forked from PaddlePaddle/Paddle
-
Notifications
You must be signed in to change notification settings - Fork 0
/
test_resnet_prim_cinn.py
192 lines (162 loc) · 5.46 KB
/
test_resnet_prim_cinn.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
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
# Copyright (c) 2023 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 time
import unittest
import numpy as np
import paddle
from paddle import fluid
from paddle.fluid import core
from paddle.vision.models import resnet50
SEED = 2020
base_lr = 0.001
momentum_rate = 0.9
l2_decay = 1e-4
batch_size = 2
epoch_num = 1
# In V100, 16G, CUDA 11.2, the results are as follows:
# DY2ST_PRIM_CINN_GT = [
# 5.8473358154296875,
# 8.322463989257812,
# 5.169863700866699,
# 8.399882316589355,
# 7.859550476074219,
# 7.4672698974609375,
# 9.828727722167969,
# 8.270355224609375,
# 8.456792831420898,
# 9.919631958007812,
# ]
# The results in ci as as follows:
DY2ST_PRIM_CINN_GT = [
5.828786849975586,
8.332863807678223,
5.0373005867004395,
8.464998245239258,
8.20099925994873,
7.576723098754883,
9.679173469543457,
8.381753921508789,
8.10612678527832,
10.124727249145508,
]
if core.is_compiled_with_cuda():
paddle.set_flags({'FLAGS_cudnn_deterministic': True})
def reader_decorator(reader):
def __reader__():
for item in reader():
img = np.array(item[0]).astype('float32').reshape(3, 224, 224)
label = np.array(item[1]).astype('int64').reshape(1)
yield img, label
return __reader__
def optimizer_setting(parameter_list=None):
optimizer = fluid.optimizer.Momentum(
learning_rate=base_lr,
momentum=momentum_rate,
regularization=paddle.regularizer.L2Decay(l2_decay),
parameter_list=parameter_list,
)
return optimizer
def run(model, data_loader, optimizer, mode):
if mode == 'train':
model.train()
end_step = 9
elif mode == 'eval':
model.eval()
end_step = 1
for epoch in range(epoch_num):
total_acc1 = 0.0
total_acc5 = 0.0
total_sample = 0
losses = []
for batch_id, data in enumerate(data_loader()):
start_time = time.time()
img, label = data
pred = model(img)
avg_loss = paddle.nn.functional.cross_entropy(
input=pred,
label=label,
soft_label=False,
reduction='mean',
use_softmax=True,
)
acc_top1 = paddle.static.accuracy(input=pred, label=label, k=1)
acc_top5 = paddle.static.accuracy(input=pred, label=label, k=5)
if mode == 'train':
avg_loss.backward()
optimizer.minimize(avg_loss)
model.clear_gradients()
total_acc1 += acc_top1
total_acc5 += acc_top5
total_sample += 1
losses.append(avg_loss.numpy().item())
end_time = time.time()
print(
"[%s]epoch %d | batch step %d, loss %0.8f, acc1 %0.3f, acc5 %0.3f, time %f"
% (
mode,
epoch,
batch_id,
avg_loss,
total_acc1.numpy() / total_sample,
total_acc5.numpy() / total_sample,
end_time - start_time,
)
)
if batch_id >= end_step:
# avoid dataloader throw abort signaal
data_loader._reset()
break
print(losses)
return losses
def train(to_static, enable_prim, enable_cinn):
if core.is_compiled_with_cuda():
paddle.set_device('gpu')
else:
paddle.set_device('cpu')
np.random.seed(SEED)
paddle.seed(SEED)
paddle.framework.random._manual_program_seed(SEED)
fluid.core._set_prim_all_enabled(enable_prim)
train_reader = paddle.batch(
reader_decorator(paddle.dataset.flowers.train(use_xmap=False)),
batch_size=batch_size,
drop_last=True,
)
data_loader = fluid.io.DataLoader.from_generator(capacity=5, iterable=True)
data_loader.set_sample_list_generator(train_reader)
resnet = resnet50(False)
if to_static:
build_strategy = paddle.static.BuildStrategy()
if enable_cinn:
build_strategy.build_cinn_pass = True
resnet = paddle.jit.to_static(resnet, build_strategy=build_strategy)
optimizer = optimizer_setting(parameter_list=resnet.parameters())
train_losses = run(resnet, data_loader, optimizer, 'train')
if to_static and enable_prim and enable_cinn:
eval_losses = run(resnet, data_loader, optimizer, 'eval')
return train_losses
class TestResnet(unittest.TestCase):
@unittest.skipIf(
not (paddle.is_compiled_with_cinn() and paddle.is_compiled_with_cuda()),
"paddle is not compiled with CINN and CUDA",
)
def test_prim_cinn(self):
dy2st_prim_cinn = train(
to_static=True, enable_prim=True, enable_cinn=True
)
np.testing.assert_allclose(
dy2st_prim_cinn, DY2ST_PRIM_CINN_GT, rtol=1e-5
)
if __name__ == '__main__':
unittest.main()