DevOps and MLOps DY0-001 Practice Questions (Page 3)

Questions 11–15

1. 11expert · hard

   A team has deployed a real-time fraud-detection model. The model is updated frequently via continuous deployment. The team wants to ensure that a new model version does not cause a significant increase in false positives, which would lead to legitimate transactions being blocked. Which monitoring approach is most effective for this specific concern?

   • AMonitor the model's confidence score distribution.
   • BMonitor the CPU utilization of the inference servers.
   • CMonitor the model's prediction latency to ensure it is within acceptable bounds.
   • DTrack the false positive rate (FPR) in production and compare it to the FPR from the previous model version.
   Reveal Solution Discussion Select an answer first
2. 12expert · hard

   A team is deploying a model that requires GPU acceleration for inference. They are using a Kubernetes cluster with a mix of CPU-only and GPU-enabled nodes. The model must be deployed as a set of replicas to handle load. How should the team configure the deployment to ensure the model runs only on GPU nodes?

   • AUse a resource request for CPU and memory in the pod spec.
   • BUse a node selector in the pod spec to target nodes with the 'gpu=true' label.
   • CUse a liveness probe to check if the GPU is available.
   • DUse a ConfigMap to specify the GPU requirements.
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   What is the primary role of a container orchestration platform like Kubernetes?

   • ATo provide a graphical interface for writing application code
   • BTo automate the deployment, scaling, and management of containerized applications
   • CTo compile container images from source code
   • DTo replace the need for virtual machines in a data center
   Reveal Solution Discussion Select an answer first
5. 14foundation · easy

   In the context of continuous deployment, what is the role of automated tests in the pipeline?

   • AThey are used only for performance testing, not functional correctness
   • BThey are replaced by manual QA testing before each release
   • CThey serve as the gate that determines whether a change is safe to deploy to production
   • DThey are optional and only run if the developer remembers to trigger them
   Reveal Solution Discussion Select an answer first
6. 15foundation · easy

   What is the purpose of model versioning in a production deployment?

   • ATo allow tracking of changes and the ability to roll back to a previous model if needed
   • BTo reduce the storage space required for the model artifacts
   • CTo increase the model's accuracy by storing multiple copies of the same model
   • DTo ensure that the model is retrained automatically every time new data arrives
   Reveal Solution Discussion Select an answer first