AI-300 Practice Question 27: Model Version Upgrade Policy
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 3 · question 7 of 10
A regulated production copilot runs on a Standard deployment of a chat model. Compliance requires that the model version behind it never change without a validation run first, so the team pins a specific version. Operations also insists the copilot must never simply stop working because the pinned version reached its retirement date. Which version upgrade option should you set on the deployment?
Show answer and explanation
Answer:
Microsoft documents three deployment upgrade options. OnceCurrentVersionExpired means "once the retirement date is reached the model deployment automatically upgrades to the current default version", so the pinned version stays put through its supported life, giving the team the validation window compliance demands, and the copilot still has a working model after retirement (option C). This is the balance between the two requirements in the stem.
- A is wrong because with OnceNewDefaultVersionAvailable the deployment "automatically upgrades to the default version within two weeks of that designation change being made". Behavior can change on Microsoft's schedule rather than after the team's validation run, which is what compliance forbids. Microsoft recommends this setting for early testing, not for a regulated production workload. - B is wrong because preview versions update automatically as new previews are released, which is even less stable than following the default, the opposite of a pinned, validated version. - D is wrong because NoAutoUpgrade means "the model deployment never automatically upgrades. Once the retirement date is reached the model deployment stops working." It satisfies compliance and breaks the operations requirement outright.
Reference: Working with Azure OpenAI models in Microsoft Foundry