一作ACL 2026Findings· Findings of ACL
R etentive KV : State-Space Memory for Uncertainty-Aware Multimodal KV Cache Eviction
Sihao Liu · YuFan Xiong · Zhonghua Jiang · Zhaode Wang (Taobao & · Chengfei Lv (Taobao & · Shengyu Zhang
摘要
Multimodal Large Language Models face severe challenges in computational efficiency and memory consumption due to the substantial expansion of the visual KV cache when processing long visual contexts. Existing KV cache compression methods typically rely on the “persistence of importance” hypothes...