Colosseum Dockerfiles for GPUs Utilization

This public repository contains a set of Dockerfiles that can be used to create basic Docker images ready to use for the Colosseum environment.

Each Dockerfile contains the minimum requirements to properly work in Colosseum together with a few best practices:

• Timezone setting to America/New York;
• OpenSSH installation;
• Enabling ssh with the root user (default password is ChangeMe);
• Deletion of apt libraries to reduce final image size;
• Various package installations based on the type of final image.

How to use

In order to build a Docker image from a Dockerfile, at least Docker needs to be installed and running on your computer.

You can build a Docker image by running the docker build command with the following parameters:

• --platform linux/amd64 to specify the correct image architecture (needed if building the image with a MAC M1/M2 for example);
• -t <image-name>:<image-tag> to set a name and a tag for the image;
• -f <dockerfile-name> to input the Dockerfile to build the image from;
• <build-context-path> to set the path for the build context, i.e. the set of files and directories that are used to build a Docker image.

As an example, a whole Docker build command is the following:

docker build --platform linux/amd64 -t example:1.0 -f Dockerfile_base .

More information on Dockerfiles can be found in [4], while [5] provides best practices on how to properly write Dockerfiles.

List of Dockerfiles

Title	Dockerfile	Notes
Base	Dockerfile_base Dockerfile_base_squashed	Base Dockerfiles (non-squashed and squashed for a lighter and compressed layered image) with just the minimum requirements for working in the Colosseum environment.
PyTorch	Dockerfile_pytorch	Base image with PyTorch installed.
TensorFlow	Dockerfile_tensorflow	Base image with TensorFlow installed.

References

[1] Colosseum Public Website

[2] Colosseum Experiments Website

[3] Colosseum FreshDesk Page

[4] Dockerfile reference

[5] Best practices for writing Dockerfiles