-
Notifications
You must be signed in to change notification settings - Fork 322
/
Copy pathtrain.py
171 lines (165 loc) · 7 KB
/
train.py
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
# Copyright (c) 2021 PPViT 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 os
import time
import random
import argparse
import numpy as np
from collections import deque
import paddle
import paddle.nn as nn
from config import *
from src.utils import logger
from src.datasets import get_dataset
from src.models import get_model
from src.transforms import *
from src.utils import TimeAverager, calculate_eta, resume, get_dataloader
from src.models.solver import get_scheduler, get_optimizer
from src.models.losses import get_loss_function
def parse_args():
parser = argparse.ArgumentParser(
description='Visual Transformer for semantic segmentation')
parser.add_argument(
"--config",
dest='cfg',
default=None,
type=str,
help="The config file."
)
parser.add_argument(
"--resume",
default=None,
type=str,
help="Training from resume checkpoint."
)
return parser.parse_args()
def main():
config = get_config()
args = parse_args()
config = update_config(config, args)
place = 'gpu' if config.TRAIN.USE_GPU else 'cpu'
paddle.set_device(place)
# build model
model = get_model(config)
model.train()
nranks = paddle.distributed.ParallelEnv().nranks
local_rank = paddle.distributed.ParallelEnv().local_rank
# build scheduler
lr_scheduler = get_scheduler(config)
# build optimizer
optimizer = get_optimizer(model, lr_scheduler, config)
# bulid train transforms
transforms_train = get_transforms(config)
# build loss function
loss_func = get_loss_function(config)
# Resume from checkpoints, and update start_iter
start_iter = 0
if args.resume is not None:
assert os.path.exists(args.resume), args.resume + "is not found!"
opt_state = paddle.load(args.resume.replace('.pdparams', '.pdopt').replace('model', 'opt'))
start_iter = opt_state['LR_Scheduler']['last_epoch']
optimizer.set_state_dict(opt_state)
model.set_state_dict(paddle.load(args.resume))
logger.info("training from checkpoint {}, start_iter= {}".format(args.resume, start_iter))
# build dataset_train
dataset_train = get_dataset(config, data_transform=transforms_train, mode='train')
train_loader = get_dataloader(dataset=dataset_train,
shuffle=True,
batch_size=config.DATA.BATCH_SIZE,
num_iters=config.TRAIN.ITERS,
num_workers=config.DATA.NUM_WORKERS,
start_iter=start_iter)
# build workspace for saving checkpoints
if not os.path.isdir(config.SAVE_DIR):
if os.path.exists(config.SAVE_DIR):
os.remove(config.SAVE_DIR)
os.makedirs(config.SAVE_DIR)
logger.info("train_loader.len= {}".format(len(train_loader)))
if nranks > 1:
# Initialize parallel environment if not done.
if not paddle.distributed.parallel.parallel_helper._is_parallel_ctx_initialized():
logger.info("using dist training")
paddle.distributed.init_parallel_env()
ddp_model = paddle.DataParallel(model)
else:
ddp_model = paddle.DataParallel(model)
avg_loss = 0.0
avg_loss_list = []
iters_per_epoch = len(dataset_train) // config.DATA.BATCH_SIZE
reader_cost_averager = TimeAverager()
batch_cost_averager = TimeAverager()
save_models = deque()
batch_start = time.time()
cur_iter = start_iter
# begin training
for data in train_loader:
cur_iter += 1
reader_cost_averager.record(time.time() - batch_start)
images = data[0]
labels = data[1].astype('int64')
if nranks > 1:
logits_list = ddp_model(images)
else:
logits_list = model(images)
loss_list = loss_func(logits_list, labels)
loss = sum(loss_list)
loss.backward()
optimizer.step()
lr = optimizer.get_lr()
if isinstance(optimizer._learning_rate,paddle.optimizer.lr.LRScheduler):
optimizer._learning_rate.step()
model.clear_gradients()
avg_loss += loss.numpy()[0]
if not avg_loss_list:
avg_loss_list = [l.numpy() for l in loss_list]
else:
for i in range(len(loss_list)):
avg_loss_list[i] += loss_list[i].numpy()
batch_cost_averager.record(
time.time() - batch_start, num_samples=config.DATA.BATCH_SIZE)
if (cur_iter) % config.LOGGING_INFO_FREQ == 0 and local_rank == 0:
avg_loss /= config.LOGGING_INFO_FREQ
avg_loss_list = [l[0] / config.LOGGING_INFO_FREQ for l in avg_loss_list]
remain_iters = config.TRAIN.ITERS - cur_iter
avg_train_batch_cost = batch_cost_averager.get_average()
avg_train_reader_cost = reader_cost_averager.get_average()
eta = calculate_eta(remain_iters, avg_train_batch_cost)
logger.info("[TRAIN] epoch: {}, iter: {}/{}, loss: {:.4f}, lr: {:.8f}, batch_cost:\
{:.4f}, reader_cost: {:.5f}, ips: {:.4f} samples/sec | ETA {}".format(
(cur_iter - 1) // iters_per_epoch + 1, cur_iter, config.TRAIN.ITERS, avg_loss,
lr, avg_train_batch_cost, avg_train_reader_cost,
batch_cost_averager.get_ips_average(), eta))
avg_loss = 0.0
avg_loss_list = []
reader_cost_averager.reset()
batch_cost_averager.reset()
if (cur_iter % config.SAVE_FREQ_CHECKPOINT == 0 or cur_iter == config.TRAIN.ITERS) and local_rank == 0:
current_save_weigth_file = os.path.join(config.SAVE_DIR,
"iter_{}_model_state.pdparams".format(cur_iter))
current_save_opt_file = os.path.join(config.SAVE_DIR,
"iter_{}_opt_state.pdopt".format(cur_iter))
paddle.save(model.state_dict(), current_save_weigth_file)
paddle.save(optimizer.state_dict(), current_save_opt_file)
save_models.append([current_save_weigth_file,
current_save_opt_file])
logger.info("saving the weights of model to {}".format(
current_save_weigth_file))
if len(save_models) > config.KEEP_CHECKPOINT_MAX > 0:
files_to_remove = save_models.popleft()
os.remove(files_to_remove[0])
os.remove(files_to_remove[1])
batch_start = time.time()
time.sleep(1.0)
if __name__ == '__main__':
main()