KServe and Model Serving on Kubernetes Quiz

Answers and explanations appear as you take the quiz.

1. What was KServe's previous name before it became a standalone project?
   • KFServing
   • KubeML
   • ModelMesh
   • Knative Models
2. What is the primary Kubernetes custom resource you define to deploy a model with KServe?
   • ModelDeployment
   • InferenceService
   • PredictPod
   • ServingChart
3. Which capability lets idle KServe models consume zero compute until a request arrives?
   • Canary rollout
   • Sharding
   • Scale-to-zero
   • Batch inference
4. Which underlying project provides KServe's request-driven autoscaling?
   • Prometheus
   • Helm
   • Argo
   • Knative
5. How does KServe typically get model artifacts into a serving pod at startup?
   • Via a storage initializer that pulls from object storage
   • By baking models into the container image only
   • By emailing the model file
   • Through a manual kubectl copy
6. What standard does KServe use so clients can call different models the same way?
   • GraphQL Federation
   • The Open Inference Protocol
   • SOAP
   • JDBC
7. In KServe’s InferenceService pipeline, which component can apply pre- or post-processing around the predictor?
   • The storage initializer
   • The Knative autoscaler
   • The transformer
   • The Kubernetes scheduler
8. What is a typical trade-off of scaling a model deployment down to zero replicas?
   • The model becomes permanently unavailable
   • Every request avoids model loading
   • The Kubernetes API is replaced
   • Idle compute is reduced, but the next request may incur a cold-start delay