@inproceedings{chen-etal-2025-diffpo,
title = "{D}iff{PO}: Diffusion-styled Preference Optimization for Inference Time Alignment of Large Language Models",
author = "Chen, Ruizhe and
Chai, Wenhao and
Yang, Zhifei and
Zhang, Xiaotian and
Wang, Ziyang and
Quek, Tony and
Zhou, Joey Tianyi and
Poria, Soujanya and
Liu, Zuozhu",
editor = "Che, Wanxiang and
Nabende, Joyce and
Shutova, Ekaterina and
Pilehvar, Mohammad Taher",
booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.acl-long.926/",
doi = "10.18653/v1/2025.acl-long.926",
pages = "18910--18925",
ISBN = "979-8-89176-251-0",
abstract = "Inference-time alignment provides an efficient alternative for aligning LLMs with humans. However, these approaches still face challenges, such as limited scalability due to policy-specific value functions and latency during the inference phase. In this paper, we propose a novel approach, Diffusion-styled Preference Optimization (DiffPO), which provides an efficient and policy-agnostic solution for aligning LLMs with humans. By directly performing alignment at sentence level, DiffPO avoids the time latency associated with token-level generation. Designed as a plug-and-play module, DiffPO can be seamlessly integrated with various base models to enhance their alignment. Extensive experiments on AlpacaEval 2, MT-bench, and HH-RLHF demonstrate that DiffPO achieves superior alignment performance across various settings, achieving a favorable trade-off between alignment quality and inference-time latency. Furthermore, DiffPO demonstrates model-agnostic scalability, significantly improving the performance of large models such as Llama-3-70B."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chen-etal-2025-diffpo">
<titleInfo>
<title>DiffPO: Diffusion-styled Preference Optimization for Inference Time Alignment of Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ruizhe</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wenhao</namePart>
<namePart type="family">Chai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhifei</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaotian</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ziyang</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tony</namePart>
<namePart type="family">Quek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joey</namePart>
<namePart type="given">Tianyi</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Soujanya</namePart>
<namePart type="family">Poria</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zuozhu</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wanxiang</namePart>
<namePart type="family">Che</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joyce</namePart>
<namePart type="family">Nabende</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Shutova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohammad</namePart>
<namePart type="given">Taher</namePart>
<namePart type="family">Pilehvar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Vienna, Austria</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-251-0</identifier>
</relatedItem>
<abstract>Inference-time alignment provides an efficient alternative for aligning LLMs with humans. However, these approaches still face challenges, such as limited scalability due to policy-specific value functions and latency during the inference phase. In this paper, we propose a novel approach, Diffusion-styled Preference Optimization (DiffPO), which provides an efficient and policy-agnostic solution for aligning LLMs with humans. By directly performing alignment at sentence level, DiffPO avoids the time latency associated with token-level generation. Designed as a plug-and-play module, DiffPO can be seamlessly integrated with various base models to enhance their alignment. Extensive experiments on AlpacaEval 2, MT-bench, and HH-RLHF demonstrate that DiffPO achieves superior alignment performance across various settings, achieving a favorable trade-off between alignment quality and inference-time latency. Furthermore, DiffPO demonstrates model-agnostic scalability, significantly improving the performance of large models such as Llama-3-70B.</abstract>
<identifier type="citekey">chen-etal-2025-diffpo</identifier>
<identifier type="doi">10.18653/v1/2025.acl-long.926</identifier>
<location>
<url>https://aclanthology.org/2025.acl-long.926/</url>
</location>
<part>
<date>2025-07</date>
<extent unit="page">
<start>18910</start>
<end>18925</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T DiffPO: Diffusion-styled Preference Optimization for Inference Time Alignment of Large Language Models
%A Chen, Ruizhe
%A Chai, Wenhao
%A Yang, Zhifei
%A Zhang, Xiaotian
%A Wang, Ziyang
%A Quek, Tony
%A Zhou, Joey Tianyi
%A Poria, Soujanya
%A Liu, Zuozhu
%Y Che, Wanxiang
%Y Nabende, Joyce
%Y Shutova, Ekaterina
%Y Pilehvar, Mohammad Taher
%S Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 979-8-89176-251-0
%F chen-etal-2025-diffpo
%X Inference-time alignment provides an efficient alternative for aligning LLMs with humans. However, these approaches still face challenges, such as limited scalability due to policy-specific value functions and latency during the inference phase. In this paper, we propose a novel approach, Diffusion-styled Preference Optimization (DiffPO), which provides an efficient and policy-agnostic solution for aligning LLMs with humans. By directly performing alignment at sentence level, DiffPO avoids the time latency associated with token-level generation. Designed as a plug-and-play module, DiffPO can be seamlessly integrated with various base models to enhance their alignment. Extensive experiments on AlpacaEval 2, MT-bench, and HH-RLHF demonstrate that DiffPO achieves superior alignment performance across various settings, achieving a favorable trade-off between alignment quality and inference-time latency. Furthermore, DiffPO demonstrates model-agnostic scalability, significantly improving the performance of large models such as Llama-3-70B.
%R 10.18653/v1/2025.acl-long.926
%U https://aclanthology.org/2025.acl-long.926/
%U https://doi.org/10.18653/v1/2025.acl-long.926
%P 18910-18925
Markdown (Informal)
[DiffPO: Diffusion-styled Preference Optimization for Inference Time Alignment of Large Language Models](https://aclanthology.org/2025.acl-long.926/) (Chen et al., ACL 2025)
ACL
- Ruizhe Chen, Wenhao Chai, Zhifei Yang, Xiaotian Zhang, Ziyang Wang, Tony Quek, Joey Tianyi Zhou, Soujanya Poria, and Zuozhu Liu. 2025. DiffPO: Diffusion-styled Preference Optimization for Inference Time Alignment of Large Language Models. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 18910–18925, Vienna, Austria. Association for Computational Linguistics.