mct-nightly 1.10.0.20231211.post417__py3-none-any.whl → 1.10.0.20231213.post410__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-1.10.0.20231211.post417.dist-info → mct_nightly-1.10.0.20231213.post410.dist-info}/METADATA +1 -1
- {mct_nightly-1.10.0.20231211.post417.dist-info → mct_nightly-1.10.0.20231213.post410.dist-info}/RECORD +8 -7
- model_compression_toolkit/core/common/graph/base_graph.py +24 -3
- model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/functional_batch_norm.py +94 -0
- model_compression_toolkit/core/pytorch/pytorch_implementation.py +4 -1
- {mct_nightly-1.10.0.20231211.post417.dist-info → mct_nightly-1.10.0.20231213.post410.dist-info}/LICENSE.md +0 -0
- {mct_nightly-1.10.0.20231211.post417.dist-info → mct_nightly-1.10.0.20231213.post410.dist-info}/WHEEL +0 -0
- {mct_nightly-1.10.0.20231211.post417.dist-info → mct_nightly-1.10.0.20231213.post410.dist-info}/top_level.txt +0 -0
|
@@ -32,7 +32,7 @@ model_compression_toolkit/core/common/collectors/statistics_collector_generator.
|
|
|
32
32
|
model_compression_toolkit/core/common/fusion/__init__.py,sha256=Rf1RcYmelmdZmBV5qOKvKWF575ofc06JFQSq83Jz99A,696
|
|
33
33
|
model_compression_toolkit/core/common/fusion/layer_fusing.py,sha256=tIsWFYc771o59uvq5fxAaBmOCnd_gd-_xMbQI9SupQA,5479
|
|
34
34
|
model_compression_toolkit/core/common/graph/__init__.py,sha256=Xr-Lt_qXMdrCnnOaUS_OJP_3iTTGfPCLf8_vSrQgCs0,773
|
|
35
|
-
model_compression_toolkit/core/common/graph/base_graph.py,sha256
|
|
35
|
+
model_compression_toolkit/core/common/graph/base_graph.py,sha256=UHhXCWXh4hK7cg2sZrNLiTYSYsvDc3yhImcdKbQ0VVs,30929
|
|
36
36
|
model_compression_toolkit/core/common/graph/base_node.py,sha256=csIgi5ex7EquQsF34w5waRIHzbg7XitvIqQgCC29azs,21118
|
|
37
37
|
model_compression_toolkit/core/common/graph/edge.py,sha256=K6Wc2hBcIqig5PbbLhbjtTgYtkyZEohfgj4Wn_J5yEA,3733
|
|
38
38
|
model_compression_toolkit/core/common/graph/functional_node.py,sha256=0TpYNa2ODZ0M9lQ2z_GsStqAbrg1Muwdni74LjphAh0,2922
|
|
@@ -197,7 +197,7 @@ model_compression_toolkit/core/pytorch/__init__.py,sha256=Rf1RcYmelmdZmBV5qOKvKW
|
|
|
197
197
|
model_compression_toolkit/core/pytorch/constants.py,sha256=Kt_GDwe3yX9oMS1DI2eXYuUT25_lpjeCkxpstsAiXCI,2472
|
|
198
198
|
model_compression_toolkit/core/pytorch/default_framework_info.py,sha256=qee8TFcDro2lfyXe_fujjX2OlxELTyKSsLlZ7QkzeXU,4200
|
|
199
199
|
model_compression_toolkit/core/pytorch/kpi_data_facade.py,sha256=J0IDOtFMVFSFyBXDzNGbwJfHu89iRBJFdid1_wFB-xQ,8482
|
|
200
|
-
model_compression_toolkit/core/pytorch/pytorch_implementation.py,sha256=
|
|
200
|
+
model_compression_toolkit/core/pytorch/pytorch_implementation.py,sha256=w4yLHjJmOxfkSgApEx9rWAEbv9vkLnZik5JQvaX55FM,26654
|
|
201
201
|
model_compression_toolkit/core/pytorch/pytorch_node_prior_info.py,sha256=n_B4a6FMwM9D2w8kzy3oenBWZgXNZuIZgTJC6JEuTy0,3250
|
|
202
202
|
model_compression_toolkit/core/pytorch/utils.py,sha256=rBQMAbWluyIMjVfeghzq6FZv3sR_khszSRpbWvwussw,2959
|
|
203
203
|
model_compression_toolkit/core/pytorch/back2framework/__init__.py,sha256=H_WixgN0elVWf3exgGYsi58imPoYDj5eYPeh6x4yfug,813
|
|
