@article{valuestategatedattentionformitigatingext, title = {Value-State Gated Attention for Mitigating Extreme-Token Phenomena in Transformers}, author = {Rui Bu and Haofeng Zhong and Wenzheng Chen and Yangyan Li}, year = {2025}, eprint = {2510.09017}, archivePrefix = {arXiv}, url = {https://arxiv.org/abs/2510.09017}, }