File size: 13,604 Bytes
97b6013 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 |
# Copyright 2018 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.
# ==============================================================================
"""Executes Keras benchmarks and accuracy tests."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import time
from absl import flags
import tensorflow as tf # pylint: disable=g-bad-import-order
from official.benchmark import keras_benchmark
from official.benchmark import benchmark_wrappers
from official.benchmark.models import resnet_cifar_main
MIN_TOP_1_ACCURACY = 0.929
MAX_TOP_1_ACCURACY = 0.938
FLAGS = flags.FLAGS
CIFAR_DATA_DIR_NAME = 'cifar-10-batches-bin'
class Resnet56KerasAccuracy(keras_benchmark.KerasBenchmark):
"""Accuracy tests for ResNet56 Keras CIFAR-10."""
def __init__(self, output_dir=None, root_data_dir=None, **kwargs):
"""A benchmark class.
Args:
output_dir: directory where to output e.g. log files
root_data_dir: directory under which to look for dataset
**kwargs: arbitrary named arguments. This is needed to make the
constructor forward compatible in case PerfZero provides more
named arguments before updating the constructor.
"""
self.data_dir = os.path.join(root_data_dir, CIFAR_DATA_DIR_NAME)
flag_methods = [resnet_cifar_main.define_cifar_flags]
super(Resnet56KerasAccuracy, self).__init__(
output_dir=output_dir, flag_methods=flag_methods)
def _setup(self):
super(Resnet56KerasAccuracy, self)._setup()
FLAGS.use_tensor_lr = False
def benchmark_graph_1_gpu(self):
"""Test keras based model with Keras fit and distribution strategies."""
self._setup()
FLAGS.num_gpus = 1
FLAGS.data_dir = self.data_dir
FLAGS.batch_size = 128
FLAGS.train_epochs = 182
FLAGS.model_dir = self._get_model_dir('benchmark_graph_1_gpu')
FLAGS.dtype = 'fp32'
self._run_and_report_benchmark()
def benchmark_1_gpu(self):
"""Test keras based model with eager and distribution strategies."""
self._setup()
FLAGS.num_gpus = 1
FLAGS.data_dir = self.data_dir
FLAGS.batch_size = 128
FLAGS.train_epochs = 182
FLAGS.model_dir = self._get_model_dir('benchmark_1_gpu')
FLAGS.dtype = 'fp32'
FLAGS.enable_eager = True
self._run_and_report_benchmark()
def benchmark_cpu(self):
"""Test keras based model on CPU."""
self._setup()
FLAGS.num_gpus = 0
FLAGS.data_dir = self.data_dir
FLAGS.batch_size = 128
FLAGS.train_epochs = 182
FLAGS.model_dir = self._get_model_dir('benchmark_cpu')
FLAGS.dtype = 'fp32'
FLAGS.enable_eager = True
FLAGS.data_format = 'channels_last'
self._run_and_report_benchmark()
def benchmark_cpu_no_dist_strat(self):
"""Test keras based model on CPU without distribution strategies."""
self._setup()
FLAGS.num_gpus = 0
FLAGS.data_dir = self.data_dir
FLAGS.batch_size = 128
FLAGS.train_epochs = 182
FLAGS.model_dir = self._get_model_dir('benchmark_cpu_no_dist_strat')
FLAGS.dtype = 'fp32'
FLAGS.enable_eager = True
FLAGS.distribution_strategy = 'off'
FLAGS.data_format = 'channels_last'
self._run_and_report_benchmark()
def benchmark_cpu_no_dist_strat_run_eagerly(self):
"""Test keras based model on CPU w/forced eager and no dist_strat."""
self._setup()
FLAGS.num_gpus = 0
FLAGS.data_dir = self.data_dir
FLAGS.batch_size = 128
FLAGS.train_epochs = 182
FLAGS.model_dir = self._get_model_dir(
'benchmark_cpu_no_dist_strat_run_eagerly')
FLAGS.dtype = 'fp32'
FLAGS.enable_eager = True
FLAGS.run_eagerly = True
FLAGS.distribution_strategy = 'off'
FLAGS.data_format = 'channels_last'
self._run_and_report_benchmark()
def benchmark_1_gpu_no_dist_strat(self):
"""Test keras based model with eager and no dist strat."""
self._setup()
FLAGS.num_gpus = 1
FLAGS.data_dir = self.data_dir
FLAGS.batch_size = 128
FLAGS.train_epochs = 182
FLAGS.model_dir = self._get_model_dir('benchmark_1_gpu_no_dist_strat')
FLAGS.dtype = 'fp32'
FLAGS.enable_eager = True
FLAGS.distribution_strategy = 'off'
self._run_and_report_benchmark()
def benchmark_1_gpu_no_dist_strat_run_eagerly(self):
"""Test keras based model w/forced eager and no dist_strat."""
