mct-nightly 2.0.0.20240403.423__tar.gz → 2.0.0.20240405.428__tar.gz

{mct-nightly-2.0.0.20240403.423 → mct-nightly-2.0.0.20240405.428}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mct-nightly
-Version: 2.0.0.20240403.423
+Version: 2.0.0.20240405.428
 Summary: A Model Compression Toolkit for neural networks
 Home-page: UNKNOWN
 License: UNKNOWN

{mct-nightly-2.0.0.20240403.423 → mct-nightly-2.0.0.20240405.428}/mct_nightly.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mct-nightly
-Version: 2.0.0.20240403.423
+Version: 2.0.0.20240405.428
 Summary: A Model Compression Toolkit for neural networks
 Home-page: UNKNOWN
 License: UNKNOWN

{mct-nightly-2.0.0.20240403.423 → mct-nightly-2.0.0.20240405.428}/mct_nightly.egg-info/SOURCES.txt

@@ -13,7 +13,6 @@ model_compression_toolkit/defaultdict.py
 model_compression_toolkit/logger.py
 model_compression_toolkit/core/__init__.py
 model_compression_toolkit/core/analyzer.py
-model_compression_toolkit/core/exporter.py
 model_compression_toolkit/core/graph_prep_runner.py
 model_compression_toolkit/core/quantization_prep_runner.py
 model_compression_toolkit/core/runner.py
@@ -182,7 +181,6 @@ model_compression_toolkit/core/keras/graph_substitutions/substitutions/linear_co
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/matmul_substitution.py
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/multi_head_attention_decomposition.py
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/relu_bound_to_power_of_2.py
-model_compression_toolkit/core/keras/graph_substitutions/substitutions/remove_relu_upper_bound.py
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/residual_collapsing.py
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/scale_equalization.py
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/separableconv_decomposition.py

{mct-nightly-2.0.0.20240403.423 → mct-nightly-2.0.0.20240405.428}/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__ = "2.0.0.20240403.000423"
+__version__ = "2.0.0.20240405.000428"

{mct-nightly-2.0.0.20240403.423 → mct-nightly-2.0.0.20240405.428}/model_compression_toolkit/core/common/framework_implementation.py

@@ -250,15 +250,6 @@ class FrameworkImplementation(ABC):
         raise NotImplemented(f'{self.__class__.__name__} have to implement the '
                              f'framework\'s get_residual_collapsing_substitution method.')  # pragma: no cover
 
-    @abstractmethod
-    def get_substitutions_pre_build(self) -> List[common.BaseSubstitution]:
-        """
-
-        Returns: A list of the framework substitutions used before we build a quantized model.
-
-        """
-        raise NotImplemented(f'{self.__class__.__name__} have to implement the '
-                             f'framework\'s get_substitutions_pre_build method.')  # pragma: no cover
 
     @abstractmethod
     def get_substitutions_post_statistics_collection(self, quant_config: QuantizationConfig) -> List[

{mct-nightly-2.0.0.20240403.423 → mct-nightly-2.0.0.20240405.428}/model_compression_toolkit/core/keras/keras_implementation.py

@@ -83,8 +83,6 @@ from model_compression_toolkit.core.keras.graph_substitutions.substitutions.inpu
     InputScalingWithPad
 from model_compression_toolkit.core.keras.graph_substitutions.substitutions.relu_bound_to_power_of_2 import \
     ReLUBoundToPowerOfTwo
-from model_compression_toolkit.core.keras.graph_substitutions.substitutions.remove_relu_upper_bound import \
-    RemoveReLUUpperBound
 from model_compression_toolkit.core.keras.graph_substitutions.substitutions.multi_head_attention_decomposition import \
     MultiHeadAttentionDecomposition
 from model_compression_toolkit.core.keras.graph_substitutions.substitutions.scale_equalization import \
@@ -321,14 +319,6 @@ class KerasImplementation(FrameworkImplementation):
             substitutions_list.append(InputScalingWithPad())
         return substitutions_list
 
-    def get_substitutions_pre_build(self) -> List[common.BaseSubstitution]:
-        """
-
-        Returns: A list of the framework substitutions used before we build a quantized model.
-
-        """
-
-        return [RemoveReLUUpperBound()]
 
     def get_substitutions_virtual_weights_activation_coupling(self) -> List[common.BaseSubstitution]:
         """

{mct-nightly-2.0.0.20240403.423 → mct-nightly-2.0.0.20240405.428}/model_compression_toolkit/core/pytorch/pytorch_implementation.py

@@ -304,11 +304,6 @@ class PytorchImplementation(FrameworkImplementation):
             Logger.critical('Input scaling is currently not supported for Pytorch.')
         return substitutions_list
 
