Skip to content

S
Serving

PaddlePaddle / Serving
大约 2 年 前同步成功

通知 187
Star 833
Fork 253
S
Serving
提交 b306084f 编写于 作者: H HexToString
浏览文件
操作

Merge branch 'develop' of https://github.com/PaddlePaddle/Serving into merge_branch

上级 241e53da 62a23aec
隐藏空白更改
内联 并排
Showing with 763 addition and 574 deletion
paddle_inference/paddle/include/paddle_engine.h
浏览文件 @ b306084f
...@@ -37,9 +37,24 @@ using paddle_infer::Tensor; ...@@ -37,9 +37,24 @@ using paddle_infer::Tensor;
using paddle_infer::CreatePredictor; using paddle_infer::CreatePredictor;
DECLARE_int32(gpuid); DECLARE_int32(gpuid);
DECLARE_string(precision);
DECLARE_bool(use_calib);
static const int max_batch = 32; static const int max_batch = 32;
static const int min_subgraph_size = 3; static const int min_subgraph_size = 3;
static PrecisionType precision_type;
PrecisionType GetPrecision(const std::string& precision_data) {
std::string precision_type = predictor::ToLower(precision_data);
if (precision_type == "fp32") {
return PrecisionType::kFloat32;
} else if (precision_type == "int8") {
return PrecisionType::kInt8;
} else if (precision_type == "fp16") {
return PrecisionType::kHalf;
}
return PrecisionType::kFloat32;
}
// Engine Base // Engine Base
class EngineCore { class EngineCore {
...@@ -137,6 +152,7 @@ class PaddleInferenceEngine : public EngineCore { ...@@ -137,6 +152,7 @@ class PaddleInferenceEngine : public EngineCore {
// 2000MB GPU memory // 2000MB GPU memory
config.EnableUseGpu(2000, FLAGS_gpuid); config.EnableUseGpu(2000, FLAGS_gpuid);
} }
precision_type = GetPrecision(FLAGS_precision);
if (engine_conf.has_use_trt() && engine_conf.use_trt()) { if (engine_conf.has_use_trt() && engine_conf.use_trt()) {
if (!engine_conf.has_use_gpu() || !engine_conf.use_gpu()) { if (!engine_conf.has_use_gpu() || !engine_conf.use_gpu()) {
...@@ -145,14 +161,24 @@ class PaddleInferenceEngine : public EngineCore { ...@@ -145,14 +161,24 @@ class PaddleInferenceEngine : public EngineCore {
config.EnableTensorRtEngine(1 << 20, config.EnableTensorRtEngine(1 << 20,
max_batch, max_batch,
min_subgraph_size, min_subgraph_size,
Config::Precision::kFloat32, precision_type,
false, false,
false); FLAGS_use_calib);
LOG(INFO) << "create TensorRT predictor"; LOG(INFO) << "create TensorRT predictor";
} }
if (engine_conf.has_use_lite() && engine_conf.use_lite()) { if (engine_conf.has_use_lite() && engine_conf.use_lite()) {
config.EnableLiteEngine(PrecisionType::kFloat32, true); config.EnableLiteEngine(precision_type, true);
}
if ((!engine_conf.has_use_lite() && !engine_conf.has_use_gpu()) ||
(engine_conf.has_use_lite() && !engine_conf.use_lite() &&
engine_conf.has_use_gpu() && !engine_conf.use_gpu())) {
if (precision_type == PrecisionType::kInt8) {
config.EnableMkldnnQuantizer();
} else if (precision_type == PrecisionType::kHalf) {
config.EnableMkldnnBfloat16();
}
} }
if (engine_conf.has_use_xpu() && engine_conf.use_xpu()) { if (engine_conf.has_use_xpu() && engine_conf.use_xpu()) {
...@@ -171,7 +197,6 @@ class PaddleInferenceEngine : public EngineCore { ...@@ -171,7 +197,6 @@ class PaddleInferenceEngine : public EngineCore {
config.EnableMemoryOptim(); config.EnableMemoryOptim();
} }
predictor::AutoLock lock(predictor::GlobalCreateMutex::instance()); predictor::AutoLock lock(predictor::GlobalCreateMutex::instance());
_predictor = CreatePredictor(config); _predictor = CreatePredictor(config);
if (NULL == _predictor.get()) { if (NULL == _predictor.get()) {
......
paddle_inference/paddle/src/paddle_engine.cpp
浏览文件 @ b306084f
...@@ -20,6 +20,8 @@ namespace paddle_serving { ...@@ -20,6 +20,8 @@ namespace paddle_serving {
namespace inference { namespace inference {
DEFINE_int32(gpuid, 0, "GPU device id to use"); DEFINE_int32(gpuid, 0, "GPU device id to use");
DEFINE_string(precision, "fp32", "precision to deploy, default is fp32");
DEFINE_bool(use_calib, false, "calibration mode, default is false");
REGIST_FACTORY_OBJECT_IMPL_WITH_NAME( REGIST_FACTORY_OBJECT_IMPL_WITH_NAME(
::baidu::paddle_serving::predictor::FluidInferEngine<PaddleInferenceEngine>, ::baidu::paddle_serving::predictor::FluidInferEngine<PaddleInferenceEngine>,
......
python/paddle_serving_server/serve.py
浏览文件 @ b306084f
...@@ -51,6 +51,16 @@ def serve_args(): ...@@ -51,6 +51,16 @@ def serve_args():
"--name", type=str, default="None", help="Default service name") "--name", type=str, default="None", help="Default service name")
parser.add_argument( parser.add_argument(
"--use_mkl", default=False, action="store_true", help="Use MKL") "--use_mkl", default=False, action="store_true", help="Use MKL")
parser.add_argument(
"--precision",
type=str,
default="fp32",
help="precision mode(fp32, int8, fp16, bf16)")
parser.add_argument(
"--use_calib",
default=False,
action="store_true",
help="Use TensorRT Calibration")
parser.add_argument( parser.add_argument(
"--mem_optim_off", "--mem_optim_off",
default=False, default=False,
...@@ -147,6 +157,8 @@ def start_standard_model(serving_port): # pylint: disable=doc-string-missing ...@@ -147,6 +157,8 @@ def start_standard_model(serving_port): # pylint: disable=doc-string-missing
server.use_mkl(use_mkl) server.use_mkl(use_mkl)
server.set_max_body_size(max_body_size) server.set_max_body_size(max_body_size)
server.set_port(port) server.set_port(port)
server.set_precision(args.precision)
server.set_use_calib(args.use_calib)
server.use_encryption_model(use_encryption_model) server.use_encryption_model(use_encryption_model)
if args.product_name != None: if args.product_name != None:
server.set_product_name(args.product_name) server.set_product_name(args.product_name)
...@@ -209,6 +221,8 @@ def start_gpu_card_model(index, gpuid, port, args): # pylint: disable=doc-strin ...@@ -209,6 +221,8 @@ def start_gpu_card_model(index, gpuid, port, args): # pylint: disable=doc-strin
server.set_op_sequence(op_seq_maker.get_op_sequence()) server.set_op_sequence(op_seq_maker.get_op_sequence())
server.set_num_threads(thread_num) server.set_num_threads(thread_num)
server.use_mkl(use_mkl) server.use_mkl(use_mkl)
server.set_precision(args.precision)
server.set_use_calib(args.use_calib)
server.set_memory_optimize(mem_optim) server.set_memory_optimize(mem_optim)
server.set_ir_optimize(ir_optim) server.set_ir_optimize(ir_optim)
server.set_max_body_size(max_body_size) server.set_max_body_size(max_body_size)
...@@ -396,7 +410,9 @@ if __name__ == "__main__": ...@@ -396,7 +410,9 @@ if __name__ == "__main__":
use_lite=args.use_lite, use_lite=args.use_lite,
use_xpu=args.use_xpu, use_xpu=args.use_xpu,
ir_optim=args.ir_optim, ir_optim=args.ir_optim,
thread_num=args.thread) thread_num=args.thread,
precision=args.precision,
use_calib=args.use_calib)
web_service.run_rpc_service() web_service.run_rpc_service()
app_instance = Flask(__name__) app_instance = Flask(__name__)
......
