MLOps Vocabulary for People Who Already Do It

1. What does a model registry in MLOps primarily serve to track?

Model deployment configurations Data drift measurements Cost and latency metrics Human judgment evaluations

2. What is meant by 'canary deployment' in the context of MLOps?

A method for live evaluations of AI outputs Serving new version to a small user base first Logging model performance metrics Rolling back to previous model versions

3. How should prompts in LLM systems be treated according to modern LLMOps?

As static artifacts without versioning Similar to schema migrations with versioning Infrequently changed without reviews As final outputs without revisions

4. What is the purpose of implementing a 'shadow deployment'?

To completely replace old models To run new logic silently alongside existing ones for comparison To enhance model registries with new features To avoid any model-related costs

Continue: The Eligibility-Checker Design Round → Review this lesson Retake quiz