self._setup()
FLAGS.num_gpus = 1
FLAGS.data_dir = self.data_dir
FLAGS.batch_size = 128
FLAGS.train_epochs = 182
FLAGS.model_dir = self._get_model_dir(
'benchmark_1_gpu_no_dist_strat_run_eagerly')
FLAGS.dtype = 'fp32'
FLAGS.enable_eager = True
FLAGS.run_eagerly = True
FLAGS.distribution_strategy = 'off'
self._run_and_report_benchmark()
def benchmark_graph_1_gpu_no_dist_strat(self):
"""Test keras based model with Keras fit but not distribution strategies."""
self._setup()
FLAGS.distribution_strategy = 'off'
FLAGS.num_gpus = 1
FLAGS.data_dir = self.data_dir
FLAGS.batch_size = 128
FLAGS.train_epochs = 182
FLAGS.model_dir = self._get_model_dir('benchmark_graph_1_gpu_no_dist_strat')
FLAGS.dtype = 'fp32'
self._run_and_report_benchmark()
def benchmark_2_gpu(self):
"""Test keras based model with eager and distribution strategies."""
self._setup()
FLAGS.num_gpus = 2
FLAGS.data_dir = self.data_dir
FLAGS.batch_size = 128
FLAGS.train_epochs = 182
FLAGS.model_dir = self._get_model_dir('benchmark_2_gpu')
FLAGS.dtype = 'fp32'
FLAGS.enable_eager = True
self._run_and_report_benchmark()
def benchmark_graph_2_gpu(self):
"""Test keras based model with Keras fit and distribution strategies."""
self._setup()
FLAGS.num_gpus = 2
FLAGS.data_dir = self.data_dir
FLAGS.batch_size = 128
FLAGS.train_epochs = 182
FLAGS.model_dir = self._get_model_dir('benchmark_graph_2_gpu')
FLAGS.dtype = 'fp32'
self._run_and_report_benchmark()
@benchmark_wrappers.enable_runtime_flags
def _run_and_report_benchmark(self):
start_time_sec = time.time()
stats = resnet_cifar_main.run(FLAGS)
wall_time_sec = time.time() - start_time_sec
super(Resnet56KerasAccuracy, self)._report_benchmark(
stats,
wall_time_sec,
top_1_min=MIN_TOP_1_ACCURACY,
top_1_max=MAX_TOP_1_ACCURACY,
total_batch_size=FLAGS.batch_size,
log_steps=100)
class Resnet56KerasBenchmarkBase(keras_benchmark.KerasBenchmark):
"""Short performance tests for ResNet56 via Keras and CIFAR-10."""
def __init__(self, output_dir=None, default_flags=None):
flag_methods = [resnet_cifar_main.define_cifar_flags]
super(Resnet56KerasBenchmarkBase, self).__init__(
output_dir=output_dir,
flag_methods=flag_methods,
default_flags=default_flags)
@benchmark_wrappers.enable_runtime_flags
def _run_and_report_benchmark(self):
start_time_sec = time.time()
stats = resnet_cifar_main.run(FLAGS)
wall_time_sec = time.time() - start_time_sec
super(Resnet56KerasBenchmarkBase, self)._report_benchmark(
stats,
wall_time_sec,
total_batch_size=FLAGS.batch_size,
log_steps=FLAGS.log_steps)
def benchmark_1_gpu(self):
"""Test 1 gpu."""
self._setup()
FLAGS.num_gpus = 1
FLAGS.enable_eager = True
FLAGS.distribution_strategy = 'one_device'
FLAGS.model_dir = self._get_model_dir('benchmark_1_gpu')
FLAGS.batch_size = 128
self._run_and_report_benchmark()
def benchmark_1_gpu_xla(self):
"""Test 1 gpu with xla enabled."""
self._setup()
FLAGS.num_gpus = 1
FLAGS.enable_eager = True
FLAGS.run_eagerly = False
FLAGS.enable_xla = True
FLAGS.distribution_strategy = 'one_device'
FLAGS.model_dir = self._get_model_dir('benchmark_1_gpu_xla')
FLAGS.batch_size = 128
self._run_and_report_benchmark()
def benchmark_graph_1_gpu(self):
"""Test 1 gpu graph."""
self._setup()
FLAGS.num_gpus = 1
FLAGS.enable_eager = False
FLAGS.run_eagerly = False
FLAGS.distribution_strategy = 'one_device'
FLAGS.model_dir = self._get_model_dir('benchmark_graph_1_gpu')
FLAGS.batch_size = 128
self._run_and_report_benchmark()
def benchmark_1_gpu_no_dist_strat(self):
"""Test 1 gpu without distribution strategies."""
self._setup()
FLAGS.num_gpus = 1
FLAGS.enable_eager = True
FLAGS.distribution_strategy = 'off'
FLAGS.model_dir = self._get_model_dir('benchmark_1_gpu_no_dist_strat')
FLAGS.batch_size = 128
self._run_and_report_benchmark()
def benchmark_graph_1_gpu_no_dist_strat(self):
"""Test 1 gpu graph mode without distribution strategies."""
