Controllable Preference Optimization: Toward Controllable Multi-Objective Alignment

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Guo, Yiju, Cui, Ganqu, Yuan, Lifan, Ding, Ning, Sun, Zexu, Sun, Bowen, Chen, Huimin, Xie, Ruobing, Zhou, Jie, Lin, Yankai, Liu, Zhiyuan, Sun, Maosong
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912068606623744
author Guo, Yiju
Cui, Ganqu
Yuan, Lifan
Ding, Ning
Sun, Zexu
Sun, Bowen
Chen, Huimin
Xie, Ruobing
Zhou, Jie
Lin, Yankai
Liu, Zhiyuan
Sun, Maosong
author_facet Guo, Yiju
Cui, Ganqu
Yuan, Lifan
Ding, Ning
Sun, Zexu
Sun, Bowen
Chen, Huimin
Xie, Ruobing
Zhou, Jie
Lin, Yankai
Liu, Zhiyuan
Sun, Maosong
contents Alignment in artificial intelligence pursues the consistency between model responses and human preferences as well as values. In practice, the multifaceted nature of human preferences inadvertently introduces what is known as the "alignment tax" -a compromise where enhancements in alignment within one objective (e.g.,harmlessness) can diminish performance in others (e.g.,helpfulness). However, existing alignment techniques are mostly unidirectional, leading to suboptimal trade-offs and poor flexibility over various objectives. To navigate this challenge, we argue the prominence of grounding LLMs with evident preferences. We introduce controllable preference optimization (CPO), which explicitly specifies preference scores for different objectives, thereby guiding the model to generate responses that meet the requirements. Our experimental analysis reveals that the aligned models can provide responses that match various preferences among the "3H" (helpfulness, honesty, harmlessness) desiderata. Furthermore, by introducing diverse data and alignment goals, we surpass baseline methods in aligning with single objectives, hence mitigating the impact of the alignment tax and achieving improvements in multi-objective alignment.
format Preprint
id arxiv_https___arxiv_org_abs_2402_19085
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Controllable Preference Optimization: Toward Controllable Multi-Objective Alignment
Guo, Yiju
Cui, Ganqu
Yuan, Lifan
Ding, Ning
Sun, Zexu
Sun, Bowen
Chen, Huimin
Xie, Ruobing
Zhou, Jie
Lin, Yankai
Liu, Zhiyuan
Sun, Maosong
Computation and Language
Artificial Intelligence
Systems and Control
Alignment in artificial intelligence pursues the consistency between model responses and human preferences as well as values. In practice, the multifaceted nature of human preferences inadvertently introduces what is known as the "alignment tax" -a compromise where enhancements in alignment within one objective (e.g.,harmlessness) can diminish performance in others (e.g.,helpfulness). However, existing alignment techniques are mostly unidirectional, leading to suboptimal trade-offs and poor flexibility over various objectives. To navigate this challenge, we argue the prominence of grounding LLMs with evident preferences. We introduce controllable preference optimization (CPO), which explicitly specifies preference scores for different objectives, thereby guiding the model to generate responses that meet the requirements. Our experimental analysis reveals that the aligned models can provide responses that match various preferences among the "3H" (helpfulness, honesty, harmlessness) desiderata. Furthermore, by introducing diverse data and alignment goals, we surpass baseline methods in aligning with single objectives, hence mitigating the impact of the alignment tax and achieving improvements in multi-objective alignment.
title Controllable Preference Optimization: Toward Controllable Multi-Objective Alignment
topic Computation and Language
Artificial Intelligence
Systems and Control
url https://arxiv.org/abs/2402.19085