-    def get_substitutions_pre_build(self) -> List[common.BaseSubstitution]:
-        """
-        Returns: A list of the framework substitutions used before we build a quantized module.
-        """
-        return []
 
     def get_substitutions_virtual_weights_activation_coupling(self) -> List[common.BaseSubstitution]:
         """

{mct-nightly-2.0.0.20240403.423 → mct-nightly-2.0.0.20240405.428}/model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_onnx_pytorch_exporter.py

@@ -103,7 +103,7 @@ class FakelyQuantONNXPyTorchExporter(BasePyTorchExporter):
         with custom quantizers.
         """
 
-        for n, m in self.model.named_children():
+        for n, m in self.model.named_modules():
             if isinstance(m, PytorchActivationQuantizationHolder):
                 assert isinstance(m.activation_holder_quantizer, pytorch_quantizers.BasePyTorchInferableQuantizer)
                 m.activation_holder_quantizer.enable_custom_impl()

{mct-nightly-2.0.0.20240403.423 → mct-nightly-2.0.0.20240405.428}/model_compression_toolkit/gptq/keras/quantization_facade.py

@@ -23,12 +23,10 @@ from model_compression_toolkit.constants import TENSORFLOW, FOUND_TF
 from model_compression_toolkit.core.common.user_info import UserInformation
 from model_compression_toolkit.gptq.common.gptq_config import GradientPTQConfig
 from model_compression_toolkit.core.common.mixed_precision.resource_utilization_tools.resource_utilization import ResourceUtilization
-from model_compression_toolkit.core.common.framework_info import FrameworkInfo
 from model_compression_toolkit.core.common.mixed_precision.mixed_precision_quantization_config import MixedPrecisionQuantizationConfig
 from model_compression_toolkit.core import CoreConfig
 from model_compression_toolkit.core.runner import core_runner
 from model_compression_toolkit.gptq.runner import gptq_runner
-from model_compression_toolkit.core.exporter import export_model
 from model_compression_toolkit.core.analyzer import analyzer_model_quantization
 from model_compression_toolkit.target_platform_capabilities.target_platform.targetplatform2framework import TargetPlatformCapabilities
 

{mct-nightly-2.0.0.20240403.423 → mct-nightly-2.0.0.20240405.428}/model_compression_toolkit/gptq/pytorch/quantization_facade.py

@@ -25,7 +25,6 @@ from model_compression_toolkit.core.common.mixed_precision.resource_utilization_
 from model_compression_toolkit.core.runner import core_runner
 from model_compression_toolkit.gptq.keras.quantization_facade import GPTQ_MOMENTUM
 from model_compression_toolkit.gptq.runner import gptq_runner
-from model_compression_toolkit.core.exporter import export_model
 from model_compression_toolkit.core.analyzer import analyzer_model_quantization
 from model_compression_toolkit.core import CoreConfig
 from model_compression_toolkit.core.common.mixed_precision.mixed_precision_quantization_config import \

{mct-nightly-2.0.0.20240403.423 → mct-nightly-2.0.0.20240405.428}/model_compression_toolkit/ptq/keras/quantization_facade.py

@@ -24,7 +24,6 @@ from model_compression_toolkit.core.common.mixed_precision.resource_utilization_
 from model_compression_toolkit.core.common.mixed_precision.mixed_precision_quantization_config import \
     MixedPrecisionQuantizationConfig
 from model_compression_toolkit.target_platform_capabilities.target_platform.targetplatform2framework import TargetPlatformCapabilities
-from model_compression_toolkit.core.exporter import export_model
 from model_compression_toolkit.core.runner import core_runner
 from model_compression_toolkit.ptq.runner import ptq_runner
 

{mct-nightly-2.0.0.20240403.423 → mct-nightly-2.0.0.20240405.428}/model_compression_toolkit/ptq/pytorch/quantization_facade.py

@@ -25,7 +25,6 @@ from model_compression_toolkit.core.common.mixed_precision.mixed_precision_quant
     MixedPrecisionQuantizationConfig
 from model_compression_toolkit.core.runner import core_runner
 from model_compression_toolkit.ptq.runner import ptq_runner
-from model_compression_toolkit.core.exporter import export_model
 from model_compression_toolkit.core.analyzer import analyzer_model_quantization
 
