Saved in:
Bibliographic Details
Main Authors: Shi, Ruizhe, Zhou, Runlong, Du, Simon S.
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2409.19605
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915133950787584
author Shi, Ruizhe
Zhou, Runlong
Du, Simon S.
author_facet Shi, Ruizhe
Zhou, Runlong
Du, Simon S.
contents Direct Preference Optimization (DPO) has emerged as a stable, scalable, and efficient solution for language model alignment. Despite its empirical success, the optimization properties, particularly the impact of samplers on its convergence rates, remain under-explored. In this paper, we provide a rigorous analysis of DPO's convergence rates with different sampling strategies under the exact gradient setting, revealing a surprising separation: uniform sampling achieves $\textbf{linear}$ convergence, while our proposed online sampler achieves $\textbf{quadratic}$ convergence. We further adapt the sampler to practical settings by incorporating posterior distributions and logit mixing, demonstrating improvements over previous methods. For example, it outperforms vanilla DPO by over $7.4$% on Safe-RLHF dataset. Our results not only offer insights into the theoretical understanding of DPO but also pave the way for further algorithm designs.
format Preprint
id arxiv_https___arxiv_org_abs_2409_19605
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle The Crucial Role of Samplers in Online Direct Preference Optimization
Shi, Ruizhe
Zhou, Runlong
Du, Simon S.
Machine Learning
Computation and Language
Direct Preference Optimization (DPO) has emerged as a stable, scalable, and efficient solution for language model alignment. Despite its empirical success, the optimization properties, particularly the impact of samplers on its convergence rates, remain under-explored. In this paper, we provide a rigorous analysis of DPO's convergence rates with different sampling strategies under the exact gradient setting, revealing a surprising separation: uniform sampling achieves $\textbf{linear}$ convergence, while our proposed online sampler achieves $\textbf{quadratic}$ convergence. We further adapt the sampler to practical settings by incorporating posterior distributions and logit mixing, demonstrating improvements over previous methods. For example, it outperforms vanilla DPO by over $7.4$% on Safe-RLHF dataset. Our results not only offer insights into the theoretical understanding of DPO but also pave the way for further algorithm designs.
title The Crucial Role of Samplers in Online Direct Preference Optimization
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2409.19605