mct-nightly 2.3.0.20250428.605__py3-none-any.whl → 2.3.0.20250430.538__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {mct_nightly-2.3.0.20250428.605.dist-info → mct_nightly-2.3.0.20250430.538.dist-info}/METADATA +3 -3
- {mct_nightly-2.3.0.20250428.605.dist-info → mct_nightly-2.3.0.20250430.538.dist-info}/RECORD +10 -10
- model_compression_toolkit/__init__.py +1 -1
- model_compression_toolkit/core/common/mixed_precision/mixed_precision_quantization_config.py +2 -0
- model_compression_toolkit/core/common/mixed_precision/mixed_precision_search_manager.py +1 -0
- model_compression_toolkit/core/common/mixed_precision/sensitivity_evaluation.py +16 -4
- model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_onnx_pytorch_exporter.py +28 -8
- {mct_nightly-2.3.0.20250428.605.dist-info → mct_nightly-2.3.0.20250430.538.dist-info}/WHEEL +0 -0
- {mct_nightly-2.3.0.20250428.605.dist-info → mct_nightly-2.3.0.20250430.538.dist-info}/licenses/LICENSE.md +0 -0
- {mct_nightly-2.3.0.20250428.605.dist-info → mct_nightly-2.3.0.20250430.538.dist-info}/top_level.txt +0 -0
{mct_nightly-2.3.0.20250428.605.dist-info → mct_nightly-2.3.0.20250430.538.dist-info}/METADATA
RENAMED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: mct-nightly
|
|
3
|
-
Version: 2.3.0.
|
|
3
|
+
Version: 2.3.0.20250430.538
|
|
4
4
|
Summary: A Model Compression Toolkit for neural networks
|
|
5
5
|
Classifier: Programming Language :: Python :: 3
|
|
6
6
|
Classifier: License :: OSI Approved :: Apache Software License
|
|
@@ -51,7 +51,7 @@ ______________________________________________________________________
|
|
|
51
51
|
</p>
|
|
52
52
|
<p align="center">
|
|
53
53
|
<a href="https://sony.github.io/model_optimization#prerequisites"><img src="https://img.shields.io/badge/pytorch-2.2%20%7C%202.3%20%7C%202.4%20%7C%202.5-blue" /></a>
|
|
54
|
-
<a href="https://sony.github.io/model_optimization#prerequisites"><img src="https://img.shields.io/badge/tensorflow-
|
|
54
|
+
<a href="https://sony.github.io/model_optimization#prerequisites"><img src="https://img.shields.io/badge/tensorflow-02.14%20%7C%202.15-blue" /></a>
|
|
55
55
|
<a href="https://sony.github.io/model_optimization#prerequisites"><img src="https://img.shields.io/badge/python-3.9%20%7C%203.10%20%7C%203.11%20%7C%203.12-blue" /></a>
|
|
56
56
|
<a href="https://github.com/sony/model_optimization/releases"><img src="https://img.shields.io/github/v/release/sony/model_optimization" /></a>
|
|
57
57
|
<a href="https://github.com/sony/model_optimization/blob/main/LICENSE.md"><img src="https://img.shields.io/badge/license-Apache%202.0-blue" /></a>
|
|
@@ -171,7 +171,7 @@ Currently, MCT is being tested on various Python, Pytorch and TensorFlow version
|
|
|
171
171
|
| Python 3.12 | [](https://github.com/sony/model_optimization/actions/workflows/run_tests_python312_pytorch22.yml) | [](https://github.com/sony/model_optimization/actions/workflows/run_tests_python312_pytorch23.yml) | [](https://github.com/sony/model_optimization/actions/workflows/run_tests_python312_pytorch24.yml) | [](https://github.com/sony/model_optimization/actions/workflows/run_tests_python312_pytorch25.yml) |
|
|
172
172
|
|
|
173
173
|
| | TensorFlow 2.14 | TensorFlow 2.15 |
|
|
174
|
-
|
|
174
|
+
|-------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
|
|
175
175
|
| Python 3.9 | [](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_keras214.yml) | [](https://github.com/sony/model_optimization/actions/workflows/run_tests_python39_keras215.yml) |
|
|
176
176
|
| Python 3.10 | [](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_keras214.yml) | [](https://github.com/sony/model_optimization/actions/workflows/run_tests_python310_keras215.yml) |
|
|
177
177
|
| Python 3.11 | [](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_keras214.yml) | [](https://github.com/sony/model_optimization/actions/workflows/run_tests_python311_keras215.yml) |
|
{mct_nightly-2.3.0.20250428.605.dist-info → mct_nightly-2.3.0.20250430.538.dist-info}/RECORD
RENAMED
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
mct_nightly-2.3.0.
