Intro

Pretty much standalone infrastructure to develop/build/run docker images which are CUDA-enabled.

Pre-reqs

The main goal behind this setup is to have as minimal dependencies as possible. That way, users can start with bare-minimal host OS installation and express complexities inside their docker images.

• python
• docker
• gpu-enabled system (preferrably >= Maxwell arch)
• nvidia-docker
• latest cuda driver

Example

Currently, the only supported docker base image is from Ubuntu 16.04. If you want to build the image ‘ml-dev:9.2’ on your system:

./do -build ml-dev:9.2

Refer to the ‘-h’ option of this script to know more about its usage.

Running containers on these images

./do -run ml:dev:9.2 /bin/bash

Refer to the ‘-h’ option of this script to know more about its usage.