Statistical Rejection Sampling Improves Preference Optimization
Auteurs : Tianqi Liu, Yao Zhao, Rishabh Joshi, Misha Khalman, Mohammad Saleh, Peter J. Liu, Jialu Liu
Résumé : Improving the alignment of language models with human preferences remains an active research challenge. Previous approaches have primarily utilized Reinforcement Learning from Human Feedback (RLHF) via online RL methods such as Proximal Policy Optimization (PPO). Recently, offline methods such as Sequence Likelihood Calibration (SLiC) and Direct Preference Optimization (DPO) have emerged as attractive alternatives, offering improvements in stability and scalability while maintaining competitive performance. SLiC refines its loss function using sequence pairs sampled from a supervised fine-tuned (SFT) policy, while DPO directly optimizes language models based on preference data, foregoing the need for a separate reward model. However, the maximum likelihood estimator (MLE) of the target optimal policy requires labeled preference pairs sampled from that policy. DPO's lack of a reward model constrains its ability to sample preference pairs from the optimal policy, and SLiC is restricted to sampling preference pairs only from the SFT policy. To address these limitations, we introduce a novel approach called Statistical Rejection Sampling Optimization (RSO) that aims to source preference data from the target optimal policy using rejection sampling, enabling a more accurate estimation of the optimal policy. We also propose a unified framework that enhances the loss functions used in both SLiC and DPO from a preference modeling standpoint. Through extensive experiments across three diverse tasks, we demonstrate that RSO consistently outperforms both SLiC and DPO on evaluations from both Large Language Model (LLM) and human raters.
Explorez l'arbre d'article
Cliquez sur les nœuds de l'arborescence pour être redirigé vers un article donné et accéder à leurs résumés et assistant virtuel
Recherchez des articles similaires (en version bêta)
En cliquant sur le bouton ci-dessus, notre algorithme analysera tous les articles de notre base de données pour trouver le plus proche en fonction du contenu des articles complets et pas seulement des métadonnées. Veuillez noter que cela ne fonctionne que pour les articles pour lesquels nous avons généré des résumés et que vous pouvez le réexécuter de temps en temps pour obtenir un résultat plus précis pendant que notre base de données s'agrandit.