@inproceedings{yao-etal-2024-pure,
title = "{PURE}: Aligning {LLM} via Pluggable Query Reformulation for Enhanced Helpfulness",
author = "Yao, Wenjin and
Wang, Yidong and
Yu, Zhuohao and
Xie, Rui and
Zhang, Shikun and
Ye, Wei",
editor = "Al-Onaizan, Yaser and
Bansal, Mohit and
Chen, Yun-Nung",
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2024",
month = nov,
year = "2024",
address = "Miami, Florida, USA",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.findings-emnlp.509/",
doi = "10.18653/v1/2024.findings-emnlp.509",
pages = "8721--8744",
abstract = "Aligning large language models (LLMs) with human values and preferences is a significant challenge. Training-based methods, such as reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO), require substantial resources and are impractical for API-based LLMs. Post-processing methods decouple alignment from training but may incur high multiple-time inference costs or rely on less knowledgeable lightweight models for response refinement. In this paper, we propose a new LLM alignment paradigm from the perspective of pre-processing. By reformulating risky queries into highly relevant yet harmless ones before feeding them into LLMs, our method eliminates the high costs of training base LLMs, efficiently applies to both open-source and proprietary LLMs, and achieves a promising balance of harmlessness and helpfulness. For example, with Vicuna-7B as the LLM to align, it enhances helpfulness by 28.52{\%} over DPO while maintaining comparable harmlessness levels. When applied to Gemini-1.5-pro, it increased harmlessness and helpfulness by 7.04{\%} and 29.37{\%}, respectively."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yao-etal-2024-pure">
<titleInfo>
<title>PURE: Aligning LLM via Pluggable Query Reformulation for Enhanced Helpfulness</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wenjin</namePart>
<namePart type="family">Yao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yidong</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhuohao</namePart>
<namePart type="family">Yu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rui</namePart>
<namePart type="family">Xie</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shikun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wei</namePart>
<namePart type="family">Ye</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2024</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yaser</namePart>
<namePart type="family">Al-Onaizan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohit</namePart>
<namePart type="family">Bansal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yun-Nung</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Miami, Florida, USA</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Aligning large language models (LLMs) with human values and preferences is a significant challenge. Training-based methods, such as reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO), require substantial resources and are impractical for API-based LLMs. Post-processing methods decouple alignment from training but may incur high multiple-time inference costs or rely on less knowledgeable lightweight models for response refinement. In this paper, we propose a new LLM alignment paradigm from the perspective of pre-processing. By reformulating risky queries into highly relevant yet harmless ones before feeding them into LLMs, our method eliminates the high costs of training base LLMs, efficiently applies to both open-source and proprietary LLMs, and achieves a promising balance of harmlessness and helpfulness. For example, with Vicuna-7B as the LLM to align, it enhances helpfulness by 28.52% over DPO while maintaining comparable harmlessness levels. When applied to Gemini-1.5-pro, it increased harmlessness and helpfulness by 7.04% and 29.37%, respectively.</abstract>
<identifier type="citekey">yao-etal-2024-pure</identifier>
<identifier type="doi">10.18653/v1/2024.findings-emnlp.509</identifier>
<location>
<url>https://aclanthology.org/2024.findings-emnlp.509/</url>
</location>
<part>
<date>2024-11</date>
<extent unit="page">
<start>8721</start>
<end>8744</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T PURE: Aligning LLM via Pluggable Query Reformulation for Enhanced Helpfulness
%A Yao, Wenjin
%A Wang, Yidong
%A Yu, Zhuohao
%A Xie, Rui
%A Zhang, Shikun
%A Ye, Wei
%Y Al-Onaizan, Yaser
%Y Bansal, Mohit
%Y Chen, Yun-Nung
%S Findings of the Association for Computational Linguistics: EMNLP 2024
%D 2024
%8 November
%I Association for Computational Linguistics
%C Miami, Florida, USA
%F yao-etal-2024-pure
%X Aligning large language models (LLMs) with human values and preferences is a significant challenge. Training-based methods, such as reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO), require substantial resources and are impractical for API-based LLMs. Post-processing methods decouple alignment from training but may incur high multiple-time inference costs or rely on less knowledgeable lightweight models for response refinement. In this paper, we propose a new LLM alignment paradigm from the perspective of pre-processing. By reformulating risky queries into highly relevant yet harmless ones before feeding them into LLMs, our method eliminates the high costs of training base LLMs, efficiently applies to both open-source and proprietary LLMs, and achieves a promising balance of harmlessness and helpfulness. For example, with Vicuna-7B as the LLM to align, it enhances helpfulness by 28.52% over DPO while maintaining comparable harmlessness levels. When applied to Gemini-1.5-pro, it increased harmlessness and helpfulness by 7.04% and 29.37%, respectively.
%R 10.18653/v1/2024.findings-emnlp.509
%U https://aclanthology.org/2024.findings-emnlp.509/
%U https://doi.org/10.18653/v1/2024.findings-emnlp.509
%P 8721-8744
Markdown (Informal)
[PURE: Aligning LLM via Pluggable Query Reformulation for Enhanced Helpfulness](https://aclanthology.org/2024.findings-emnlp.509/) (Yao et al., Findings 2024)
ACL