Adversarial Preference Optimization: Enhancing Your Alignment via RM-LLM Game

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Cheng, Pengyu, Yang, Yifan, Li, Jian, Dai, Yong, Hu, Tianhao, Cao, Peixin, Du, Nan, Li, Xiaolong
Formato: Preprint
Publicado: 2023
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866909214995120128
author Cheng, Pengyu
Yang, Yifan
Li, Jian
Dai, Yong
Hu, Tianhao
Cao, Peixin
Du, Nan
Li, Xiaolong
author_facet Cheng, Pengyu
Yang, Yifan
Li, Jian
Dai, Yong
Hu, Tianhao
Cao, Peixin
Du, Nan
Li, Xiaolong
contents Human preference alignment is essential to improve the interaction quality of large language models (LLMs). Existing alignment methods depend on manually annotated preference data to guide the LLM optimization directions. However, continuously updating LLMs for alignment raises a distribution gap between model-generated samples and human-annotated responses, hindering training effectiveness. To mitigate this issue, previous methods require additional preference annotation on newly generated samples to adapt to the shifted distribution, which consumes a large amount of annotation resources. Targeting more efficient human preference optimization, we propose an Adversarial Preference Optimization (APO) framework, in which the LLM and the reward model update alternatively via a min-max game. Through adversarial training, the reward model can adapt to the shifted generation distribution of the LLM without any additional annotation. With comprehensive experiments, we find the proposed adversarial training framework further enhances existing alignment baselines in terms of LLM helpfulness and harmlessness. The code is at https://github.com/Linear95/APO.
format Preprint
id arxiv_https___arxiv_org_abs_2311_08045
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Adversarial Preference Optimization: Enhancing Your Alignment via RM-LLM Game
Cheng, Pengyu
Yang, Yifan
Li, Jian
Dai, Yong
Hu, Tianhao
Cao, Peixin
Du, Nan
Li, Xiaolong
Computation and Language
Artificial Intelligence
Machine Learning
Human preference alignment is essential to improve the interaction quality of large language models (LLMs). Existing alignment methods depend on manually annotated preference data to guide the LLM optimization directions. However, continuously updating LLMs for alignment raises a distribution gap between model-generated samples and human-annotated responses, hindering training effectiveness. To mitigate this issue, previous methods require additional preference annotation on newly generated samples to adapt to the shifted distribution, which consumes a large amount of annotation resources. Targeting more efficient human preference optimization, we propose an Adversarial Preference Optimization (APO) framework, in which the LLM and the reward model update alternatively via a min-max game. Through adversarial training, the reward model can adapt to the shifted generation distribution of the LLM without any additional annotation. With comprehensive experiments, we find the proposed adversarial training framework further enhances existing alignment baselines in terms of LLM helpfulness and harmlessness. The code is at https://github.com/Linear95/APO.
title Adversarial Preference Optimization: Enhancing Your Alignment via RM-LLM Game
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2311.08045