Core Concepts
Neeko is an innovative framework designed for efficient multi-character role-playing, utilizing dynamic LoRA strategy to adapt seamlessly to diverse characters.
Abstract
Neeko introduces a novel task of Multi-Character Role-Playing (MCRP) and presents a framework that breaks down the role-playing process into agent pre-training, multiple characters playing, and character incremental learning. The dynamic approach enhances adaptability to unique attributes, personalities, and speaking patterns. Neeko demonstrates superior performance in MCRP over existing methods by offering engaging user interaction experiences.
Stats
Neeko surpasses most existing methods in MCRP performance.
Neeko utilizes a dynamic gating network to activate role-specific LoRA blocks.
Neekofus does not require additional data for incremental learning.
LoRA exhibits poor performance in handling new characters during incremental learning.