mct-nightly 2.3.0.20250304.517__py3-none-any.whl → 2.3.0.20250306.517__py3-none-any.whl

{mct_nightly-2.3.0.20250304.517.dist-info → mct_nightly-2.3.0.20250306.517.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: mct-nightly
-Version: 2.3.0.20250304.517
+Version: 2.3.0.20250306.517
 Summary: A Model Compression Toolkit for neural networks
 Classifier: Programming Language :: Python :: 3
 Classifier: License :: OSI Approved :: Apache Software License

{mct_nightly-2.3.0.20250304.517.dist-info → mct_nightly-2.3.0.20250306.517.dist-info}/RECORD

@@ -1,4 +1,4 @@
-model_compression_toolkit/__init__.py,sha256=ovorAudQVUZkkSP9sCAZa5w2M5hCKGvczDcgMWGS8iQ,1557
+model_compression_toolkit/__init__.py,sha256=KAly1dabQ2mmOhCqkNoDMZrsXtV-JE-zjtsSqNdp6bg,1557
 model_compression_toolkit/constants.py,sha256=i_R6uXBfO1ph_X6DNJych2x59SUojfJbn7dNjs_mZnc,3846
 model_compression_toolkit/defaultdict.py,sha256=LSc-sbZYXENMCw3U9F4GiXuv67IKpdn0Qm7Fr11jy-4,2277
 model_compression_toolkit/logger.py,sha256=L3q7tn3Uht0i_7phnlOWMR2Te2zvzrt2HOz9vYEInts,4529
@@ -42,7 +42,7 @@ model_compression_toolkit/core/common/graph/graph_searches.py,sha256=2oKuW6L8hP-
 model_compression_toolkit/core/common/graph/virtual_activation_weights_node.py,sha256=3el-A7j1oyoo1_9zq3faQp7IeRsFXFCvnrb3zZFXpU0,9803
 model_compression_toolkit/core/common/graph/memory_graph/__init__.py,sha256=cco4TmeIDIh32nj9ZZXVkws4dd9F2UDrmjKzTN8G0V0,697
 model_compression_toolkit/core/common/graph/memory_graph/bipartite_graph.py,sha256=X6FK3C3y8ixFRPjC_wm3ClloCX8_06SOdA1TRi7o_LA,3800
-model_compression_toolkit/core/common/graph/memory_graph/compute_graph_max_cut.py,sha256=sUGp9GnKBI5NL7Y6d9pCyAL6Nv_3Htf2wInUtCtNMpU,3497
+model_compression_toolkit/core/common/graph/memory_graph/compute_graph_max_cut.py,sha256=oyz260JXDbvL8aI-DVtUvLHtLRWC2Yu4SBYlGL68c2Y,3498
 model_compression_toolkit/core/common/graph/memory_graph/cut.py,sha256=7Dfq4TVJIrnencHLJqjhxYKhY7ooUo_ml33WH2IIAgc,2576
 model_compression_toolkit/core/common/graph/memory_graph/max_cut_astar.py,sha256=E8xKMUxtEF0GjztUk-3CmMtivPPBcADnZTUaSN24o6A,17816
 model_compression_toolkit/core/common/graph/memory_graph/memory_element.py,sha256=ISD2BvJWj5mB91jrFjG8VQb0oOoLBoita_thCZWzCPI,4238
@@ -255,7 +255,7 @@ model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/remove_
 model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/reshape_with_static_shapes.py,sha256=hAZXzrEinHa-dJHLj39Hy_9Q-13QyO95rtYVSLrhvT8,4915
 model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/residual_collapsing.py,sha256=DcJEIkGvBdIMOelNIwaJUZ5UsAHiGnDJPR20I464vWo,2929
 model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/scale_equalization.py,sha256=XFtU9yuBmoZlX0f0mS6otMPWMk-RcWs94XdvvTNhW8Y,3303
-model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/scaled_dot_product_attention.py,sha256=SBtIuxb1Q2oUMJKSrAyN2wuaY4k1tsKt7qql0dP_PE0,12473
+model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/scaled_dot_product_attention.py,sha256=WG7MyYTP5JhMZHYxj4PB-7TTuvUDjFQScG4_Ce1mQDY,12476
 model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/shift_negative_activation.py,sha256=3WCLvPyx7tVkM0rwYhYq-gntCzW9R_DcImR1ucKlPac,10772
 model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/softmax_shift.py,sha256=05lV4pIL3hJkZl4JQPV4wk_EFD0eYLG5b8cdzvZk4P8,1588
 model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/transform_function_call_method.py,sha256=EC9Dvp-_UlpDWnipnf8ds65wh_Y-T8pXAFIwRScWpiY,2044
@@ -439,7 +439,7 @@ model_compression_toolkit/target_platform_capabilities/schema/v1.py,sha256=4CGpW
 model_compression_toolkit/target_platform_capabilities/targetplatform2framework/__init__.py,sha256=XjNws3zoiJkeH4ixKqrLA5xBvpv5rq31qX7wYQjNpZM,1447
