mct-nightly 2.0.0.20240403.423__py3-none-any.whl → 2.0.0.20240404.424__py3-none-any.whl

{mct_nightly-2.0.0.20240403.423.dist-info → mct_nightly-2.0.0.20240404.424.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: mct-nightly
-Version: 2.0.0.20240403.423
+Version: 2.0.0.20240404.424
 Summary: A Model Compression Toolkit for neural networks
 Home-page: UNKNOWN
 License: UNKNOWN

{mct_nightly-2.0.0.20240403.423.dist-info → mct_nightly-2.0.0.20240404.424.dist-info}/RECORD

@@ -1,16 +1,15 @@
-model_compression_toolkit/__init__.py,sha256=rr-lU1cO0RuS_7QNw41uShlq2s0ILvvtATqaUT-NKzo,1573
+model_compression_toolkit/__init__.py,sha256=lJlwXgpNCKdbNOfoMIxzBiVzQu9sxWFTlnhSNS2ky00,1573
 model_compression_toolkit/constants.py,sha256=KW_HUEPmQEYqCvWGyORqkYxpvO7w5LViB5J5D-pm_6o,3648
 model_compression_toolkit/defaultdict.py,sha256=LSc-sbZYXENMCw3U9F4GiXuv67IKpdn0Qm7Fr11jy-4,2277
 model_compression_toolkit/logger.py,sha256=3DByV41XHRR3kLTJNbpaMmikL8icd9e1N-nkQAY9oDk,4567
 model_compression_toolkit/core/__init__.py,sha256=TrRgkWpT1AN2Faw1M_1HXyJkJnbxfn9p-RigDZl7pg0,1982
 model_compression_toolkit/core/analyzer.py,sha256=dbsD61pakp_9JXNyAScLdtJvcXny9jr_cMbET0Bd3Sg,2975
-model_compression_toolkit/core/exporter.py,sha256=Zo_C5GjIzihtJOyGp-xeCVhY_qohkVz_EGyrSZCbWRM,4115
 model_compression_toolkit/core/graph_prep_runner.py,sha256=Ftqm59hT5TGWmSNkY9bFZkVfCacpGyZfCe-6yZR5WY0,10100
 model_compression_toolkit/core/quantization_prep_runner.py,sha256=hFhDkS8GwzXZ7Ho_9qbbb8DAAWs3OONOfMSD5OU_b0o,6153
 model_compression_toolkit/core/runner.py,sha256=NKSC6ujfQPy6dKtJVwxyK2zNDd64eyR5csYy9lBrCPA,11836
 model_compression_toolkit/core/common/__init__.py,sha256=Wh127PbXcETZX_d1PQqZ71ETK3J9XO5A-HpadGUbj6o,1447
 model_compression_toolkit/core/common/base_substitutions.py,sha256=xDFSmVVs_iFSZfajytI0cuQaNRNcwHX3uqOoHgVUvxQ,1666
-model_compression_toolkit/core/common/framework_implementation.py,sha256=njBP3elCFiSHyT2Y9zT1ISyDeQFR3j5UzPR4XQK2KnY,21152
+model_compression_toolkit/core/common/framework_implementation.py,sha256=pOT9ZmRFL9FY92uUtigrO3sbWGiyVDhHAM1fbA4b5yo,20752
 model_compression_toolkit/core/common/framework_info.py,sha256=1ZMMGS9ip-kSflqkartyNRt9aQ5ub1WepuTRcTy-YSQ,6337
 model_compression_toolkit/core/common/memory_computation.py,sha256=ixoSpV5ZYZGyzhre3kQcvR2sNA8KBsPZ3lgbkDnw9Cs,1205
 model_compression_toolkit/core/common/model_builder_mode.py,sha256=jll9-59OPaE3ug7Y9-lLyV99_FoNHxkGZMgcm0Vkpss,1324
@@ -149,7 +148,7 @@ model_compression_toolkit/core/keras/__init__.py,sha256=mjbqLD-KcG3eNeCYpu1GBS7V
 model_compression_toolkit/core/keras/constants.py,sha256=Uv3c0UdW55pIVQNW_1HQlgl-dHXREkltOLyzp8G1mTQ,3163
 model_compression_toolkit/core/keras/custom_layer_validation.py,sha256=f-b14wuiIgitBe7d0MmofYhDCTO3IhwJgwrh-Hq_t_U,1192
 model_compression_toolkit/core/keras/default_framework_info.py,sha256=Ha4HTHuiw_KTS5Po1Xnv6GyK9eprpDhYWf-eooS62Ys,4961
-model_compression_toolkit/core/keras/keras_implementation.py,sha256=wg5ZkmQZc15w8k_VkXwAhfHQbZL3gpZZblYv_70-qxA,29399
+model_compression_toolkit/core/keras/keras_implementation.py,sha256=NDHLl19I-xQrQGcsAwTcFjnIjCRn31xaPrqDYm8g_dg,29027
