Skip to content

Workloads Overview


Run:ai schedules Workloads. Run:ai workloads are comprised of:

  • The Kubernetes object (Job, Deployment, etc) which is used to launch the container, inside which the data science code runs.
  • A set of additional resources that are required to run the Workload. Examples: a service entry point that allows access to the Job, a persistent volume claim to access data on the network, and more.

All of these components are created together and deleted together when the Workload ends.

Run:ai currently supports the following Workloads types:

Workload Type Kubernetes Name Description
Interactive InteractiveWorkload Submit an interactive workload
Training TrainingWorkload Submit a training workload
Distributed Training DistributedWorkload Submit a distributed training workload using TensorFlow, PyTorch or MPI
Inference InferenceWorkload Submit an inference workload


A Workload will typically have a list of values (sometimes called flags), such as name, image, and resources. A full list of values is available in the runai-submit Command-line reference.

How to Submit

A Workload can be submitted via various channels:

Workload Policies

As an administrator, you can set Policies on Workloads. Policies allow administrators to impose restrictions and set default values for Researcher Workloads. For more information see Workload Policies.