forked from PaddlePaddle/FastDeploy
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathbenchmark_ppcls.py
187 lines (166 loc) · 6.25 KB
/
benchmark_ppcls.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
# Copyright (c) 2022 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 fastdeploy as fd
import cv2
import os
import numpy as np
import pynvml
import psutil
import GPUtil
import time
def parse_arguments():
import argparse
import ast
parser = argparse.ArgumentParser()
parser.add_argument(
"--model", required=True, help="Path of PaddleClas model.")
parser.add_argument(
"--image", type=str, required=False, help="Path of test image file.")
parser.add_argument(
"--cpu_num_thread",
type=int,
default=8,
help="default number of cpu thread.")
parser.add_argument(
"--device_id", type=int, default=0, help="device(gpu) id")
parser.add_argument(
"--iter_num",
required=True,
type=int,
default=300,
help="number of iterations for computing performace.")
parser.add_argument(
"--device",
default="cpu",
help="Type of inference device, support 'cpu' or 'gpu'.")
parser.add_argument(
"--backend",
type=str,
default="ort",
help="inference backend, ort, ov, trt, paddle, paddle_trt.")
parser.add_argument(
"--enable_trt_fp16",
type=bool,
default=False,
help="whether enable fp16 in trt backend")
parser.add_argument(
"--enable_collect_memory_info",
type=bool,
default=False,
help="whether enable collect memory info")
args = parser.parse_args()
return args
def build_option(args):
option = fd.RuntimeOption()
device = args.device
backend = args.backend
option.set_cpu_thread_num(args.cpu_num_thread)
if device == "gpu":
option.use_gpu(args.device_id)
if backend == "trt" or backend == "paddle_trt":
assert device == "gpu", "the trt backend need device==gpu"
option.use_trt_backend()
if backend == "paddle_trt":
option.enable_paddle_to_trt()
if args.enable_trt_fp16:
option.enable_trt_fp16()
elif backend == "ov":
assert device == "cpu", "the openvino backend need device==cpu"
option.use_openvino_backend()
elif backend == "paddle":
option.use_paddle_backend()
elif backend == "ort":
option.use_ort_backend()
else:
print("%s is an unsupported backend" % backend)
return option
def get_current_memory_mb(gpu_id=None):
pid = os.getpid()
p = psutil.Process(pid)
info = p.memory_full_info()
cpu_mem = info.uss / 1024. / 1024.
gpu_mem = 0
if gpu_id is not None:
pynvml.nvmlInit()
handle = pynvml.nvmlDeviceGetHandleByIndex(0)
meminfo = pynvml.nvmlDeviceGetMemoryInfo(handle)
gpu_mem = meminfo.used / 1024. / 1024.
return cpu_mem, gpu_mem
def get_current_gputil(gpu_id):
GPUs = GPUtil.getGPUs()
gpu_load = GPUs[gpu_id].load
return gpu_load
if __name__ == '__main__':
args = parse_arguments()
option = build_option(args)
model_file = os.path.join(args.model, "inference.pdmodel")
params_file = os.path.join(args.model, "inference.pdiparams")
config_file = os.path.join(args.model, "inference_cls.yaml")
gpu_id = args.device_id
end2end_statis = list()
cpu_mem = list()
gpu_mem = list()
gpu_util = list()
if args.device == "cpu":
file_path = args.model + "_model_" + args.backend + "_" + \
args.device + "_" + str(args.cpu_num_thread) + ".txt"
else:
if args.enable_trt_fp16:
file_path = args.model + "_model_" + \
args.backend + "_fp16_" + args.device + ".txt"
else:
file_path = args.model + "_model_" + args.backend + "_" + args.device + ".txt"
f = open(file_path, "w")
f.writelines("===={}====: \n".format(os.path.split(file_path)[-1][:-4]))
try:
model = fd.vision.classification.PaddleClasModel(
model_file, params_file, config_file, runtime_option=option)
model.enable_record_time_of_runtime()
im_ori = cv2.imread(args.image)
for i in range(args.iter_num):
im = im_ori
start = time.time()
result = model.predict(im)
end2end_statis.append(time.time() - start)
if args.enable_collect_memory_info:
gpu_util.append(get_current_gputil(gpu_id))
cm, gm = get_current_memory_mb(gpu_id)
cpu_mem.append(cm)
gpu_mem.append(gm)
runtime_statis = model.print_statis_info_of_runtime()
warmup_iter = args.iter_num // 5
end2end_statis_repeat = end2end_statis[warmup_iter:]
if args.enable_collect_memory_info:
cpu_mem_repeat = cpu_mem[warmup_iter:]
gpu_mem_repeat = gpu_mem[warmup_iter:]
gpu_util_repeat = gpu_util[warmup_iter:]
dump_result = dict()
dump_result["runtime"] = runtime_statis["avg_time"] * 1000
dump_result["end2end"] = np.mean(end2end_statis_repeat) * 1000
if args.enable_collect_memory_info:
dump_result["cpu_rss_mb"] = np.mean(cpu_mem_repeat)
dump_result["gpu_rss_mb"] = np.mean(gpu_mem_repeat)
dump_result["gpu_util"] = np.mean(gpu_util_repeat)
f.writelines("Runtime(ms): {} \n".format(str(dump_result["runtime"])))
f.writelines("End2End(ms): {} \n".format(str(dump_result["end2end"])))
if args.enable_collect_memory_info:
f.writelines("cpu_rss_mb: {} \n".format(
str(dump_result["cpu_rss_mb"])))
f.writelines("gpu_rss_mb: {} \n".format(
str(dump_result["gpu_rss_mb"])))
f.writelines("gpu_util: {} \n".format(
str(dump_result["gpu_util"])))
except:
f.writelines("!!!!!Infer Failed\n")
f.close()