 model_compression_toolkit/core/keras/keras_model_validation.py,sha256=1wNV2clFdC9BzIELRLSO2uKf0xqjLqlkTJudwtCeaJk,1722
 model_compression_toolkit/core/keras/keras_node_prior_info.py,sha256=Aqh31wOPaiZcJIOm-uJwzev0eTMdJyXaOk97rs4z7BU,3879
 model_compression_toolkit/core/keras/resource_utilization_data_facade.py,sha256=Xmk2ZL5CaYdb7iG62HdtZ1F64vap7ffnrsuR3e3G5hc,4851
@@ -173,7 +172,6 @@ model_compression_toolkit/core/keras/graph_substitutions/substitutions/linear_co
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/matmul_substitution.py,sha256=kjwlKtm5yhNgWVVcW6mN-hn7enwAnn_8-TUZvxZBiQs,4112
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/multi_head_attention_decomposition.py,sha256=l9PUREBf4aRwWILiybdteveeUbh7js-i-hLt8Ma0e4c,26771
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/relu_bound_to_power_of_2.py,sha256=IdKOg6AWZWMcmDbOuNdxetS5_zTarXIIffdYL7JTdvk,3872
-model_compression_toolkit/core/keras/graph_substitutions/substitutions/remove_relu_upper_bound.py,sha256=cJQTDzTDQKAJ7EQ20tfsmReGA_OoTIN793MwVe1Ok8g,2387
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/residual_collapsing.py,sha256=gSqUYh76tP7NcZfqFSnuPIrUpyBh6UjjcPJtJxZtOZk,3181
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/scale_equalization.py,sha256=ryes9y1ie-vjBGso2TeO4EXxVk69Ew3iSAhshPz1Ou4,5542
 model_compression_toolkit/core/keras/graph_substitutions/substitutions/separableconv_decomposition.py,sha256=TEaHlIbXj_ZjIdT5TmAICD3WLD3u_7g0fLWQcNzTJuM,7941
@@ -211,7 +209,7 @@ model_compression_toolkit/core/pytorch/__init__.py,sha256=Rf1RcYmelmdZmBV5qOKvKW
 model_compression_toolkit/core/pytorch/constants.py,sha256=NI-J7REuxn06oEIHsmJ4GqtNC3TbV8xlkJjt5Ar-c4U,2626
 model_compression_toolkit/core/pytorch/default_framework_info.py,sha256=r1XyzUFvrjGcJHQM5ETLsMZIG2yHCr9HMjqf0ti9inw,4175
 model_compression_toolkit/core/pytorch/pytorch_device_config.py,sha256=IoMvTch5awAEPvB6Tg6ANhFGXvfSgv7JLsUBlxpMwk4,4330
-model_compression_toolkit/core/pytorch/pytorch_implementation.py,sha256=Yr5ldjroFaoImOlPw3DhqL5uWV-WHyn7TS3kGRWyTjI,27082
+model_compression_toolkit/core/pytorch/pytorch_implementation.py,sha256=dakO4Nj-tFfs53y6dJyXbpoljx2n3ZqmMoB4CFWGNSQ,26868
 model_compression_toolkit/core/pytorch/pytorch_node_prior_info.py,sha256=n_B4a6FMwM9D2w8kzy3oenBWZgXNZuIZgTJC6JEuTy0,3250
 model_compression_toolkit/core/pytorch/resource_utilization_data_facade.py,sha256=E6ifk1HdO60k4IRH2EFBzAYWtwUlrGqJoQ66nknpHoQ,4983
 model_compression_toolkit/core/pytorch/utils.py,sha256=dRPiteBg2dBNsHwZyYzXiCIAjnelSoeZZsDXlsTw5JQ,2880
@@ -311,7 +309,7 @@ model_compression_toolkit/exporter/model_exporter/keras/mctq_keras_exporter.py,s
 model_compression_toolkit/exporter/model_exporter/pytorch/__init__.py,sha256=uZ2RigbY9O2PJ0Il8wPpS_s7frgg9WUGd_SHeKGyl1A,699
 model_compression_toolkit/exporter/model_exporter/pytorch/base_pytorch_exporter.py,sha256=UPVkEUQCMZ4Lld6CRnEOPEmlfe5vcQZG0Q3FwRBodD4,4021
 model_compression_toolkit/exporter/model_exporter/pytorch/export_serialization_format.py,sha256=bPevy6OBqng41PqytBR55e6cBEuyrUS0H8dWX4zgjQ4,967
