Multi-Personality Generation of LLMs at Decoding-time

TL;DR

This paper introduces a flexible, decoding-time method for multi-personality generation in large language models, using implicit density ratios and a novel rejection sampling technique to efficiently produce personalized responses without additional training.

Contribution

It proposes a novel decoding-time framework for multi-personality generation that avoids costly retraining and external models, utilizing implicit density ratios and a chunk-level rejection sampling method.

Findings

Achieves 16-18% improvement in personality consistency.

Reduces computational overhead with Speculative Chunk-level Rejection sampling.

Demonstrates effectiveness on MBTI and Role-Playing tasks.

Abstract

Multi-personality generation for LLMs, enabling simultaneous embodiment of multiple personalization attributes, is a fundamental challenge. Existing retraining-based approaches are costly and poorly scalable, while decoding-time methods often rely on external models or heuristics, limiting flexibility and robustness. In this paper, we propose a novel Multi-Personality Generation (MPG) framework under the decoding-time combination paradigm. It flexibly controls multi-personality without relying on scarce multi-dimensional models or extra training, leveraging implicit density ratios in single-dimensional models as a "free lunch" to reformulate the task as sampling from a target strategy aggregating these ratios. To implement MPG efficiently, we design Speculative Chunk-level based Rejection sampling (SCR), which generates responses in chunks and parallelly validates them via estimated thresholds within a sliding window. This significantly reduces computational overhead while maintaining high-quality generation. Experiments on MBTI personality and Role-Playing demonstrate the effectiveness of MPG, showing improvements up to 16%-18%. Code and data are available at https://github.com/Libra117/MPG .