Ubuntu 源码编译安装 GPU版 TensorFlow

- 从github下载 Tensorflow 源码

$ git clone https://github.com/tensorflow/tensorflow

查看当前分支, 根据需求切换到指定分支, e.g. r1.3

$ cd tensorflow
$ git checkout Branch # where Branch is the desired branch
$ git checkout r1.3

- 安装Bazel

- 安装python依赖包

numpy, which is a numerical processing package that TensorFlow requires.

dev, which enables adding extensions to Python.

pip, which enables you to install and manage certain Python packages.

wheel, which enables you to manage Python compressed packages in the wheel (.whl) format.

python 2 安装命令如下

$ sudo apt-get install python-numpy python-dev python-pip python-wheel

python 3 安装命令如下

$ sudo apt-get install python3-numpy python3-dev python3-pip python3-wheel

-安装 cuda

cuda 下载路径 CUDA Toolkit

$ sudo dpkg -i cuda-repo-ubuntu1404-8-0-local-ga2_8.0.61-1_amd64.deb
$ sudo apt-get update
$ sudo apt-get install -y cuda

cudnn 下载路径 cudnn

注册个账号,然后根据自己的系统选择下载相应的cudnn.

$ tar zxvf cudnn-7.5-linux-x64-v5.1.tgz
$ cd cuda
$ sudo cp include/cudnn.h /usr/local/include
$ sudo cp lib64/libcudnn.* /usr/local/lib
$ sudo ln -sf /usr/local/lib/libcudnn.so.5.1.3 /usr/local/lib/libcudnn.so.5
$ sudo ln -sf /usr/local/lib/libcudnn.so.5 /usr/local/lib/libcudnn.so
$ sudo ldconfig -v

最后安装 libcupti-dev

$ sudo apt-get install libcupti-dev

-设置安装配置

根据自己python版本, cuda, cudnn版本和GPU计算力设置

$ cd tensorflow  # cd to the top-level directory created
$ ./configure

-编译 pip package

- TensorFlow compiled to use SSE, AXV, FMA

$ bazel build -c opt --copt=-march=native --copt=-mfpmath=both --config=cuda -k //tensorflow/tools/pip_package:build_pip_package

生成 .whl 文件

$ bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg

-安装 .whl

$ sudo pip install /tmp/tensorflow_pkg/tensorflow-1.3.0-py2-none-any.whl

- 问题汇总

tensorflow-*-any.whl not a supported wheel on this platform

使用 python -m pip install 代替 pip install 安装

sudo python -m pip install /tmp/tensorflow_pkg/tensorflow-1.3.0-py2-none-any.whl