Preprint
Aug 2026
TwinKV: A Composable Repair Pass for KV Cache Eviction via Pairwise Key Redundancy
TwinKV, a training-free method that discounts value energy by nonlocal post-RoPE key frequency, is introduced, showing that modest frequency corrections can change retention and answering outcomes, with effects that depend on the model and task.
Hong Chen, Yuan Zeng, Yong-Wei Huang et al.
· 0 citations