Distributing training jobs

Earlier, you trained the fraud detection model directly in a notebook and then in a pipeline. You can also distribute the training of a machine learning model across many CPUs.

Distributing training is not necessary for a simple model. However, by applying it to the example fraud model, you learn how to train more complex models that require more compute power.

NOTE: Distributed training in OpenShift AI uses the Red Hat build of Kueue for admission and scheduling. Before you run the Ray or Training Operator examples in this tutorial, complete the setup tasks in Setting up Kueue resources.

You can try one or both of the following options:

The Ray distributed computing framework, as described in Distributing training jobs with Ray.
The Training Operator, as described in Distributing training jobs with the Training Operator.