Spaces:
Runtime error
Runtime error
File size: 9,942 Bytes
5672777 |
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 |
# Copyright 2023 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Tests for train_ctl_lib."""
import json
import os
from absl import flags
from absl.testing import flagsaver
from absl.testing import parameterized
import numpy as np
import tensorflow as tf, tf_keras
from tensorflow.python.distribute import combinations
from tensorflow.python.distribute import strategy_combinations
from official.common import flags as tfm_flags
# pylint: disable=unused-import
from official.common import registry_imports
# pylint: enable=unused-import
from official.core import task_factory
from official.core import train_lib
from official.core import train_utils
from official.utils.testing import mock_task
FLAGS = flags.FLAGS
tfm_flags.define_flags()
class TrainTest(tf.test.TestCase, parameterized.TestCase):
def setUp(self):
super(TrainTest, self).setUp()
self._test_config = {
'trainer': {
'checkpoint_interval': 10,
'steps_per_loop': 10,
'summary_interval': 10,
'train_steps': 10,
'validation_steps': 5,
'validation_interval': 10,
'continuous_eval_timeout': 1,
'validation_summary_subdir': 'validation',
'optimizer_config': {
'optimizer': {
'type': 'sgd',
},
'learning_rate': {
'type': 'constant'
}
}
},
}
@combinations.generate(
combinations.combine(
distribution_strategy=[
strategy_combinations.default_strategy,
strategy_combinations.cloud_tpu_strategy,
strategy_combinations.one_device_strategy_gpu,
],
flag_mode=['train', 'eval', 'train_and_eval'],
run_post_eval=[True, False]))
def test_end_to_end(self, distribution_strategy, flag_mode, run_post_eval):
model_dir = self.get_temp_dir()
flags_dict = dict(
experiment='mock',
mode=flag_mode,
model_dir=model_dir,
params_override=json.dumps(self._test_config))
with flagsaver.flagsaver(**flags_dict):
params = train_utils.parse_configuration(flags.FLAGS)
train_utils.serialize_config(params, model_dir)
with distribution_strategy.scope():
task = task_factory.get_task(params.task, logging_dir=model_dir)
_, logs = train_lib.run_experiment(
distribution_strategy=distribution_strategy,
task=task,
mode=flag_mode,
params=params,
model_dir=model_dir,
run_post_eval=run_post_eval)
if 'eval' in flag_mode:
self.assertTrue(
tf.io.gfile.exists(
os.path.join(model_dir,
params.trainer.validation_summary_subdir)))
if run_post_eval:
self.assertNotEmpty(logs)
else:
self.assertEmpty(logs)
self.assertNotEmpty(
tf.io.gfile.glob(os.path.join(model_dir, 'params.yaml')))
if flag_mode == 'eval':
return
self.assertNotEmpty(
tf.io.gfile.glob(os.path.join(model_dir, 'checkpoint')))
# Tests continuous evaluation.
_, logs = train_lib.run_experiment(
distribution_strategy=distribution_strategy,
task=task,
mode='continuous_eval',
params=params,
model_dir=model_dir,
run_post_eval=run_post_eval)
@combinations.generate(
combinations.combine(
distribution_strategy=[
strategy_combinations.default_strategy,
strategy_combinations.cloud_tpu_strategy,
strategy_combinations.one_device_strategy_gpu,
],
flag_mode=['train', 'eval', 'train_and_eval'],
run_post_eval=[True, False]))
def test_end_to_end_class(self, distribution_strategy, flag_mode,
run_post_eval):
model_dir = self.get_temp_dir()
flags_dict = dict(
experiment='mock',
mode=flag_mode,
model_dir=model_dir,
params_override=json.dumps(self._test_config))
with flagsaver.flagsaver(**flags_dict):
params = train_utils.parse_configuration(flags.FLAGS)
train_utils.serialize_config(params, model_dir)
with distribution_strategy.scope():
task = task_factory.get_task(params.task, logging_dir=model_dir)
_, logs = train_lib.OrbitExperimentRunner(
distribution_strategy=distribution_strategy,
task=task,
mode=flag_mode,
params=params,
model_dir=model_dir,
run_post_eval=run_post_eval).run()
if 'eval' in flag_mode:
self.assertTrue(
tf.io.gfile.exists(
os.path.join(model_dir,
params.trainer.validation_summary_subdir)))
if run_post_eval:
self.assertNotEmpty(logs)
else:
self.assertEmpty(logs)
self.assertNotEmpty(
tf.io.gfile.glob(os.path.join(model_dir, 'params.yaml')))
if flag_mode == 'eval':
return
self.assertNotEmpty(
tf.io.gfile.glob(os.path.join(model_dir, 'checkpoint')))
