Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhang, Jia, Liu, Yao, Zhang, Chen-Xi, Liu, Yi, Jin, Yi-Xuan, Guo, Lan-Zhe, Li, Yu-Feng
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2508.07638
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914362618281984
author Zhang, Jia
Liu, Yao
Zhang, Chen-Xi
Liu, Yi
Jin, Yi-Xuan
Guo, Lan-Zhe
Li, Yu-Feng
author_facet Zhang, Jia
Liu, Yao
Zhang, Chen-Xi
Liu, Yi
Jin, Yi-Xuan
Guo, Lan-Zhe
Li, Yu-Feng
contents Large language models (LLMs) alignment aims to ensure that the behavior of LLMs meets human preferences. While collecting data from multiple fine-grained, aspect-specific preferences becomes more and more feasible, existing alignment methods typically work on a single preference and thus struggle with conflicts inherent in such aggregated datasets. As one early attempt, in this paper, we propose a data-centric approach to align LLMs through the effective use of fine-grained preferences. Specifically, we formulate the problem as a direct fine-grained preference optimization and introduce preference divergence (PD) that quantifies inter-aspect preference conflicts. Instead of directly tackling the consequent complicated optimization, we recast it as a data selection problem and propose a simple yet effective strategy, which identifies a subset of data corresponding to the most negative PD values, for efficient training. We theoretically analyze the loss-bound optimality of our selection strategy and conduct extensive empirical studies on varied settings and datasets to demonstrate that our practical selection method could achieve consistent improvement against standard full-data alignment, using even just 30% of the data. Our work shares a line that LLM alignment using fine-grained preferences is highly feasible.
format Preprint
id arxiv_https___arxiv_org_abs_2508_07638
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Data Selection for LLM Alignment Using Fine-Grained Preferences
Zhang, Jia
Liu, Yao
Zhang, Chen-Xi
Liu, Yi
Jin, Yi-Xuan
Guo, Lan-Zhe
Li, Yu-Feng
Machine Learning
Large language models (LLMs) alignment aims to ensure that the behavior of LLMs meets human preferences. While collecting data from multiple fine-grained, aspect-specific preferences becomes more and more feasible, existing alignment methods typically work on a single preference and thus struggle with conflicts inherent in such aggregated datasets. As one early attempt, in this paper, we propose a data-centric approach to align LLMs through the effective use of fine-grained preferences. Specifically, we formulate the problem as a direct fine-grained preference optimization and introduce preference divergence (PD) that quantifies inter-aspect preference conflicts. Instead of directly tackling the consequent complicated optimization, we recast it as a data selection problem and propose a simple yet effective strategy, which identifies a subset of data corresponding to the most negative PD values, for efficient training. We theoretically analyze the loss-bound optimality of our selection strategy and conduct extensive empirical studies on varied settings and datasets to demonstrate that our practical selection method could achieve consistent improvement against standard full-data alignment, using even just 30% of the data. Our work shares a line that LLM alignment using fine-grained preferences is highly feasible.
title Data Selection for LLM Alignment Using Fine-Grained Preferences
topic Machine Learning
url https://arxiv.org/abs/2508.07638