Optimizing Preference Alignment with Differentiable NDCG Ranking