Papers · arXiv · 计算机视觉

Beyond the Sink Token: Sink-Guided Key Editing in Vision-Language Models

作者 Hui Zhu, Saleh Zare Zade, Rafi Ibn Sultan, Xiangyu Zhou, Dongxiao Zhu 2026-11-19 · 4 小时前 English
Summary · 摘要

arXiv:2611.23516v1 Announce Type: new Abstract: In vision-language models (VLMs), a few sink tokens …

阅读原文 arxiv.org