Question 10
Domain 4: NVIDIA Platform Implementation and Production OperationsWhat automated retraining and versioning strategy would best meet these requirements?
Correct answer: B
Explanation
An MLOps pipeline fits because it automates the full model lifecycle: retraining when the knowledge base changes, evaluation against a benchmark suite, and deployment with canary testing. A versioned model registry and rollback capability provide traceability and safe recovery, which are standard requirements for controlled model updates.
Why each option is right or wrong
A. Manually retrain and deploy models weekly, keeping the previous version as backup.
B. Implement MLOps pipeline with: automated retraining triggered by knowledge base updates, automated evaluation against benchmark suite, versioned model registry, automated deployment with canary testing, and rollback capability.
The requirement set points to a controlled model-lifecycle workflow, which is exactly what an MLOps pipeline formalizes: retraining can be event-driven off knowledge-base updates, then gated by automated validation against a benchmark suite before release. A versioned model registry is the mechanism that preserves model lineage and reproducibility, while canary deployment and rollback are standard release controls for limiting blast radius and restoring the prior approved version if post-deployment metrics degrade.
C. Continuously retrain the model every hour with latest data and deploy immediately.
D. Use the same model indefinitely and only update the knowledge base without retraining.