Docker中编译Caffe

 

基础环境信息：

OS：CentOS Linux release 7.1.1503

NV驱动：nvidia-diag-driver-local-repo-rhel7-384.66-1.0-1.x86_64

CUDA：8.0，cudnn5

Docker：Docker version 1.12.6, build 3e8e77d/1.12.6

NV-Docker：nvidia-docker-1.0.1-1

 

1. 选择BaseImage，拉取Docker镜像

docker pull nvidia/cuda:8.0-cudnn7-devel-centos7

2. 启动Docker Container

NV_GPU=0 nvidia-docker run --it --name docker.io/nvidia/cuda:8.0-cudnn5-devel-centos7

3. 确认Docker Container中GPU Device可见

nvidia-smi

4. 编译Caffe（http://caffe.berkeleyvision.org/installation.html）

• yum -y install epel-release
• yum -y install findutils make automake autoconf libtool gcc gcc-c++ cmake gdb git wget atlas-devel boost-devel gflags-devel glog-devel hdf5-devel leveldb-devel lmdb-devel opencv-devel protobuf-devel snappy-devel protobuf-compiler freetype-devel libpng-devel python-devel python-numpy python-pip python-scipy gcc-gfortran libjpeg-turbo-devel
• 下载caffe源码（https://github.com/BVLC/caffe.git，多GPU支持还需要nccl）
• pip install --upgrade pip &&  for req in $(cat python/requirements.txt) pydot; do pip --no-cache-dir install $req; done

5. Commit镜像

转载于:https://my.oschina.net/u/3869269/blog/1818883