@@ -216,6 +216,7 @@ model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/batchno
|
|
|
216
216
|
model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/batchnorm_reconstruction.py,sha256=B7aC2TZNrQJ2oQVGBFhKAVqdUU5lYVJSMmwKhjxOHWk,2822
|
|
217
217
|
model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/batchnorm_refusing.py,sha256=JDWOaNwYrZG0zTwd3HwoZUM3tKu7zPbzLOrqNQsu8xA,2162
|
|
218
218
|
model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/const_holder_conv.py,sha256=4mnowFmfDQjKlhHqsNto1iL4WbHyh4cM3Lf67Z-Cnzc,4804
|
|
219
|
+
model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/functional_batch_norm.py,sha256=bSexq2A7WxLLm13v67SgVbb4T1Y6nrKQDZfk4iSj_ec,3941
|
|
219
220
|
model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/linear_collapsing.py,sha256=qCNT3L4mnZtIP75c8YwImvsTWdPIdsEvO4pc3SE4y6s,5797
|
|
220
221
|
model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/multi_head_attention_decomposition.py,sha256=gGlVE1xUa1Pv2NbRx0I2y5Okg3kneBWSx9JwULuTWz0,38353
|
|
221
222
|
model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/permute_call_method.py,sha256=EMCviyFyJFLEKuAUz3rZHLfB9MAU1kywSBL2XQNzLlg,1953
|
|
@@ -450,8 +451,8 @@ model_compression_toolkit/trainable_infrastructure/keras/quantize_wrapper.py,sha
|
|
|
450
451
|
model_compression_toolkit/trainable_infrastructure/keras/quantizer_utils.py,sha256=MVwXNymmFRB2NXIBx4e2mdJ1RfoHxRPYRgjb1MQP5kY,1797
|
|
451
452
|
model_compression_toolkit/trainable_infrastructure/pytorch/__init__.py,sha256=huHoBUcKNB6BnY6YaUCcFvdyBtBI172ZoUD8ZYeNc6o,696
|
|
452
453
|
model_compression_toolkit/trainable_infrastructure/pytorch/base_pytorch_quantizer.py,sha256=SbvRlIdE32PEBsINt1bhSqvrKL_zbM9V-aeSkOn-sw4,3083
|
|
453
|
-
mct_nightly-1.10.0.
|
|
454
|
-
mct_nightly-1.10.0.
|
|
455
|
-
mct_nightly-1.10.0.
|
|
456
|
-
mct_nightly-1.10.0.
|
|
457
|
-
mct_nightly-1.10.0.
|
|
454
|
+
mct_nightly-1.10.0.20231213.post410.dist-info/LICENSE.md,sha256=aYSSIb-5AFPeITTvXm1UAoe0uYBiMmSS8flvXaaFUks,10174
|
|
455
|
+
mct_nightly-1.10.0.20231213.post410.dist-info/METADATA,sha256=udqlXh19QUWj8Kl3urGNDSPuK2qjV7ZEY_8OFS-gtvw,16232
|
|
456
|
+
mct_nightly-1.10.0.20231213.post410.dist-info/WHEEL,sha256=oiQVh_5PnQM0E3gPdiz09WCNmwiHDMaGer_elqB3coM,92
|
|
457
|
+
mct_nightly-1.10.0.20231213.post410.dist-info/top_level.txt,sha256=gsYA8juk0Z-ZmQRKULkb3JLGdOdz8jW_cMRjisn9ga4,26
|
|
458
|
+
mct_nightly-1.10.0.20231213.post410.dist-info/RECORD,,
|
|
@@ -299,19 +299,24 @@ class Graph(nx.MultiDiGraph, GraphSearches):
|
|
|
299
299
|
return [edges_list.sink_node for edges_list in self.out_edges(node_obj)]
|
|
300
300
|
|
|
301
301
|
def get_prev_nodes(self,
|
|
302
|
-
node_obj: BaseNode
|
|
302
|
+
node_obj: BaseNode,
|
|
303
|
+
sink_index_sorted: bool = False) -> List[BaseNode]:
|
|
303
304
|
"""
|
|
304
305
|
Get previous nodes (in a topological order) of a node.
|
|
305
306
|
|
|
306
307
|
Args:
|
|
307
308
|
node_obj: Node to get its previous nodes.
|
|
309
|
+
sink_index_sorted: Whether to sort the returned list by the sink_index of the edges.
|
|
308
310
|
|
|
309
311
|
Returns:
|
|
310
312
|
List of input nodes objects.
