Add some dist-training robust cases into fluid benchmark test (#11207)
* 1. add weight decay feature into fluid benchmark test 2. add learning rate decay feature into fluid benchmark test 3. add L1&L2 regularization feature into fluid benchmark test 4. add error clipping feature into fluid benchmark test 5. add gradient clipping feature into fluid benchmark test * Add some document to README.md under benchmark/fluid/ repo * Add model_base.py * Fix bugs in test_listen_and_serv_op * 1. remove args out of fluid_benchmark.py 2. remove lr_decay, regularization, clipping out of fluid_benchmark.py * add async_mode description to doc and remove the clipping description out * for restart build * to restart build * remove optimization args from args.py * 1. remove optimization from models 2. fix bug in test_listen_and_serv_op * change the name retry_times to left_time * change retry_times to the pserver start left timewangkuiyi-patch-1
Qiyang Min
7 years ago
committed by
Wu Yi
parent
3fd3e500cc
commit
1cfd3cb13b
@ -0,0 +1,126 @@
|
||||
# Copyright (c) 2018 PaddlePaddle 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.
|
||||
|
||||
import argparse
|
||||
|
||||
__all__ = ['parse_args', ]
|
||||
|
||||
BENCHMARK_MODELS = [
|
||||
"machine_translation", "resnet", "vgg", "mnist", "stacked_dynamic_lstm"
|
||||
]
|
||||
|
||||
|
||||
def parse_args():
|
||||
parser = argparse.ArgumentParser('Fluid model benchmarks.')
|
||||
parser.add_argument(
|
||||
'--model',
|
||||
type=str,
|
||||
choices=BENCHMARK_MODELS,
|
||||
default='resnet',
|
||||
help='The model to run benchmark with.')
|
||||
parser.add_argument(
|
||||
'--batch_size', type=int, default=32, help='The minibatch size.')
|
||||
# args related to learning rate
|
||||
parser.add_argument(
|
||||
'--learning_rate', type=float, default=0.001, help='The learning rate.')
|
||||
# TODO(wuyi): add "--use_fake_data" option back.
|
||||
parser.add_argument(
|
||||
'--skip_batch_num',
|
||||
type=int,
|
||||
default=5,
|
||||
help='The first num of minibatch num to skip, for better performance test'
|
||||
)
|
||||
parser.add_argument(
|
||||
'--iterations', type=int, default=80, help='The number of minibatches.')
|
||||
parser.add_argument(
|
||||
'--pass_num', type=int, default=100, help='The number of passes.')
|
||||
parser.add_argument(
|
||||
'--data_format',
|
||||
type=str,
|
||||
default='NCHW',
|
||||
choices=['NCHW', 'NHWC'],
|
||||
help='The data data_format, now only support NCHW.')
|
||||
parser.add_argument(
|
||||
'--device',
|
||||
type=str,
|
||||
default='GPU',
|
||||
choices=['CPU', 'GPU'],
|
||||
help='The device type.')
|
||||
parser.add_argument(
|
||||
'--gpus',
|
||||
type=int,
|
||||
default=1,
|
||||
help='If gpus > 1, will use ParallelExecutor to run, else use Executor.')
|
||||
# this option is available only for vgg and resnet.
|
||||
parser.add_argument(
|
||||
'--cpus',
|
||||
type=int,
|
||||
default=1,
|
||||
help='If cpus > 1, will use ParallelDo to run, else use Executor.')
|
||||
parser.add_argument(
|
||||
'--data_set',
|
||||
type=str,
|
||||
default='flowers',
|
||||
choices=['cifar10', 'flowers'],
|
||||
help='Optional dataset for benchmark.')
|
||||
parser.add_argument(
|
||||
'--infer_only', action='store_true', help='If set, run forward only.')
|
||||
parser.add_argument(
|
||||
'--use_cprof', action='store_true', help='If set, use cProfile.')
|
||||
parser.add_argument(
|
||||
'--use_nvprof',
|
||||
action='store_true',
|
||||
help='If set, use nvprof for CUDA.')
|
||||
parser.add_argument(
|
||||
'--no_test',
|
||||
action='store_true',
|
||||
help='If set, do not test the testset during training.')
|
||||
parser.add_argument(
|
||||
'--memory_optimize',
|
||||
action='store_true',
|
||||
help='If set, optimize runtime memory before start.')
|
||||
parser.add_argument(
|
||||
'--use_fake_data',
|
||||
action='store_true',
|
||||
help='If set ommit the actual read data operators.')
|
||||
parser.add_argument(
|
||||
'--profile', action='store_true', help='If set, profile a few steps.')
|
||||
parser.add_argument(
|
||||
'--update_method',
|
||||
type=str,
|
||||
default='local',
|
||||
choices=['local', 'pserver', 'nccl2'],
|
||||
help='Choose parameter update method, can be local, pserver, nccl2.')
|
||||
parser.add_argument(
|
||||
'--no_split_var',
|
||||
action='store_true',
|
||||
default=False,
|
||||
help='Whether split variables into blocks when update_method is pserver')
|
||||
parser.add_argument(
|
||||
'--async_mode',
|
||||
action='store_true',
|
||||
default=False,
|
||||
help='Whether start pserver in async mode to support ASGD')
|
||||
parser.add_argument(
|
||||
'--use_reader_op',
|
||||
action='store_true',
|
||||
help='Whether to use reader op, and must specify the data path if set this to true.'
|
||||
)
|
||||
parser.add_argument(
|
||||
'--data_path',
|
||||
type=str,
|
||||
default="",
|
||||
help='Directory that contains all the training recordio files.')
|
||||
args = parser.parse_args()
|
||||
return args
|
||||
Loading…
Reference in new issue