1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
|
# SPDX-FileCopyrightText: Copyright 2022-2024, Arm Limited and/or its affiliates.
# SPDX-License-Identifier: Apache-2.0
"""Test for module optimizations/clustering."""
from __future__ import annotations
import math
from pathlib import Path
import pytest
import tensorflow as tf
from flaky import flaky
from mlia.nn.tensorflow.optimizations.clustering import Clusterer
from mlia.nn.tensorflow.optimizations.clustering import ClusteringConfiguration
from mlia.nn.tensorflow.optimizations.pruning import Pruner
from mlia.nn.tensorflow.optimizations.pruning import PruningConfiguration
from mlia.nn.tensorflow.tflite_convert import convert_to_tflite
from mlia.nn.tensorflow.tflite_metrics import ReportClusterMode
from mlia.nn.tensorflow.tflite_metrics import TFLiteMetrics
from tests.utils.common import get_dataset
from tests.utils.common import train_model
def _prune_model(
model: tf.keras.Model, target_sparsity: float, layers_to_prune: list[str] | None
) -> tf.keras.Model:
x_train, y_train = get_dataset()
batch_size = 1
num_epochs = 1
pruner = Pruner(
model,
PruningConfiguration(
target_sparsity,
layers_to_prune,
x_train,
y_train,
batch_size,
num_epochs,
),
)
pruner.apply_optimization()
pruned_model = pruner.get_model()
return pruned_model
def _test_num_unique_weights(
metrics: TFLiteMetrics,
target_num_clusters: int,
layers_to_cluster: list[str] | None,
) -> None:
clustered_uniqueness = metrics.num_unique_weights(
ReportClusterMode.NUM_CLUSTERS_PER_AXIS
)
num_clustered_layers = 0
for layer_num_clusters in clustered_uniqueness.values():
if layer_num_clusters[0] <= target_num_clusters:
num_clustered_layers += 1
expected_num_clustered_layers = len(layers_to_cluster or clustered_uniqueness)
assert num_clustered_layers == expected_num_clustered_layers
def _test_sparsity(
metrics: TFLiteMetrics,
target_sparsity: float,
layers_to_cluster: list[str] | None,
) -> None:
error_margin = 0.03
pruned_sparsity = metrics.sparsity_per_layer()
num_sparse_layers = 0
for layer_sparsity in pruned_sparsity.values():
if math.isclose(layer_sparsity, target_sparsity, abs_tol=error_margin):
num_sparse_layers += 1
# make sure we are having exactly as many sparse layers as we wanted
expected_num_sparse_layers = len(layers_to_cluster or pruned_sparsity)
assert num_sparse_layers == expected_num_sparse_layers
# This test fails sporadically for stochastic reasons, due to a threshold not being met.
# Re-running the test will help. We are yet to find a more deterministic approach
# to run the test, and in the meantime we classify it as a known issue.
# Additionally, flaky is (as of 2023) untyped and thus we need to silence the
# warning from mypy.
@flaky(max_runs=4, min_passes=1) # type: ignore
@pytest.mark.parametrize("target_num_clusters", (32, 4))
@pytest.mark.parametrize("sparsity_aware", (False, True))
@pytest.mark.parametrize("layers_to_cluster", (["conv1"], ["conv1", "conv2"], None))
def test_cluster_simple_model_fully(
target_num_clusters: int,
sparsity_aware: bool,
layers_to_cluster: list[str] | None,
tmp_path: Path,
test_keras_model: Path,
) -> None:
"""Simple MNIST test to see if clustering works correctly."""
target_sparsity = 0.5
base_model = tf.keras.models.load_model(str(test_keras_model))
train_model(base_model)
if sparsity_aware:
base_model = _prune_model(base_model, target_sparsity, layers_to_cluster)
clusterer = Clusterer(
base_model,
ClusteringConfiguration(
target_num_clusters,
layers_to_cluster,
),
)
clusterer.apply_optimization()
clustered_model = clusterer.get_model()
temp_file = tmp_path / "test_cluster_simple_model_fully_after.tflite"
convert_to_tflite(clustered_model, output_path=temp_file)
clustered_tflite_metrics = TFLiteMetrics(str(temp_file))
_test_num_unique_weights(
clustered_tflite_metrics, target_num_clusters, layers_to_cluster
)
if sparsity_aware:
_test_sparsity(clustered_tflite_metrics, target_sparsity, layers_to_cluster)
|