|
|
311
313
|
|
|
312
314
|
"""
|
|
313
|
-
|
|
314
|
-
|
|
315
|
+
if sink_index_sorted:
|
|
316
|
+
sort_attr = 'sink_index'
|
|
317
|
+
else:
|
|
318
|
+
sort_attr = None
|
|
319
|
+
return [edges_list.source_node for edges_list in self.incoming_edges(node_obj, sort_by_attr=sort_attr)]
|
|
315
320
|
|
|
316
321
|
def reconnect_out_edges(self,
|
|
317
322
|
current_node: BaseNode,
|
|
@@ -705,3 +710,19 @@ class Graph(nx.MultiDiGraph, GraphSearches):
|
|
|
705
710
|
|
|
706
711
|
"""
|
|
707
712
|
return all([n.is_all_activation_candidates_equal() for n in self.nodes])
|
|
713
|
+
|
|
714
|
+
def replace_node(self, node_to_replace: BaseNode, new_node: BaseNode):
|
|
715
|
+
"""
|
|
716
|
+
Replaces a node in the graph with a new node.
|
|
717
|
+
|
|
718
|
+
Args:
|
|
719
|
+
node_to_replace: The node to replace.
|
|
720
|
+
new_node: The new node to replace with.
|
|
721
|
+
|
|
722
|
+
"""
|
|
723
|
+
self.add_node(new_node)
|
|
724
|
+
self.reconnect_out_edges(node_to_replace, new_node)
|
|
725
|
+
self.reconnect_in_edges(node_to_replace, new_node)
|
|
726
|
+
self.replace_output_node(node_to_replace, new_node)
|
|
727
|
+
self.replace_input_node(node_to_replace, new_node)
|
|
728
|
+
self.remove_node(node_to_replace)
|
model_compression_toolkit/core/pytorch/graph_substitutions/substitutions/functional_batch_norm.py
ADDED
|
@@ -0,0 +1,94 @@
|
|
|
1
|
+
# Copyright 2023 Sony Semiconductor Israel, Inc. All rights reserved.
|
|
2
|
+
#
|
|
3
|
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
4
|
+
# you may not use this file except in compliance with the License.
|
|
5
|
+
# You may obtain a copy of the License at
|
|
6
|
+
#
|
|
7
|
+
# http://www.apache.org/licenses/LICENSE-2.0
|
|
8
|
+
#
|
|
9
|
+
# Unless required by applicable law or agreed to in writing, software
|
|
10
|
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
11
|
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
12
|
+
# See the License for the specific language governing permissions and
|
|
13
|
+
# limitations under the License.
|
|
14
|
+
# ==============================================================================
|
|
15
|
+
from torch import nn
|
|
16
|
+
import torch.nn.functional as F
|
|
17
|
+
|
|
18
|
+
from model_compression_toolkit.core.common.graph.graph_matchers import NodeOperationMatcher
|
|
19
|
+
from model_compression_toolkit.core import common
|
|
20
|
+
from model_compression_toolkit.core.common import BaseNode, Graph
|
|
21
|
+
from model_compression_toolkit.core.pytorch.constants import *
|
|
22
|
+
from model_compression_toolkit.logger import Logger
|
|
23
|
+
|
|
24
|
+
|
|
25
|
+
class FunctionalBatchNorm(common.BaseSubstitution):
|
|
26
|
+
"""
|
|
27
|
+
Replace functional batch_norm with BatchNorm2d.
|
|
28
|
+
"""
|
|
29
|
+
|
|
30
|
+
def __init__(self):
|
|
31
|
+
"""
|
|
32
|
+
Matches: functional batch_norm
|
|
33
|
+
"""
|
|
34
|
+
bn_node = NodeOperationMatcher(F.batch_norm)
|
|
35
|
+
super().__init__(matcher_instance=bn_node)
|
|
36
|
+
|
|
37
|
+
def get_attributes_from_inputs(self, graph: Graph, node: BaseNode) -> dict:
|
|
38
|
+
input_nodes = graph.get_prev_nodes(node, sink_index_sorted=True)
|
|
39
|
+
|
|
40
|
+
if len(input_nodes) == 5:
|
|
41
|
+
return {
|
|
42
|
+
MOVING_MEAN: list(input_nodes[1].weights.values())[0],
|
|
43
|
+
MOVING_VARIANCE: list(input_nodes[2].weights.values())[0],
|
|
44
|
+
GAMMA: list(input_nodes[3].weights.values())[0],
|
|
45
|
+
BETA: list(input_nodes[4].weights.values())[0]
|
|
46
|
+
}
|
|
47
|
+
else:
|
|
48
|
+
Logger.warning(f'functional batch_norm is only folded in the 5 inputs case (input, mean, var, gamma, beta),'
|
|
49
|
+
f'got {len(input_nodes)}')
|
|
50
|
+
return {}
|
|
51
|
+
|
|
52
|
+
def substitute(self,
|
|
53
|
+
graph: Graph,
|
|
54
|
+
node: BaseNode) -> Graph:
|
|
55
|
+
"""
|
|
56
|
+
Substitute functional.batch_norm and its inputs with BatchNorm2d.
