PAD: Personalized Alignment of LLMs at Decoding-Time

TL;DR

PAD introduces a decoding-time framework for aligning large language models with personalized preferences without additional training, enabling real-time, scalable, and preference-specific text generation.

Contribution

This paper presents a novel decoding-time personalized alignment method that decouples preference modeling from training, allowing real-time, scalable, and generalizable alignment of LLMs.

Findings

Outperforms training-based alignment methods in preference alignment accuracy.

Demonstrates strong generalization to unseen preferences.

Scales effectively across different base models.

Abstract

Aligning with personalized preferences, which vary significantly across cultural, educational, and political differences, poses a significant challenge due to the computational costs and data demands of traditional alignment methods. In response, this paper presents Personalized Alignment at Decoding-time (PAD), a novel framework designed to align LLM outputs with diverse personalized preferences during the inference phase, eliminating the need for additional training. By introducing a unique personalized reward modeling strategy, this framework decouples the text generation process from personalized preferences, facilitating the generation of generalizable token-level personalized rewards. The PAD algorithm leverages these rewards to guide the decoding process, dynamically tailoring the base model's predictions to personalized preferences. Extensive experimental results demonstrate that PAD not only outperforms existing training-based alignment methods in terms of aligning with diverse preferences but also shows significant generalizability to preferences unseen during training and scalability across different base models. This work advances the capability of LLMs to meet user needs in real-time applications, presenting a substantial step forward in personalized LLM alignment.