizationConfig()),
                          fw_info: FrameworkInfo = DEFAULT_PYTORCH_INFO,
                          target_platform_capabilities: TargetPlatformCapabilities = PYTORCH_DEFAULT_TPC) -> KPI:
         """
@@ -75,9 +75,9 @@ if FOUND_TORCH:
 
         """
 
-        if not isinstance(core_config.mixed_precision_config, MixedPrecisionQuantizationConfigV2):
-            Logger.error("KPI data computation can't be executed without MixedPrecisionQuantizationConfigV2 object."
-                         "Given quant_config is not of type MixedPrecisionQuantizationConfigV2.")
+        if not isinstance(core_config.mixed_precision_config, MixedPrecisionQuantizationConfig):
+            Logger.error("KPI data computation can't be executed without MixedPrecisionQuantizationConfig object."
+                         "Given quant_config is not of type MixedPrecisionQuantizationConfig.")
 
         fw_impl = PytorchImplementation()
 

mct-nightly-1.11.0.20240305.post352/model_compression_toolkit/core/pytorch/pruning/__init__.py

@@ -0,0 +1,14 @@
+# Copyright 2024 Sony Semiconductor Israel, Inc. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================