ICDPO: Effectively Borrowing Alignment Capability of Others via In-context Direct Preference Optimization

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Song, Feifan, Fan, Yuxuan, Zhang, Xin, Wang, Peiyi, Wang, Houfeng
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910331374141440
author Song, Feifan
Fan, Yuxuan
Zhang, Xin
Wang, Peiyi
Wang, Houfeng
author_facet Song, Feifan
Fan, Yuxuan
Zhang, Xin
Wang, Peiyi
Wang, Houfeng
contents Large Language Models (LLMs) rely on Human Preference Alignment (HPA) to ensure the generation of safe content. Due to the heavy cost associated with fine-tuning, fine-tuning-free methods have emerged, typically modifying LLM decoding with external auxiliary methods. However, these methods do not essentially enhance the LLM itself. In this paper, we rethink the derivation procedures of DPO, based on which we conversely build an instant scorer using the states of the LLM before and after In-context Learning (ICL). Accordingly, we propose a novel approach called In-Context Direct Preference Optimization (ICDPO). It enables LLMs to borrow the HPA capabilities from superior LLMs with ICL, generating well-aligned responses as estimated by the aforementioned instant scorer, thereby enhancing the final performance. ICDPO can be further enhanced with a two-stage retriever and an upgraded scorer, both offering benefits. Extensive experiments show its effectiveness, particularly in outperforming two fine-tuning-free baselines, and it exhibits competitiveness with SFT + LoRA. We also conduct detailed analyses to offer comprehensive insights into ICDPO.
format Preprint
id arxiv_https___arxiv_org_abs_2402_09320
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle ICDPO: Effectively Borrowing Alignment Capability of Others via In-context Direct Preference Optimization
Song, Feifan
Fan, Yuxuan
Zhang, Xin
Wang, Peiyi
Wang, Houfeng
Computation and Language
Artificial Intelligence
Large Language Models (LLMs) rely on Human Preference Alignment (HPA) to ensure the generation of safe content. Due to the heavy cost associated with fine-tuning, fine-tuning-free methods have emerged, typically modifying LLM decoding with external auxiliary methods. However, these methods do not essentially enhance the LLM itself. In this paper, we rethink the derivation procedures of DPO, based on which we conversely build an instant scorer using the states of the LLM before and after In-context Learning (ICL). Accordingly, we propose a novel approach called In-Context Direct Preference Optimization (ICDPO). It enables LLMs to borrow the HPA capabilities from superior LLMs with ICL, generating well-aligned responses as estimated by the aforementioned instant scorer, thereby enhancing the final performance. ICDPO can be further enhanced with a two-stage retriever and an upgraded scorer, both offering benefits. Extensive experiments show its effectiveness, particularly in outperforming two fine-tuning-free baselines, and it exhibits competitiveness with SFT + LoRA. We also conduct detailed analyses to offer comprehensive insights into ICDPO.
title ICDPO: Effectively Borrowing Alignment Capability of Others via In-context Direct Preference Optimization
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2402.09320