|
|
57
|
+
Args:
|
|
58
|
+
graph: Graph we apply the substitution on.
|
|
59
|
+
node: node that match the pattern in the substitution init.
|
|
60
|
+
|
|
61
|
+
Returns:
|
|
62
|
+
Graph after applying the substitution.
|
|
63
|
+
"""
|
|
64
|
+
# if the input is not a 4D tensor, we can't substitute it with BatchNorm2d
|
|
65
|
+
if len(node.input_shape[0]) != 4:
|
|
66
|
+
return graph
|
|
67
|
+
out_channels = node.output_shape[0][1]
|
|
68
|
+
|
|
69
|
+
bn_node_weights = self.get_attributes_from_inputs(graph, node)
|
|
70
|
+
if not bn_node_weights:
|
|
71
|
+
return graph
|
|
72
|
+
new_batchnorm2d = BaseNode(name=node.name + '_into_BatchNorm2d',
|
|
73
|
+
framework_attr={NUM_FEATURES: out_channels,
|
|
74
|
+
EPSILON: EPSILON_VAL,
|
|
75
|
+
MOMENTUM: MOMENTUM_VAL},
|
|
76
|
+
input_shape=node.output_shape,
|
|
77
|
+
output_shape=node.output_shape,
|
|
78
|
+
weights=bn_node_weights,
|
|
79
|
+
layer_class=nn.BatchNorm2d)
|
|
80
|
+
|
|
81
|
+
num_nodes_before_substitution = len(graph.nodes)
|
|
82
|
+
num_edges_before_substitution = len(graph.edges)
|
|
83
|
+
|
|
84
|
+
batch_norm_consts = graph.get_prev_nodes(node)[1:]
|
|
85
|
+
for const in batch_norm_consts:
|
|
86
|
+
graph.remove_edge(const, node)
|
|
87
|
+
graph.remove_node(const)
|
|
88
|
+
|
|
89
|
+
graph.replace_node(node, new_batchnorm2d)
|
|
90
|
+
|
|
91
|
+
assert num_nodes_before_substitution - len(graph.nodes) == len(batch_norm_consts)
|
|
92
|
+
assert num_edges_before_substitution - len(graph.edges) == len(batch_norm_consts)
|
|
93
|
+
|
|
94
|
+
return graph
|
|
@@ -48,6 +48,8 @@ from model_compression_toolkit.core.pytorch.graph_substitutions.substitutions.ba
|
|
|
48
48
|
pytorch_batchnorm_reconstruction
|
|
49
49
|
from model_compression_toolkit.core.pytorch.graph_substitutions.substitutions.batchnorm_refusing import \
|
|
50
50
|
pytorch_batchnorm_refusing
|
|
51
|
+
from model_compression_toolkit.core.pytorch.graph_substitutions.substitutions.functional_batch_norm import \
|
|
52
|
+
FunctionalBatchNorm
|
|
51
53
|
from model_compression_toolkit.core.pytorch.graph_substitutions.substitutions.linear_collapsing import \
|
|
52
54
|
pytorch_linear_collapsing
|
|
53
55
|
from model_compression_toolkit.core.pytorch.graph_substitutions.substitutions.multi_head_attention_decomposition \
|
|
@@ -243,7 +245,8 @@ class PytorchImplementation(FrameworkImplementation):
|
|
|
243
245
|
return [ReshapeWithStaticShapes(),
|
|
244
246
|
MultiHeadAttentionDecomposition(),
|
|
245
247
|
PermuteCallMethod(),
|
|
246
|
-
ConstantHolderConv(fw_info)
|
|
248
|
+
ConstantHolderConv(fw_info),
|
|
249
|
+
FunctionalBatchNorm()]
|
|
247
250
|
|
|
248
251
|
def get_substitutions_pre_statistics_collection(self,
|
|
249
252
|
quant_config: QuantizationConfig
|
|
File without changes
|
|
File without changes
|
|
File without changes
|