Cent OS 7 安装TensorFlow

安装相关依赖

# yum install wget

# yum -y install gcc

# yum -y install gcc-c++ 

# yum install glibc-headers

# yum install zlib-devel bzip2-devel openssl-devel ncurses-devel sqlite-devel readline-devel tk-devel make

可以选装gcc 8.3.0，也可以用上边装的默认版本，下载安装包
https://mirrors.ustc.edu.cn/gnu/gcc/gcc-8.3.0/gcc-8.3.0.tar.xz

# tar xvJf gcc-8.3.0.tar.xz

# cd ~/gcc-8.3.0

# yum search bzip2

# yum install bzip2.x86_64

# ./contrib/download_prerequisites

# mkdir build-gcc-8.3.0

# cd build-gcc-8.3.0

# ../configure --enable-checking=release --enable-languages=c,c++ --disable-multilib

# make -j4

# make install

 

安装Python3

访问Python官网 https://www.python.org/downloads/source/

下载安装包

https://www.python.org/ftp/python/3.7.4/Python-3.7.4.tar.xz

# tar -xvJf Python-3.7.4.tar.xz

# cd Python-3.7.4

# ./configure

# make && make install


若报 _ctypes 有关错误

# yum install libffi-devel -y

# make && make install

若报ssl module安装失败，用libressl替代openssl

下载最新稳定版本 2.7.4

https://ftp.openbsd.org/pub/OpenBSD/LibreSSL/libressl-2.7.4.tar.gz

# tar -zxvf libressl-2.7.4.tar.gz

# cd libressl-2.7.4
# ./configure
# make
# sudo make install
# vim /etc/ld.so.conf.d/local.conf 
添加如下内容：
    /usr/local/lib
    
即将 /usr/local/lib 目录加入到模块加载目录。

重新加载共享模块：
# ldconfig -v
# openssl version
    LibreSSL 2.7.4

回到 Python-3.7.4 目录，编辑安装文件 
# vim /Modules/Setup

删除有关 ssl 编译代码的注释，共 4 行，并修改 SSL 目录为 SSL=/usr/local, 如下所示:

# Socket module helper for SSL support; you must comment out the other
# socket line above, and possibly edit the SSL variable:
SSL=/usr/local
_ssl _ssl.c \
        -DUSE_SSL -I$(SSL)/include -I$(SSL)/include/openssl \
        -L$(SSL)/lib -lssl -lcrypto

重新编译
# ./configure
# make
# make install

链接新版本Python
# unlink /usr/bin/python

# ln -s /usr/local/bin/python3 /usr/bin/python


修改yum，否则不可使用

# vim /usr/bin/yum

首行改为 #!/usr/bin/python2.7

# vim /usr/libexec/urlgrabber-ext-down
首行改为 #!/usr/bin/python2.7

Python安装完毕

 

开始安装TensorFlow

升级pip

# pip3 install --upgrade pip

升级setuptools

# pip3 uninstall setuptools

# pip3 install setuptools==41.0.0

安装TensorFlow CPU版本 
# pip3 install tensorflow==2.0.0-beta0

或者TensorFlow GPU版本
# pip install tensorflow-gpu==2.0.0-beta0

升级libstdc.so.6

下载 libstdc++.so.6.0.21，记住一定要是这个版本，高一点低一点tensorflow都不能运行

备份旧的链接
# mv /lib64/libstdc++.so.6 /lib64/libstdc++.so.6.bak

# cp libstdc++.so.6.0.21 /lib64/

链接新的libstdc++

# ln -s /lib64/libstdc++.so.6.0.21 /lib64/libstdc++.so.6

去 https://download.csdn.net/download/qq_33160790/10241415 下载glibc-2.23.tar.bz2

# tar -xvf glibc-2.23.tar.bz2

# cd glibc-2.23

# mkdir build-glibc-2.23

# cd build-glibc-2.23

# ../configure --disable-sanity-checks --disable-werror --prefix /usr/

# make 
# make install

如果发生如下错误
gawk: error while loading shared libraries: /lib64/libm.so.6: invalid ELF header

make[2]: *** [/disk1/software/gcc/glibc-2.23/build/math/stubs] Error 127

make[2]: Leaving directory `/disk1/software/gcc/glibc-2.23/math'

make[1]: *** [math/subdir_install] Error 2

make[1]: Leaving directory `/disk1/software/gcc/glibc-2.23'

# unlink /lib64/libm.so.6
# ln -s /lib64/libm-2.23.so /lib64/libm.so.6
# make install