mct-nightly 2.2.0.20250113.134913__py3-none-any.whl → 2.2.0.20250114.134534__py3-none-any.whl

model_compression_toolkit/core/common/quantization/quantization_config.py

@@ -25,7 +25,7 @@ from model_compression_toolkit.constants import MIN_THRESHOLD
 class CustomOpsetLayers(NamedTuple):
     """
     This struct defines a set of operators from a specific framework, which will be used to configure a custom operator
-    set in the TPC.
+    set in the FQC.
 
     Args:
         operators: a list of framework operators to map to a certain custom opset name.

model_compression_toolkit/core/common/quantization/quantization_fn_selection.py

@@ -16,8 +16,8 @@
 from collections.abc import Callable
 from functools import partial
 
+from mct_quantizers import QuantizationMethod
 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.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

model_compression_toolkit/core/common/quantization/quantization_params_fn_selection.py

@@ -16,8 +16,8 @@
 from collections.abc import Callable
 from functools import partial
 
+from mct_quantizers import QuantizationMethod
 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.lut_kmeans_params import \
     lut_kmeans_tensor, lut_kmeans_histogram
 from model_compression_toolkit.core.common.quantization.quantization_params_generation.symmetric_selection import \

model_compression_toolkit/core/common/quantization/quantization_params_generation/error_functions.py

@@ -16,11 +16,11 @@ from copy import deepcopy
 from typing import Tuple, Callable, List, Iterable, Optional
 import numpy as np
 import model_compression_toolkit.core.common.quantization.quantization_config as qc
+from mct_quantizers import QuantizationMethod
 from model_compression_toolkit.core.common.hessian import HessianScoresRequest, HessianMode, HessianScoresGranularity, \
     HessianInfoService
 from model_compression_toolkit.core.common.similarity_analyzer import compute_mse, compute_mae, compute_lp_norm
 from model_compression_toolkit.logger import Logger
-from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
 from model_compression_toolkit.constants import FLOAT_32, NUM_QPARAM_HESSIAN_SAMPLES
 from model_compression_toolkit.core.common.quantization.quantizers.quantizers_helpers import uniform_quantize_tensor, \
     reshape_tensor_for_per_channel_search

model_compression_toolkit/core/common/quantization/quantization_params_generation/power_of_two_selection.py

@@ -16,6 +16,7 @@ import numpy as np
 from typing import Union, Tuple, Dict
 
 import model_compression_toolkit.core.common.quantization.quantization_config as qc
+from mct_quantizers import QuantizationMethod
 from model_compression_toolkit.constants import MIN_THRESHOLD, THRESHOLD, NUM_QPARAM_HESSIAN_SAMPLES, SIGNED
 from model_compression_toolkit.core.common.hessian import HessianInfoService
 from model_compression_toolkit.core.common.quantization.quantization_params_generation.qparams_search import \
@@ -23,7 +24,6 @@ from model_compression_toolkit.core.common.quantization.quantization_params_gene
 from model_compression_toolkit.core.common.quantization.quantizers.quantizers_helpers import max_power_of_two, get_tensor_max
 from model_compression_toolkit.core.common.quantization.quantization_params_generation.error_functions import \
     get_threshold_selection_tensor_error_function, get_threshold_selection_histogram_error_function
-from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
 from model_compression_toolkit.core.common.similarity_analyzer import compute_mse
 from model_compression_toolkit.core.common.quantization.quantizers.quantizers_helpers import quantize_tensor
 

model_compression_toolkit/core/common/quantization/quantization_params_generation/qparams_activations_computation.py

@@ -15,7 +15,7 @@
 import numpy as np
 from typing import Dict, Union
 
-from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
+from mct_quantizers import QuantizationMethod
 from model_compression_toolkit.target_platform_capabilities.schema.mct_current_schema import Signedness
 from model_compression_toolkit.core.common.collectors.statistics_collector import BaseStatsCollector
 from model_compression_toolkit.core.common.quantization import quantization_params_generation

model_compression_toolkit/core/common/quantization/quantization_params_generation/symmetric_selection.py