|
|
2
|
-
model_compression_toolkit/__init__.py,sha256=
|
|
1
|
+
mct_nightly-2.3.0.20250430.538.dist-info/licenses/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
|
|
2
|
+
model_compression_toolkit/__init__.py,sha256=_uN22zhCLe0oS_Ex9lxOL4jmaLZUIfp9zJ82XXgBrqQ,1557
|
|
3
3
|
model_compression_toolkit/constants.py,sha256=iJ6vfTjC2oFIZWt8wvHoxEw5YJi3yl0Hd4q30_8q0Zc,3958
|
|
4
4
|
model_compression_toolkit/defaultdict.py,sha256=LSc-sbZYXENMCw3U9F4GiXuv67IKpdn0Qm7Fr11jy-4,2277
|
|
5
5
|
model_compression_toolkit/logger.py,sha256=L3q7tn3Uht0i_7phnlOWMR2Te2zvzrt2HOz9vYEInts,4529
|
|
@@ -66,11 +66,11 @@ model_compression_toolkit/core/common/mixed_precision/configurable_quant_id.py,s
|
|
|
66
66
|
model_compression_toolkit/core/common/mixed_precision/configurable_quantizer_utils.py,sha256=7dKMi5S0zQZ16m8NWn1XIuoXsKuZUg64G4-uK8-j1PQ,5177
|
|
67
67
|
model_compression_toolkit/core/common/mixed_precision/distance_weighting.py,sha256=-x8edUyudu1EAEM66AuXPtgayLpzbxoLNubfEbFM5kU,2867
|
|
68
68
|
model_compression_toolkit/core/common/mixed_precision/mixed_precision_candidates_filter.py,sha256=6pLUEEIqRTVIlCYQC4JIvY55KAvuBHEX8uTOQ-1Ac4Q,3859
|
|
69
|
-
model_compression_toolkit/core/common/mixed_precision/mixed_precision_quantization_config.py,sha256=
|
|
69
|
+
model_compression_toolkit/core/common/mixed_precision/mixed_precision_quantization_config.py,sha256=onHgDwfw8CUbZFNU-RYit9eqA6FrzAtFA3akVZ2d7IM,4533
|
|
70
70
|
model_compression_toolkit/core/common/mixed_precision/mixed_precision_ru_helper.py,sha256=-hOMBucYn12ePyLd0b1KxniPOIRu4b53SwEzv0bWToI,4943
|
|
71
71
|
model_compression_toolkit/core/common/mixed_precision/mixed_precision_search_facade.py,sha256=d5-3j2e_rdcQOT7c4s0p7640i3nSetjJ6MgMhhMM7dc,6152
|
|
72
|
-
model_compression_toolkit/core/common/mixed_precision/mixed_precision_search_manager.py,sha256=
|
|
73
|
-
model_compression_toolkit/core/common/mixed_precision/sensitivity_evaluation.py,sha256=
|
|
72
|
+
model_compression_toolkit/core/common/mixed_precision/mixed_precision_search_manager.py,sha256=J8io_axti6gRoch9QR0FmKOP8JSHGeKqX95rf-nG6fI,37719
|
|
73
|
+
model_compression_toolkit/core/common/mixed_precision/sensitivity_evaluation.py,sha256=R3UIO9lKf-lpEGfJOqgpQAXdP1IWMatWxXKYDkhWj_E,28096
|
|
74
74
|