 model_compression_toolkit/target_platform_capabilities/targetplatform2framework/attach2fw.py,sha256=HJ8uc3PFfyxg-WpVXPBg4mGaox8Z9bRqtQNbRfIyAk4,3745
 model_compression_toolkit/target_platform_capabilities/targetplatform2framework/attach2keras.py,sha256=mxc3DBbUi-HDFgSx8Nmnyxr8SIdbx8lmtcRMsQl1BLE,7578
-model_compression_toolkit/target_platform_capabilities/targetplatform2framework/attach2pytorch.py,sha256=swbfYmAbvQ7yDfyPhDWZxeN2P18WohI8uzz8zR1z0cg,6301
+model_compression_toolkit/target_platform_capabilities/targetplatform2framework/attach2pytorch.py,sha256=WPCqs_aFGE28XJf7KKB-SlrYoUNOcD9epgoaqQMCJMw,6320
 model_compression_toolkit/target_platform_capabilities/targetplatform2framework/attribute_filter.py,sha256=jfhszvuD2Fyy6W2KjlLzXBQKFzTqGAaDZeFVr4-ONQw,8776
 model_compression_toolkit/target_platform_capabilities/targetplatform2framework/current_tpc.py,sha256=_kFG0USYa6yzvLsi82_Vusv_KR8Hi7J1u680pPXECuo,2192
 model_compression_toolkit/target_platform_capabilities/targetplatform2framework/framework_quantization_capabilities.py,sha256=UKzckLYLdBcFAptyKnVMwpPpfRkmF0SK1Kl0g0eGjQA,9710
@@ -459,7 +459,7 @@ model_compression_toolkit/target_platform_capabilities/tpc_models/qnnpack_tpc/v1
 model_compression_toolkit/target_platform_capabilities/tpc_models/tflite_tpc/__init__.py,sha256=cco4TmeIDIh32nj9ZZXVkws4dd9F2UDrmjKzTN8G0V0,697
 model_compression_toolkit/target_platform_capabilities/tpc_models/tflite_tpc/latest/__init__.py,sha256=NGnmaFTeLINIdAZ1svx-_OiF6vIs8upH-tz3q9jWBQ4,1554
 model_compression_toolkit/target_platform_capabilities/tpc_models/tflite_tpc/v1/__init__.py,sha256=t4JKsPcor-7KSCKzIwuaBv0NLNwfhuewAQGlDl6iBeo,717
-model_compression_toolkit/target_platform_capabilities/tpc_models/tflite_tpc/v1/tpc.py,sha256=2UXcjIBKL0sVdJ2zN_L6JyX9P37ErD95dDN0TUzbFz0,12781
+model_compression_toolkit/target_platform_capabilities/tpc_models/tflite_tpc/v1/tpc.py,sha256=sd1oN7mYH6fHqJwU2QZf6WU2NJ8EKzQ9-o4-JGU6Plc,12958
 model_compression_toolkit/trainable_infrastructure/__init__.py,sha256=uewpvlPkH9mBFt8IxoAgIfz6iEcvWbOImm_fb6_BxD8,1543
 model_compression_toolkit/trainable_infrastructure/common/__init__.py,sha256=huHoBUcKNB6BnY6YaUCcFvdyBtBI172ZoUD8ZYeNc6o,696
 model_compression_toolkit/trainable_infrastructure/common/annealing_schedulers.py,sha256=qm2_wa61nga08Jdcl3RkgTsJ0zyHNjZ_A6I2--oVOig,2455
@@ -524,8 +524,8 @@ model_compression_toolkit/xquant/pytorch/model_analyzer.py,sha256=b93o800yVB3Z-i
 model_compression_toolkit/xquant/pytorch/pytorch_report_utils.py,sha256=UVN_S9ULHBEldBpShCOt8-soT8YTQ5oE362y96qF_FA,3950
 model_compression_toolkit/xquant/pytorch/similarity_functions.py,sha256=CERxq5K8rqaiE-DlwhZBTUd9x69dtYJlkHOPLB54vm8,2354
 model_compression_toolkit/xquant/pytorch/tensorboard_utils.py,sha256=mkoEktLFFHtEKzzFRn_jCnxjhJolK12TZ5AQeDHzUO8,9767
-mct_nightly-2.3.0.20250304.517.dist-info/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
-mct_nightly-2.3.0.20250304.517.dist-info/METADATA,sha256=ohIw0TcK_CUbaSVCUGc9YVprMJap8YJyC9kKYVoFxfA,27079
-mct_nightly-2.3.0.20250304.517.dist-info/WHEEL,sha256=jB7zZ3N9hIM9adW7qlTAyycLYW9npaWKLRzaoVcLKcM,91
-mct_nightly-2.3.0.20250304.517.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
-mct_nightly-2.3.0.20250304.517.dist-info/RECORD,,
+mct_nightly-2.3.0.20250306.517.dist-info/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
+mct_nightly-2.3.0.20250306.517.dist-info/METADATA,sha256=oluoSGV3e3blg9xNQ0KiWtwivNTNNTSDl6zzjLJSSFI,27079
+mct_nightly-2.3.0.20250306.517.dist-info/WHEEL,sha256=jB7zZ3N9hIM9adW7qlTAyycLYW9npaWKLRzaoVcLKcM,91
+mct_nightly-2.3.0.20250306.517.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
+mct_nightly-2.3.0.20250306.517.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__ = "2.3.0.20250304.000517"
+__version__ = "2.3.0.20250306.000517"

