container

Scripts to help instantiate docker containers with GPU acceleration and python deep learning frameworks.

1. Install Docker

Run:

bash setup/dockerinstall.sh 

Give to docker sudo permissions:

bash setup/dockersudo.sh

To get docker with GPU support run (nvidia-docker2):

bash setup/nvidiadocker.sh

For any issue regarding nvidia-docker please refer to the official page.

2. Build and run containers

To get custom container running on your home directory:

bash build.sh -n ContainerNameYouPrefer -f DockerfileYouDefined -i DeepLearningFrameworkImageName -r RequirementsFileName
bash run.sh -n ContainerNameYouPrefer 

Please notice that RequirementsFileName must be a path in the current docker context (current build folder - cd ./).

Examples

To get container with latest Tensorflow and GPU support running on your home directory:

bash main.sh 

Use different frameworks version

Visit docker hub, search for the framework and select the tag corresponding the version you need (e.g. visit tensorflow of pytorch). Then replace the name in the DeepLearningFrameworkImageName, e.g. if you want to pass to 2.8.0-gpu-jupyter version, change this:

bash build.sh -i tensorflow/tensorflow:latest-gpu-jupyter 

into this:

bash build.sh -i tensorflow/tensorflow:2.8.0-gpu-jupyter 

How to use in projects

You can define a requirements.txt file in YourProjectFolder and run:

cd YourProjectFolder/
bash container/main.sh # this will automatically search for requirements.txt file in the current docker context (YourProjectFolder)