model_compression_toolkit/core/common/mixed_precision/set_layer_to_bitwidth.py,sha256=P8QtKgFXtt5b2RoubzI5OGlCfbEfZsAirjyrkFzK26A,2846
|
|
75
75
|
model_compression_toolkit/core/common/mixed_precision/solution_refinement_procedure.py,sha256=S1ChgxtUjzXJufNWyRbKoNdyNC6fGUjPeComDMx8ZCo,9479
|
|
76
76
|
model_compression_toolkit/core/common/mixed_precision/resource_utilization_tools/__init__.py,sha256=Rf1RcYmelmdZmBV5qOKvKWF575ofc06JFQSq83Jz99A,696
|
|
@@ -335,7 +335,7 @@ model_compression_toolkit/exporter/model_exporter/keras/mctq_keras_exporter.py,s
|
|
|
335
335
|
model_compression_toolkit/exporter/model_exporter/pytorch/__init__.py,sha256=uZ2RigbY9O2PJ0Il8wPpS_s7frgg9WUGd_SHeKGyl1A,699
|
|
336
336
|
model_compression_toolkit/exporter/model_exporter/pytorch/base_pytorch_exporter.py,sha256=UPVkEUQCMZ4Lld6CRnEOPEmlfe5vcQZG0Q3FwRBodD4,4021
|
|
337
337
|
model_compression_toolkit/exporter/model_exporter/pytorch/export_serialization_format.py,sha256=bPevy6OBqng41PqytBR55e6cBEuyrUS0H8dWX4zgjQ4,967
|
|
338
|
-
model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_onnx_pytorch_exporter.py,sha256=
|
|
338
|
+
model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_onnx_pytorch_exporter.py,sha256=5DHg8ch8_DtSQ7M5Aw3LcZLSVFqFH556cKcosp3Ik-I,8720
|
|
339
339
|
model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_torchscript_pytorch_exporter.py,sha256=ksWV2A-Njo-wAxQ_Ye2sLIZXBWJ_WNyjT7-qFFwvV2o,2897
|
|
340
340
|
model_compression_toolkit/exporter/model_exporter/pytorch/pytorch_export_facade.py,sha256=8vYGKa58BkasvoHejYaPwubOJPcW0s-RY79_Kkw0Hy8,6236
|
|
341
341
|
model_compression_toolkit/exporter/model_wrapper/__init__.py,sha256=7CF2zvpTrIEm8qnbuHnLZyTZkwBBxV24V8QA0oxGbh0,1187
|
|
@@ -528,7 +528,7 @@ model_compression_toolkit/xquant/pytorch/model_analyzer.py,sha256=b93o800yVB3Z-i
|
|
|
528
528
|
model_compression_toolkit/xquant/pytorch/pytorch_report_utils.py,sha256=UVN_S9ULHBEldBpShCOt8-soT8YTQ5oE362y96qF_FA,3950
|
|
529
529
|
model_compression_toolkit/xquant/pytorch/similarity_functions.py,sha256=CERxq5K8rqaiE-DlwhZBTUd9x69dtYJlkHOPLB54vm8,2354
|
|
530
530
|
model_compression_toolkit/xquant/pytorch/tensorboard_utils.py,sha256=mkoEktLFFHtEKzzFRn_jCnxjhJolK12TZ5AQeDHzUO8,9767
|
|
531
|
-
mct_nightly-2.3.0.
|
|
532
|
-
mct_nightly-2.3.0.
|
|
533
|
-
mct_nightly-2.3.0.
|
|
534
|
-
mct_nightly-2.3.0.
