AI-300 Practice Question 2: Cross Workspace Registry
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 1 · question 2 of 10
Your organization trains models in a development Azure Machine Learning workspace and serves them from a production workspace that sits in a different Azure subscription and region. When a production model shows accuracy problems, auditors must be able to trace the deployed model back to the training job, its code, and its environment. What should you use to promote models between the two workspaces?
Show answer and explanation
Answer:
An Azure Machine Learning registry decouples assets from any single workspace and hosts them centrally, so every workspace in the organization can use them, across subscriptions and regions. Registries store models, environments, components, and data assets, which is exactly the promotion path Microsoft documents: develop iteratively in the dev workspace, publish the candidate to a registry, then deploy from the registry to endpoints in other workspaces (option A). Because the model is promoted as a registered asset, the lineage back to the training job is preserved.
- B is wrong because a manual download and re-register breaks lineage. The new registration in production is a fresh artifact with no link to the originating job, metrics, code, or data. - C is wrong because a shared datastore moves *files*, not assets. There is no versioning, no environment, and no job lineage attached to a blob, and cross-subscription storage access still has to be solved separately. - D is wrong because it doesn't promote anything. Production would be calling into the development workspace, which defeats the isolation between environments that motivated the separate workspaces in the first place.
Reference: Machine Learning registries - Azure Machine Learning