|
- # 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 squeezenet."""
- import ast
- import os
- import argparse
- import glob
- import numpy as np
-
- from mindspore import context
- from mindspore import Tensor
- from mindspore import export
- from mindspore.nn.optim.momentum import Momentum
- from mindspore.train.model import Model
- from mindspore.context import ParallelMode
- from mindspore.train.callback import ModelCheckpoint, CheckpointConfig, LossMonitor, TimeMonitor
- from mindspore.nn.loss import SoftmaxCrossEntropyWithLogits
- from mindspore.train.loss_scale_manager import FixedLossScaleManager
- from mindspore.train.serialization import load_checkpoint, load_param_into_net
- from mindspore.common import set_seed
- from mindspore.nn.metrics import Accuracy
- from mindspore.communication.management import init
- from src.lr_generator import get_lr
- from src.CrossEntropySmooth import CrossEntropySmooth
- from src.squeezenet import SqueezeNet as squeezenet
-
- parser = argparse.ArgumentParser(description='SqueezeNet1_1')
- parser.add_argument('--net', type=str, default='squeezenet', help='Model.')
- parser.add_argument('--dataset', type=str, default='imagenet', help='Dataset.')
- parser.add_argument('--run_cloudbrain', type=ast.literal_eval, default=False,
- help='Whether it is running on CloudBrain platform.')
- parser.add_argument('--run_distribute', type=bool, default=False, help='Run distribute')
- parser.add_argument('--device_num', type=int, default=1, help='Device num.')
- parser.add_argument('--dataset_path', type=str, default='', help='Dataset path')
- parser.add_argument('--device_target', type=str, default='Ascend', help='Device target')
- parser.add_argument('--pre_trained', type=str, default="None", help='Pretrained checkpoint path')
- parser.add_argument('--data_url', type=str, default="None", help='Datapath')
- parser.add_argument('--train_url', type=str, default="None", help='Train output path')
- parser.add_argument('--num_classes', type=int, default="1000", help="classes")
- parser.add_argument('--epoch_size', type=int, default="200", help="epoch_size")
- parser.add_argument('--batch_size', type=int, default="32", help="batch_size")
- args_opt = parser.parse_args()
-
- local_data_url = '/cache/data'
- local_train_url = '/cache/ckpt'
- local_pretrain_url = '/cache/preckpt.ckpt'
-
- set_seed(1)
-
- def filter_checkpoint_parameter_by_list(origin_dict, param_filter):
- """remove useless parameters according to filter_list"""
- for key in list(origin_dict.keys()):
- for name in param_filter:
- if name in key:
- print("Delete parameter from checkpoint: ", key)
- del origin_dict[key]
- break
-
-
- def frozen_to_air(network, args):
- paramdict = load_checkpoint(args.get("ckpt_file"))
- load_param_into_net(network, paramdict)
- input_arr = Tensor(np.zeros([args.get("batch_size"), 3, args.get("height"), args.get("width")], np.float32))
- export(network, input_arr, file_name=args.get("file_name"), file_format=args.get("file_format"))
-
- if __name__ == '__main__':
-
- target = args_opt.device_target
- if args_opt.device_target != "Ascend":
- raise ValueError("Unsupported device target.")