# Tests continuous evaluation.
_, logs = train_lib.OrbitExperimentRunner(
distribution_strategy=distribution_strategy,
task=task,
mode='continuous_eval',
params=params,
model_dir=model_dir,
run_post_eval=run_post_eval).run()
@combinations.generate(
combinations.combine(
distribution_strategy=[
strategy_combinations.default_strategy,
strategy_combinations.cloud_tpu_strategy,
strategy_combinations.one_device_strategy_gpu,
],
flag_mode=['train', 'train_and_eval'],
))
def test_recovery_nan_error(self, distribution_strategy, flag_mode):
model_dir = self.get_temp_dir()
flags_dict = dict(
experiment='mock',
mode=flag_mode,
model_dir=model_dir,
params_override=json.dumps(self._test_config))
with flagsaver.flagsaver(**flags_dict):
params = train_utils.parse_configuration(flags.FLAGS)
train_utils.serialize_config(params, model_dir)
with distribution_strategy.scope():
# task = task_factory.get_task(params.task, logging_dir=model_dir)
task = mock_task.MockTask(params.task, logging_dir=model_dir)
# Set the loss to NaN to trigger RunTimeError.
def build_losses(labels, model_outputs, aux_losses=None):
del labels, model_outputs
return tf.constant([np.nan], tf.float32) + aux_losses
task.build_losses = build_losses
with self.assertRaises(RuntimeError):
train_lib.OrbitExperimentRunner(
distribution_strategy=distribution_strategy,
task=task,
mode=flag_mode,
params=params,
model_dir=model_dir).run()
@combinations.generate(
combinations.combine(
distribution_strategy=[
strategy_combinations.default_strategy,
strategy_combinations.cloud_tpu_strategy,
strategy_combinations.one_device_strategy_gpu,
],
flag_mode=['train'],
))
def test_recovery(self, distribution_strategy, flag_mode):
loss_threshold = 1.0
model_dir = self.get_temp_dir()
flags_dict = dict(
experiment='mock',
mode=flag_mode,
model_dir=model_dir,
params_override=json.dumps(self._test_config))
with flagsaver.flagsaver(**flags_dict):
params = train_utils.parse_configuration(flags.FLAGS)
params.trainer.loss_upper_bound = loss_threshold
params.trainer.recovery_max_trials = 1
train_utils.serialize_config(params, model_dir)
with distribution_strategy.scope():
task = task_factory.get_task(params.task, logging_dir=model_dir)
# Saves a checkpoint for reference.
model = task.build_model()
checkpoint = tf.train.Checkpoint(model=model)
checkpoint_manager = tf.train.CheckpointManager(
checkpoint, self.get_temp_dir(), max_to_keep=2)
checkpoint_manager.save()
before_weights = model.get_weights()
def build_losses(labels, model_outputs, aux_losses=None):
del labels, model_outputs
return tf.constant([loss_threshold], tf.float32) + aux_losses
task.build_losses = build_losses
model, _ = train_lib.OrbitExperimentRunner(
distribution_strategy=distribution_strategy,
task=task,
mode=flag_mode,
params=params,
model_dir=model_dir).run()
after_weights = model.get_weights()
for left, right in zip(before_weights, after_weights):
self.assertAllEqual(left, right)
def test_parse_configuration(self):
model_dir = self.get_temp_dir()
flags_dict = dict(
experiment='mock',
mode='train',
model_dir=model_dir,
params_override=json.dumps(self._test_config))
with flagsaver.flagsaver(**flags_dict):
params = train_utils.parse_configuration(flags.FLAGS, lock_return=True)
with self.assertRaises(ValueError):
params.override({'task': {'init_checkpoint': 'Foo'}})
params = train_utils.parse_configuration(flags.FLAGS, lock_return=False)
params.override({'task': {'init_checkpoint': 'Bar'}})
self.assertEqual(params.task.init_checkpoint, 'Bar')
if __name__ == '__main__':
tf.test.main()
|