AI-300 Certification Practice Question #25

Question

You are reviewing a dataset that will be used for an advanced fine-tuning job in Microsoft Foundry.
The fine-tuning job uses preference comparison data.
You review the following dataset excerpt.

For each of the following statements, select Yes if the statement is true. Otherwise, select No.
NOTE: Each correct selection is worth one point.

Accepted Answer

DPO trains on the **contrast** between a preferred and a non-preferred completion. In Record 1 the two completions are word-for-word identical, so there's no preference gradient — it provides **no** usable signal (S1 = No). Record 2's completions differ ("seven years" vs "five years"), giving a valid signal (S2 = Yes). Both records' outputs are well-formed `assistant` messages, satisfying the DPO format requirement (S3 = Yes). And it is generally true that DPO needs *differing* preferred/non-preferred outputs to learn (S4 = Yes). This matches rveney and the documented DPO format.

More AI-300 practice questions