@@ -25,7 +25,7 @@ from model_compression_toolkit.core.common.quantization.quantization_params_gene
     qparams_symmetric_selection_histogram_search, kl_qparams_symmetric_selection_histogram_search
 from model_compression_toolkit.core.common.quantization.quantizers.quantizers_helpers import \
     get_tensor_max
-from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
+from mct_quantizers import QuantizationMethod
 from model_compression_toolkit.core.common.similarity_analyzer import compute_mse
 from model_compression_toolkit.core.common.quantization.quantizers.quantizers_helpers import quantize_tensor
 

model_compression_toolkit/core/common/quantization/quantization_params_generation/uniform_selection.py

@@ -24,7 +24,7 @@ from model_compression_toolkit.core.common.quantization.quantization_params_gene
     get_threshold_selection_tensor_error_function, get_threshold_selection_histogram_error_function
 from model_compression_toolkit.core.common.quantization.quantizers.quantizers_helpers import get_tensor_max, \
     get_tensor_min
-from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
+from mct_quantizers import QuantizationMethod
 from model_compression_toolkit.core.common.similarity_analyzer import compute_mse
 from model_compression_toolkit.core.common.quantization.quantizers.quantizers_helpers import uniform_quantize_tensor
 

model_compression_toolkit/core/common/quantization/set_node_quantization_config.py

@@ -33,9 +33,10 @@ from model_compression_toolkit.core.common.quantization.quantization_params_fn_s
 from model_compression_toolkit.core.common.quantization.quantization_fn_selection import \
     get_weights_quantization_fn
 from model_compression_toolkit.target_platform_capabilities.schema.schema_functions import max_input_activation_n_bits
-from model_compression_toolkit.target_platform_capabilities.target_platform.targetplatform2framework import TargetPlatformCapabilities
 from model_compression_toolkit.target_platform_capabilities.schema.mct_current_schema import OpQuantizationConfig, \
     QuantizationConfigOptions
+from model_compression_toolkit.target_platform_capabilities.targetplatform2framework.framework_quantization_capabilities import \
+    FrameworkQuantizationCapabilities
 
 
 def set_quantization_configuration_to_graph(graph: Graph,
@@ -71,14 +72,14 @@ def set_quantization_configuration_to_graph(graph: Graph,
                                          graph=graph,
                                          quant_config=quant_config,
                                          fw_info=graph.fw_info,
-                                         tpc=graph.tpc,
+                                         fqc=graph.fqc,
                                          mixed_precision_enable=mixed_precision_enable,
                                          manual_bit_width_override=nodes_to_manipulate_bit_widths.get(n))
     return graph
 
 
 def filter_node_qco_by_graph(node: BaseNode,
-                             tpc: TargetPlatformCapabilities,
+                             fqc: FrameworkQuantizationCapabilities,
                              graph: Graph,
                              node_qc_options: QuantizationConfigOptions
                              ) -> Tuple[OpQuantizationConfig, List[OpQuantizationConfig]]:
@@ -90,7 +91,7 @@ def filter_node_qco_by_graph(node: BaseNode,
 
     Args:
         node: Node for filtering.
-        tpc: TPC to extract the QuantizationConfigOptions for the next nodes.
+        fqc: FQC to extract the QuantizationConfigOptions for the next nodes.
         graph: Graph object.
         node_qc_options: Node's QuantizationConfigOptions.
 
@@ -108,7 +109,7 @@ def filter_node_qco_by_graph(node: BaseNode,
     next_nodes = []
     while len(_next_nodes):
         n = _next_nodes.pop(0)
-        qco = n.get_qco(tpc)
+        qco = n.get_qco(fqc)
         qp = [qc.quantization_preserving for qc in qco.quantization_configurations]
         if not all(qp) and any(qp):
             Logger.error(f'Attribute "quantization_preserving" should be the same for all QuantizaionConfigOptions in {n}.')
@@ -117,7 +118,7 @@ def filter_node_qco_by_graph(node: BaseNode,
         next_nodes.append(n)
 
     if len(next_nodes):
-        next_nodes_qc_options = [_node.get_qco(tpc) for _node in next_nodes]
+        next_nodes_qc_options = [_node.get_qco(fqc) for _node in next_nodes]
         next_nodes_supported_input_bitwidth = min([max_input_activation_n_bits(op_cfg)
                                                    for qc_opts in next_nodes_qc_options
                                                    for op_cfg in qc_opts.quantization_configurations])
@@ -126,7 +127,7 @@ def filter_node_qco_by_graph(node: BaseNode,
         _node_qc_options = [_option for _option in _node_qc_options
                             if _option.activation_n_bits <= next_nodes_supported_input_bitwidth]
         if len(_node_qc_options) == 0:
-            Logger.critical(f"Graph doesn't match TPC bit configurations: {node} -> {next_nodes}.")
+            Logger.critical(f"Graph doesn't match FQC bit configurations: {node} -> {next_nodes}.")
 
