Salvato in:
Dettagli Bibliografici
Autori principali: Wang, Tianze, Gui, Dongnan, Hu, Yifan, Lin, Shuhang, Zhang, Linjun
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2502.18699
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908459913445376
author Wang, Tianze
Gui, Dongnan
Hu, Yifan
Lin, Shuhang
Zhang, Linjun
author_facet Wang, Tianze
Gui, Dongnan
Hu, Yifan
Lin, Shuhang
Zhang, Linjun
contents Reinforcement Learning from Human Feedback (RLHF) has shown promise in aligning large language models (LLMs). Yet its reliance on a singular reward model often overlooks the diversity of human preferences. Recent approaches address this limitation by leveraging multi-dimensional feedback to fine-tune corresponding reward models and train LLMs using reinforcement learning. However, the process is costly and unstable, especially given the competing and heterogeneous nature of human preferences. In this paper, we propose Mixing Preference Optimization (MPO), a post-processing framework for aggregating single-objective policies as an alternative to both multi-objective RLHF (MORLHF) and MaxMin-RLHF. MPO avoids alignment from scratch. Instead, it log-linearly combines existing policies into a unified one with the weight of each policy computed via a batch stochastic mirror descent. Empirical results demonstrate that MPO achieves balanced performance across diverse preferences, outperforming or matching existing models with significantly reduced computational costs.
format Preprint
id arxiv_https___arxiv_org_abs_2502_18699
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MPO: An Efficient Post-Processing Framework for Mixing Diverse Preference Alignment
Wang, Tianze
Gui, Dongnan
Hu, Yifan
Lin, Shuhang
Zhang, Linjun
Computation and Language
Machine Learning
Methodology
Reinforcement Learning from Human Feedback (RLHF) has shown promise in aligning large language models (LLMs). Yet its reliance on a singular reward model often overlooks the diversity of human preferences. Recent approaches address this limitation by leveraging multi-dimensional feedback to fine-tune corresponding reward models and train LLMs using reinforcement learning. However, the process is costly and unstable, especially given the competing and heterogeneous nature of human preferences. In this paper, we propose Mixing Preference Optimization (MPO), a post-processing framework for aggregating single-objective policies as an alternative to both multi-objective RLHF (MORLHF) and MaxMin-RLHF. MPO avoids alignment from scratch. Instead, it log-linearly combines existing policies into a unified one with the weight of each policy computed via a batch stochastic mirror descent. Empirical results demonstrate that MPO achieves balanced performance across diverse preferences, outperforming or matching existing models with significantly reduced computational costs.
title MPO: An Efficient Post-Processing Framework for Mixing Diverse Preference Alignment
topic Computation and Language
Machine Learning
Methodology
url https://arxiv.org/abs/2502.18699