def __init__(self, resnet_size_choices=None):
super(ResnetArgParser, self).__init__(parents=[
parsers.BaseParser(multi_gpu=False),
parsers.PerformanceParser(num_parallel_calls=False),
parsers.ImageModelParser(),
parsers.ExportParser(),
parsers.BenchmarkParser(),
])
self.add_argument(
'--version',
'-v',
type=int,
choices=[1, 2],
default=resnet_model.DEFAULT_VERSION,
help='Version of ResNet. (1 or 2) See README.md for details.')
self.add_argument(
'--resnet_size',
'-rs',
type=int,
default=50,
choices=resnet_size_choices,
help='[default: %(default)s] The size of the ResNet model to use.',
metavar='<RS>' if resnet_size_choices is None else None)
def __init__(self):
super(TestParser, self).__init__(parents=[
parsers.BaseParser(),
parsers.PerformanceParser(num_parallel_calls=True, inter_op=True,
intra_op=True, use_synthetic_data=True),
parsers.ImageModelParser(data_format=True),
parsers.BenchmarkParser(benchmark_log_dir=True)
])
def main(argv):
parser = parsers.BenchmarkParser()
flags = parser.parse_args(args=argv[1:])
if not flags.benchmark_log_dir:
print("Usage: benchmark_uploader.py --benchmark_log_dir=/some/dir")
sys.exit(1)
uploader = BigQueryUploader(flags.benchmark_log_dir,
gcp_project=flags.gcp_project)
run_id = str(uuid.uuid4())
uploader.upload_benchmark_run(flags.bigquery_data_set,
flags.bigquery_run_table, run_id)
uploader.upload_metric(flags.bigquery_data_set,
flags.bigquery_metric_table, run_id)
def __init__(self, resnet_size_choices=None):
super(ResnetArgParser, self).__init__(parents=[
parsers.BaseParser(),
parsers.PerformanceParser(),
parsers.ImageModelParser(),
parsers.ExportParser(),
parsers.BenchmarkParser(),
])
self.add_argument('--dataset','-d',default="cifar10",
help='Which dataset to use (currently cifar10/cifar100)'
)
self.add_argument(
'--version', '-v', type=int, choices=[1, 2],
default=rncm.RESNET_DEFAULT_VERSION,
help='Version of ResNet. (1 or 2) See README.md for details.'
)
self.add_argument(
'--resnet_size', '-rs', type=int, default=50,
choices=resnet_size_choices,
help='[default: %(default)s] The size of the ResNet model to use.',
metavar='<RS>' if resnet_size_choices is None else None
)
self.add_argument(
'--continu',type=int,default=0,
help='Continue with an existing model, or start from scratch'
)
self.add_argument(
'--scratch',type=int,default=0,
help='Start from scratch even if model exist'
)
self.add_argument(
'--ncmmethod', default=rncm.NCM_DEFAULT_METHOD,
help='[default: %(default)s] Which NCM method to use',
)
self.add_argument(
'--ncmparam', default=rncm.NCM_DEFAULT_PARAMETER, type=float,
help='[default: %(default)s] additional NCM parameter to use',
)
self.add_argument(
'--initial_learning_scale', '-l', default=0.1, type=float,
help='Intial Learning Scale (default: %(default)s)',
)
def __init__(self, resnet_size_choices=None):
super(ResnetArgParser, self).__init__(parents=[
parsers.BaseParser(),
parsers.PerformanceParser(),
parsers.ImageModelParser(),
parsers.ExportParser(),
parsers.BenchmarkParser(),
])
self.add_argument(
'--version',
'-v',
type=int,
choices=[1, 2],
default=resnet_model.DEFAULT_VERSION,
help='Version of ResNet. (1 or 2) See README.md for details.')
self.add_argument(
'--resnet_size',
'-rs',
type=int,
default=50,
choices=resnet_size_choices,
help='[default: %(default)s] The size of the ResNet model to use.',
metavar='<RS>' if resnet_size_choices is None else None)
self.add_argument(
'--enable_ml_comm',
'-mc',
type=int,
choices=[0, 1],
default=1,
help=
'[default: %(default)s] Whether to use Cray ML-Comm Distributed Training Plugin'
)
self.add_argument(
'--global_perf_log_freq',
'-pf',
type=int,
default=50,
help=
'[default: %(default)s] Number of steps after which to report global (all process averages) training loss and performance'
)
self.add_argument(
'--warmup_epochs',
'-we',
type=int,
default=0,
help=
'[default: %(default)s] Number of warmup epochs when using LARS')
self.add_argument(
'--base_lr',
'-blr',
type=float,
default=1.0,
help=
'[default: %(default)s] Learning rate to start after warmup epochs finish when using LARS'
)
self.add_argument(
'--init_lr',
'-ilr',
type=float,
default=0.1,
help=
'[default: %(default)s] Learning rate to start warmup with when using LARS'
)
self.add_argument(
'--weight_decay',
'-wd',
type=float,
default=1e-4,
help='[default: %(default)s] Weight decay to use during training')