         # Verify base config match
         if any([node_qc_options.base_config.activation_n_bits > max_input_activation_n_bits(qc_opt.base_config)
@@ -136,9 +137,9 @@ def filter_node_qco_by_graph(node: BaseNode,
             if len(_node_qc_options) > 0:
                 output_act_bitwidth = {qco.activation_n_bits: i for i, qco in enumerate(_node_qc_options)}
                 _base_config = _node_qc_options[output_act_bitwidth[max(output_act_bitwidth)]]
-                Logger.warning(f"Node {node} base quantization config changed to match Graph and TPC configuration.\nCause: {node} -> {next_nodes}.")
+                Logger.warning(f"Node {node} base quantization config changed to match Graph and FQC configuration.\nCause: {node} -> {next_nodes}.")
             else:
-                Logger.critical(f"Graph doesn't match TPC bit configurations: {node} -> {next_nodes}.")  # pragma: no cover
+                Logger.critical(f"Graph doesn't match FQC bit configurations: {node} -> {next_nodes}.")  # pragma: no cover
 
     return _base_config, _node_qc_options
 
@@ -147,7 +148,7 @@ def set_quantization_configs_to_node(node: BaseNode,
                                      graph: Graph,
                                      quant_config: QuantizationConfig,
                                      fw_info: FrameworkInfo,
-                                     tpc: TargetPlatformCapabilities,
+                                     fqc: FrameworkQuantizationCapabilities,
                                      mixed_precision_enable: bool = False,
                                      manual_bit_width_override: Optional[int] = None):
     """
@@ -158,12 +159,12 @@ def set_quantization_configs_to_node(node: BaseNode,
         graph (Graph): Model's internal representation graph.
         quant_config (QuantizationConfig): Quantization configuration to generate the node's configurations from.
         fw_info (FrameworkInfo): Information needed for quantization about the specific framework.
-        tpc (TargetPlatformCapabilities): TargetPlatformCapabilities to get default OpQuantizationConfig.
+        fqc (FrameworkQuantizationCapabilities): FrameworkQuantizationCapabilities to get default OpQuantizationConfig.
         mixed_precision_enable (bool): Whether mixed precision is enabled. Defaults to False.
         manual_bit_width_override (Optional[int]): Specifies a custom bit-width to override the node's activation bit-width. Defaults to None.
     """
-    node_qc_options = node.get_qco(tpc)
-    base_config, node_qc_options_list = filter_node_qco_by_graph(node, tpc, graph, node_qc_options)
+    node_qc_options = node.get_qco(fqc)
+    base_config, node_qc_options_list = filter_node_qco_by_graph(node, fqc, graph, node_qc_options)
 
     # If a manual_bit_width_override is given, filter node_qc_options_list to retain only the options with activation bits equal to manual_bit_width_override,
     # and update base_config accordingly.
@@ -257,7 +258,7 @@ def _create_node_single_candidate_qc(qc: QuantizationConfig,
     attrs_with_enabled_quantization = [attr for attr, cfg in op_cfg.attr_weights_configs_mapping.items()
                                        if cfg.enable_weights_quantization]
     if len(attrs_with_enabled_quantization) > 1:
-        Logger.warning(f"Multiple weights attributes quantization is enabled via the provided TPC."
+        Logger.warning(f"Multiple weights attributes quantization is enabled via the provided FQC."
                        f"Quantizing any attribute other than the kernel is experimental "
                        f"and may be subject to unstable behavior."
                        f"Attributes with enabled weights quantization: {attrs_with_enabled_quantization}.")

model_compression_toolkit/core/common/substitutions/batchnorm_reconstruction.py

@@ -26,7 +26,7 @@ from model_compression_toolkit.logger import Logger
 from model_compression_toolkit.core.common.graph.base_graph import Graph
 from model_compression_toolkit.core.common.graph.base_node import BaseNode
 from model_compression_toolkit.core.common.graph.graph_matchers import NodeOperationMatcher
-from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
+from mct_quantizers import QuantizationMethod
 from model_compression_toolkit.target_platform_capabilities.schema.mct_current_schema import AttributeQuantizationConfig
 
