@article{hardkvheadadaptiveregularizationfordecod, title = {HARD-KV: Head-Adaptive Regularization for Decoding-time KV Compression}, author = {Yuxuan Yang and Feiyang Ren and Bowen Zeng and Dalin Zhang and Jinpeng Chen and Gang Chen and Huan Li}, year = {2026}, eprint = {2606.28831}, archivePrefix = {arXiv}, url = {https://arxiv.org/abs/2606.28831}, }