-model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_onnx_pytorch_exporter.py,sha256=JmxSa8mwL-HaTY1fS0UEpKJVlg95W_eBYudpeVRnRDo,5137
+model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_onnx_pytorch_exporter.py,sha256=TedEyQN9TQzRnb8afQVF4Kx28IeyEJRWiUloQxkBHAY,5136
 model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_torchscript_pytorch_exporter.py,sha256=ksWV2A-Njo-wAxQ_Ye2sLIZXBWJ_WNyjT7-qFFwvV2o,2897
 model_compression_toolkit/exporter/model_exporter/pytorch/pytorch_export_facade.py,sha256=iVKanLUALkK7WIDmNLpcvHu02EO47GM-T8iLr4srp-Q,6351
 model_compression_toolkit/exporter/model_wrapper/__init__.py,sha256=7CF2zvpTrIEm8qnbuHnLZyTZkwBBxV24V8QA0oxGbh0,1187
@@ -340,7 +338,7 @@ model_compression_toolkit/gptq/keras/gptq_keras_implementation.py,sha256=axBwnCS
 model_compression_toolkit/gptq/keras/gptq_loss.py,sha256=rbRkF15MYd6nq4G49kcjb_dPTa-XNq9cTkrb93mXawo,6241
 model_compression_toolkit/gptq/keras/gptq_training.py,sha256=6TseqtBzZkLqyc3hiRVdA1dv01us6Y_Su05CBboGjjc,18438
 model_compression_toolkit/gptq/keras/graph_info.py,sha256=5IvgGlJlgOmQYmldjdCBv7tuzAoY0HazatG5Pedrg0Q,4639
-model_compression_toolkit/gptq/keras/quantization_facade.py,sha256=l4UNddc9tX1YxfvPB_4NsqqkRh79d2YwR2L2rnVzYBU,14057
+model_compression_toolkit/gptq/keras/quantization_facade.py,sha256=yRgiikyeELar4jlsdcf5pO9HQcxiyhKiAXY3lsMixew,13913
 model_compression_toolkit/gptq/keras/quantizer/__init__.py,sha256=-DK1CDXvlsnEbki4lukZLpl6Xrbo91_jcqxXlG5Eg6Q,963
 model_compression_toolkit/gptq/keras/quantizer/base_keras_gptq_quantizer.py,sha256=2YU-x4-Q5f6hkUJf0tw6vcwdNwRMHdefrFjhhyHYsvA,4782
 model_compression_toolkit/gptq/keras/quantizer/quant_utils.py,sha256=Vt7Qb8i4JsE4sFtcjpfM4FTXTtfV1t6SwfoNH8a_Iaw,5055
@@ -357,7 +355,7 @@ model_compression_toolkit/gptq/pytorch/gptq_loss.py,sha256=kDuWw-6zh17wZpYWh4Xa9
 model_compression_toolkit/gptq/pytorch/gptq_pytorch_implementation.py,sha256=tECPTavxn8EEwgLaP2zvxdJH6Vg9jC0YOIMJ7857Sdc,1268
 model_compression_toolkit/gptq/pytorch/gptq_training.py,sha256=ksFXtepAsk-66xk7OPciG05kU9sgAUrWqjOgplsGSnw,15808
 model_compression_toolkit/gptq/pytorch/graph_info.py,sha256=yXJzDd24zfGs2_vfMovxD1WSh1RxXoPxN4GztOf3P5c,3967
-model_compression_toolkit/gptq/pytorch/quantization_facade.py,sha256=7haHpYlD-qWwk06wpbPMOFPs_nWS4QnWNaeGIrLKAm0,12495
+model_compression_toolkit/gptq/pytorch/quantization_facade.py,sha256=bImrTw9rrAVc3VD4nmrXmBo_K4fuf5m5XPPf8ybOThs,12430
 model_compression_toolkit/gptq/pytorch/quantizer/__init__.py,sha256=ZHNHo1yzye44m9_ht4UUZfTpK01RiVR3Tr74-vtnOGI,968
 model_compression_toolkit/gptq/pytorch/quantizer/base_pytorch_gptq_quantizer.py,sha256=TCA1hAc7raPnrjl06sjFtVM4XUtLtuwAhCGX4U3KGZo,4137
 model_compression_toolkit/gptq/pytorch/quantizer/quant_utils.py,sha256=OocYYRqvl7rZ37QT0hTzfJnWGiNCPskg7cziTlR7TRk,3893
@@ -377,9 +375,9 @@ model_compression_toolkit/pruning/pytorch/pruning_facade.py,sha256=cSuvHHCqgr7k9
 model_compression_toolkit/ptq/__init__.py,sha256=Z_hkmTh7aLFei1DJKV0oNVUbrv_Q_0CTw-qD85Xf8UM,904
 model_compression_toolkit/ptq/runner.py,sha256=_c1dSjlPPpsx59Vbg1buhG9bZq__OORz1VlPkwjJzoc,2552
