Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
PaddlePaddle
Serving
提交
cfc75929
S
Serving
项目概览
PaddlePaddle
/
Serving
大约 1 年 前同步成功
通知
185
Star
833
Fork
253
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
105
列表
看板
标记
里程碑
合并请求
10
Wiki
2
Wiki
分析
仓库
DevOps
项目成员
Pages
S
Serving
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
105
Issue
105
列表
看板
标记
里程碑
合并请求
10
合并请求
10
Pages
分析
分析
仓库分析
DevOps
Wiki
2
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
提交
cfc75929
编写于
11月 30, 2020
作者:
C
chengang06
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
add unet benckmark scripts
上级
6dbc6ccd
变更
5
隐藏空白更改
内联
并排
Showing
5 changed file
with
220 addition
and
0 deletion
+220
-0
doc/FAQ.md
doc/FAQ.md
+50
-0
python/examples/unet_for_image_seg/unet_benchmark/README.md
python/examples/unet_for_image_seg/unet_benchmark/README.md
+8
-0
python/examples/unet_for_image_seg/unet_benchmark/img_data/N0060.jpg
...ples/unet_for_image_seg/unet_benchmark/img_data/N0060.jpg
+0
-0
python/examples/unet_for_image_seg/unet_benchmark/launch_benckmark.sh
...les/unet_for_image_seg/unet_benchmark/launch_benckmark.sh
+3
-0
python/examples/unet_for_image_seg/unet_benchmark/unet_benchmark.py
...mples/unet_for_image_seg/unet_benchmark/unet_benchmark.py
+159
-0
未找到文件。
doc/FAQ.md
浏览文件 @
cfc75929
...
@@ -4,6 +4,35 @@
...
@@ -4,6 +4,35 @@
## 基础知识
## 基础知识
#### Q: Paddle Serving 、Paddle Inference、PaddleHub Serving三者的区别及联系?
**A:**
paddle serving是远程服务,即发起预测的设备(手机、浏览器、客户端等)与实际预测的硬件不在一起。 paddle inference是一个library,适合嵌入到一个大系统中保证预测效率,paddle serving调用了paddle inference做远程服务。paddlehub serving可以认为是一个示例,都会使用paddle serving作为统一预测服务入口。如果在web端交互,一般是调用远程服务的形式,可以使用paddle serving的web service搭建。
#### Q: paddle-serving是否支持Int32支持
**A:**
在protobuf定feed_type和fetch_type编号与数据类型对应如下
0-int64
1-float32
2-int32
#### Q: paddle-serving是否支持windows和Linux环境下的多线程调用
**A:**
客户端可以发起多线程访问调用服务端
#### Q: paddle-serving如何修改消息大小限制
**A:**
在server端和client但通过FLAGS_max_body_size来扩大数据量限制,单位为字节,默认为64MB
#### Q: paddle-serving客户端目前支持哪些语言
**A:**
java c++ python
#### Q: paddle-serving目前支持哪些协议
**A:**
http rpc
## 编译问题
## 编译问题
...
@@ -46,7 +75,15 @@ InvalidArgumentError: Device id must be less than GPU count, but received id is:
...
@@ -46,7 +75,15 @@ InvalidArgumentError: Device id must be less than GPU count, but received id is:
**A:**
目前(0.4.0)仅支持CentOS,具体列表查阅
[
这里
](
https://github.com/PaddlePaddle/Serving/blob/develop/doc/DOCKER_IMAGES.md
)
**A:**
目前(0.4.0)仅支持CentOS,具体列表查阅
[
这里
](
https://github.com/PaddlePaddle/Serving/blob/develop/doc/DOCKER_IMAGES.md
)
#### Q: python编译的GCC版本与serving的版本不匹配
**A:**
:1)使用
[
GPU docker
](
https://github.com/PaddlePaddle/Serving/blob/develop/doc/RUN_IN_DOCKER.md#gpunvidia-docker
)
解决环境问题
2)修改anaconda的虚拟环境下安装的python的gcc版本
[
参考
](
https://www.jianshu.com/p/c498b3d86f77
)
#### Q: paddle-serving是否支持本地离线安装
**A:**
支持离线部署,需要把一些相关的
[
依赖包
](
https://github.com/PaddlePaddle/Serving/blob/develop/doc/COMPILE.md
)
提前准备安装好
## 预测问题
## 预测问题
...
