TensorFlow

安装

系统 ubuntu 16.04 amd64

安装 pip3

这里选择 python 3 作为开发环境

sudo apt-get install python-dev build-essential libcupti-dev
sudo apt-get install -y python3-pip
sudo pip3 install --upgrade pip
sudo mv /usr/bin/pip3 /usr/bin/pip3.bak
sudo ln -s /usr/local/bin/pip3 /usr/bin/

下面安装过程中，如果有 nvidia GPU 显卡，安装 gpu 版本，否则安装 cpu 版本。

安装CPU版本

sudo pip3 install tensorflow

安装GPU版本

安装 tensorflow gpu 版本

sudo pip3 install tensorflow-gpu
sudo pip3 install gunicorn flask image

加入 显卡 ppa 源

sudo apt-get install software-properties-common
sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt update

安装 Nvidia Toolkit 8.0

来源： https://developer.nvidia.com/cuda-toolkit

wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-repo-ubuntu1604_8.0.61-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1604_8.0.61-1_amd64.deb
sudo apt-get update
sudo apt-get install cuda

安装 CudNN

来源： https://developer.nvidia.com/cudnn 需要注册和下载

下载： https://developer.nvidia.com/compute/machine-learning/cudnn/secure/v5.1/prod_20161129/8.0/libcudnn5_5.1.10-1+cuda8.0_amd64-deb

sudo dpkg -i libcudnn5_5.1.10-1+cuda8.0_amd64-deb

设置环境变量

echo 'export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64"' >> .bashrc
echo 'export CUDA_HOME=/usr/local/cuda' >> .bashrc
source ~/.bashrc

查看显卡信息

nvidia-smi

参考链接

http://wiki.jikexueyuan.com/project/tensorflow-zh/tutorials/mnist_beginners.html

http://blog.topspeedsnail.com/archives/10858

http://blog.topspeedsnail.com/archives/10897