Machine Learning Professional · 12% of the exam

Model Deployment: free practice questions

5 sample questions from our 10-question bank for this domain — answers and explanations included. These are the same scenario-based style as the real Databricks exam.

1. A continuous stream of events must be scored as they arrive, with results written out incrementally. Which deployment paradigm fits?

  • A. Streaming inference (e.g., Structured Streaming applying the model to a streaming source)✓ Correct
  • B. Nightly batch scoring only
  • C. A static one-time prediction
  • D. A SQL dashboard
Explanation

Scoring events continuously as they arrive is streaming inference, typically via Structured Streaming applying the model to a streaming DataFrame. Nightly batch (B) adds latency, a one-time prediction (C) isn't continuous, and a dashboard (D) doesn't score events.

2. An application needs sub-second predictions per request from a web front end. Which Databricks deployment should be used?

  • A. Databricks Model Serving (a real-time serving endpoint)✓ Correct
  • B. A nightly batch job
  • C. A one-time notebook run
  • D. A Delta Live Tables pipeline for reporting
Explanation

Databricks Model Serving exposes a registered model as a low-latency REST endpoint for real-time, per-request predictions. Batch (B) and a one-time notebook (C) can't meet real-time per-request needs, and DLT (D) is for ETL/reporting.

3. A canary deployment of a new model version shows elevated error rates shortly after receiving 5% of traffic. What is the appropriate action?

  • A. Route traffic back to the previous stable version (roll back) and investigate before proceeding✓ Correct
  • B. Immediately send 100% of traffic to the new version
  • C. Ignore the errors and continue the rollout
  • D. Delete the serving endpoint
Explanation

The point of a canary is to limit blast radius: on elevated errors, roll traffic back to the stable version and investigate before continuing. Sending 100% to the failing version (B) or ignoring errors (C) harms users, and deleting the endpoint (D) causes an outage.

4. A team wants to roll out a new model version to a small percentage of live traffic first, monitor it, then increase its share. Which Model Serving capability supports this?

  • A. Traffic splitting across served model versions (A/B or canary rollout)✓ Correct
  • B. Deleting the old version before deploying the new one
  • C. Only ever serving one version with no control
  • D. A batch job
Explanation

Databricks Model Serving endpoints can split traffic across multiple served model versions, enabling A/B testing or canary rollouts where a small share goes to the new version first. Deleting the old version (B) removes the fallback, single-version-only (C) offers no control, and a batch job (D) isn't real-time rollout.

5. Which statement about serving a model registered in the MLflow Model Registry is TRUE?

  • A. The serving endpoint can load a specific registered model version (or alias), so deployments are governed and traceable to a registry entry✓ Correct
  • B. Serving requires copying the model file manually to the endpoint each time
  • C. Only unregistered models can be served
  • D. Serving bypasses the registry entirely and needs no versioning
Explanation

Model Serving deploys a registered model version (or alias), keeping deployments governed and traceable to the registry. Manual file copying (B) isn't required, registered models are exactly what you serve (C is false), and governed serving relies on the registry's versioning (D is false).

5 more questions in this domain

Practice the full bank with instant grading, flashcards, and a timed mock exam.

Start practicing free