Alauda AI

Run an Experiment

Label the namespace for Katib's metrics collector webhook, then create a minimal Experiment:

kubectl label namespace <namespace> katib.kubeflow.org/metrics-collector-injection=enabled --overwrite
kubectl apply -f experiment.yaml
kubectl get experiments,suggestions,trials -n <namespace>

The Experiment controller creates a Trial and a training Job. Check workload creation and logs with:

kubectl get jobs,pods -n <namespace> -l katib.kubeflow.org/experiment=<experiment-name>
kubectl logs -n <namespace> job/<trial-name>

Use the Katib API or UI to inspect metrics and the best trial. Delete the Experiment when finished to clean up generated Trials and Jobs.