 

model_compression_toolkit/core/common/substitutions/batchnorm_refusing.py

@@ -22,7 +22,7 @@ from model_compression_toolkit.core import common
 from model_compression_toolkit.core.common.graph.base_graph import Graph
 from model_compression_toolkit.core.common.graph.graph_matchers import EdgeMatcher, NodeOperationMatcher
 from model_compression_toolkit.core.common.graph.base_node import BaseNode
-from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
+from mct_quantizers import QuantizationMethod
 from model_compression_toolkit.constants import THRESHOLD, RANGE_MIN, RANGE_MAX
 from model_compression_toolkit.logger import Logger
 

model_compression_toolkit/core/common/substitutions/shift_negative_activation.py

@@ -22,7 +22,7 @@ from model_compression_toolkit.logger import Logger
 from model_compression_toolkit.core.common import FrameworkInfo, Graph, BaseNode
 from model_compression_toolkit.constants import THRESHOLD, SIGNED, SHIFT_NEGATIVE_NON_LINEAR_NUM_BITS
 from model_compression_toolkit.core.common.graph.graph_matchers import NodeOperationMatcher
-from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
+from mct_quantizers import QuantizationMethod
 from model_compression_toolkit.target_platform_capabilities.schema.mct_current_schema import AttributeQuantizationConfig
 from model_compression_toolkit.core.common.quantization.set_node_quantization_config import create_node_activation_qc, \
     set_quantization_configs_to_node
@@ -359,7 +359,7 @@ def shift_negative_function(graph: Graph,
                                          node=pad_node,
                                          graph=graph,
                                          quant_config=core_config.quantization_config,
-                                         tpc=graph.tpc,
+                                         fqc=graph.fqc,
                                          mixed_precision_enable=core_config.is_mixed_precision_enabled)
 
         for candidate_qc in pad_node.candidates_quantization_cfg:
@@ -376,7 +376,7 @@ def shift_negative_function(graph: Graph,
                                      node=add_node,
                                      graph=graph,
                                      quant_config=core_config.quantization_config,
-                                     tpc=graph.tpc,
+                                     fqc=graph.fqc,
                                      mixed_precision_enable=core_config.is_mixed_precision_enabled)
 
     original_non_linear_activation_nbits = non_linear_node_cfg_candidate.activation_n_bits
@@ -392,7 +392,7 @@ def shift_negative_function(graph: Graph,
                     bypass_candidate_qc.activation_quantization_cfg.activation_quantization_params[SIGNED] = False
                     graph.shift_stats_collector(bypass_node, np.array(shift_value))
 
-    add_node_qco = add_node.get_qco(graph.tpc).quantization_configurations
+    add_node_qco = add_node.get_qco(graph.fqc).quantization_configurations
     for op_qc_idx, candidate_qc in enumerate(add_node.candidates_quantization_cfg):
         for attr in add_node.get_node_weights_attributes():
             candidate_qc.weights_quantization_cfg.get_attr_config(attr).enable_weights_quantization = False
@@ -533,7 +533,7 @@ def apply_shift_negative_correction(graph: Graph,
     nodes = list(graph.nodes())
     for n in nodes:
         # Skip substitution if QuantizationMethod is uniform.
-        node_qco = n.get_qco(graph.tpc)
+        node_qco = n.get_qco(graph.fqc)
         if any([op_qc.activation_quantization_method is QuantizationMethod.UNIFORM
                 for op_qc in node_qco.quantization_configurations]):
             continue

model_compression_toolkit/core/graph_prep_runner.py

@@ -29,8 +29,9 @@ from model_compression_toolkit.core.common.quantization.set_node_quantization_co
 from model_compression_toolkit.core.common.substitutions.apply_substitutions import substitute
 from model_compression_toolkit.core.common.substitutions.linear_collapsing_substitution import \
     linear_collapsing_substitute
-from model_compression_toolkit.target_platform_capabilities.target_platform.targetplatform2framework import TargetPlatformCapabilities
 from model_compression_toolkit.core.common.visualization.tensorboard_writer import TensorboardWriter
+from model_compression_toolkit.target_platform_capabilities.targetplatform2framework.framework_quantization_capabilities import \
+    FrameworkQuantizationCapabilities
 
