Installation

This guide describes how to install Kubeflow Trainer control plane on a Kubernetes cluster.

You can skip these steps if the Kubeflow platform is already deployed using manifests or package distributions, as it includes Kubeflow Trainer by default.

Prerequisites

These are the minimal requirements to install Kubeflow Trainer control plane:

• Kubernetes >= 1.28
• kubectl >= 1.28

kind create cluster # or minikube start

Installing the Kubeflow Trainer Controller Manager

Run the following command to deploy the Kubeflow Trainer controller manager:

kubectl apply --server-side -k "https://github.com/kubeflow/trainer.git/manifests/overlays/manager?ref=master"

Ensure that the JobSet and Trainer controller manager pods are running:

$ kubectl get pods -n kubeflow-system

NAME                                                  READY   STATUS    RESTARTS   AGE
jobset-controller-manager-54968bd57b-88dk4            2/2     Running   0          65s
kubeflow-trainer-controller-manager-cc6468559-dblnw   1/1     Running   0          65s

Installing the Kubeflow Training Runtimes

Run the following command to deploy the Kubeflow Training Runtimes:

kubectl apply --server-side -k "https://github.com/kubeflow/trainer.git/manifests/overlays/runtimes?ref=master"

Next Steps

• How to migrate from Kubeflow Training Operator V1.