|
- # Copyright 2021 Huawei Technologies Co., Ltd
- #
- # 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.
- # ============================================================================
- """
- #################train advanced_east on dataset########################
- """
- import argparse
- import datetime
- import os
- import ast
-
- from mindspore import context, Model
- from mindspore.common import set_seed
- from mindspore.communication.management import init, get_rank, get_group_size
- from mindspore.context import ParallelMode
- from mindspore.nn.optim import AdamWeightDecay
- from mindspore.train.callback import ModelCheckpoint, CheckpointConfig, LossMonitor, TimeMonitor
- from mindspore.train.serialization import load_param_into_net, load_checkpoint
- from src.logger import get_logger
- from src.config import config as cfg
- from src.dataset import load_adEAST_dataset
- from src.model import get_AdvancedEast_net
-
- set_seed(1)
-
-
- def parse_args(cloud_args=None):
- """parameters"""
- parser = argparse.ArgumentParser('mindspore adveast training')
- parser.add_argument('--device_target', type=str, default='Ascend', choices=['Ascend', 'GPU'],
- help='device where the code will be implemented. (Default: Ascend)')
- parser.add_argument('--device_id', type=int, default=0, help='device id of GPU or Ascend.')
-
- # network related
- parser.add_argument('--pre_trained', default=False, type=ast.literal_eval,
- help='model_path, local pretrained model to load')
- parser.add_argument('--data_path', type=str)
- parser.add_argument('--pre_trained_ckpt', type=str)
-
- # logging and checkpoint related
- parser.add_argument('--ckpt_path', type=str, default='outputs/', help='checkpoint save location')
- parser.add_argument('--ckpt_interval', type=int, default=1, help='ckpt_interval')
- parser.add_argument('--is_save_on_master', type=int, default=1, help='save ckpt on master or all rank')
-
- # distributed related
- parser.add_argument('--is_distributed', type=int, default=0, help='if multi device')
- parser.add_argument('--rank', type=int, default=0, help='local rank of distributed')
- parser.add_argument('--group_size', type=int, default=1, help='world size of distributed')
- args_opt = parser.parse_args()
-
- args_opt.epoch_num = cfg.epoch_num
- args_opt.batch_size = cfg.batch_size
- args_opt.ckpt_save_max = cfg.ckpt_save_max
- args_opt.data_dir = cfg.data_dir
- args_opt.mindsrecord_train_file = cfg.mindsrecord_train_file
- args_opt.mindsrecord_test_file = cfg.mindsrecord_test_file
- args_opt.results_dir = cfg.results_dir
- args_opt.last_model_name = cfg.last_model_name
- args_opt.saved_model_file_path = cfg.saved_model_file_path
- args_opt.is_train = True
- return args_opt
-
-
- if __name__ == '__main__':
- args = parse_args()
- context.set_context(device_target=args.device_target, device_id=args.device_id, mode=context.GRAPH_MODE)
- workers = 32
- device_num = 1
- if args.is_distributed:
- init()
- if args.device_target == "Ascend":
- device_num = int(os.environ.get("RANK_SIZE"))
- rank = int(os.environ.get("RANK_ID"))
- args.rank = args.device_id
- elif args.device_target == "GPU":
- context.set_context(device_target=args.device_target)
- device_num = get_group_size()
- args.rank = get_rank()
- args.group_size = device_num
- context.reset_auto_parallel_context()
- context.set_auto_parallel_context(device_num=8, gradients_mean=True, parallel_mode=ParallelMode.DATA_PARALLEL)
- else:
- context.set_context(device_id=args.device_id)
-
- # logger
- args.outputs_dir = os.path.join(args.ckpt_path,
- datetime.datetime.now().strftime('%Y-%m-%d_time_%H_%M_%S'))
- args.logger = get_logger(args.outputs_dir, args.rank)
-
- # dataset
-
- mindrecordfile = args.data_path
-
- train_dataset, batch_num = load_adEAST_dataset(mindrecordfile, batch_size=args.batch_size,
- device_num=device_num, rank_id=args.rank, is_training=True,
- num_parallel_workers=workers)
-
- args.logger.save_args(args)
- # network
- args.logger.important_info('start create network')
- # select for master rank save ckpt or all rank save, compatible for model parallel
- args.rank_save_ckpt_flag = 0
- if args.is_save_on_master:
- if args.rank == 0:
- args.rank_save_ckpt_flag = 1
- else:
- args.rank_save_ckpt_flag = 1
-
- # get network and init
- loss_net, train_net = get_AdvancedEast_net(args)
- loss_net.add_flags_recursive(fp32=True)
- train_net.set_train(True)
- # pre_trained
- if args.pre_trained:
- load_param_into_net(train_net, load_checkpoint(args.pre_trained_ckpt))
-
- learning_rate = cfg.learning_rate_ascend if args.device_target == 'Ascend' else cfg.learning_rate_gpu
- decay = cfg.decay_ascend if args.device_target == 'Ascend' else cfg.decay_gpu
-
- train_net.optimizer = AdamWeightDecay(train_net.weights, learning_rate=learning_rate
- , eps=1e-7, weight_decay=decay)
- model = Model(train_net)
- time_cb = TimeMonitor(data_size=batch_num)
- loss_cb = LossMonitor(per_print_times=batch_num)
- callbacks = [time_cb, loss_cb]
- ckpt_config = CheckpointConfig(save_checkpoint_steps=args.ckpt_interval * batch_num,
- keep_checkpoint_max=args.ckpt_save_max)
- save_ckpt_path = args.saved_model_file_path
- if args.rank_save_ckpt_flag:
- ckpt_cb = ModelCheckpoint(config=ckpt_config,
- directory=save_ckpt_path,
- prefix='Epoch_{}'.format(args.rank))
- callbacks.append(ckpt_cb)
- model.train(epoch=cfg.epoch_num, train_dataset=train_dataset, callbacks=callbacks, dataset_sink_mode=True)
|