本文记录Ubuntu 16.04安装Tensorflow步骤,也包括怎么从源码编译安装Tensorflow。
要想安装Tensorflow GPU版本,你需要有一个新一点的Nvidia显卡。
Tensorflow CPU版本的安装
1
2
$sudoapt-getinstall-pippython-devpython-virtualenv# python 2.7
$sudoapt-getinstallpython3-pippython3-devpython3-virtualenv# python 3.4+
1
2
$virtualenv--system-site-packages~/tensorflow
$source~/tensorflow/bin/activate
1
2
3
4
5
6
7
8
# Ubuntu/Linux 64-bit, CPU only, Python 2.7
$exportTF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.11.0rc1-cp27-none-linux_x86_64.whl
# Ubuntu/Linux 64-bit, CPU only, Python 3.4
$exportTF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.11.0rc1-cp34-cp34m-linux_x86_64.whl
# Ubuntu/Linux 64-bit, CPU only, Python 3.5
$exportTF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.11.0rc1-cp35-cp35m-linux_x86_64.whl
安装Tensorflow:
1
2
3
4
5
# Python 2
$sudopipinstall--upgrade$TF_BINARY_URL
# Python 3
$sudopip3install--upgrade$TF_BINARY_URL
如果要升级Tensorflow,替换新版本的TF_BINARY_URL。https://www.tensorflow.org
编译安装Tensorflow(GPU支持)
安装NVidia显卡驱动,你可以在Ubuntu内置的附加驱动中安装。
1
2
$sudoadd-apt-repositoryppa:graphics-drivers/ppa
$sudoaptupdate
安装CUDA:
1
2
3
4
5
6
#下载安装:
#https://developer.nvidia.com/cuda-toolkit
$sudoshcuda_8.0.44_linux.run--override# 安装位置: /usr/local/cuda
# 默认仓库中的版本较旧
#$ sudo apt install nvidia-cuda-toolkit nvidia-cuda-dev # 安装位置: /usr
1
2
3
4
5
# 下载CudNN 5.1 for Cuda 8.0
$sudotar-xzvfcudnn-8.0-linux-x64-v5.1.tgz
$sudocpcuda/include/cudnn.h/usr/local/cuda/include
$sudocpcuda/lib64/libcudnn*/usr/local/cuda/lib64
$sudochmoda+r/usr/local/cuda/include/cudnn.h/usr/local/cuda/lib64/libcudnn*
在~/.bashrc文件中添加环境变量:
1
2
exportLD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64"
exportCUDA_HOME=/usr/local/cuda
使环境变量生效:
1
$source~/.bashrc
下载tensorflow源码:
1
2
$cd~
$gitclonehttps://github.com/tensorflow/tensorflow
安装一些编译和依赖工具:
1
$sudoapt-getinstalldefault-jdkpython-devpython3-devpython-numpypython3-numpybuild-essentialpython-pippython3-pippython-virtualenvswigpython-wheellibcurl3-dev
安装Bazel:
1
2
3
4
5
$echo"deb [arch=amd64] http://storage.googleapis.com/bazel-apt stable jdk1.8"|sudotee/etc/apt/sources.list.d/bazel.list
$curlhttps://storage.googleapis.com/bazel-apt/doc/apt-key.pub.gpg|sudoapt-keyadd-
$sudoapt-getupdate
$sudoapt-getinstallbazel
$sudoapt-getupgradebazel
配置编译选项:
1
2
$cd~/tensorflow
$./configure
需要输入Python路径,默认是 /usr/bin/python。如果你使用Python3,输入:/usr/bin/python3.5。
输入Python模块路径,默认是/usr/local/lib/python2.7/dist-packages。如果你使用Python3,输入:/usr/local/lib/python3.5/dist-packages。
输入Cuda SDK版本和Cudnn版本:8.0、5.1.5。
配置完成,输入如下信息:
1
2
INFO:Allexternaldependenciesfetchedsuccessfully.
Configurationfinished
编译tensorflow:
1
2
3
$bazelbuild-copt--config=cuda# GPU支持
# CPU支持
#$ bazel build -c opt
构建pip包:
1
$bazel-bin/tensorflow/tools/pip_package/build_pip_package/tmp/tensorflow_pkg
安装pip包:
1
2
$sudopipinstall/tmp/tensorflow_pkg/tensorflow# python2
$sudopip3install/tmp/tensorflow_pkg/tensorflow# python3
参考:
https://www.tensorflow.org/versions/r0.11/get_started/os_setup.html#download-and-setup