Redirecting to original paper in 30 seconds...
Click below to go immediately or wait for automatic redirect
📄 Abstract
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 .
Key Contributions
This paper proposes a novel Multi-Personality Generation (MPG) framework operating at decoding time, which flexibly controls multiple LLM personalities without retraining or relying on external models. It leverages implicit density ratios and introduces Speculative Chunk-level based Rejection sampling (SCR) to efficiently generate responses in chunks, significantly reducing computational overhead while maintaining flexibility and robustness.
Business Value
Enables the creation of more engaging and personalized AI experiences (e.g., chatbots, virtual assistants) at a lower computational cost, improving user satisfaction and retention.