python/paddle_serving_server/web_service.py
浏览文件 @ b306084f
...@@ -115,7 +115,9 @@ class WebService(object): ...@@ -115,7 +115,9 @@ class WebService(object):
mem_optim=True, mem_optim=True,
use_lite=False, use_lite=False,
use_xpu=False, use_xpu=False,
ir_optim=False): ir_optim=False,
precision="fp32",
use_calib=False):
device = "gpu" device = "gpu"
if gpuid == -1: if gpuid == -1:
if use_lite: if use_lite:
...@@ -146,6 +148,8 @@ class WebService(object): ...@@ -146,6 +148,8 @@ class WebService(object):
server.set_memory_optimize(mem_optim) server.set_memory_optimize(mem_optim)
server.set_ir_optimize(ir_optim) server.set_ir_optimize(ir_optim)
server.set_device(device) server.set_device(device)
server.set_precision(precision)
server.set_use_calib(use_calib)
if use_lite: if use_lite:
server.set_lite() server.set_lite()
...@@ -166,6 +170,8 @@ class WebService(object): ...@@ -166,6 +170,8 @@ class WebService(object):
workdir="", workdir="",
port=9393, port=9393,
device="gpu", device="gpu",
precision="fp32",
use_calib=False,
use_lite=False, use_lite=False,
use_xpu=False, use_xpu=False,
ir_optim=False, ir_optim=False,
...@@ -197,7 +203,9 @@ class WebService(object): ...@@ -197,7 +203,9 @@ class WebService(object):
mem_optim=mem_optim, mem_optim=mem_optim,
use_lite=use_lite, use_lite=use_lite,
use_xpu=use_xpu, use_xpu=use_xpu,
ir_optim=ir_optim)) ir_optim=ir_optim,
precision=precision,
use_calib=use_calib))
else: else:
for i, gpuid in enumerate(self.gpus): for i, gpuid in enumerate(self.gpus):
self.rpc_service_list.append( self.rpc_service_list.append(
...@@ -209,7 +217,9 @@ class WebService(object): ...@@ -209,7 +217,9 @@ class WebService(object):
mem_optim=mem_optim, mem_optim=mem_optim,
use_lite=use_lite, use_lite=use_lite,
use_xpu=use_xpu, use_xpu=use_xpu,
ir_optim=ir_optim)) ir_optim=ir_optim,
precision=precision,
use_calib=use_calib))
def _launch_web_service(self): def _launch_web_service(self):
gpu_num = len(self.gpus) gpu_num = len(self.gpus)
......
tools/scripts/ipipe_py3.sh
浏览文件 @ b306084f
...@@ -8,11 +8,26 @@ echo "# #" ...@@ -8,11 +8,26 @@ echo "# #"
echo "# #" echo "# #"
echo "# #" echo "# #"
echo "################################################################" echo "################################################################"
export GOPATH=$HOME/go export GOPATH=$HOME/go
export PATH=$PATH:$GOROOT/bin:$GOPATH/bin export PATH=$PATH:$GOROOT/bin:$GOPATH/bin
export CUDA_INCLUDE_DIRS=/usr/local/cuda-10.2/include export CUDA_INCLUDE_DIRS=/usr/local/cuda-10.2/include
export PYTHONROOT=/usr/local export PYTHONROOT=/usr/local
export PYTHONIOENCODING=utf-8
build_path=/workspace/Serving/
error_words="Fail|DENIED|UNKNOWN|None"
log_dir=${build_path}logs/
data=/root/.cache/serving_data/
dir=`pwd`
RED_COLOR='\E[1;31m'
GREEN_COLOR='\E[1;32m'
YELOW_COLOR='\E[1;33m'
RES='\E[0m'
cuda_version=`cat /usr/local/cuda/version.txt`
if [ $? -ne 0 ]; then
cuda_version=11
fi
go env -w GO111MODULE=on go env -w GO111MODULE=on
go env -w GOPROXY=https://goproxy.cn,direct go env -w GOPROXY=https://goproxy.cn,direct
...@@ -21,90 +36,141 @@ go get -u github.com/grpc-ecosystem/grpc-gateway/protoc-gen-swagger@v1.15.2 ...@@ -21,90 +36,141 @@ go get -u github.com/grpc-ecosystem/grpc-gateway/protoc-gen-swagger@v1.15.2
go get -u github.com/golang/protobuf/protoc-gen-go@v1.4.3 go get -u github.com/golang/protobuf/protoc-gen-go@v1.4.3
go get -u google.golang.org/grpc@v1.33.0 go get -u google.golang.org/grpc@v1.33.0
build_path=/workspace/Serving/ build_whl_list=(build_cpu_server build_gpu_server build_client build_app)
build_whl_list=(build_gpu_server build_client build_cpu_server build_app) rpc_model_list=(grpc_fit_a_line grpc_yolov4 pipeline_imagenet bert_rpc_gpu bert_rpc_cpu ResNet50_rpc \
rpc_model_list=(grpc_impl pipeline_imagenet bert_rpc_gpu bert_rpc_cpu ResNet50_rpc lac_rpc \ lac_rpc cnn_rpc bow_rpc lstm_rpc fit_a_line_rpc deeplabv3_rpc mobilenet_rpc unet_rpc resnetv2_rpc \
cnn_rpc bow_rpc lstm_rpc fit_a_line_rpc deeplabv3_rpc mobilenet_rpc unet_rpc resnetv2_rpc \ criteo_ctr_rpc_cpu criteo_ctr_rpc_gpu ocr_rpc yolov4_rpc_gpu faster_rcnn_hrnetv2p_w18_1x_encrypt)
criteo_ctr_rpc_cpu criteo_ctr_rpc_gpu ocr_rpc yolov4_rpc_gpu) http_model_list=(fit_a_line_http lac_http cnn_http bow_http lstm_http ResNet50_http bert_http\
http_model_list=(fit_a_line_http lac_http cnn_http bow_http lstm_http ResNet50_http bert_http) pipeline_ocr_cpu_http)
function setproxy(){ function setproxy() {
export http_proxy=${proxy} export http_proxy=${proxy}
export https_proxy=${proxy} export https_proxy=${proxy}
} }
function unsetproxy(){ function unsetproxy() {
unset http_proxy unset http_proxy
unset https_proxy unset https_proxy
} }
function kill_server_process(){ function kill_server_process() {
kill `ps -ef|grep $1 |awk '{print $2}'` kill `ps -ef | grep serving | awk '{print $2}'` > /dev/null 2>&1
kill `ps -ef|grep serving |awk '{print $2}'` kill `ps -ef | grep python | awk '{print $2}'` > /dev/null 2>&1
echo -e "${GREEN_COLOR}process killed...${RES}"
} }
function check() { function check() {
cd ${build_path} cd ${build_path}
if [ ! -f paddle_serving_app* ]; then if [ ! -f paddle_serving_app* ]; then
echo "paddle_serving_app is compiled failed, please check your pull request" echo "paddle_serving_app is compiled failed, please check your pull request"
exit 1 exit 1
elif [ ! -f paddle_serving_server-* ]; then elif [ ! -f paddle_serving_server-* ]; then
echo "paddle_serving_server-cpu is compiled failed, please check your pull request" echo "paddle_serving_server-cpu is compiled failed, please check your pull request"
exit 1 exit 1
elif [ ! -f paddle_serving_server_* ]; then elif [ ! -f paddle_serving_server_* ]; then
echo "paddle_serving_server_gpu is compiled failed, please check your pull request" echo "paddle_serving_server_gpu is compiled failed, please check your pull request"
exit 1 exit 1
elif [ ! -f paddle_serving_client* ]; then elif [ ! -f paddle_serving_client* ]; then
echo "paddle_serving_server_client is compiled failed, please check your pull request" echo "paddle_serving_server_client is compiled failed, please check your pull request"
exit 1 exit 1
else else
echo "paddle serving build passed" echo "paddle serving build passed"
fi fi
} }
function check_result() { function check_result() {
if [ $? -ne 0 ];then if [ $? == 0 ]; then
echo -e "\033[4;31;42m$1 model runs failed, please check your pull request or modify test case! \033[0m" echo -e "${GREEN_COLOR}$1 execute normally${RES}"
exit 1 if [ $1 == "server" ]; then
sleep $2
tail ${dir}server_log.txt | tee -a ${log_dir}server_total.txt
fi
if [ $1 == "client" ]; then
tail ${dir}client_log.txt | tee -a ${log_dir}client_total.txt
grep -E "${error_words}" ${dir}client_log.txt > /dev/null
if [ $? == 0 ]; then
echo -e "${RED_COLOR}$1 error command${RES}\n" | tee -a ${log_dir}server_total.txt ${log_dir}client_total.txt
error_log $2
else
echo -e "${GREEN_COLOR}$2${RES}\n" | tee -a ${log_dir}server_total.txt ${log_dir}client_total.txt
fi
fi
else else
echo -e "\033[4;37;42m$1 model runs successfully, congratulations! \033[0m" echo -e "${RED_COLOR}$1 error command${RES}\n" | tee -a ${log_dir}server_total.txt ${log_dir}client_total.txt
tail ${dir}client_log.txt | tee -a ${log_dir}client_total.txt
error_log $2
fi fi
} }
function before_hook(){ function error_log() {
setproxy arg=${1//\//_}
