AI-300: Operationalizing Machine Learning and Generative AI SolutionsDemo

Question

You are examining a dataset intended for an advanced fine-tuning job in Microsoft Foundry.

The fine-tuning job uses preference comparison data. Review the following dataset excerpt.

Question Image

For each of the following statements, select Yes if the statement is true. Otherwise, select No.

Yes or No

Statements	Yes	No
The first record provides a usable preference signal for preference-based fine-tuning.
The second record provides a usable preference signal for preference-based fine-tuning.
Each preferred and non-preferred output in the dataset contains a valid assistant message as required for direct preference optimization.
Both records must contain different preferred and non-preferred outputs to produce usable training signals.

Explanation

Direct preference optimization uses paired preferred and non-preferred completions. Identical completions provide no preference contrast, whereas distinct completions do. Each displayed completion is an assistant-role message, satisfying the assistant-message requirement for both output fields.

Learn more

Direct preference optimization - Microsoft Foundry

AI-300: Operationalizing Machine Learning and Generative AI SolutionsDemo

AstroTutor

Notes

Readiness

QuestionQ32

Community Discussion