 
 def graph_preparation_runner(in_model: Any,
@@ -38,7 +39,7 @@ def graph_preparation_runner(in_model: Any,
                              quantization_config: QuantizationConfig,
                              fw_info: FrameworkInfo,
                              fw_impl: FrameworkImplementation,
-                             tpc: TargetPlatformCapabilities,
+                             fqc: FrameworkQuantizationCapabilities,
                              bit_width_config: BitWidthConfig = None,
                              tb_w: TensorboardWriter = None,
                              mixed_precision_enable: bool = False,
@@ -58,7 +59,7 @@ def graph_preparation_runner(in_model: Any,
         fw_info (FrameworkInfo): 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): FrameworkImplementation object with a specific framework methods implementation.
-        tpc (TargetPlatformCapabilities): TargetPlatformCapabilities object that models the inference target platform and
+        fqc (FrameworkQuantizationCapabilities): FrameworkQuantizationCapabilities object that models the inference target platform and
             the attached framework operator's information.
         bit_width_config (BitWidthConfig): Config for bit-width selection. Defaults to None.
         tb_w (TensorboardWriter): TensorboardWriter object for logging.
@@ -71,7 +72,7 @@ def graph_preparation_runner(in_model: Any,
 
     graph = read_model_to_graph(in_model,
                                 representative_data_gen,
-                                tpc,
+                                fqc,
                                 fw_info,
                                 fw_impl)
 
@@ -79,7 +80,7 @@ def graph_preparation_runner(in_model: Any,
         tb_w.add_graph(graph, 'initial_graph')
 
     transformed_graph = get_finalized_graph(graph,
-                                            tpc,
+                                            fqc,
                                             quantization_config,
                                             bit_width_config,
                                             fw_info,
@@ -92,7 +93,7 @@ def graph_preparation_runner(in_model: Any,
 
 
 def get_finalized_graph(initial_graph: Graph,
-                        tpc: TargetPlatformCapabilities,
+                        fqc: FrameworkQuantizationCapabilities,
                         quant_config: QuantizationConfig = DEFAULTCONFIG,
                         bit_width_config: BitWidthConfig = None,
                         fw_info: FrameworkInfo = None,
@@ -106,7 +107,7 @@ def get_finalized_graph(initial_graph: Graph,
 
     Args:
         initial_graph (Graph): Graph to apply the changes to.
-        tpc (TargetPlatformCapabilities): TargetPlatformCapabilities object that describes the desired inference target platform (includes fusing patterns MCT should handle).
+        fqc (FrameworkQuantizationCapabilities): FrameworkQuantizationCapabilities object that describes the desired inference target platform (includes fusing patterns MCT should handle).
         quant_config (QuantizationConfig): QuantizationConfig containing parameters of how the model should be
             quantized.
         bit_width_config (BitWidthConfig): Config for bit-width selection. Defaults to None.
@@ -160,7 +161,7 @@ def get_finalized_graph(initial_graph: Graph,
     ######################################
     # Layer fusing
     ######################################
-    transformed_graph = fusion(transformed_graph, tpc)
+    transformed_graph = fusion(transformed_graph, fqc)
 
     ######################################
     # Channel equalization
@@ -185,7 +186,7 @@ def get_finalized_graph(initial_graph: Graph,
 
 def read_model_to_graph(in_model: Any,
                         representative_data_gen: Callable,
-                        tpc: TargetPlatformCapabilities,
+                        fqc: FrameworkQuantizationCapabilities,
                         fw_info: FrameworkInfo = None,
                         fw_impl: FrameworkImplementation = None) -> Graph:
 
@@ -195,7 +196,7 @@ def read_model_to_graph(in_model: Any,
     Args:
         in_model: Model to optimize and prepare for quantization.
         representative_data_gen: Dataset used for calibration.
-        tpc: TargetPlatformCapabilities object that models the inference target platform and
+        fqc: FrameworkQuantizationCapabilities object that models the inference target platform and
                       the attached framework operator's information.
         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.)
@@ -207,5 +208,5 @@ def read_model_to_graph(in_model: Any,
     graph = fw_impl.model_reader(in_model,
                                  representative_data_gen)
     graph.set_fw_info(fw_info)
-    graph.set_tpc(tpc)
+    graph.set_fqc(fqc)
     return graph

model_compression_toolkit/core/keras/default_framework_info.py

@@ -26,7 +26,7 @@ else:
 
 from model_compression_toolkit.defaultdict import DefaultDict
 from model_compression_toolkit.core.common.framework_info import FrameworkInfo
-from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
+from mct_quantizers import QuantizationMethod
 from model_compression_toolkit.constants import SOFTMAX_THRESHOLD
 from model_compression_toolkit.core.keras.constants import SOFTMAX, LINEAR, RELU, SWISH, SIGMOID, IDENTITY, TANH, SELU, \
     KERNEL, DEPTHWISE_KERNEL, GELU

model_compression_toolkit/core/keras/mixed_precision/configurable_weights_quantizer.py