cd ${build_path}/python echo "-----------------------------" | tee -a ${log_dir}error_models.txt
pip3.6 install --upgrade pip arg=${arg%% *}
pip3.6 install requests arr=(${arg//_/ })
pip3.6 install -r requirements.txt if [ ${arr[@]: -1} == 1 -o ${arr[@]: -1} == 2 ]; then
pip3.6 install numpy==1.16.4 model=${arr[*]:0:${#arr[*]}-3}
echo "before hook configuration is successful.... " deployment=${arr[*]: -3}
else
model=${arr[*]:0:${#arr[*]}-2}
deployment=${arr[*]: -2}
fi
echo "model: ${model// /_}" | tee -a ${log_dir}error_models.txt
echo "deployment: ${deployment// /_}" | tee -a ${log_dir}error_models.txt
echo "py_version: python3.6" | tee -a ${log_dir}error_models.txt
echo "cuda_version: ${cuda_version}" | tee -a ${log_dir}error_models.txt
echo "status: Failed" | tee -a ${log_dir}error_models.txt
echo -e "-----------------------------\n\n" | tee -a ${log_dir}error_models.txt
prefix=${arg//\//_}
for file in ${dir}*
do
cp ${file} ${log_dir}error/${prefix}_${file##*/}
done
}
function check_dir() {
if [ ! -d "$1" ]
then
mkdir -p $1
fi
} }
function run_env(){ function link_data() {
setproxy for file in $1*
pip3.6 install --upgrade nltk==3.4 do
pip3.6 install --upgrade scipy==1.2.1 if [ ! -h ${file##*/} ]
pip3.6 install --upgrade setuptools==41.0.0 then
pip3.6 install paddlehub ujson paddlepaddle==2.0.0 ln -s ${file} ./${file##*/}
echo "run env configuration is successful.... " fi
done
} }
function run_gpu_env(){ function before_hook() {
cd ${build_path} setproxy
export LD_LIBRARY_PATH=/usr/local/lib64/python3.6/site-packages/paddle/libs/:$LD_LIBRARY_PATH unsetproxy
export LD_LIBRARY_PATH=/workspace/Serving/build_gpu/third_party/install/Paddle/lib/:/workspace/Serving/build_gpu/third_party/Paddle/src/extern_paddle/third_party/install/mklml/lib/:/workspace/Serving/build_gpu/third_party/Paddle/src/extern_paddle/third_party/install/mkldnn/lib/:$LD_LIBRARY_PATH cd ${build_path}/python
export SERVING_BIN=${build_path}/build_gpu/core/general-server/serving python3.6 -m pip install --upgrade pip
echo "run gpu env configuration is successful.... " python3.6 -m pip install requests
python3.6 -m pip install -r requirements.txt -i https://mirror.baidu.com/pypi/simple
python3.6 -m pip install numpy==1.16.4
python3.6 -m pip install paddlehub -i https://mirror.baidu.com/pypi/simple
echo "before hook configuration is successful.... "
} }
function run_cpu_env(){ function run_env() {
cd ${build_path} setproxy
export LD_LIBRARY_PATH=/usr/local/lib64/python3.6/site-packages/paddle/libs/:$LD_LIBRARY_PATH python3.6 -m pip install --upgrade nltk==3.4
export LD_LIBRARY_PATH=/workspace/Serving/build_cpu/third_party/install/Paddle/lib/:$LD_LIBRARY_PATH python3.6 -m pip install --upgrade scipy==1.2.1
export SERVING_BIN=${build_path}/build_cpu/core/general-server/serving python3.6 -m pip install --upgrade setuptools==41.0.0
echo "run cpu env configuration is successful.... " python3.6 -m pip install paddlehub ujson paddlepaddle==2.0.0
echo "run env configuration is successful.... "
}
function run_gpu_env() {
cd ${build_path}
export LD_LIBRARY_PATH=/usr/local/lib64/python3.6/site-packages/paddle/libs/:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=/workspace/Serving/build_gpu/third_party/install/Paddle/lib/:/workspace/Serving/build_gpu/third_party/Paddle/src/extern_paddle/third_party/install/mklml/lib/:/workspace/Serving/build_gpu/third_party/Paddle/src/extern_paddle/third_party/install/mkldnn/lib/:$LD_LIBRARY_PATH
export SERVING_BIN=${build_path}/build_gpu/core/general-server/serving
echo "run gpu env configuration is successful.... "
} }
function build_gpu_server() { function build_gpu_server() {
...@@ -124,573 +190,643 @@ function build_gpu_server() { ...@@ -124,573 +190,643 @@ function build_gpu_server() {
-DSERVER=ON \ -DSERVER=ON \
-DTENSORRT_ROOT=/usr \ -DTENSORRT_ROOT=/usr \
-DWITH_GPU=ON .. -DWITH_GPU=ON ..
make -j18 make -j32
make -j18 make -j32
make install -j18 make install -j32
pip3.6 uninstall paddle-serving-server-gpu -y python3.6 -m pip uninstall paddle-serving-server-gpu -y
pip3.6 install ${build_path}/build/python/dist/* python3.6 -m pip install ${build_path}/build/python/dist/*
cp ${build_path}/build/python/dist/* ../ cp ${build_path}/build/python/dist/* ../
cp -r ${build_path}/build/ ${build_path}/build_gpu cp -r ${build_path}/build/ ${build_path}/build_gpu
} }
function build_client() { function build_cpu_server(){
setproxy setproxy
cd ${build_path} cd ${build_path}
if [ -d build ];then if [ -d build_cpu ];then
rm -rf build rm -rf build_cpu
fi fi
mkdir build && cd build if [ -d build ];then
cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python3.6m/ \ rm -rf build
-DPYTHON_LIBRARIES=$PYTHONROOT/lib64/libpython3.6.so \ fi
-DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python3.6 \ mkdir build && cd build
-DCLIENT=ON .. cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python3.6m/ \
make -j18 -DPYTHON_LIBRARIES=$PYTHONROOT/lib64/libpython3.6.so \
make -j18 -DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python3.6 \
cp ${build_path}/build/python/dist/* ../ -DWITH_GPU=OFF \
pip3.6 uninstall paddle-serving-client -y -DSERVER=ON ..
pip3.6 install ${build_path}/build/python/dist/* make -j32
make -j32
make install -j32
cp ${build_path}/build/python/dist/* ../
python3.6 -m pip uninstall paddle-serving-server -y
python3.6 -m pip install ${build_path}/build/python/dist/*
cp -r ${build_path}/build/ ${build_path}/build_cpu
} }
function build_cpu_server(){ function build_client() {
setproxy setproxy
cd ${build_path} cd ${build_path}
if [ -d build_cpu ];then if [ -d build ];then
rm -rf build_cpu rm -rf build
fi fi
if [ -d build ];then mkdir build && cd build
rm -rf build cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python3.6m/ \
fi -DPYTHON_LIBRARIES=$PYTHONROOT/lib64/libpython3.6.so \
mkdir build && cd build -DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python3.6 \
cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python3.6m/ \ -DCLIENT=ON ..
-DPYTHON_LIBRARIES=$PYTHONROOT/lib64/libpython3.6.so \ make -j32
-DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python3.6 \ make -j32
-DWITH_GPU=OFF \ cp ${build_path}/build/python/dist/* ../
-DSERVER=ON .. python3.6 -m pip uninstall paddle-serving-client -y
make -j18 python3.6 -m pip install ${build_path}/build/python/dist/*
make -j18
make install -j18
cp ${build_path}/build/python/dist/* ../
pip3.6 uninstall paddle-serving-server -y
pip3.6 install ${build_path}/build/python/dist/*
cp -r ${build_path}/build/ ${build_path}/build_cpu
} }
function build_app() { function build_app() {
setproxy setproxy
pip3.6 install paddlehub ujson Pillow python3.6 -m pip install paddlehub ujson Pillow
pip3.6 install paddlepaddle==2.0.0 python3.6 -m pip install paddlepaddle==2.0.0
cd ${build_path} cd ${build_path}
if [ -d build ];then if [ -d build ];then
rm -rf build rm -rf build
fi fi
mkdir build && cd build mkdir build && cd build
cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python3.6m/ \ cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python3.6m/ \
-DPYTHON_LIBRARIES=$PYTHONROOT/lib/libpython3.6.so \ -DPYTHON_LIBRARIES=$PYTHONROOT/lib/libpython3.6.so \
-DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python3.6 \ -DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python3.6 \
-DCMAKE_INSTALL_PREFIX=./output -DAPP=ON .. -DCMAKE_INSTALL_PREFIX=./output -DAPP=ON ..
