A note of points that get stuck when you try to use tensorflow in an environment where you do not have root privileges. The environment is CentOS 6.7, but I think it's the same everywhere. I don't use CUDA.
It will not work unless 2.7 series is included. I used pyenv. Please see here. However, at that time, you need to specify the option as shown in here. By the way, if you put it in with pip without options, you need to re-install the library. Re-inserted numpy, sympy, tensorflow.
Since I am using a compiled binary, I cannot use it unless the libc version matches. So download the rpm and replace it. It seems that there is nothing like packages.debian.org on CentOS, so pull it out directly from the repository. I will come. For example, for RIKEN, go to here.
python
$ mkdir ~/tflib
$ cd ~/tflib
$ wget http://ftp.riken.jp/Linux/centos/7/os/x86_64/Packages/glibc-2.17-78.el7.x86_64.rpm
$ rpm2cpio glibc-2.17-78.el7.x86_64.rpm|cpio -idv
$ wget http://ftp.riken.jp/Linux/centos/7/os/x86_64/Packages/libstdc++-4.8.3-9.el7.x86_64.rpm
$ rpm2cpio libstdc++-4.8.3-9.el7.x86_64.rpm|cpio -idv
I will leave it as. If it is Debian / Ubuntu, you should find it appropriately and unzip it. [^ 1] [^ 1]: Maybe there is a dependency, but it worked without anything else. For Debian / Ubuntu, you can see dependencies in packages.debian.org, so please add them as appropriate. Now that I have everything, but the version of ld-linux.so doesn't match, so I'm afraid.
python
$ LD_LIBRARY_PATH=~/tflib/lib64/:~/tflib/usr/lib64/ ~/tflib/lib64/ld-2.17.so ~/.pyenv/versions/2.7.10/bin/python tutorial1.py
You can finally start it. I'm troublesome
~/.bashrc
alias tfpy='LD_LIBRARY_PATH=~/tflib/lib64/:~/tflib/usr/lib64/ ~/tflib/lib64/ld-2.17.so ~/.pyenv/versions/2.7.10/bin/python'
I left it as.
python
$ tfpy tutorial1.py
Actually, I feel that maybe numpy or Python itself will have to be recompiled with the new libc. .. There are no errors so far.
Recommended Posts