SnowPro Specialty: Gen AI free question 6 / 10 · Domain 3: Snowflake Gen AI Governance
A run was invoked and its metrics were computed. A week later the team wants a metric it did not request the first time, for the same outputs. What does the reference say?
- Trigger a new computation with the added metric on that run
- Copy the traces to a new table and compute the metric there
- Create a new run, since metrics are fixed when a run starts
- Delete the run and invoke it again with the full metric list
Show the answer and explanation
Answer: A
Why this is correct
In the run lifecycle under "Runs", invocation reads inputs from the dataset, invokes the application for each input, and stores the traces; computation then computes the metrics you specify. The reference says you can trigger multiple computations and add new metrics later for an existing run. The new metric is computed on the outputs already stored, without invoking the application again.
Why the other options are wrong
"Create a new run, since metrics are fixed when a run starts" … Metrics are not fixed at the start. Computation is a separate step after invocation, and it can be repeated.
"Delete the run and invoke it again with the full metric list" … Deleting and invoking again calls the application a second time and may produce different outputs. The reference allows adding metrics to the existing run.
"Copy the traces to a new table and compute the metric there" … The reference does not describe copying traces. Computation works on the run itself.
Four steps of a run
Creation, invocation, computation, and visualization. Keeping invocation separate from computation is what lets metrics be added without calling the application again.
Go to the SnowPro Specialty: Gen AI practice app (10 free questions; purchase unlocks all questions)