mct-nightly 2.2.0.20250113.527__py3-none-any.whl → 2.2.0.20250113.134913__py3-none-any.whl

{mct_nightly-2.2.0.20250113.527.dist-info → mct_nightly-2.2.0.20250113.134913.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: mct-nightly
-Version: 2.2.0.20250113.527
+Version: 2.2.0.20250113.134913
 Summary: A Model Compression Toolkit for neural networks
 Classifier: Programming Language :: Python :: 3
 Classifier: License :: OSI Approved :: Apache Software License

{mct_nightly-2.2.0.20250113.527.dist-info → mct_nightly-2.2.0.20250113.134913.dist-info}/RECORD

@@ -1,4 +1,4 @@
-model_compression_toolkit/__init__.py,sha256=Mk_Xga1TGuHC9ARKoo8EqPPyhJTicRHOES48xvekDQ8,1573
+model_compression_toolkit/__init__.py,sha256=lAJGvoR9rqdv9OkbbBQa5WQeqHFUM7X0gMUeE_3A4Vc,1573
 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=3DByV41XHRR3kLTJNbpaMmikL8icd9e1N-nkQAY9oDk,4567
@@ -155,7 +155,7 @@ model_compression_toolkit/core/common/visualization/tensorboard_writer.py,sha256
 model_compression_toolkit/core/keras/__init__.py,sha256=mjbqLD-KcG3eNeCYpu1GBS7VclGVOQ63x2p6mAAuba4,698
 model_compression_toolkit/core/keras/constants.py,sha256=dh4elQWt6Q6NYRht5k5RiiOcnLAq1v0MMBCJqMJzzFk,3225
 model_compression_toolkit/core/keras/custom_layer_validation.py,sha256=f-b14wuiIgitBe7d0MmofYhDCTO3IhwJgwrh-Hq_t_U,1192
-model_compression_toolkit/core/keras/data_util.py,sha256=-fqhXTzlA3RybWp0M5phPkzVbSJ2vPLrjFcCazWMYHk,7300
+model_compression_toolkit/core/keras/data_util.py,sha256=jm54o-SlI1DJ-sEvRuX9OyLN68tEt0VxcqrdIjR98Ag,8366
 model_compression_toolkit/core/keras/default_framework_info.py,sha256=PYcER89eEXjKtR0T7-2Y4f7cckqoD5OQbpHePoRkMec,5030
 model_compression_toolkit/core/keras/keras_implementation.py,sha256=HwbIR7x4t-TBNbWHVvVNFk8z-KFt6zM0LWAUXQuNZrk,31753
 model_compression_toolkit/core/keras/keras_model_validation.py,sha256=1wNV2clFdC9BzIELRLSO2uKf0xqjLqlkTJudwtCeaJk,1722
@@ -525,8 +525,8 @@ model_compression_toolkit/xquant/pytorch/model_analyzer.py,sha256=b93o800yVB3Z-i
 model_compression_toolkit/xquant/pytorch/pytorch_report_utils.py,sha256=3jNiV5Z4BVw9cEWuLKNOlLuLdr0EMuKg6eYnSiAq3LU,3952
 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-2.2.0.20250113.527.dist-info/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
-mct_nightly-2.2.0.20250113.527.dist-info/METADATA,sha256=KXLhoCPM8Ox59dSevH9A6kiCOfE8xFtr-NDX5sxctcs,26601
-mct_nightly-2.2.0.20250113.527.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
-mct_nightly-2.2.0.20250113.527.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
-mct_nightly-2.2.0.20250113.527.dist-info/RECORD,,
+mct_nightly-2.2.0.20250113.134913.dist-info/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
+mct_nightly-2.2.0.20250113.134913.dist-info/METADATA,sha256=MRbQVy655Qf1QPZeBJj9K8Hc4p0GzYq8tAfuIvtwNAk,26604
+mct_nightly-2.2.0.20250113.134913.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
+mct_nightly-2.2.0.20250113.134913.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
+mct_nightly-2.2.0.20250113.134913.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__ = "2.2.0.20250113.000527"
+__version__ = "2.2.0.20250113.134913"

model_compression_toolkit/core/keras/data_util.py

@@ -134,11 +134,30 @@ class FixedSampleInfoDataset:
         self.samples = samples
         self.sample_info = sample_info
 
-        # Create a TensorFlow dataset that holds (sample, sample_info) tuples
-        self.tf_dataset = tf.data.Dataset.from_tensor_slices((
-            tf.convert_to_tensor(self.samples),
-            tuple(tf.convert_to_tensor(info) for info in self.sample_info)
-        ))
+        # Get the number of tensors in each tuple (corresponds to the number of input layers the model has)
+        num_tensors = len(samples[0])
+
+        # Create separate lists: one for each input layer and separate the tuples into lists
+        sample_tensor_lists = [[] for _ in range(num_tensors)]
+        for s in samples:
+            for i, data_tensor in enumerate(s):
+                sample_tensor_lists[i].append(data_tensor)
+
+        # In order to deal with models that have different input shapes for different layers, we need first to
+        # organize the data in a dictionary in order to use tf.data.Dataset.from_tensor_slices
+        samples_dict = {f'tensor_{i}': tensors for i, tensors in enumerate(sample_tensor_lists)}
+        info_dict = {f'info_{i}': tf.convert_to_tensor(info) for i, info in enumerate(self.sample_info)}
+        combined_dict = {**samples_dict, **info_dict}
+
+        tf_dataset = tf.data.Dataset.from_tensor_slices(combined_dict)
+
+        # Map the dataset to return tuples instead of dict
+        def reorganize_ds_outputs(ds_output):
+            tensors = tuple(ds_output[f'tensor_{i}'] for i in range(num_tensors))
+            infos = tuple(ds_output[f'info_{i}'] for i in range(len(sample_info)))
+            return tensors, infos
+
+        self.tf_dataset = tf_dataset.map(reorganize_ds_outputs)
 
     def __len__(self):
         return len(self.samples)