@inproceedings{zhou-etal-2024-beyond,
title = "Beyond One-Preference-Fits-All Alignment: Multi-Objective Direct Preference Optimization",
author = "Zhou, Zhanhui and
Liu, Jie and
Shao, Jing and
Yue, Xiangyu and
Yang, Chao and
Ouyang, Wanli and
Qiao, Yu",
editor = "Ku, Lun-Wei and
Martins, Andre and
Srikumar, Vivek",
booktitle = "Findings of the Association for Computational Linguistics: ACL 2024",
month = aug,
year = "2024",
address = "Bangkok, Thailand",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.findings-acl.630",
doi = "10.18653/v1/2024.findings-acl.630",
pages = "10586--10613",
abstract = "A single language model, even when aligned with labelers through reinforcement learning from human feedback (RLHF), may not suit all human preferences. Recent approaches therefore prefer customization, gathering multi-dimensional feedback, and creating distinct reward models for each dimension.Different language models are then optimized for various preferences using multi-objective RLHF (MORLHF) with varying reward weights.However, RL fine-tuning is unstable and resource-heavy, especially with diverse and usually conflicting objectives.In this paper, we present Multi-Objective Direct Preference Optimization (MODPO), an RL-free extension of Direct Preference Optimization (DPO) for multiple alignment objectives.Essentially, MODPO folds language modeling directly into reward modeling, training language models as implicit collective reward models that combine all objectives with specific weights. MODPO theoretically yields the same optimal solutions as MORLHF but is practically more stable and efficient.Empirical results in safety alignment and long-form question answering show that MODPO matches or outperforms existing methods, producing a Pareto front of language models catering to diverse preferences with three times less computational resources compared to MORLHF.Code is available at https://github.com/ZHZisZZ/modpo.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhou-etal-2024-beyond">
<titleInfo>
<title>Beyond One-Preference-Fits-All Alignment: Multi-Objective Direct Preference Optimization</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zhanhui</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jie</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jing</namePart>
<namePart type="family">Shao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiangyu</namePart>
<namePart type="family">Yue</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chao</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wanli</namePart>
<namePart type="family">Ouyang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yu</namePart>
<namePart type="family">Qiao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-08</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2024</title>
</titleInfo>
<name type="personal">
<namePart type="given">Lun-Wei</namePart>
<namePart type="family">Ku</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Andre</namePart>
<namePart type="family">Martins</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vivek</namePart>
<namePart type="family">Srikumar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Bangkok, Thailand</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>A single language model, even when aligned with labelers through reinforcement learning from human feedback (RLHF), may not suit all human preferences. Recent approaches therefore prefer customization, gathering multi-dimensional feedback, and creating distinct reward models for each dimension.Different language models are then optimized for various preferences using multi-objective RLHF (MORLHF) with varying reward weights.However, RL fine-tuning is unstable and resource-heavy, especially with diverse and usually conflicting objectives.In this paper, we present Multi-Objective Direct Preference Optimization (MODPO), an RL-free extension of Direct Preference Optimization (DPO) for multiple alignment objectives.Essentially, MODPO folds language modeling directly into reward modeling, training language models as implicit collective reward models that combine all objectives with specific weights. MODPO theoretically yields the same optimal solutions as MORLHF but is practically more stable and efficient.Empirical results in safety alignment and long-form question answering show that MODPO matches or outperforms existing methods, producing a Pareto front of language models catering to diverse preferences with three times less computational resources compared to MORLHF.Code is available at https://github.com/ZHZisZZ/modpo.</abstract>
<identifier type="citekey">zhou-etal-2024-beyond</identifier>
<identifier type="doi">10.18653/v1/2024.findings-acl.630</identifier>
<location>
<url>https://aclanthology.org/2024.findings-acl.630</url>
</location>
<part>
<date>2024-08</date>
<extent unit="page">
<start>10586</start>
<end>10613</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Beyond One-Preference-Fits-All Alignment: Multi-Objective Direct Preference Optimization
%A Zhou, Zhanhui
%A Liu, Jie
%A Shao, Jing
%A Yue, Xiangyu
%A Yang, Chao
%A Ouyang, Wanli
%A Qiao, Yu
%Y Ku, Lun-Wei
%Y Martins, Andre
%Y Srikumar, Vivek
%S Findings of the Association for Computational Linguistics: ACL 2024
%D 2024
%8 August
%I Association for Computational Linguistics
%C Bangkok, Thailand
%F zhou-etal-2024-beyond
%X A single language model, even when aligned with labelers through reinforcement learning from human feedback (RLHF), may not suit all human preferences. Recent approaches therefore prefer customization, gathering multi-dimensional feedback, and creating distinct reward models for each dimension.Different language models are then optimized for various preferences using multi-objective RLHF (MORLHF) with varying reward weights.However, RL fine-tuning is unstable and resource-heavy, especially with diverse and usually conflicting objectives.In this paper, we present Multi-Objective Direct Preference Optimization (MODPO), an RL-free extension of Direct Preference Optimization (DPO) for multiple alignment objectives.Essentially, MODPO folds language modeling directly into reward modeling, training language models as implicit collective reward models that combine all objectives with specific weights. MODPO theoretically yields the same optimal solutions as MORLHF but is practically more stable and efficient.Empirical results in safety alignment and long-form question answering show that MODPO matches or outperforms existing methods, producing a Pareto front of language models catering to diverse preferences with three times less computational resources compared to MORLHF.Code is available at https://github.com/ZHZisZZ/modpo.
%R 10.18653/v1/2024.findings-acl.630
%U https://aclanthology.org/2024.findings-acl.630
%U https://doi.org/10.18653/v1/2024.findings-acl.630
%P 10586-10613
Markdown (Informal)
[Beyond One-Preference-Fits-All Alignment: Multi-Objective Direct Preference Optimization](https://aclanthology.org/2024.findings-acl.630) (Zhou et al., Findings 2024)
ACL