 model_compression_toolkit/ptq/keras/__init__.py,sha256=cco4TmeIDIh32nj9ZZXVkws4dd9F2UDrmjKzTN8G0V0,697
-model_compression_toolkit/ptq/keras/quantization_facade.py,sha256=QkZHN_jsC3WDuKT6FbJfPcRlq6XddYeAnZRBhRBH7eI,9057
+model_compression_toolkit/ptq/keras/quantization_facade.py,sha256=5Doa5Rwer84DRgJxLa2e6aX9B4yGYdmFGgiv71_wD9o,8992
 model_compression_toolkit/ptq/pytorch/__init__.py,sha256=cco4TmeIDIh32nj9ZZXVkws4dd9F2UDrmjKzTN8G0V0,697
-model_compression_toolkit/ptq/pytorch/quantization_facade.py,sha256=4No5_g7wReXNDhj4Vcco-DLDWjZoqaJLOTBJQyEVtv4,7583
+model_compression_toolkit/ptq/pytorch/quantization_facade.py,sha256=QW7lMRuuonupLPZ2w2PDIQd7qpDZ_euLInhskTc1Yes,7518
 model_compression_toolkit/qat/__init__.py,sha256=kj2qsZh_Ca7PncsHKcaL5EVT2H8g4hYtvaQ3KFxOkwE,1143
 model_compression_toolkit/qat/common/__init__.py,sha256=6tLZ4R4pYP6QVztLVQC_jik2nES3l4uhML0qUxZrezk,829
 model_compression_toolkit/qat/common/qat_config.py,sha256=zoq0Vb74vCY7WlWD8JH_KPrHDoUHSvMc3gcO53u7L2U,3394
@@ -471,8 +469,8 @@ model_compression_toolkit/trainable_infrastructure/keras/quantize_wrapper.py,sha
 model_compression_toolkit/trainable_infrastructure/keras/quantizer_utils.py,sha256=MVwXNymmFRB2NXIBx4e2mdJ1RfoHxRPYRgjb1MQP5kY,1797
 model_compression_toolkit/trainable_infrastructure/pytorch/__init__.py,sha256=huHoBUcKNB6BnY6YaUCcFvdyBtBI172ZoUD8ZYeNc6o,696
 model_compression_toolkit/trainable_infrastructure/pytorch/base_pytorch_quantizer.py,sha256=7bbzqJN8ZAycVDvZr_5xC-niTAR5df8f03Kooev_pfg,3047
-mct_nightly-2.0.0.20240403.423.dist-info/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
-mct_nightly-2.0.0.20240403.423.dist-info/METADATA,sha256=dHTEKAHtl1x-dI2jaHQiDYi7QJInTciTrVMfjF2DzxQ,18795
-mct_nightly-2.0.0.20240403.423.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
-mct_nightly-2.0.0.20240403.423.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
-mct_nightly-2.0.0.20240403.423.dist-info/RECORD,,
+mct_nightly-2.0.0.20240404.424.dist-info/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
+mct_nightly-2.0.0.20240404.424.dist-info/METADATA,sha256=4ID8UlB0QTcYADss82tFhwAwGOBOPZdmbuT7kIJLWm8,18795
+mct_nightly-2.0.0.20240404.424.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
+mct_nightly-2.0.0.20240404.424.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
+mct_nightly-2.0.0.20240404.424.dist-info/RECORD,,

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__ = "2.0.0.20240404.000424"

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[

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]:
         """

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]:
         """

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()

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
 

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 \

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
 

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
 
 

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
-

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