ImportError: /usr/lib64/libm.so.6: version 'GLIBC_2.23' not found

[VisionTheta]

Thanks for your great work on providing us prebuild binaries against cuda 8.0 and cudnn 6 etc.

However, when I install from your wheel file from release page(centos 7. py35+tf1.8+cuda8+cudnn6), it encountered the following error:

ImportError: /usr/lib64/libm.so.6: version 'GLIBC_2.23' not found

The reason I guess is that you used ubuntu 16.04 as your docker base image, which default glibc version is 2.23, but on centos 7, the default glibc version was 2.17

Can you help me to solve this problem? upgrading glibc to 2.23 maybe dangerous I think.

[ghostplant]

Thanks, I will look into this issue and update the prebuild binaries if possible.

Some alternative suggestions:

1. Run your tensorflow in docker container based on Ubuntu 16.04, this can be managed by nvidia-docker;
2. The Dockerfile (https://github.com/ghostplant/tensorflow-cuda8-optimized/blob/master/Dockerfile.tf18-py35-cuda8-cudnn6021) already shows how to build a native binary, you can reference each step and build it directly on centos7.