βοΈ Kubeflow Flashcards
π What is Kubeflow?
Kubeflow is an open-source platform designed to run machine learning workflows on Kubernetes seamlessly.
π§± What is a Kubeflow Pipeline?
Kubeflow Pipelines are reusable ML workflows defined as a sequence of steps that are containerized and run on Kubernetes.
π§ͺ What components are in Kubeflow?
It includes Pipelines, Notebooks, Katib (AutoML), TFJob, PyTorchJob, MPIJob, and more.
π What is Katib?
Katib is Kubeflow’s component for hyperparameter tuning and automated model search (AutoML).
π» Does Kubeflow support Jupyter Notebooks?
Yes, Kubeflow provides Notebook Servers for running Jupyter Notebooks inside the Kubernetes cluster.
βοΈ Kubeflow vs MLflow?
Kubeflow is Kubernetes-native and full stack. MLflow is language-agnostic and simpler to set up for experiment tracking.
π Is Kubeflow cloud-agnostic?
Yes. Kubeflow runs on any Kubernetes clusterβon-premises or on cloud (GKE, EKS, AKS, etc.).
π Does Kubeflow support RBAC?
Yes, Kubeflow supports multi-user environments with namespaces and Role-Based Access Control (RBAC).
π Monitoring in Kubeflow?
Monitoring is achieved via Prometheus, Grafana, and integration with tools like Kiali and Istio for service mesh observability.
π οΈ Can I use Kubeflow for training & serving?
Yes, it supports training via TFJob, MPIJob, etc., and serving via KFServing (now KServe).