arXiv 2505.14009
Activation-Guided Consensus Merging for Large Language Models
By Yuxuan Yao, Shuqi Liu, et al.
Published 2025-05-20
Mindmap
Browse the paper's core ideas, clusters, and relationships in a structured outline.
Recent research has increasingly focused on reconciling the reasoning capabilities of System 2 with the efficiency of System 1. While existing training-based and prompt-based approaches face significant challenges in terms of efficiency and stability, model merging emerges as a promising strategy to integrate the diverse capabilities of different Large Language Models (LLMs) into a unified model. However, convention…