 

mct-nightly-2.0.0.20240403.423/model_compression_toolkit/core/exporter.py

@@ -1,90 +0,0 @@
-# Copyright 2022 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.
-# ==============================================================================
-
-
-from typing import Tuple, Any
-
-from model_compression_toolkit.core.common.framework_implementation import FrameworkImplementation
-from model_compression_toolkit.core.common import FrameworkInfo
-from model_compression_toolkit.core.common.graph.base_graph import Graph
-from model_compression_toolkit.core.common.model_builder_mode import ModelBuilderMode
-from model_compression_toolkit.core.common.quantization.quantize_graph_weights import quantize_graph_weights
-
-from model_compression_toolkit.core.common.substitutions.apply_substitutions import substitute
-from model_compression_toolkit.core.common.user_info import UserInformation
-
-from model_compression_toolkit.core.common.visualization.tensorboard_writer import TensorboardWriter
-
-
-def _quantize_model(tg: Graph,
-                    fw_info: FrameworkInfo,
-                    fw_impl: FrameworkImplementation,
-                    tb_w: TensorboardWriter) -> Tuple[Any, UserInformation]:
-    """
-    Quantize graph's weights, and build a quantized framework model from it.
-
-    Args:
-        tg: A prepared for quantization graph.
-        fw_info: Information needed for quantization about the specific framework (e.g., kernel channels indices, groups of layers by how they should be quantized, etc.).
-        fw_impl: FrameworkImplementation object with a specific framework methods implementation.
-        tb_w: TensorBoardWriter object to log events.
-
-    Returns:
-        Quantized model in the input framework, and information the user may need in order to use the quantized model.
-    """
-
-    quantized_tg = quantize_graph_weights(tg)
-    if tb_w is not None:
-        tb_w.add_graph(quantized_tg, 'after_quantization')
-    ######################################
-    # Back2Framework
-    ######################################
-    # Before building a quantized model, first apply some substitutions.
-    quantized_tg = substitute(quantized_tg,
-                              fw_impl.get_substitutions_pre_build())
-
-    quantized_model, user_info = fw_impl.model_builder(quantized_tg,
-                                                       mode=ModelBuilderMode.QUANTIZED,
-                                                       fw_info=fw_info)
-    return quantized_model, user_info
-
-
-def export_model(tg,
-                 fw_info,
-                 fw_impl,
-                 tb_w,
-                 bit_widths_config):
-
-    """
-    A function for quantizing the graph's weights and build a quantized framework model from it.
-
-    Args:
-        tg: A prepared for quantization graph.
-        fw_info: Information needed for quantization about the specific framework (e.g., kernel channels indices, groups of layers by how they should be quantized, etc.).
-        fw_impl: FrameworkImplementation object with a specific framework methods implementation.
-        tb_w: TensorBoardWriter object to log events.
-        bit_widths_config: mixed-precision bit configuration to be added to model user_info
-
-    Returns:
-        Quantized model in the input framework, and information the user may need in order to use the quantized model.
-    """
-    quantized_model, user_info = _quantize_model(tg,
-                                                 fw_info,
-                                                 fw_impl,
-                                                 tb_w)
-    user_info.mixed_precision_cfg = bit_widths_config
-
-    return quantized_model, user_info
-

mct-nightly-2.0.0.20240403.423/model_compression_toolkit/core/keras/graph_substitutions/substitutions/remove_relu_upper_bound.py

@@ -1,61 +0,0 @@
-# Copyright 2021 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.
-# ==============================================================================
-
-
-from tensorflow.keras.layers import ReLU
-
-from model_compression_toolkit.core import common
-from model_compression_toolkit.core.common import Graph, BaseNode
-from model_compression_toolkit.core.common.graph.graph_matchers import NodeOperationMatcher,NodeFrameworkAttrMatcher
-from model_compression_toolkit.core.keras.constants import RELU_MAX_VALUE
-from model_compression_toolkit.constants import THRESHOLD
-from model_compression_toolkit.logger import Logger
-
-MATCHER = NodeOperationMatcher(ReLU) & NodeFrameworkAttrMatcher(RELU_MAX_VALUE, None).logic_not()
-
-
-class RemoveReLUUpperBound(common.BaseSubstitution):
-    """
-    Remove ReLU upper bound if its activation threshold bounds it anyway at
-    the same value.
-    """
-
-
-    def __init__(self):
-        """
-        Initialize a RemoveReLUUpperBound object.
-        """
-        super().__init__(matcher_instance=MATCHER)
-
-    def substitute(self,
-                   graph: Graph,
-                   node: BaseNode) -> Graph:
-        """
-        Remove ReLU upper bound if its activation threshold bounds it anyway at
-        the same value.
-
-        Args:
-            graph: Graph we apply the substitution on.
-            node: Node to remove its bound.
-
-        Returns:
-            Graph after applying the substitution.
-        """
-        if node.final_activation_quantization_cfg and \
-                node.final_activation_quantization_cfg.activation_quantization_params.get(THRESHOLD) == \
-                node.framework_attr.get(RELU_MAX_VALUE):
-            node.framework_attr[RELU_MAX_VALUE] = None
-            Logger.info(f'Removing upper bound of {node.name}. Threshold and upper bound are equal.')
-        return graph