make make
cp ${build_path}/build/python/dist/* ../ cp ${build_path}/build/python/dist/* ../
pip3.6 uninstall paddle-serving-app -y python3.6 -m pip uninstall paddle-serving-app -y
pip3.6 install ${build_path}/build/python/dist/* python3.6 -m pip install ${build_path}/build/python/dist/*
} }
function bert_rpc_gpu(){ function faster_rcnn_hrnetv2p_w18_1x_encrypt() {
run_gpu_env dir=${log_dir}rpc_model/faster_rcnn_hrnetv2p_w18_1x/
unsetproxy cd ${build_path}/python/examples/detection/faster_rcnn_hrnetv2p_w18_1x
cd ${build_path}/python/examples/bert check_dir ${dir}
sh get_data.sh >/dev/null 2>&1 data_dir=${data}detection/faster_rcnn_hrnetv2p_w18_1x/
sed -i 's/9292/8860/g' bert_client.py link_data ${data_dir}
sed -i '$aprint(result)' bert_client.py python3.6 encrypt.py
cp -r /root/.cache/dist_data/serving/bert/bert_seq128_* ./ unsetproxy
ls -hlst echo -e "${GREEN_COLOR}faster_rcnn_hrnetv2p_w18_1x_ENCRYPTION_GPU_RPC server started${RES}" | tee -a ${log_dir}server_total.txt
python3.6 -m paddle_serving_server_gpu.serve --model bert_seq128_model/ --port 8860 --gpu_ids 0 & python3.6 -m paddle_serving_server.serve --model encrypt_server/ --port 9494 --use_trt --gpu_ids 0 --use_encryption_model > ${dir}server_log.txt 2>&1 &
sleep 15 check_result server 3
nvidia-smi echo -e "${GREEN_COLOR}faster_rcnn_hrnetv2p_w18_1x_ENCRYPTION_GPU_RPC client started${RES}" | tee -a ${log_dir}client_total.txt
head data-c.txt | python3.6 bert_client.py --model bert_seq128_client/serving_client_conf.prototxt python3.6 test_encryption.py 000000570688.jpg > ${dir}client_log.txt 2>&1
nvidia-smi check_result client "faster_rcnn_hrnetv2p_w18_1x_ENCRYPTION_GPU_RPC server test completed"
check_result $FUNCNAME kill_server_process
kill_server_process serving }
}
function pipeline_ocr_cpu_http() {
function bert_rpc_cpu(){ dir=${log_dir}rpc_model/pipeline_ocr_cpu_http/
run_cpu_env check_dir ${dir}
unsetproxy cd ${build_path}/python/examples/pipeline/ocr
cd ${build_path}/python/examples/bert data_dir=${data}ocr/
sed -i 's/8860/8861/g' bert_client.py link_data ${data_dir}
python3.6 -m paddle_serving_server.serve --model bert_seq128_model/ --port 8861 & unsetproxy
sleep 3 echo -e "${GREEN_COLOR}pipeline_ocr_CPU_HTTP server started${RES}" | tee -a ${log_dir}server_total.txt
cp data-c.txt.1 data-c.txt $py_version web_service.py > ${dir}server_log.txt 2>&1 &
head data-c.txt | python3.6 bert_client.py --model bert_seq128_client/serving_client_conf.prototxt check_result server 5
check_result $FUNCNAME echo -e "${GREEN_COLOR}pipeline_ocr_CPU_HTTP client started${RES}" | tee -a ${log_dir}client_total.txt
kill_server_process serving timeout 15s $py_version pipeline_http_client.py > ${dir}client_log.txt 2>&1
} check_result client "pipeline_ocr_CPU_HTTP server test completed"
kill_server_process
function criteo_ctr_with_cube_rpc(){ }
unsetproxy
run_cpu_env function bert_rpc_gpu() {
cd ${build_path}/python/examples/criteo_ctr_with_cube dir=${log_dir}rpc_model/bert_rpc_gpu/
ln -s /root/.cache/dist_data/serving/criteo_ctr_with_cube/raw_data ./ check_dir ${dir}
sed -i "s/9292/8888/g" test_server.py run_gpu_env
sed -i "s/9292/8888/g" test_client.py unsetproxy
wget https://paddle-serving.bj.bcebos.com/unittest/ctr_cube_unittest.tar.gz >/dev/null 2>&1 cd ${build_path}/python/examples/bert
tar xf ctr_cube_unittest.tar.gz data_dir=${data}bert/
mv models/ctr_client_conf ./ link_data ${data_dir}
mv models/ctr_serving_model_kv ./ sh get_data.sh >/dev/null 2>&1
mv models/data ./cube/ sed -i 's/9292/8860/g' bert_client.py
wget https://paddle-serving.bj.bcebos.com/others/cube_app.tar.gz >/dev/null 2>&1 sed -i '$aprint(result)' bert_client.py
tar xf cube_app.tar.gz ls -hlst
mv cube_app/cube* ./cube/ python3.6 -m paddle_serving_server.serve --model bert_seq128_model/ --port 8860 --gpu_ids 0 > ${dir}server_log.txt 2>&1 &
sh cube_prepare.sh > haha 2>&1 & check_result server 15
sleep 5 nvidia-smi
python3.6 test_server.py ctr_serving_model_kv & head data-c.txt | python3.6 bert_client.py --model bert_seq128_client/serving_client_conf.prototxt > ${dir}client_log.txt 2>&1
sleep 5 check_result client "bert_GPU_RPC server test completed"
python3.6 test_client.py ctr_client_conf/serving_client_conf.prototxt ./raw_data nvidia-smi
check_result $FUNCNAME kill_server_process
kill `ps -ef|grep cube|awk '{print $2}'` }
kill_server_process test_server
} function bert_rpc_cpu() {
dir=${log_dir}rpc_model/bert_rpc_cpu/
function pipeline_imagenet(){ check_dir ${dir}
run_gpu_env unsetproxy
unsetproxy cd ${build_path}/python/examples/bert
cd ${build_path}/python/examples/pipeline/imagenet data_dir=${data}bert/
cp -r /root/.cache/dist_data/serving/imagenet/* ./ link_data ${data_dir}
ls -a sed -i 's/8860/8861/g' bert_client.py
python3.6 resnet50_web_service.py & python3.6 -m paddle_serving_server.serve --model bert_seq128_model/ --port 8861 > ${dir}server_log.txt 2>&1 &
sleep 5 check_result server 3
nvidia-smi cp data-c.txt.1 data-c.txt
python3.6 pipeline_rpc_client.py head data-c.txt | python3.6 bert_client.py --model bert_seq128_client/serving_client_conf.prototxt > ${dir}client_log.txt 2>&1
nvidia-smi check_result client "bert_CPU_RPC server test completed"
# check_result $FUNCNAME kill_server_process
kill_server_process resnet50_web_service }
}
function pipeline_imagenet() {
function ResNet50_rpc(){ dir=${log_dir}rpc_model/pipeline_imagenet/
run_gpu_env check_dir ${dir}
unsetproxy unsetproxy
cd ${build_path}/python/examples/imagenet cd ${build_path}/python/examples/pipeline/imagenet
cp -r /root/.cache/dist_data/serving/imagenet/* ./ data_dir=${data}imagenet/
sed -i 's/9696/8863/g' resnet50_rpc_client.py link_data ${data_dir}
python3.6 -m paddle_serving_server_gpu.serve --model ResNet50_vd_model --port 8863 --gpu_ids 0 & python3.6 resnet50_web_service.py > ${dir}server_log.txt 2>&1 &
sleep 5 check_result server 5
nvidia-smi nvidia-smi
python3.6 resnet50_rpc_client.py ResNet50_vd_client_config/serving_client_conf.prototxt python3.6 pipeline_rpc_client.py > ${dir}client_log.txt 2>&1
nvidia-smi check_result client "pipeline_imagenet_GPU_RPC server test completed"
check_result $FUNCNAME nvidia-smi
kill_server_process serving kill_server_process
} }
function ResNet101_rpc(){ function ResNet50_rpc() {
run_gpu_env dir=${log_dir}rpc_model/ResNet50_rpc/
unsetproxy check_dir ${dir}
cd ${build_path}/python/examples/imagenet unsetproxy
sed -i "22cclient.connect(['${host}:8864'])" image_rpc_client.py cd ${build_path}/python/examples/imagenet
python3.6 -m paddle_serving_server_gpu.serve --model ResNet101_vd_model --port 8864 --gpu_ids 0 & data_dir=${data}imagenet/
sleep 5 link_data ${data_dir}
nvidia-smi sed -i 's/9696/8863/g' resnet50_rpc_client.py
python3.6 image_rpc_client.py ResNet101_vd_client_config/serving_client_conf.prototxt python3.6 -m paddle_serving_server.serve --model ResNet50_vd_model --port 8863 --gpu_ids 0 > ${dir}server_log.txt 2>&1 &
nvidia-smi check_result server 5
check_result $FUNCNAME nvidia-smi
kill_server_process serving python3.6 resnet50_rpc_client.py ResNet50_vd_client_config/serving_client_conf.prototxt > ${dir}client_log.txt 2>&1
sleep 5 check_result client "ResNet50_GPU_RPC server test completed"
} nvidia-smi
kill_server_process
function cnn_rpc(){ }
unsetproxy
run_cpu_env function ResNet101_rpc() {
cd ${build_path}/python/examples/imdb dir=${log_dir}rpc_model/ResNet101_rpc/
cp -r /root/.cache/dist_data/serving/imdb/* ./ check_dir ${dir}
tar xf imdb_model.tar.gz && tar xf text_classification_data.tar.gz unsetproxy
sed -i 's/9292/8865/g' test_client.py cd ${build_path}/python/examples/imagenet
python3.6 -m paddle_serving_server.serve --model imdb_cnn_model/ --port 8865 & data_dir=${data}imagenet/
sleep 5 link_data ${data_dir}
head test_data/part-0 | python3.6 test_client.py imdb_cnn_client_conf/serving_client_conf.prototxt imdb.vocab sed -i "22cclient.connect(['127.0.0.1:8864'])" image_rpc_client.py
check_result $FUNCNAME python3.6 -m paddle_serving_server.serve --model ResNet101_vd_model --port 8864 --gpu_ids 0 > ${dir}server_log.txt 2>&1 &
kill_server_process serving check_result server 5
} nvidia-smi
python3.6 image_rpc_client.py ResNet101_vd_client_config/serving_client_conf.prototxt > ${dir}client_log.txt 2>&1
function bow_rpc(){ check_result client "ResNet101_GPU_RPC server test completed"
unsetproxy nvidia-smi
run_cpu_env kill_server_process
cd ${build_path}/python/examples/imdb }
sed -i 's/8865/8866/g' test_client.py
python3.6 -m paddle_serving_server.serve --model imdb_bow_model/ --port 8866 & function cnn_rpc() {
sleep 5 dir=${log_dir}rpc_model/cnn_rpc/
head test_data/part-0 | python3.6 test_client.py imdb_bow_client_conf/serving_client_conf.prototxt imdb.vocab check_dir ${dir}
check_result $FUNCNAME unsetproxy
kill_server_process serving cd ${build_path}/python/examples/imdb
} data_dir=${data}imdb/
link_data ${data_dir}
function lstm_rpc(){ sed -i 's/9292/8865/g' test_client.py
unsetproxy python3.6 -m paddle_serving_server.serve --model imdb_cnn_model/ --port 8865 > ${dir}server_log.txt 2>&1 &
run_cpu_env check_result server 5
cd ${build_path}/python/examples/imdb head test_data/part-0 | python3.6 test_client.py imdb_cnn_client_conf/serving_client_conf.prototxt imdb.vocab > ${dir}client_log.txt 2>&1
sed -i 's/8866/8867/g' test_client.py check_result client "cnn_CPU_RPC server test completed"
python3.6 -m paddle_serving_server.serve --model imdb_lstm_model/ --port 8867 & kill_server_process
sleep 5 }
head test_data/part-0 | python3.6 test_client.py imdb_lstm_client_conf/serving_client_conf.prototxt imdb.vocab
check_result $FUNCNAME function bow_rpc() {
kill_server_process serving dir=${log_dir}rpc_model/bow_rpc/
} check_dir ${dir}
unsetproxy
function lac_rpc(){ cd ${build_path}/python/examples/imdb
unsetproxy data_dir=${data}imdb/
run_cpu_env link_data ${data_dir}
cd ${build_path}/python/examples/lac sed -i 's/8865/8866/g' test_client.py
python3.6 -m paddle_serving_app.package --get_model lac >/dev/null 2>&1 python3.6 -m paddle_serving_server.serve --model imdb_bow_model/ --port 8866 > ${dir}server_log.txt 2>&1 &
tar xf lac.tar.gz check_result server 5
sed -i 's/9292/8868/g' lac_client.py head test_data/part-0 | python3.6 test_client.py imdb_bow_client_conf/serving_client_conf.prototxt imdb.vocab > ${dir}client_log.txt 2>&1
python3.6 -m paddle_serving_server.serve --model lac_model/ --port 8868 & check_result client "bow_CPU_RPC server test completed"
sleep 5 kill_server_process
echo "我爱北京天安门" | python3.6 lac_client.py lac_client/serving_client_conf.prototxt lac_dict/ }
check_result $FUNCNAME
kill_server_process serving function lstm_rpc() {
} dir=${log_dir}rpc_model/lstm_rpc/
check_dir ${dir}
function fit_a_line_rpc(){ unsetproxy
unsetproxy cd ${build_path}/python/examples/imdb
run_cpu_env data_dir=${data}imdb/
cd ${build_path}/python/examples/fit_a_line link_data ${data_dir}
sh get_data.sh >/dev/null 2>&1 sed -i 's/8866/8867/g' test_client.py
sed -i 's/9393/8869/g' test_client.py python3.6 -m paddle_serving_server.serve --model imdb_lstm_model/ --port 8867 > ${dir}server_log.txt 2>&1 &
python3.6 -m paddle_serving_server.serve --model uci_housing_model --port 8869 & check_result server 5
sleep 5 head test_data/part-0 | python3.6 test_client.py imdb_lstm_client_conf/serving_client_conf.prototxt imdb.vocab > ${dir}client_log.txt 2>&1
python3.6 test_client.py uci_housing_client/serving_client_conf.prototxt check_result client "lstm_CPU_RPC server test completed"
check_result $FUNCNAME kill_server_process
kill_server_process serving }
}
function lac_rpc() {
function faster_rcnn_model_rpc(){ dir=${log_dir}rpc_model/lac_rpc/
unsetproxy check_dir ${dir}
run_gpu_env unsetproxy
cd ${build_path}/python/examples/faster_rcnn cd ${build_path}/python/examples/lac
cp -r /root/.cache/dist_data/serving/faster_rcnn/faster_rcnn_model.tar.gz ./ data_dir=${data}lac/
tar xf faster_rcnn_model.tar.gz link_data ${data_dir}
wget https://paddle-serving.bj.bcebos.com/pddet_demo/infer_cfg.yml >/dev/null 2>&1 sed -i 's/9292/8868/g' lac_client.py
mv faster_rcnn_model/pddet* ./ python3.6 -m paddle_serving_server.serve --model lac_model/ --port 8868 > ${dir}server_log.txt 2>&1 &
sed -i 's/9494/8870/g' test_client.py check_result server 5
python3.6 -m paddle_serving_server_gpu.serve --model pddet_serving_model --port 8870 --gpu_id 0 --thread 2 & echo "我爱北京天安门" | python3.6 lac_client.py lac_client/serving_client_conf.prototxt lac_dict/ > ${dir}client_log.txt 2>&1
echo "faster rcnn running ..." check_result client "lac_CPU_RPC server test completed"
nvidia-smi kill_server_process
sleep 5 }
python3.6 test_client.py pddet_client_conf/serving_client_conf.prototxt infer_cfg.yml 000000570688.jpg
nvidia-smi function fit_a_line_rpc() {
check_result $FUNCNAME dir=${log_dir}rpc_model/fit_a_line_rpc/
kill_server_process serving check_dir ${dir}
} unsetproxy
cd ${build_path}/python/examples/fit_a_line
function cascade_rcnn_rpc(){ data_dir=${data}fit_a_line/
unsetproxy link_data ${data_dir}
run_gpu_env sed -i 's/9393/8869/g' test_client.py
cd ${build_path}/python/examples/cascade_rcnn python3.6 -m paddle_serving_server.serve --model uci_housing_model --port 8869 > ${dir}server_log.txt 2>&1 &
cp -r /root/.cache/dist_data/serving/cascade_rcnn/cascade_rcnn_r50_fpx_1x_serving.tar.gz ./ check_result server 5
tar xf cascade_rcnn_r50_fpx_1x_serving.tar.gz python3.6 test_client.py uci_housing_client/serving_client_conf.prototxt > ${dir}client_log.txt 2>&1
sed -i "s/9292/8879/g" test_client.py check_result client "fit_a_line_CPU_RPC server test completed"
python3.6 -m paddle_serving_server_gpu.serve --model serving_server --port 8879 --gpu_id 0 --thread 2 & kill_server_process
sleep 5 }
nvidia-smi
python3.6 test_client.py function faster_rcnn_model_rpc() {
nvidia-smi dir=${log_dir}rpc_model/faster_rcnn_rpc/
check_result $FUNCNAME check_dir ${dir}
kill_server_process serving unsetproxy
cd ${build_path}/python/examples/detection/faster_rcnn_r50_fpn_1x_coco
data_dir=${data}detection/faster_rcnn_r50_fpn_1x_coco/
link_data ${data_dir}
sed -i 's/9494/8870/g' test_client.py
python3.6 -m paddle_serving_server.serve --model serving_server --port 8870 --gpu_ids 0 --thread 2 --use_trt > ${dir}server_log.txt 2>&1 &
echo "faster rcnn running ..."
nvidia-smi
check_result server 10
python3.6 test_client.py 000000570688.jpg > ${dir}client_log.txt 2>&1
nvidia-smi
check_result client "faster_rcnn_GPU_RPC server test completed"
kill_server_process
}
function cascade_rcnn_rpc() {
dir=${log_dir}rpc_model/cascade_rcnn_rpc/
check_dir ${dir}
unsetproxy
cd ${build_path}/python/examples/cascade_rcnn
data_dir=${data}cascade_rcnn/
link_data ${data_dir}
sed -i "s/9292/8879/g" test_client.py
python3.6 -m paddle_serving_server.serve --model serving_server --port 8879 --gpu_ids 0 --thread 2 > ${dir}server_log.txt 2>&1 &
check_result server 5
nvidia-smi
python3.6 test_client.py > ${dir}client_log.txt 2>&1
nvidia-smi
check_result client "cascade_rcnn_GPU_RPC server test completed"
kill_server_process
} }
function deeplabv3_rpc() { function deeplabv3_rpc() {
unsetproxy dir=${log_dir}rpc_model/deeplabv3_rpc/
run_gpu_env check_dir ${dir}
cd ${build_path}/python/examples/deeplabv3 unsetproxy
cp -r /root/.cache/dist_data/serving/deeplabv3/deeplabv3.tar.gz ./ cd ${build_path}/python/examples/deeplabv3
tar xf deeplabv3.tar.gz data_dir=${data}deeplabv3/
sed -i "s/9494/8880/g" deeplabv3_client.py link_data ${data_dir}
python3.6 -m paddle_serving_server_gpu.serve --model deeplabv3_server --gpu_ids 0 --port 8880 --thread 2 & sed -i "s/9494/8880/g" deeplabv3_client.py
sleep 5 python3.6 -m paddle_serving_server.serve --model deeplabv3_server --gpu_ids 0 --port 8880 --thread 2 > ${dir}server_log.txt 2>&1 &
nvidia-smi check_result server 5
python3.6 deeplabv3_client.py nvidia-smi
nvidia-smi python3.6 deeplabv3_client.py > ${dir}client_log.txt 2>&1
check_result $FUNCNAME nvidia-smi
kill_server_process serving check_result client "deeplabv3_GPU_RPC server test completed"
kill_server_process
} }
function mobilenet_rpc() { function mobilenet_rpc() {
unsetproxy dir=${log_dir}rpc_model/mobilenet_rpc/
run_gpu_env check_dir ${dir}
cd ${build_path}/python/examples/mobilenet unsetproxy
python3.6 -m paddle_serving_app.package --get_model mobilenet_v2_imagenet >/dev/null 2>&1 cd ${build_path}/python/examples/mobilenet
tar xf mobilenet_v2_imagenet.tar.gz python3.6 -m paddle_serving_app.package --get_model mobilenet_v2_imagenet >/dev/null 2>&1
sed -i "s/9393/8881/g" mobilenet_tutorial.py tar xf mobilenet_v2_imagenet.tar.gz
python3.6 -m paddle_serving_server_gpu.serve --model mobilenet_v2_imagenet_model --gpu_ids 0 --port 8881 & sed -i "s/9393/8881/g" mobilenet_tutorial.py
sleep 5 python3.6 -m paddle_serving_server.serve --model mobilenet_v2_imagenet_model --gpu_ids 0 --port 8881 > ${dir}server_log.txt 2>&1 &
nvidia-smi check_result server 5
python3.6 mobilenet_tutorial.py nvidia-smi
nvidia-smi python3.6 mobilenet_tutorial.py > ${dir}client_log.txt 2>&1
check_result $FUNCNAME nvidia-smi
kill_server_process serving check_result client "mobilenet_GPU_RPC server test completed"
kill_server_process
} }
function unet_rpc() { function unet_rpc() {
unsetproxy dir=${log_dir}rpc_model/unet_rpc/
run_gpu_env check_dir ${dir}
cd ${build_path}/python/examples/unet_for_image_seg unsetproxy
python3.6 -m paddle_serving_app.package --get_model unet >/dev/null 2>&1 cd ${build_path}/python/examples/unet_for_image_seg
tar xf unet.tar.gz data_dir=${data}unet_for_image_seg/
sed -i "s/9494/8882/g" seg_client.py link_data ${data_dir}
python3.6 -m paddle_serving_server_gpu.serve --model unet_model --gpu_ids 0 --port 8882 & sed -i "s/9494/8882/g" seg_client.py
sleep 5 python3.6 -m paddle_serving_server.serve --model unet_model --gpu_ids 0 --port 8882 > ${dir}server_log.txt 2>&1 &
nvidia-smi check_result server 5
python3.6 seg_client.py nvidia-smi
nvidia-smi python3.6 seg_client.py > ${dir}client_log.txt 2>&1
check_result $FUNCNAME nvidia-smi
kill_server_process serving check_result client "unet_GPU_RPC server test completed"
kill_server_process
} }
function resnetv2_rpc() { function resnetv2_rpc() {
unsetproxy dir=${log_dir}rpc_model/resnetv2_rpc/
run_gpu_env check_dir ${dir}
cd ${build_path}/python/examples/resnet_v2_50 unsetproxy
cp /root/.cache/dist_data/serving/resnet_v2_50/resnet_v2_50_imagenet.tar.gz ./ cd ${build_path}/python/examples/resnet_v2_50
tar xf resnet_v2_50_imagenet.tar.gz data_dir=${data}resnet_v2_50/
sed -i 's/9393/8883/g' resnet50_v2_tutorial.py link_data ${data_dir}
python3.6 -m paddle_serving_server_gpu.serve --model resnet_v2_50_imagenet_model --gpu_ids 0 --port 8883 & sed -i 's/9393/8883/g' resnet50_v2_tutorial.py
sleep 10 python3.6 -m paddle_serving_server.serve --model resnet_v2_50_imagenet_model --gpu_ids 0 --port 8883 > ${dir}server_log.txt 2>&1 &
nvidia-smi check_result server 10
python3.6 resnet50_v2_tutorial.py nvidia-smi
nvidia-smi python3.6 resnet50_v2_tutorial.py > ${dir}client_log.txt 2>&1
check_result $FUNCNAME nvidia-smi
kill_server_process serving check_result client "resnetv2_GPU_RPC server test completed"
kill_server_process
} }
function ocr_rpc() { function ocr_rpc() {
unsetproxy dir=${log_dir}rpc_model/ocr_rpc/
run_cpu_env check_dir ${dir}
cd ${build_path}/python/examples/ocr unsetproxy
cp -r /root/.cache/dist_data/serving/ocr/test_imgs ./ cd ${build_path}/python/examples/ocr
python3.6 -m paddle_serving_app.package --get_model ocr_rec >/dev/null 2>&1 data_dir=${data}ocr/
tar xf ocr_rec.tar.gz link_data ${data_dir}
sed -i 's/9292/8884/g' test_ocr_rec_client.py python3.6 -m paddle_serving_app.package --get_model ocr_rec >/dev/null 2>&1
python3.6 -m paddle_serving_server.serve --model ocr_rec_model --port 8884 & tar xf ocr_rec.tar.gz
sleep 5 sed -i 's/9292/8884/g' test_ocr_rec_client.py
python3.6 test_ocr_rec_client.py python3.6 -m paddle_serving_server.serve --model ocr_rec_model --port 8884 > ${dir}server_log.txt 2>&1 &
# check_result $FUNCNAME check_result server 5
kill_server_process serving python3.6 test_ocr_rec_client.py > ${dir}client_log.txt 2>&1
check_result client "ocr_CPU_RPC server test completed"
kill_server_process
} }
function criteo_ctr_rpc_cpu() { function criteo_ctr_rpc_cpu() {
unsetproxy dir=${log_dir}rpc_model/criteo_ctr_rpc_cpu/
run_cpu_env check_dir ${dir}
cd ${build_path}/python/examples/criteo_ctr unsetproxy
sed -i "s/9292/8885/g" test_client.py cd ${build_path}/python/examples/criteo_ctr
ln -s /root/.cache/dist_data/serving/criteo_ctr_with_cube/raw_data ./ data_dir=${data}criteo_ctr/
wget https://paddle-serving.bj.bcebos.com/criteo_ctr_example/criteo_ctr_demo_model.tar.gz >/dev/null 2>&1 link_data ${data_dir}
tar xf criteo_ctr_demo_model.tar.gz sed -i "s/9292/8885/g" test_client.py
mv models/ctr_client_conf . python3.6 -m paddle_serving_server.serve --model ctr_serving_model/ --port 8885 > ${dir}server_log.txt 2>&1 &
mv models/ctr_serving_model . check_result server 5
python3.6 -m paddle_serving_server.serve --model ctr_serving_model/ --port 8885 & python3.6 test_client.py ctr_client_conf/serving_client_conf.prototxt raw_data/part-0 > ${dir}client_log.txt 2>&1
sleep 5 check_result client "criteo_ctr_CPU_RPC server test completed"