-
- # init context
- if args_opt.run_distribute:
- device_num = int(os.getenv("RANK_SIZE"))
- device_id = int(os.getenv("DEVICE_ID"))
- context.set_context(mode=context.GRAPH_MODE,
- device_target=target)
- context.set_context(device_id=device_id,
- enable_auto_mixed_precision=True)
- context.set_auto_parallel_context(
- device_num=device_num,
- parallel_mode=ParallelMode.DATA_PARALLEL,
- gradients_mean=True)
- init()
- local_data_url = os.path.join(local_data_url, str(device_id))
-
- else:
- device_id = 0
- context.set_context(mode=context.GRAPH_MODE,
- device_target=target)
-
- # create dataset
- if args_opt.dataset == "cifar10":
- from src.config import config_cifar as config
- from src.dataset import create_dataset_cifar as create_dataset
- else:
- from src.config import config_imagenet as config
- from src.dataset import create_dataset_imagenet as create_dataset
-
- if args_opt.run_cloudbrain:
- import moxing as mox
- mox.file.copy_parallel(args_opt.data_url, local_data_url)
- dataset = create_dataset(dataset_path=local_data_url,
- do_train=True,
- repeat_num=1,
- batch_size=args_opt.batch_size,
- target=target,
- run_distribute=args_opt.run_distribute)
-
-
- step_size = dataset.get_dataset_size()
-
- # define net
- net = squeezenet(num_classes=args_opt.num_classes)
-
- # load checkpoint
- if args_opt.pre_trained != "None":
- if args_opt.run_cloudbrain:
- import moxing as mox
- mox.file.copy_parallel(args_opt.pre_trained, local_pretrain_url)
- param_dict = load_checkpoint(local_pretrain_url)
- filter_list = [x.name for x in net.final_conv.get_parameters()]
- filter_checkpoint_parameter_by_list(param_dict, filter_list)
- load_param_into_net(net, param_dict)
-
-
- # init lr
- lr = get_lr(lr_init=config.lr_init,
- lr_end=config.lr_end,
- lr_max=config.lr_max,
- total_epochs=args_opt.epoch_size,
- warmup_epochs=config.warmup_epochs,
- pretrain_epochs=config.pretrain_epoch_size,
- steps_per_epoch=step_size,
- lr_decay_mode=config.lr_decay_mode)
- lr = Tensor(lr)
-
- # define loss
- if args_opt.dataset == "imagenet":
- if not config.use_label_smooth:
- config.label_smooth_factor = 0.0
- loss = CrossEntropySmooth(sparse=True,
- reduction='mean',
- smooth_factor=config.label_smooth_factor,
- num_classes=args_opt.num_classes)
- else:
- loss = SoftmaxCrossEntropyWithLogits(sparse=True, reduction='mean')
-
- # define opt, model
- loss_scale = FixedLossScaleManager(config.loss_scale,
- drop_overflow_update=False)
- opt = Momentum(filter(lambda x: x.requires_grad, net.get_parameters()),
- lr,
- config.momentum,
- config.weight_decay,
- config.loss_scale,
- use_nesterov=True)
- model = Model(net,
- loss_fn=loss,
- optimizer=opt,
- loss_scale_manager=loss_scale,
- metrics={'acc': Accuracy()},
- amp_level="O2",
- keep_batchnorm_fp32=False)
-
- # define callbacks
- time_cb = TimeMonitor(data_size=step_size)
- loss_cb = LossMonitor()
- cb = [time_cb, loss_cb]
- if config.save_checkpoint and device_id == 0:
- config_ck = CheckpointConfig(
- save_checkpoint_steps=config.save_checkpoint_epochs * step_size,
- keep_checkpoint_max=config.keep_checkpoint_max)
- ckpt_cb = ModelCheckpoint(prefix=args_opt.net,
- directory=local_train_url,
- config=config_ck)
- cb += [ckpt_cb]
-
- # train model
- model.train(args_opt.epoch_size - config.pretrain_epoch_size,
- dataset,
- callbacks=cb)
- if device_id == 0:
- ckpt_list = glob.glob("/cache/ckpt/squeezenet*.ckpt")
-
- if not ckpt_list:
- print("ckpt file not generated.")
-
- ckpt_list.sort(key=os.path.getmtime)
- ckpt_model = ckpt_list[-1]
- print("checkpoint path", ckpt_model)
-
- net = squeezenet(args_opt.num_classes)
-
- frozen_to_air_args = {'ckpt_file': ckpt_model,
- 'batch_size': 1,
- 'height': 227,
- 'width': 227,
- 'file_name': '/cache/ckpt/squeezenet',
- 'file_format': 'AIR'}
- frozen_to_air(net, frozen_to_air_args)
-
- if args_opt.run_cloudbrain:
- mox.file.copy_parallel(local_train_url, args_opt.train_url)
|