self._setup()
FLAGS.num_gpus = 1
FLAGS.enable_eager = False
FLAGS.distribution_strategy = 'off'
FLAGS.model_dir = self._get_model_dir('benchmark_graph_1_gpu_no_dist_strat')
FLAGS.batch_size = 128
self._run_and_report_benchmark()
def benchmark_1_gpu_no_dist_strat_run_eagerly(self):
"""Test 1 gpu without distribution strategy and forced eager."""
self._setup()
FLAGS.num_gpus = 1
FLAGS.batch_size = 128
FLAGS.model_dir = self._get_model_dir(
'benchmark_1_gpu_no_dist_strat_run_eagerly')
FLAGS.dtype = 'fp32'
FLAGS.enable_eager = True
FLAGS.run_eagerly = True
FLAGS.distribution_strategy = 'off'
self._run_and_report_benchmark()
def benchmark_2_gpu(self):
"""Test 2 gpu."""
self._setup()
FLAGS.num_gpus = 2
FLAGS.enable_eager = True
FLAGS.run_eagerly = False
FLAGS.distribution_strategy = 'mirrored'
FLAGS.model_dir = self._get_model_dir('benchmark_2_gpu')
FLAGS.batch_size = 128 * 2 # 2 GPUs
self._run_and_report_benchmark()
def benchmark_graph_2_gpu(self):
"""Test 2 gpu graph mode."""
self._setup()
FLAGS.num_gpus = 2
FLAGS.enable_eager = False
FLAGS.run_eagerly = False
FLAGS.distribution_strategy = 'mirrored'
FLAGS.model_dir = self._get_model_dir('benchmark_graph_2_gpu')
FLAGS.batch_size = 128 * 2 # 2 GPUs
self._run_and_report_benchmark()
def benchmark_cpu(self):
"""Test cpu."""
self._setup()
FLAGS.num_gpus = 0
FLAGS.enable_eager = True
FLAGS.model_dir = self._get_model_dir('benchmark_cpu')
FLAGS.batch_size = 128
FLAGS.data_format = 'channels_last'
self._run_and_report_benchmark()
def benchmark_graph_cpu(self):
"""Test cpu graph mode."""
self._setup()
FLAGS.num_gpus = 0
FLAGS.enable_eager = False
FLAGS.model_dir = self._get_model_dir('benchmark_graph_cpu')
FLAGS.batch_size = 128
FLAGS.data_format = 'channels_last'
self._run_and_report_benchmark()
def benchmark_cpu_no_dist_strat_run_eagerly(self):
"""Test cpu without distribution strategy and forced eager."""
self._setup()
FLAGS.num_gpus = 0
FLAGS.distribution_strategy = 'off'
FLAGS.enable_eager = True
FLAGS.run_eagerly = True
FLAGS.model_dir = self._get_model_dir(
'benchmark_cpu_no_dist_strat_run_eagerly')
FLAGS.batch_size = 128
FLAGS.data_format = 'channels_last'
self._run_and_report_benchmark()
def benchmark_cpu_no_dist_strat(self):
"""Test cpu without distribution strategies."""
self._setup()
FLAGS.num_gpus = 0
FLAGS.enable_eager = True
FLAGS.distribution_strategy = 'off'
FLAGS.model_dir = self._get_model_dir('benchmark_cpu_no_dist_strat')
FLAGS.batch_size = 128
FLAGS.data_format = 'channels_last'
self._run_and_report_benchmark()
def benchmark_graph_cpu_no_dist_strat(self):
"""Test cpu graph mode without distribution strategies."""
self._setup()
FLAGS.num_gpus = 0
FLAGS.enable_eager = False
FLAGS.distribution_strategy = 'off'
FLAGS.model_dir = self._get_model_dir('benchmark_graph_cpu_no_dist_strat')
FLAGS.batch_size = 128
FLAGS.data_format = 'channels_last'
self._run_and_report_benchmark()
class Resnet56KerasBenchmarkSynth(Resnet56KerasBenchmarkBase):
"""Synthetic benchmarks for ResNet56 and Keras."""
def __init__(self, output_dir=None, root_data_dir=None, **kwargs):
default_flags = {}
default_flags['skip_eval'] = True
default_flags['use_synthetic_data'] = True
default_flags['train_steps'] = 110
default_flags['log_steps'] = 10
default_flags['use_tensor_lr'] = False
super(Resnet56KerasBenchmarkSynth, self).__init__(
output_dir=output_dir, default_flags=default_flags)
class Resnet56KerasBenchmarkReal(Resnet56KerasBenchmarkBase):
"""Real data benchmarks for ResNet56 and Keras."""
def __init__(self, output_dir=None, root_data_dir=None, **kwargs):
default_flags = {}
default_flags['skip_eval'] = True
default_flags['data_dir'] = os.path.join(root_data_dir, CIFAR_DATA_DIR_NAME)
default_flags['train_steps'] = 110
default_flags['log_steps'] = 10
default_flags['use_tensor_lr'] = False
super(Resnet56KerasBenchmarkReal, self).__init__(
output_dir=output_dir, default_flags=default_flags)
if __name__ == '__main__':
tf.test.main()
|