|
|
531
|
+
mct_nightly-2.3.0.20250430.538.dist-info/METADATA,sha256=8l9P7bLANjVXo-Z5_aR8x8VtcIWHhdYNNHQmoOS6cig,25101
|
|
532
|
+
mct_nightly-2.3.0.20250430.538.dist-info/WHEEL,sha256=ck4Vq1_RXyvS4Jt6SI0Vz6fyVs4GWg7AINwpsaGEgPE,91
|
|
533
|
+
mct_nightly-2.3.0.20250430.538.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
|
|
534
|
+
mct_nightly-2.3.0.20250430.538.dist-info/RECORD,,
|
|
@@ -27,4 +27,4 @@ from model_compression_toolkit import data_generation
|
|
|
27
27
|
from model_compression_toolkit import pruning
|
|
28
28
|
from model_compression_toolkit.trainable_infrastructure.keras.load_model import keras_load_quantized_model
|
|
29
29
|
|
|
30
|
-
__version__ = "2.3.0.
|
|
30
|
+
__version__ = "2.3.0.20250430.000538"
|
model_compression_toolkit/core/common/mixed_precision/mixed_precision_quantization_config.py
CHANGED
|
@@ -27,6 +27,7 @@ class MixedPrecisionQuantizationConfig:
|
|
|
27
27
|
Args:
|
|
28
28
|
compute_distance_fn (Callable): Function to compute a distance between two tensors. If None, using pre-defined distance methods based on the layer type for each layer.
|
|
29
29
|
distance_weighting_method (MpDistanceWeighting): MpDistanceWeighting enum value that provides a function to use when weighting the distances among different layers when computing the sensitivity metric.
|
|
30
|
+
custom_metric_fn (Callable): Function to compute a custom metric. As input gets the model_mp and returns a float value for metric. If None, uses interest point metric.
|
|
30
31
|
num_of_images (int): Number of images to use to evaluate the sensitivity of a mixed-precision model comparing to the float model.
|
|
31
32
|
configuration_overwrite (List[int]): A list of integers that enables overwrite of mixed precision with a predefined one.
|
|
32
33
|
num_interest_points_factor (float): A multiplication factor between zero and one (represents percentage) to reduce the number of interest points used to calculate the distance metric.
|
|
@@ -39,6 +40,7 @@ class MixedPrecisionQuantizationConfig:
|
|
|
39
40
|
|
|
40
41
|
compute_distance_fn: Optional[Callable] = None
|
|
41
42
|
distance_weighting_method: MpDistanceWeighting = MpDistanceWeighting.AVG
|
|
43
|
+
custom_metric_fn: Optional[Callable] = None
|
|
42
44
|
num_of_images: int = MP_DEFAULT_NUM_SAMPLES
|
|
43
45
|
configuration_overwrite: Optional[List[int]] = None
|
|
44
46
|
num_interest_points_factor: float = field(default=1.0, metadata={"description": "Should be between 0.0 and 1.0"})
|
|
@@ -169,6 +169,7 @@ class MixedPrecisionSearchManager:
|
|
|
169
169
|
return self.sensitivity_evaluator.compute_metric(topo_cfg(cfg),
|
|
170
170
|
node_idx,
|
|
171
171
|
topo_cfg(baseline_cfg) if baseline_cfg else None)
|
|
172
|
+
|
|
172
173
|
if self.using_virtual_graph:
|
|
173
174
|
origin_max_config = self.config_reconstruction_helper.reconstruct_config_from_virtual_graph(
|
|
174
175
|
self.max_ru_config)
|
|
@@ -89,6 +89,9 @@ class SensitivityEvaluation:
|
|
|
89
89
|
self.interest_points = get_mp_interest_points(graph,
|
|
90
90
|
fw_impl.count_node_for_mixed_precision_interest_points,
|
|
91
91
|
quant_config.num_interest_points_factor)
|
|
92
|
+
# If using a custom metric - return only model outputs
|
|
93
|
+
if self.quant_config.custom_metric_fn is not None:
|
|
94
|
+
self.interest_points = []
