Customizing Clusterone

Last updated 6 months ago

Clusterone runs NVIDIA NGC containers by default, but it can run any user-define container.


  • the container must be located on a public registry (contact for info regarding private registries)

Adding a container to Clusterone

The administration dashboard enables to add a container hosted on a public registry, or a container hosted on a private registry )provided its secret has already been stored on Clusterone). This functionality is not yet exposed to administrators, please contact your account manager.

To add an ML Framework, navigate to Jobs / Docker images and click Add docker image.

Input the following

  • Framework ID: a unique, sluggified identifier for the container that will be visible to the users. Exclude the version, it should be provided as Framework Version Number.

  • Framework name: this identifier will be used for selecting the Kubernetes template, which defines environment variables, requirements, network configuration, etc.. Currently tensorflow, pythorch, openmpi and xgboost frameworks are supported with custom templates - all other framework use a generic template.

  • Framework Version Number: the version of the framework, for example the version of TensorFlow 1.8 is 1.8

  • Path to the docker image for this framework: the URL of the docker image, eg.

  • Name of the Kubernetes Secret: (private container only) name of a Kubernetes secret - keep in mind that this secret has to be present in the Kubernetes cluster.