model_compression_toolkit/core/common/graph/memory_graph/compute_graph_max_cut.py

@@ -27,7 +27,7 @@ SchedulerInfo = namedtuple('SchedulerInfo', [OPERATORS_SCHEDULING, MAX_CUT, CUTS
 
 def compute_graph_max_cut(memory_graph: MemoryGraph,
                           n_iter: int = 50,
-                          astar_n_iter: int = 500,
+                          astar_n_iter: int = 1000,
                           eps: float = 1e-2) -> Tuple[List[BaseNode], float, List[Cut]]:
     """
     A wrapper function to compute max cut and schedule for a given model.

model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/scaled_dot_product_attention.py

@@ -103,7 +103,7 @@ class ScaledDotProductDecomposition(BaseSubstitution):
         matmul_name = f'{attention_node_name}_matmul1'
         return FunctionalNode(name=matmul_name,
                               framework_attr={},
-                              input_shape=(tuple(q_node.output_shape[0]), tuple(transposed_k_node.output_shape)),
+                              input_shape=(tuple(q_node.output_shape[0]), tuple(transposed_k_node.output_shape[0])),
                               output_shape=tuple(matmul1_output_shape),
                               weights={},
                               layer_class=torch.matmul,

model_compression_toolkit/target_platform_capabilities/targetplatform2framework/attach2pytorch.py

@@ -70,7 +70,7 @@ class AttachTpcToPytorch(AttachTpcToFramework):
             OperatorSetNames.HARDSWISH: [Hardswish, hardswish],
             OperatorSetNames.FLATTEN: [Flatten, flatten],
             OperatorSetNames.GET_ITEM: [operator.getitem],
-            OperatorSetNames.RESHAPE: [reshape],
+            OperatorSetNames.RESHAPE: [reshape, torch.Tensor.view],
             OperatorSetNames.UNSQUEEZE: [unsqueeze],
             OperatorSetNames.SQUEEZE: [squeeze],
             OperatorSetNames.PERMUTE: [permute],

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

@@ -137,6 +137,7 @@ def generate_tpc(default_config: OpQuantizationConfig,
     # If the QuantizationConfigOptions contains only one configuration,
     # this configuration will be used for the operation quantization:
     default_configuration_options = schema.QuantizationConfigOptions(quantization_configurations=tuple([default_config]))
+    base_configuration_options = schema.QuantizationConfigOptions(quantization_configurations=tuple([base_config]), base_config=base_config)
 
     # In TFLite, the quantized operator specifications constraint operators quantization
     # differently. For more details:
@@ -179,13 +180,13 @@ def generate_tpc(default_config: OpQuantizationConfig,
                                qc_options=default_configuration_options.clone_and_edit(
                                    fixed_zero_point=-128, fixed_scale=1 / 256))
     fc = schema.OperatorsSet(name=schema.OperatorSetNames.FULLY_CONNECTED,
-                             qc_options=default_configuration_options.clone_and_edit_weight_attribute(
+                             qc_options=base_configuration_options.clone_and_edit_weight_attribute(
                                  weights_per_channel_threshold=False))
     squeeze = schema.OperatorsSet(name=schema.OperatorSetNames.SQUEEZE,
                                   qc_options=default_configuration_options.clone_and_edit(
                                       quantization_preserving=True))
 
-    conv2d = schema.OperatorsSet(name=schema.OperatorSetNames.CONV)
+    conv2d = schema.OperatorsSet(name=schema.OperatorSetNames.CONV, qc_options=base_configuration_options)
     relu = schema.OperatorsSet(name=schema.OperatorSetNames.RELU)
     relu6 = schema.OperatorsSet(name=schema.OperatorSetNames.RELU6)
     elu = schema.OperatorsSet(name=schema.OperatorSetNames.ELU)