File size: 8,914 Bytes
05bcc9e |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 |
"""
This module contains unit tests for the `freeze_layers_except` function.
The `freeze_layers_except` function is used to freeze layers in a model, except for the specified layers.
The unit tests in this module verify the behavior of the `freeze_layers_except` function in different scenarios.
"""
import unittest
import torch
from torch import nn
from axolotl.utils.freeze import freeze_layers_except
ZERO = [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]
ONE_TO_TEN = [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0]
class TestFreezeLayersExcept(unittest.TestCase):
"""
A test case class for the `freeze_layers_except` function.
"""
def setUp(self):
self.model = _TestModel()
def test_freeze_layers_with_dots_in_name(self):
freeze_layers_except(self.model, ["features.layer"])
self.assertTrue(
self.model.features.layer.weight.requires_grad,
"model.features.layer should be trainable.",
)
self.assertFalse(
self.model.classifier.weight.requires_grad,
"model.classifier should be frozen.",
)
def test_freeze_layers_without_dots_in_name(self):
freeze_layers_except(self.model, ["classifier"])
self.assertFalse(
self.model.features.layer.weight.requires_grad,
"model.features.layer should be trainable.",
)
self.assertTrue(
self.model.classifier.weight.requires_grad,
"model.classifier should be frozen.",
)
def test_freeze_layers_regex_patterns(self):
# The second pattern cannot match because only characters 'a' to 'c' are allowed after the word 'class', whereas it should be matching the character 'i'.
freeze_layers_except(self.model, [r"^features.[a-z]+.weight$", r"class[a-c]+"])
self.assertTrue(
self.model.features.layer.weight.requires_grad,
"model.features.layer should be trainable.",
)
self.assertFalse(
self.model.classifier.weight.requires_grad,
"model.classifier should be frozen.",
)
def test_all_layers_frozen(self):
freeze_layers_except(self.model, [])
self.assertFalse(
self.model.features.layer.weight.requires_grad,
"model.features.layer should be frozen.",
)
self.assertFalse(
self.model.classifier.weight.requires_grad,
"model.classifier should be frozen.",
)
def test_all_layers_unfrozen(self):
freeze_layers_except(self.model, ["features.layer", "classifier"])
self.assertTrue(
self.model.features.layer.weight.requires_grad,
"model.features.layer should be trainable.",
)
self.assertTrue(
self.model.classifier.weight.requires_grad,
"model.classifier should be trainable.",
)
def test_freeze_layers_with_range_pattern_start_end(self):
freeze_layers_except(self.model, ["features.layer[1:5]"])
self.assertTrue(
self.model.features.layer.weight.requires_grad,
"model.features.layer should be trainable.",
)
self.assertFalse(
self.model.classifier.weight.requires_grad,
"model.classifier should be frozen.",
)
self._assert_gradient_output(
[
ZERO,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
ZERO,
ZERO,
ZERO,
ZERO,
ZERO,
]
)
def test_freeze_layers_with_range_pattern_single_index(self):
freeze_layers_except(self.model, ["features.layer[5]"])
self.assertTrue(
self.model.features.layer.weight.requires_grad,
"model.features.layer should be trainable.",
)
self.assertFalse(
self.model.classifier.weight.requires_grad,
"model.classifier should be frozen.",
)
self._assert_gradient_output(
[ZERO, ZERO, ZERO, ZERO, ZERO, ONE_TO_TEN, ZERO, ZERO, ZERO, ZERO]
)
def test_freeze_layers_with_range_pattern_start_omitted(self):
freeze_layers_except(self.model, ["features.layer[:5]"])
self.assertTrue(
self.model.features.layer.weight.requires_grad,
"model.features.layer should be trainable.",
)
self.assertFalse(
self.model.classifier.weight.requires_grad,
"model.classifier should be frozen.",
)
self._assert_gradient_output(
[
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
ZERO,
ZERO,
ZERO,
ZERO,
ZERO,
]
)
def test_freeze_layers_with_range_pattern_end_omitted(self):
freeze_layers_except(self.model, ["features.layer[4:]"])
self.assertTrue(
self.model.features.layer.weight.requires_grad,
"model.features.layer should be trainable.",
)
self.assertFalse(
self.model.classifier.weight.requires_grad,
"model.classifier should be frozen.",
)
self._assert_gradient_output(
[
ZERO,
ZERO,
ZERO,
ZERO,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
]
)
def test_freeze_layers_with_range_pattern_merge_included(self):
freeze_layers_except(self.model, ["features.layer[4:]", "features.layer[5:6]"])
self.assertTrue(
self.model.features.layer.weight.requires_grad,
"model.features.layer should be trainable.",
)
self.assertFalse(
self.model.classifier.weight.requires_grad,
"model.classifier should be frozen.",
)
self._assert_gradient_output(
[
ZERO,
ZERO,
ZERO,
ZERO,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
]
)
def test_freeze_layers_with_range_pattern_merge_intersect(self):
freeze_layers_except(self.model, ["features.layer[4:7]", "features.layer[6:8]"])
self.assertTrue(
self.model.features.layer.weight.requires_grad,
"model.features.layer should be trainable.",
)
self.assertFalse(
self.model.classifier.weight.requires_grad,
"model.classifier should be frozen.",
)
self._assert_gradient_output(
[
ZERO,
ZERO,
ZERO,
ZERO,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
ONE_TO_TEN,
ZERO,
ZERO,
]
)
def test_freeze_layers_with_range_pattern_merge_separate(self):
freeze_layers_except(
self.model,
["features.layer[1:2]", "features.layer[3:4]", "features.layer[5:6]"],
)
self.assertTrue(
self.model.features.layer.weight.requires_grad,
"model.features.layer should be trainable.",
)
self.assertFalse(
self.model.classifier.weight.requires_grad,
"model.classifier should be frozen.",
)
self._assert_gradient_output(
[
ZERO,
ONE_TO_TEN,
ZERO,
ONE_TO_TEN,
ZERO,
ONE_TO_TEN,
ZERO,
ZERO,
ZERO,
ZERO,
]
)
def _assert_gradient_output(self, expected):
input_tensor = torch.tensor([ONE_TO_TEN], dtype=torch.float32)
self.model.features.layer.weight.grad = None # Reset gradients
output = self.model.features.layer(input_tensor)
loss = output.sum()
loss.backward()
expected_grads = torch.tensor(expected)
torch.testing.assert_close(
self.model.features.layer.weight.grad, expected_grads
)
class _SubLayerModule(nn.Module):
def __init__(self):
super().__init__()
self.layer = nn.Linear(10, 10)
class _TestModel(nn.Module):
def __init__(self):
super().__init__()
self.features = _SubLayerModule()
self.classifier = nn.Linear(10, 2)
if __name__ == "__main__":
unittest.main()
|