Skip to main content
These components cover experiment tracking, notebooks, training, model serving, and data labeling. MLflow and Label Studio can also run as separate project instances.

MLflow

MLflow tracks experiments and stores models in a model registry.
  • URL: https://<base-hostname>/mlflow/

Kubeflow

Kubeflow provides notebooks, pipelines, and hyperparameter tuning. Each platform user gets their own Kubeflow namespace, prefixed with cake- by default.
  • Notebooks: https://<base-hostname>/jupyter/
  • Pipelines: https://<base-hostname>/pipeline/
  • Katib (hyperparameter tuning): https://<base-hostname>/katib/

KServe

KServe serves models as scalable inference endpoints in the cluster.

Ray

Ray runs distributed training and serving workloads.
  • Ray dashboard: https://<base-hostname>/ray/
  • Ray Serve: https://<base-hostname>/rayserve/

SkyPilot

SkyPilot launches and manages compute jobs on the cluster.
  • URL: https://skypilot.<base-hostname>

Label Studio

Label Studio labels data for training and evaluation.
  • URL: https://<base-hostname>/labelstudio/

ClearML

ClearML tracks experiments and orchestrates ML jobs.
  • URL: https://clearml.<base-hostname>