Agents, Tool Use & Product Post-Training · Problem 5 of 7
Implement a preference-dataset builder from accepted/rejected suggestions, including a position-bias correction.
Implement the function/class skeleton in the editor. Any correct approach is accepted.
def build_preference_dataset(logs, exam_propensity, exam_threshold=0.1):
raise NotImplementedErrorReady when you are
Submit your solution and a structured review appears here — verdict, score, and concrete feedback. Any correct approach passes.
Implement a preference-dataset builder from accepted/rejected suggestions, including a position-bias correction.
Implement the function/class skeleton in the editor. Any correct approach is accepted.