@@ -20,8 +20,7 @@ from model_compression_toolkit.core.common.mixed_precision.configurable_quantize
 from model_compression_toolkit.core.common.quantization.candidate_node_quantization_config import \
     CandidateNodeQuantizationConfig
 from model_compression_toolkit.logger import Logger
-from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
-from mct_quantizers import QuantizationTarget
+from mct_quantizers import QuantizationTarget, QuantizationMethod
 from mct_quantizers import mark_quantizer
 
 import tensorflow as tf

model_compression_toolkit/core/keras/resource_utilization_data_facade.py

@@ -18,18 +18,17 @@ from model_compression_toolkit.core import MixedPrecisionQuantizationConfig, Cor
 from model_compression_toolkit.core.common.mixed_precision.resource_utilization_tools.resource_utilization import ResourceUtilization
 from model_compression_toolkit.logger import Logger
 from model_compression_toolkit.constants import TENSORFLOW
-from model_compression_toolkit.target_platform_capabilities.schema.mct_current_schema import TargetPlatformModel
-from model_compression_toolkit.target_platform_capabilities.target_platform import TargetPlatformCapabilities
+from model_compression_toolkit.target_platform_capabilities.schema.mct_current_schema import TargetPlatformCapabilities
 from model_compression_toolkit.core.common.mixed_precision.resource_utilization_tools.resource_utilization_data import compute_resource_utilization_data
 from model_compression_toolkit.verify_packages import FOUND_TF
 
 if FOUND_TF:
+    from model_compression_toolkit.target_platform_capabilities.targetplatform2framework.attach2keras import \
+        AttachTpcToKeras
     from model_compression_toolkit.target_platform_capabilities.constants import DEFAULT_TP_MODEL
     from model_compression_toolkit.core.keras.default_framework_info import DEFAULT_KERAS_INFO
     from model_compression_toolkit.core.keras.keras_implementation import KerasImplementation
     from tensorflow.keras.models import Model
-    from model_compression_toolkit.target_platform_capabilities.target_platform.targetplatform2framework.attach2keras import \
-        AttachTpcToKeras
 
     from model_compression_toolkit import get_target_platform_capabilities
 
@@ -39,7 +38,7 @@ if FOUND_TF:
                                         representative_data_gen: Callable,
                                         core_config: CoreConfig = CoreConfig(
                                             mixed_precision_config=MixedPrecisionQuantizationConfig()),
-                                        target_platform_capabilities: TargetPlatformModel = KERAS_DEFAULT_TPC
+                                        target_platform_capabilities: TargetPlatformCapabilities = KERAS_DEFAULT_TPC
                                         ) -> ResourceUtilization:
         """
         Computes resource utilization data that can be used to calculate the desired target resource utilization
@@ -51,7 +50,7 @@ if FOUND_TF:
             in_model (Model): Keras model to quantize.
             representative_data_gen (Callable): Dataset used for calibration.
             core_config (CoreConfig): CoreConfig containing parameters for quantization and mixed precision of how the model should be quantized.
-            target_platform_capabilities (TargetPlatformCapabilities): TargetPlatformCapabilities to optimize the Keras model according to.
+            target_platform_capabilities (FrameworkQuantizationCapabilities): FrameworkQuantizationCapabilities to optimize the Keras model according to.
 