|
|
92
95
|
|
|
93
96
|
# We use normalized MSE when not running hessian-based. For Hessian-based normalized MSE is not needed
|
|
94
97
|
# because hessian weights already do normalization.
|
|
@@ -96,6 +99,9 @@ class SensitivityEvaluation:
|
|
|
96
99
|
self.ips_distance_fns, self.ips_axis = self._init_metric_points_lists(self.interest_points, use_normalized_mse)
|
|
97
100
|
|
|
98
101
|
self.output_points = get_output_nodes_for_metric(graph)
|
|
102
|
+
# If using a custom metric - return all model outputs
|
|
103
|
+
if self.quant_config.custom_metric_fn is not None:
|
|
104
|
+
self.output_points = [n.node for n in graph.get_outputs()]
|
|
99
105
|
self.out_ps_distance_fns, self.out_ps_axis = self._init_metric_points_lists(self.output_points,
|
|
100
106
|
use_normalized_mse)
|
|
101
107
|
|
|
@@ -160,7 +166,7 @@ class SensitivityEvaluation:
|
|
|
160
166
|
"""
|
|
161
167
|
Compute the sensitivity metric of the MP model for a given configuration (the sensitivity
|
|
162
168
|
is computed based on the similarity of the interest points' outputs between the MP model
|
|
163
|
-
and the float model).
|
|
169
|
+
and the float model or a custom metric if given).
|
|
164
170
|
|
|
165
171
|
Args:
|
|
166
172
|
mp_model_configuration: Bitwidth configuration to use to configure the MP model.
|
|
@@ -177,15 +183,21 @@ class SensitivityEvaluation:
|
|
|
177
183
|
node_idx)
|
|
178
184
|
|
|
179
185
|
# Compute the distance metric
|
|
180
|
-
|
|
186
|
+
if self.quant_config.custom_metric_fn is None:
|
|
187
|
+
ipts_distances, out_pts_distances = self._compute_distance()
|
|
188
|
+
sensitivity_metric = self._compute_mp_distance_measure(ipts_distances, out_pts_distances,
|
|
189
|
+
self.quant_config.distance_weighting_method)
|
|
190
|
+
else:
|
|
191
|
+
sensitivity_metric = self.quant_config.custom_metric_fn(self.model_mp)
|
|
192
|
+
if not isinstance(sensitivity_metric, (float, np.floating)):
|
|
193
|
+
raise TypeError(f'The custom_metric_fn is expected to return float or numpy float, got {type(sensitivity_metric).__name__}')
|
|
181
194
|
|
|
182
195
|
# Configure MP model back to the same configuration as the baseline model if baseline provided
|
|
183
196
|
if baseline_mp_configuration is not None:
|
|
184
197
|
self._configure_bitwidths_model(baseline_mp_configuration,
|
|
185
198
|
node_idx)
|
|
186
199
|
|
|
187
|
-
return
|
|
188
|
-
self.quant_config.distance_weighting_method)
|
|
200
|
+
return sensitivity_metric
|
|
189
201
|
|
|
190
202
|
def _init_baseline_tensors_list(self):
|
|
191
203
|
"""
|
model_compression_toolkit/exporter/model_exporter/pytorch/fakely_quant_onnx_pytorch_exporter.py
CHANGED
|
@@ -24,11 +24,11 @@ from model_compression_toolkit.core.pytorch.utils import to_torch_tensor
|
|
|
24
24
|
from model_compression_toolkit.exporter.model_exporter.pytorch.base_pytorch_exporter import BasePyTorchExporter
|
|
25
25
|
from mct_quantizers import pytorch_quantizers
|
|
26
26
|
|
|
27
|
-
|
|
28
27
|
if FOUND_ONNX:
|
|
29
28
|
import onnx
|
|
30
29
|
from mct_quantizers.pytorch.metadata import add_onnx_metadata
|
|
31
30
|
|
|
31
|
+
|
|
32
32
|
class FakelyQuantONNXPyTorchExporter(BasePyTorchExporter):
|
|
33
33
|
"""
|
|
34
34
|
Exporter for fakely-quant PyTorch models.
|
|
@@ -63,7 +63,7 @@ if FOUND_ONNX:
|
|
|
63
63
|
self._use_onnx_custom_quantizer_ops = use_onnx_custom_quantizer_ops
|
|
64
64
|
self._onnx_opset_version = onnx_opset_version
|
|
65
65
|
|
|
66
|
-
def export(self) -> None:
|
|
66
|
+
def export(self, output_names=None) -> None:
|
|
67
67
|
"""
|
|
68
68
|
Convert an exportable (fully-quantized) PyTorch model to a fakely-quant model
|
|
69
69
|
(namely, weights that are in fake-quant format) and fake-quant layers for the activations.
