mct-nightly 1.8.0.22042023.post414__py3-none-any.whl → 1.8.0.22052023.post408__py3-none-any.whl

model_compression_toolkit/target_platform_capabilities/tpc_models/tflite_tpc/v1/tp_model.py

@@ -15,8 +15,12 @@
 from typing import List, Tuple
 
 import model_compression_toolkit as mct
-from model_compression_toolkit.target_platform_capabilities.target_platform import OpQuantizationConfig, TargetPlatformModel
-from model_compression_toolkit.target_platform_capabilities.target_platform.op_quantization_config import QuantizationMethod
+from model_compression_toolkit.target_platform_capabilities.target_platform import OpQuantizationConfig, \
+    TargetPlatformModel
+from model_compression_toolkit.target_platform_capabilities.target_platform.op_quantization_config import \
+    QuantizationMethod
+from model_compression_toolkit.target_platform_capabilities.target_platform.quantization_format import \
+    QuantizationFormat
 
 tp = mct.target_platform
 
@@ -65,7 +69,7 @@ def get_op_quantization_configs() -> Tuple[OpQuantizationConfig, List[OpQuantiza
         weights_multiplier_nbits=None
     )
 
-    mixed_precision_cfg_list = [] # No mixed precision
+    mixed_precision_cfg_list = []  # No mixed precision
 
     return eight_bits, mixed_precision_cfg_list
 
@@ -106,28 +110,28 @@ def generate_tp_model(default_config: OpQuantizationConfig,
         # differently. For more details:
         # https://www.tensorflow.org/lite/performance/quantization_spec#int8_quantized_operator_specifications
         tp.OperatorsSet("NoQuantization",
-                         tp.get_default_quantization_config_options().clone_and_edit(
-                                  quantization_preserving=True))
+                        tp.get_default_quantization_config_options().clone_and_edit(
+                            quantization_preserving=True))
 
         fc = tp.OperatorsSet("FullyConnected",
-                              tp.get_default_quantization_config_options().clone_and_edit(
-                                       weights_per_channel_threshold=False))
+                             tp.get_default_quantization_config_options().clone_and_edit(
+                                 weights_per_channel_threshold=False))
 
         tp.OperatorsSet("L2Normalization",
-                         tp.get_default_quantization_config_options().clone_and_edit(
-                                  fixed_zero_point=0, fixed_scale=1 / 128))
+                        tp.get_default_quantization_config_options().clone_and_edit(
+                            fixed_zero_point=0, fixed_scale=1 / 128))
         tp.OperatorsSet("LogSoftmax",
-                         tp.get_default_quantization_config_options().clone_and_edit(
-                                  fixed_zero_point=127, fixed_scale=16 / 256))
+                        tp.get_default_quantization_config_options().clone_and_edit(
+                            fixed_zero_point=127, fixed_scale=16 / 256))
         tp.OperatorsSet("Tanh",
-                         tp.get_default_quantization_config_options().clone_and_edit(
-                                  fixed_zero_point=0, fixed_scale=1 / 128))
+                        tp.get_default_quantization_config_options().clone_and_edit(
+                            fixed_zero_point=0, fixed_scale=1 / 128))
         tp.OperatorsSet("Softmax",
-                         tp.get_default_quantization_config_options().clone_and_edit(
-                                  fixed_zero_point=-128, fixed_scale=1 / 256))
+                        tp.get_default_quantization_config_options().clone_and_edit(
+                            fixed_zero_point=-128, fixed_scale=1 / 256))
         tp.OperatorsSet("Logistic",
-                         tp.get_default_quantization_config_options().clone_and_edit(
-                                  fixed_zero_point=-128, fixed_scale=1 / 256))
+                        tp.get_default_quantization_config_options().clone_and_edit(
+                            fixed_zero_point=-128, fixed_scale=1 / 256))
 
         conv2d = tp.OperatorsSet("Conv2d")
         kernel = tp.OperatorSetConcat(conv2d, fc)
@@ -140,7 +144,8 @@ def generate_tp_model(default_config: OpQuantizationConfig,
         bias_add = tp.OperatorsSet("BiasAdd")
         add = tp.OperatorsSet("Add")
         squeeze = tp.OperatorsSet("Squeeze",
-                                   qc_options=tp.get_default_quantization_config_options().clone_and_edit(quantization_preserving=True))
+                                  qc_options=tp.get_default_quantization_config_options().clone_and_edit(
+                                      quantization_preserving=True))
         # ------------------- #
         # Fusions
         # ------------------- #
@@ -152,4 +157,7 @@ def generate_tp_model(default_config: OpQuantizationConfig,
         tp.Fusing([batch_norm, activations_to_fuse])
         tp.Fusing([batch_norm, add, activations_to_fuse])
 
+        # Set quantization format to int8
+        generated_tpc.set_quantization_format(QuantizationFormat.INT8)
+
     return generated_tpc

model_compression_toolkit/exporter/model_exporter/tflite/tflite_export_facade.py

@@ -1,73 +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 enum import Enum
-from typing import Callable
-
-from model_compression_toolkit.core.common import Logger
-from model_compression_toolkit.core.common.constants import FOUND_TF
-
-
-class TFLiteExportMode(Enum):
-    FAKELY_QUANT = 0
-    INT8 = 1
-
-if FOUND_TF:
-    import keras
-    from model_compression_toolkit.exporter.model_exporter.tflite.fakely_quant_tflite_exporter import FakelyQuantTFLiteExporter
-    from model_compression_toolkit.exporter.model_exporter.tflite.int8_tflite_exporter import INT8TFLiteExporter
-    from model_compression_toolkit.exporter.model_wrapper.keras.validate_layer import is_keras_layer_exportable
-
-    def tflite_export_model(model: keras.models.Model,
-                            save_model_path: str,
-                            mode: TFLiteExportMode = TFLiteExportMode.FAKELY_QUANT,
-                            is_layer_exportable_fn: Callable = is_keras_layer_exportable
-                            ):
-        """
-        Export a Keras quantized model to a tflite model.
-        The model will be saved to the path in save_model_path.
-        Mode can be used for different exported files. Currently, tflite_export_model
-        supports TFLiteExportMode.FAKELY_QUANT (where weights and activations are
-        float fakely-quantized values), and TFLiteExportMode.INT8 (where weights
-        and activations are represented using 8bits integers).
-
-        Args:
-            model: Model to export.
-            is_layer_exportable_fn: Callable to check whether a layer can be exported or not.
-            mode: Mode to export the model according to.
-            save_model_path: Path to save the model.
-
-        """
-
-        if mode == TFLiteExportMode.FAKELY_QUANT:
-            exporter = FakelyQuantTFLiteExporter(model,
-                                                 is_layer_exportable_fn,
-                                                 save_model_path)
-        elif mode == TFLiteExportMode.INT8:
-            exporter = INT8TFLiteExporter(model,
-                                          is_layer_exportable_fn,
-                                          save_model_path)
-
-        else:
-            Logger.critical(
-                f'Unsupported mode was used {mode.name} to export TFLite model.'
-                f' Please see API for supported modes.')  # pragma: no cover
-
-        exporter.export()
-
-else:
-    def tflite_export_model(*args, **kwargs):
-        Logger.error('Installing tensorflow and tensorflow_model_optimization is mandatory '
-                     'when using tflite_export_model. '
-                     'Could not find some or all of TensorFlow packages.')  # pragma: no cover