AI-300 Practice Question 55: Pipeline Step Caching
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 6 · question 5 of 10
A pipeline's first step is a command component that calls a REST API and writes the latest exchange rates to its output. The component's inputs, code and environment never change. After the first run, later runs finish quickly but keep using the same exchange rates, because the step is reused from the earlier run. The team still wants reuse to work for the other steps. What should they do?
Show answer and explanation
Answer:
By default a component is treated as deterministic: if its inputs, code and environment are unchanged, Azure Machine Learning reuses the earlier run's result instead of running it again. A component that reads changing external data is not deterministic. Setting is_deterministic: false on that one component forces it to run every time, while other steps keep their reuse (option C).
- A is wrong because reuse depends on the component, its inputs and its environment. It is not tied to which compute cluster the job runs on. - B is wrong because registering an output does not affect how the next run decides to reuse the step. The step would still be skipped. - D is wrong because running the pipeline more often only starts more runs that reuse the same cached result.
Reference: CLI (v2) command component YAML schema