@@ -105,6 +142,19 @@ client端的日志直接打印到标准输出。
...
@@ -105,6 +142,19 @@ client端的日志直接打印到标准输出。
通过在部署服务之前 'export GLOG_v=3'可以输出更为详细的日志信息。
通过在部署服务之前 'export GLOG_v=3'可以输出更为详细的日志信息。
#### Q: paddle-serving启动成功后,相关的日志在哪里设置
**A:**
1)警告是glog组件打印的,告知glog初始化之前日志打印在STDERR
2)一般采用GLOG_v方式启动服务同时设置日志级别。
例如:
```
GLOG_v=2 python -m paddle_serving_server.serve --model xxx_conf/ --port 9999
```
#### Q: (GLOG_v=2下)Server端日志一切正常,但Client端始终得不到正确的预测结果
#### Q: (GLOG_v=2下)Server端日志一切正常,但Client端始终得不到正确的预测结果
**A:**
可能是配置文件有问题,检查下配置文件(is_load_tensor,fetch_type等有没有问题)
**A:**
可能是配置文件有问题,检查下配置文件(is_load_tensor,fetch_type等有没有问题)
...
...
python/examples/unet_for_image_seg/unet_benchmark/README.md
0 → 100644
浏览文件 @
cfc75929
#UNET_BENCHMARK 使用说明
## 功能
*
benchmark测试
## 注意事项
*
示例图片(可以有多张)请放置于与img_data路径中,支持jpg,jpeg
*
图片张数应该大于等于并发数量
## TODO
*
http benchmark
python/examples/unet_for_image_seg/unet_benchmark/img_data/N0060.jpg
0 → 100644
浏览文件 @
cfc75929
48.4 KB
python/examples/unet_for_image_seg/unet_benchmark/launch_benckmark.sh
0 → 100644
浏览文件 @
cfc75929
#!/bin/bash
python unet_benchmark.py
--thread
1
--batch_size
1
--model
../unet_client/serving_client_conf.prototxt
# thread/batch can be modified as you wish
python/examples/unet_for_image_seg/unet_benchmark/unet_benchmark.py
0 → 100644
浏览文件 @
cfc75929
# -*- coding: utf-8 -*-
#
# Copyright (c) 2020 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.
"""
unet bench mark script
20201130 first edition by cg82616424
"""
from
__future__
import
unicode_literals
,
absolute_import
import
os
import
time
import
json
import
requests
from
paddle_serving_client
import
Client
from
paddle_serving_client.utils
import
MultiThreadRunner
from
paddle_serving_client.utils
import
benchmark_args
,
show_latency
from
paddle_serving_app.reader
import
Sequential
,
File2Image
,
Resize
,
Transpose
,
BGR2RGB
,
SegPostprocess
args
=
benchmark_args
()
def
get_img_names
(
path
):
"""
Brief:
get img files(jpg) under this path
if any exception happened return None
Args:
path (string): image file path
Returns:
list: images names under this folder
"""
if
not
os
.
path
.
exists
(
path
):
return
None
if
not
os
.
path
.
isdir
(
path
):
return
None
list_name
=
[]
for
f_handler
in
os
.
listdir
(
path
):
file_path
=
os
.
path
.
join
(
path
,
f_handler
)
if
os
.
path
.
isdir
(
file_path
):
continue
else
:
if
not
file_path
.
endswith
(
".jpeg"
)
and
not
file_path
.
endswith
(
".jpg"
):
continue
list_name
.