         Returns:
 

model_compression_toolkit/core/pytorch/back2framework/pytorch_model_builder.py

@@ -225,7 +225,7 @@ class PytorchModel(torch.nn.Module):
         """
         super(PytorchModel, self).__init__()
         self.graph = copy.deepcopy(graph)
-        delattr(self.graph, 'tpc')
+        delattr(self.graph, 'fqc')
 
         self.node_sort = list(topological_sort(self.graph))
         self.node_to_activation_quantization_holder = {}

model_compression_toolkit/core/pytorch/default_framework_info.py

@@ -19,7 +19,7 @@ from torch import sigmoid
 
 from model_compression_toolkit.defaultdict import DefaultDict
 from model_compression_toolkit.core.common.framework_info import FrameworkInfo
-from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
+from mct_quantizers import QuantizationMethod
 from model_compression_toolkit.constants import SOFTMAX_THRESHOLD
 from model_compression_toolkit.core.pytorch.constants import KERNEL
 from model_compression_toolkit.core.pytorch.quantizer.fake_quant_builder import power_of_two_quantization, \

model_compression_toolkit/core/pytorch/mixed_precision/configurable_activation_quantizer.py

@@ -21,7 +21,7 @@ from model_compression_toolkit.core.common.mixed_precision.configurable_quantize
 from model_compression_toolkit.core.common.quantization.candidate_node_quantization_config import \
     CandidateNodeQuantizationConfig
 from model_compression_toolkit.logger import Logger
-from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
+from mct_quantizers import QuantizationMethod
 from mct_quantizers import QuantizationTarget
 from mct_quantizers import mark_quantizer
 

model_compression_toolkit/core/pytorch/mixed_precision/configurable_weights_quantizer.py

@@ -20,7 +20,7 @@ from model_compression_toolkit.core.common.mixed_precision.configurable_quantize
 from model_compression_toolkit.core.common.quantization.candidate_node_quantization_config import \
     CandidateNodeQuantizationConfig
 from model_compression_toolkit.logger import Logger
-from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
+from mct_quantizers import QuantizationMethod
 from mct_quantizers import QuantizationTarget
 
 from mct_quantizers import mark_quantizer

model_compression_toolkit/core/pytorch/resource_utilization_data_facade.py

@@ -17,8 +17,7 @@ from typing import Callable
 
 from model_compression_toolkit.logger import Logger
 from model_compression_toolkit.constants import PYTORCH
-from model_compression_toolkit.target_platform_capabilities.schema.mct_current_schema import TargetPlatformModel
-from model_compression_toolkit.target_platform_capabilities.target_platform import TargetPlatformCapabilities
+from model_compression_toolkit.target_platform_capabilities.schema.mct_current_schema import TargetPlatformCapabilities
 from model_compression_toolkit.core.common.mixed_precision.resource_utilization_tools.resource_utilization import ResourceUtilization
 from model_compression_toolkit.core.common.mixed_precision.resource_utilization_tools.resource_utilization_data import compute_resource_utilization_data
 from model_compression_toolkit.core.common.quantization.core_config import CoreConfig
@@ -30,7 +29,7 @@ if FOUND_TORCH:
     from model_compression_toolkit.core.pytorch.default_framework_info import DEFAULT_PYTORCH_INFO
     from model_compression_toolkit.core.pytorch.pytorch_implementation import PytorchImplementation
     from torch.nn import Module
-    from model_compression_toolkit.target_platform_capabilities.target_platform.targetplatform2framework.attach2pytorch import \
+    from model_compression_toolkit.target_platform_capabilities.targetplatform2framework.attach2pytorch import \
         AttachTpcToPytorch
 
     from model_compression_toolkit import get_target_platform_capabilities
@@ -41,7 +40,7 @@ if FOUND_TORCH:
     def pytorch_resource_utilization_data(in_model: Module,
                                           representative_data_gen: Callable,
                                           core_config: CoreConfig = CoreConfig(),
-                                          target_platform_capabilities: TargetPlatformModel= PYTORCH_DEFAULT_TPC
+                                          target_platform_capabilities: TargetPlatformCapabilities= PYTORCH_DEFAULT_TPC
                                           ) -> ResourceUtilization:
         """
         Computes resource utilization data that can be used to calculate the desired target resource utilization for mixed-precision quantization.
@@ -51,7 +50,7 @@ if FOUND_TORCH:
             in_model (Model): PyTorch model to quantize.
             representative_data_gen (Callable): Dataset used for calibration.
             core_config (CoreConfig): CoreConfig containing parameters for quantization and mixed precision
-            target_platform_capabilities (TargetPlatformCapabilities): TargetPlatformCapabilities to optimize the PyTorch model according to.
+            target_platform_capabilities (FrameworkQuantizationCapabilities): FrameworkQuantizationCapabilities to optimize the PyTorch model according to.
 
         Returns: