编译TensorFlow-serving GPU版本

编译TensorFlow-serving GPU版本

TensorFlow Serving 介绍

编译GPU版本

1. 下载源码
   git clone https://github.com/tensorflow/serving.git
2. 创建镜像
   nvidia-docker build --pull -t $USER/tensorflow-serving-devel -f serving/tensorflow_serving/tools/docker/Dockerfile.devel-gpu .
3. 创建并进入容器
4. 下载TensorFlow源码
   git clone https://github.com/tensorflow/tensorflow
   cd tensorflow
   git checkout v1.8.0 #检出v1.8.0
   ./configure
5. 修改代码
   vi tensorflow_serving/util/net_http/client/testing/BUILD
   vi tensorflow_serving/util/net_http/server/testing/BUILD
   添加cc_binary( ... linkopts = ["-lm"], )
   https://github.com/tensorflow/serving/issues/971
6. 编译
   cd ..
   bazel build -c opt --config=cuda tensorflow_serving/...