Attention Sink in Transformers: A Survey on Utilization, Interpretation, and Mitigation
Abstract
As the foundational architecture of modern machine learning, Transformers have driven remarkable progress across diverse AI domains. Despite their transformative impact, a persistent challenge across various Transformers is Attention Sink (AS), in which a disproportionate amount of attention is focused on a small subset of specific yet uninformative tokens. AS complicates interpretability, significantly affecting the training and inference dynamics, and exacerbates issues such as hallucinations. In recent years, substantial research has been dedicated to understanding and harnessing AS. However, a comprehensive survey that systematically consolidates AS-related research and offers guidance for future advancements remains lacking. To address this gap, we present the first survey on AS, structured around three key dimensions that define the current research landscape: Fundamental Utilization, Mechanistic Interpretation, and Strategic Mitigation. Our work provides a pivotal contribution by clarifying key concepts and guiding researchers through the evolution and trends of the field. We envision this survey as a definitive resource, empowering researchers and practitioners to effectively manage AS within the current Transformer paradigm, while simultaneously inspiring innovative advancements for the next generation of Transformers. The paper list of this work is available at https://github.com/ZunhaiSu/Awesome-Attention-Sink.
Cite
@article{arxiv.2604.10098,
title = {Attention Sink in Transformers: A Survey on Utilization, Interpretation, and Mitigation},
author = {Zunhai Su and Hengyuan Zhang and Wei Wu and Yifan Zhang and Yaxiu Liu and He Xiao and Qingyao Yang and Yuxuan Sun and Rui Yang and Chao Zhang and Keyu Fan and Weihao Ye and Jing Xiong and Hui Shen and Chaofan Tao and Taiqiang Wu and Zhongwei Wan and Yulei Qian and Yuchen Xie and Ngai Wong},
journal= {arXiv preprint arXiv:2604.10098},
year = {2026}
}