python3.6 test_client.py ctr_client_conf/serving_client_conf.prototxt raw_data/part-0 kill_server_process
check_result $FUNCNAME
kill_server_process serving
} }
function criteo_ctr_rpc_gpu() { function criteo_ctr_rpc_gpu() {
unsetproxy dir=${log_dir}rpc_model/criteo_ctr_rpc_gpu/
run_gpu_env check_dir ${dir}
cd ${build_path}/python/examples/criteo_ctr unsetproxy
sed -i "s/8885/8886/g" test_client.py cd ${build_path}/python/examples/criteo_ctr
wget https://paddle-serving.bj.bcebos.com/criteo_ctr_example/criteo_ctr_demo_model.tar.gz >/dev/null 2>&1 data_dir=${data}criteo_ctr/
python3.6 -m paddle_serving_server_gpu.serve --model ctr_serving_model/ --port 8886 --gpu_ids 0 & link_data ${data_dir}
sleep 5 sed -i "s/8885/8886/g" test_client.py
nvidia-smi python3.6 -m paddle_serving_server.serve --model ctr_serving_model/ --port 8886 --gpu_ids 0 > ${dir}server_log.txt 2>&1 &
python3.6 test_client.py ctr_client_conf/serving_client_conf.prototxt raw_data/ check_result server 5
nvidia-smi nvidia-smi
check_result $FUNCNAME python3.6 test_client.py ctr_client_conf/serving_client_conf.prototxt raw_data/part-0 > ${dir}client_log.txt 2>&1
kill `ps -ef|grep ctr|awk '{print $2}'` nvidia-smi
kill_server_process serving check_result client "criteo_ctr_GPU_RPC server test completed"
kill_server_process
} }
function yolov4_rpc_gpu() { function yolov4_rpc_gpu() {
unsetproxy dir=${log_dir}rpc_model/yolov4_rpc_gpu/
run_gpu_env check_dir ${dir}
cd ${build_path}/python/examples/yolov4 unsetproxy
sed -i "s/9393/8887/g" test_client.py cd ${build_path}/python/examples/yolov4
cp -r /root/.cache/dist_data/serving/yolov4/yolov4.tar.gz ./ data_dir=${data}yolov4/
tar xf yolov4.tar.gz link_data ${data_dir}
python3.6 -m paddle_serving_server_gpu.serve --model yolov4_model --port 8887 --gpu_ids 0 & sed -i "s/9393/8887/g" test_client.py
nvidia-smi python3.6 -m paddle_serving_server.serve --model yolov4_model --port 8887 --gpu_ids 0 > ${dir}server_log.txt 2>&1 &
sleep 5 nvidia-smi
python3.6 test_client.py 000000570688.jpg check_result server 5
nvidia-smi python3.6 test_client.py 000000570688.jpg > ${dir}client_log.txt 2>&1
# check_result $FUNCNAME nvidia-smi
kill_server_process serving check_result client "yolov4_GPU_RPC server test completed"
kill_server_process
} }
function senta_rpc_cpu() { function senta_rpc_cpu() {
unsetproxy dir=${log_dir}rpc_model/senta_rpc_cpu/
run_gpu_env check_dir ${dir}
cd ${build_path}/python/examples/senta unsetproxy
sed -i "s/9393/8887/g" test_client.py cd ${build_path}/python/examples/senta
cp -r /data/.cache/dist_data/serving/yolov4/yolov4.tar.gz ./ data_dir=${data}senta/
tar xf yolov4.tar.gz link_data ${data_dir}
python3.6 -m paddle_serving_server_gpu.serve --model yolov4_model --port 8887 --gpu_ids 0 & sed -i "s/9393/8887/g" test_client.py
nvidia-smi python3.6 -m paddle_serving_server.serve --model yolov4_model --port 8887 --gpu_ids 0 > ${dir}server_log.txt 2>&1 &
sleep 5 nvidia-smi
python3.6 test_client.py 000000570688.jpg check_result server 5
nvidia-smi python3.6 test_client.py 000000570688.jpg > ${dir}client_log.txt 2>&1
check_result $FUNCNAME nvidia-smi
kill_server_process serving check_result client "senta_GPU_RPC server test completed"
kill_server_process
} }
function fit_a_line_http() { function fit_a_line_http() {
unsetproxy dir=${log_dir}http_model/fit_a_line_http/
run_cpu_env check_dir ${dir}
cd ${build_path}/python/examples/fit_a_line unsetproxy
sed -i "s/9292/8871/g" test_server.py cd ${build_path}/python/examples/fit_a_line
python3.6 test_server.py & sed -i "s/9393/8871/g" test_server.py
sleep 10 python3.6 test_server.py > ${dir}server_log.txt 2>&1 &
curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"x": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]}], "fetch":["price"]}' http://${host}:8871/uci/prediction check_result server 10
check_result $FUNCNAME curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"x": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332]}], "fetch":["price"]}' http://127.0.0.1:8871/uci/prediction > ${dir}client_log.txt 2>&1
kill_server_process test_server check_result client "fit_a_line_CPU_HTTP server test completed"
kill_server_process
} }
function lac_http() { function lac_http() {
unsetproxy dir=${log_dir}http_model/lac_http/
run_cpu_env check_dir ${dir}
cd ${build_path}/python/examples/lac unsetproxy
python3.6 lac_web_service.py lac_model/ lac_workdir 8872 & cd ${build_path}/python/examples/lac
sleep 10 python3.6 lac_web_service.py lac_model/ lac_workdir 8872 > ${dir}server_log.txt 2>&1 &
curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"words": "我爱北京天安门"}], "fetch":["word_seg"]}' http://${host}:8872/lac/prediction check_result server 10
check_result $FUNCNAME curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"words": "我爱北京天安门"}], "fetch":["word_seg"]}' http://127.0.0.1:8872/lac/prediction > ${dir}client_log.txt 2>&1
kill_server_process lac_web_service check_result client "lac_CPU_HTTP server test completed"
kill_server_process
} }
function cnn_http() { function cnn_http() {
unsetproxy dir=${log_dir}http_model/cnn_http/
run_cpu_env check_dir ${dir}
cd ${build_path}/python/examples/imdb unsetproxy
python3.6 text_classify_service.py imdb_cnn_model/ workdir/ 8873 imdb.vocab & cd ${build_path}/python/examples/imdb
sleep 10 python3.6 text_classify_service.py imdb_cnn_model/ workdir/ 8873 imdb.vocab > ${dir}server_log.txt 2>&1 &
curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"words": "i am very sad | 0"}], "fetch":["prediction"]}' http://${host}:8873/imdb/prediction check_result server 10
check_result $FUNCNAME curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"words": "i am very sad | 0"}], "fetch":["prediction"]}' http://127.0.0.1:8873/imdb/prediction > ${dir}client_log.txt 2>&1
kill_server_process text_classify_service check_result client "cnn_CPU_HTTP server test completed"
kill_server_process
} }
function bow_http() { function bow_http() {
unsetproxy dir=${log_dir}http_model/bow_http/
run_cpu_env check_dir ${dir}
cd ${build_path}/python/examples/imdb unsetproxy
python3.6 text_classify_service.py imdb_bow_model/ workdir/ 8874 imdb.vocab & cd ${build_path}/python/examples/imdb
sleep 10 python3.6 text_classify_service.py imdb_bow_model/ workdir/ 8874 imdb.vocab > ${dir}server_log.txt 2>&1 &
curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"words": "i am very sad | 0"}], "fetch":["prediction"]}' http://${host}:8874/imdb/prediction check_result server 10
check_result $FUNCNAME curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"words": "i am very sad | 0"}], "fetch":["prediction"]}' http://127.0.0.1:8874/imdb/prediction > ${dir}client_log.txt 2>&1
kill_server_process text_classify_service check_result client "bow_CPU_HTTP server test completed"
kill_server_process
} }
function lstm_http() { function lstm_http() {
unsetproxy dir=${log_dir}http_model/lstm_http/
run_cpu_env check_dir ${dir}
cd ${build_path}/python/examples/imdb unsetproxy