append
(
file_path
)
return
list_name
def
preprocess_img
(
img_list
):
"""
Brief:
prepare img data for benchmark
Args:
img_list(list): list for img file path
Returns:
image content binary list after preprocess
"""
preprocess
=
Sequential
([
File2Image
(),
Resize
((
512
,
512
))])
result_list
=
[]
for
img
in
img_list
:
img_tmp
=
preprocess
(
img
)
result_list
.
append
(
img_tmp
)
return
result_list
def
benckmark_worker
(
idx
,
resource
):
"""
Brief:
benchmark single worker for unet
Args:
idx(int): worker idx ,use idx to select backend unet service
resource(dict): unet serving endpoint dict
Returns:
latency
TODO:
http benckmarks
"""
profile_flags
=
False
latency_flags
=
False
postprocess
=
SegPostprocess
(
2
)
if
os
.
getenv
(
"FLAGS_profile_client"
):
profile_flags
=
True
if
os
.
getenv
(
"FLAGS_serving_latency"
):
latency_flags
=
True
latency_list
=
[]
client_handler
=
Client
()
client_handler
.
load_client_config
(
args
.
model
)
client_handler
.
connect
(
[
resource
[
"endpoint"
][
idx
%
len
(
resource
[
"endpoint"
])]])
start
=
time
.
time
()
turns
=
resource
[
"turns"
]
img_list
=
resource
[
"img_list"
]
for
i
in
range
(
turns
):
if
args
.
batch_size
>=
1
:
l_start
=
time
.
time
()
feed_batch
=
[]
b_start
=
time
.
time
()
for
bi
in
range
(
args
.
batch_size
):
feed_batch
.
append
({
"image"
:
img_list
[
bi
]})
b_end
=
time
.
time
()
if
profile_flags
:
sys
.
stderr
.
write
(
"PROFILE
\t
pid:{}
\t
unt_pre_0:{} unet_pre_1:{}
\n
"
.
format
(
os
.
getpid
(),
int
(
round
(
b_start
*
1000000
)),
int
(
round
(
b_end
*
1000000
))))
result
=
client_handler
.
predict
(
feed
=
{
"image"
:
img_list
[
bi
]},
fetch
=
[
"output"
])
#result["filename"] = "./img_data/N0060.jpg" % (os.getpid(), idx, time.time())
#postprocess(result) # if you want to measure post process time, you have to uncomment this line
l_end
=
time
.
time
()
if
latency_flags
:
latency_list
.
append
(
l_end
*
1000
-
l_start
*
1000
)
else
:
print
(
"unsupport batch size {}"
.
format
(
args
.
batch_size
))
end
=
time
.
time
()
if
latency_flags
:
return
[[
end
-
start
],
latency_list
]
else
:
return
[[
end
-
start
]]
if
__name__
==
'__main__'
:
"""
usage:
"""
img_file_list
=
get_img_names
(
"./img_data"
)
img_content_list
=
preprocess_img
(
img_file_list
)
multi_thread_runner
=
MultiThreadRunner
()
endpoint_list
=
[
"127.0.0.1:9494"
]
turns
=
1
start
=
time
.
time
()
result
=
multi_thread_runner
.
run
(
benckmark_worker
,
args
.
thread
,
{
"endpoint"
:
endpoint_list
,
"turns"
:
turns
,
"img_list"
:
img_content_list
})
end
=
time
.
time
()
total_cost
=
end
-
start
avg_cost
=
0
for
i
in
range
(
args
.
thread
):
avg_cost
+=
result
[
0
][
i
]
avg_cost
=
avg_cost
/
args
.
thread
print
(
"total cost: {}s"
.
format
(
total_cost
))
print
(
"each thread cost: {}s. "
.
format
(
avg_cost
))
print
(
"qps: {}samples/s"
.
format
(
args
.
batch_size
*
args
.
thread
*
turns
/
total_cost
))
if
os
.
getenv
(
"FLAGS_serving_latency"
):
show_latency
(
result
[
1
])
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录