CLEAR-MoE: Shared-Basis Expert Extraction from Frozen Vision Transformers via Calibration-Driven Layer Selection figure
AlphaXiv 中文概览(可滚动查看)