python3.6 text_classify_service.py imdb_bow_model/ workdir/ 8875 imdb.vocab & cd ${build_path}/python/examples/imdb
sleep 10 python3.6 text_classify_service.py imdb_bow_model/ workdir/ 8875 imdb.vocab > ${dir}server_log.txt 2>&1 &
curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"words": "i am very sad | 0"}], "fetch":["prediction"]}' http://${host}:8875/imdb/prediction check_result server 10
check_result $FUNCNAME curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"words": "i am very sad | 0"}], "fetch":["prediction"]}' http://127.0.0.1:8875/imdb/prediction > ${dir}client_log.txt 2>&1
kill `ps -ef|grep imdb|awk '{print $2}'` check_result client "lstm_CPU_HTTP server test completed"
kill_server_process text_classify_service kill_server_process
} }
function ResNet50_http() { function ResNet50_http() {
unsetproxy dir=${log_dir}http_model/ResNet50_http/
run_gpu_env check_dir ${dir}
cd ${build_path}/python/examples/imagenet unsetproxy
python3.6 resnet50_web_service.py ResNet50_vd_model gpu 8876 & cd ${build_path}/python/examples/imagenet
sleep 10 python3.6 resnet50_web_service.py ResNet50_vd_model gpu 8876 > ${dir}server_log.txt 2>&1 &
curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"image": "https://paddle-serving.bj.bcebos.com/imagenet-example/daisy.jpg"}], "fetch": ["score"]}' http://${host}:8876/image/prediction check_result server 10
check_result $FUNCNAME curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"image": "https://paddle-serving.bj.bcebos.com/imagenet-example/daisy.jpg"}], "fetch": ["score"]}' http://127.0.0.1:8876/image/prediction > ${dir}client_log.txt 2>&1
kill_server_process resnet50_web_service check_result client "ResNet50_GPU_HTTP server test completed"
} kill_server_process
}
function bert_http(){
unsetproxy function bert_http() {
run_gpu_env dir=${log_dir}http_model/ResNet50_http/
cd ${build_path}/python/examples/bert check_dir ${dir}
cp data-c.txt.1 data-c.txt unsetproxy
cp vocab.txt.1 vocab.txt cd ${build_path}/python/examples/bert
export CUDA_VISIBLE_DEVICES=0 cp data-c.txt.1 data-c.txt
python3.6 bert_web_service.py bert_seq128_model/ 8878 & cp vocab.txt.1 vocab.txt
sleep 5 export CUDA_VISIBLE_DEVICES=0
curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"words": "hello"}], "fetch":["pooled_output"]}' http://127.0.0.1:8878/bert/prediction python3.6 bert_web_service.py bert_seq128_model/ 8878 > ${dir}server_log.txt 2>&1 &
check_result $FUNCNAME check_result server 5
kill_server_process bert_web_service curl -H "Content-Type:application/json" -X POST -d '{"feed":[{"words": "hello"}], "fetch":["pooled_output"]}' http://127.0.0.1:8878/bert/prediction > ${dir}client_log.txt 2>&1
} check_result client "bert_GPU_HTTP server test completed"
kill_server_process
grpc_impl(){ }
unsetproxy
run_gpu_env function grpc_fit_a_line() {
cd ${build_path}/python/examples/grpc_impl_example/fit_a_line dir=${log_dir}rpc_model/grpc_fit_a_line/
sh get_data.sh >/dev/null 2>&1 check_dir ${dir}
python3.6 test_server.py uci_housing_model/ & unsetproxy
sleep 5 cd ${build_path}/python/examples/grpc_impl_example/fit_a_line
echo "sync predict" data_dir=${data}fit_a_line/
python3.6 test_sync_client.py link_data ${data_dir}
echo "async predict" python3.6 test_server.py uci_housing_model/ > ${dir}server_log.txt 2>&1 &
python3.6 test_asyn_client.py check_result server 5
echo "batch predict" echo "sync predict" > ${dir}client_log.txt 2>&1
python3.6 test_batch_client.py python3.6 test_sync_client.py >> ${dir}client_log.txt 2>&1
echo "timeout predict" check_result client "grpc_impl_example_fit_a_line_sync_CPU_gRPC server sync test completed"
python3.6 test_timeout_client.py echo "async predict" >> ${dir}client_log.txt 2>&1
# check_result $FUNCNAME python3.6 test_asyn_client.py >> ${dir}client_log.txt 2>&1
kill_server_process test_server check_result client "grpc_impl_example_fit_a_line_asyn_CPU_gRPC server asyn test completed"
} echo "batch predict" >> ${dir}client_log.txt 2>&1
python3.6 test_batch_client.py >> ${dir}client_log.txt 2>&1
function build_all_whl(){ check_result client "grpc_impl_example_fit_a_line_batch_CPU_gRPC server batch test completed"
for whl in ${build_whl_list[@]} echo "timeout predict" >> ${dir}client_log.txt 2>&1
do python3.6 test_timeout_client.py >> ${dir}client_log.txt 2>&1
echo "===========${whl} begin build===========" check_result client "grpc_impl_example_fit_a_line_timeout_CPU_gRPC server timeout test completed"
$whl kill_server_process
sleep 3 }
echo "===========${whl} build over ==========="
done function grpc_yolov4() {
} dir=${log_dir}rpc_model/grpc_yolov4/
cd ${build_path}/python/examples/grpc_impl_example/yolov4
function run_rpc_models(){ check_dir ${dir}
for model in ${rpc_model_list[@]} data_dir=${data}yolov4/
do link_data ${data_dir}
echo "===========${model} run begin===========" echo -e "${GREEN_COLOR}grpc_impl_example_yolov4_GPU_gRPC server started${RES}"
$model python3.6 -m paddle_serving_server.serve --model yolov4_model --port 9393 --gpu_ids 0 --use_multilang > ${dir}server_log.txt 2>&1 &
sleep 3 check_result server 5
echo "===========${model} run end ===========" echo -e "${GREEN_COLOR}grpc_impl_example_yolov4_GPU_gRPC client started${RES}"
done python3.6 test_client.py 000000570688.jpg > ${dir}client_log.txt 2>&1
} check_result client "grpc_yolov4_GPU_GRPC server test completed"
kill_server_process
function run_http_models(){ }
for model in ${http_model_list[@]}
do function build_all_whl() {
echo "===========${model} run begin===========" for whl in ${build_whl_list[@]}
$model do
sleep 3 echo "===========${whl} begin build==========="
echo "===========${model} run end ===========" $whl
done sleep 3
} echo "===========${whl} build over ==========="
done
function end_hook(){ }
cd ${build_path}
kill_server_process function run_rpc_models() {
kill `ps -ef|grep python|awk '{print $2}'` for model in ${rpc_model_list[@]}
sleep 5 do
echo "===========files===========" echo "===========${model} run begin==========="
ls -hlst $model
echo "=========== end ===========" sleep 3
echo "===========${model} run end ==========="
done
}
function run_http_models() {
for model in ${http_model_list[@]}
do
echo "===========${model} run begin==========="
$model
sleep 3
echo "===========${model} run end ==========="
done
}
function end_hook() {
cd ${build_path}
kill_server_process
kill `ps -ef|grep python|awk '{print $2}'`
sleep 5
echo "===========files==========="
ls -hlst
echo "=========== end ==========="
} }
function main() { function main() {
before_hook before_hook
build_all_whl build_all_whl
check check
run_env run_env
run_rpc_models unsetproxy
# run_http_models run_gpu_env
end_hook check_dir ${log_dir}rpc_model/
check_dir ${log_dir}http_model/
check_dir ${log_dir}error/
run_rpc_models
run_http_models
end_hook
if [ -f ${log_dir}error_models.txt ]; then
cat ${log_dir}error_models.txt
echo "error occurred!"
# exit 1
fi
} }
main$@ main$@
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或