|
|
@@ -95,6 +95,28 @@ if FOUND_ONNX:
|
|
|
95
95
|
Logger.info(f"Exporting fake-quant onnx model: {self.save_model_path}")
|
|
96
96
|
|
|
97
97
|
model_input = to_torch_tensor(next(self.repr_dataset()))
|
|
98
|
+
model_output = self.model(*model_input) if isinstance(model_input, (list, tuple)) else self.model(
|
|
99
|
+
model_input)
|
|
100
|
+
|
|
101
|
+
if output_names is None:
|
|
102
|
+
# Determine number of outputs and prepare output_names and dynamic_axes
|
|
103
|
+
if isinstance(model_output, (list, tuple)):
|
|
104
|
+
output_names = [f"output_{i}" for i in range(len(model_output))]
|
|
105
|
+
dynamic_axes = {'input': {0: 'batch_size'}}
|
|
106
|
+
dynamic_axes.update({name: {0: 'batch_size'} for name in output_names})
|
|
107
|
+
else:
|
|
108
|
+
output_names = ['output']
|
|
109
|
+
dynamic_axes = {'input': {0: 'batch_size'}, 'output': {0: 'batch_size'}}
|
|
110
|
+
else:
|
|
111
|
+
if isinstance(model_output, (list, tuple)):
|
|
112
|
+
num_of_outputs = len(model_output)
|
|
113
|
+
else:
|
|
114
|
+
num_of_outputs = 1
|
|
115
|
+
assert len(output_names) == num_of_outputs, (f"Mismatch between number of requested output names "
|
|
116
|
+
f"({output_names}) and model output count "
|
|
117
|
+
f"({num_of_outputs}):\n")
|
|
118
|
+
dynamic_axes = {'input': {0: 'batch_size'}}
|
|
119
|
+
dynamic_axes.update({name: {0: 'batch_size'} for name in output_names})
|
|
98
120
|
|
|
99
121
|
if hasattr(self.model, 'metadata'):
|
|
100
122
|
onnx_bytes = BytesIO()
|
|
@@ -104,9 +126,8 @@ if FOUND_ONNX:
|
|
|
104
126
|
opset_version=self._onnx_opset_version,
|
|
105
127
|
verbose=False,
|
|
106
128
|
input_names=['input'],
|
|
107
|
-
output_names=
|
|
108
|
-
dynamic_axes=
|
|
109
|
-
'output': {0: 'batch_size'}})
|
|
129
|
+
output_names=output_names,
|
|
130
|
+
dynamic_axes=dynamic_axes)
|
|
110
131
|
onnx_model = onnx.load_from_string(onnx_bytes.getvalue())
|
|
111
132
|
onnx_model = add_onnx_metadata(onnx_model, self.model.metadata)
|
|
112
133
|
onnx.save_model(onnx_model, self.save_model_path)
|
|
@@ -117,9 +138,8 @@ if FOUND_ONNX:
|
|
|
117
138
|
opset_version=self._onnx_opset_version,
|
|
118
139
|
verbose=False,
|
|
119
140
|
input_names=['input'],
|
|
120
|
-
output_names=
|
|
121
|
-
dynamic_axes=
|
|
122
|
-
'output': {0: 'batch_size'}})
|
|
141
|
+
output_names=output_names,
|
|
142
|
+
dynamic_axes=dynamic_axes)
|
|
123
143
|
|
|
124
144
|
for layer in self.model.children():
|
|
125
145
|
# Set disable for reuse for weight quantizers if quantizer was reused
|
|
File without changes
|
|
File without changes
|
{mct_nightly-2.3.0.20250428.605.dist-info → mct_nightly-2.3.0.20250430.538.dist-info}/top_level.txt
RENAMED
|
File without changes
|