1 paper
Hong Chen, Xiang Liu, Bo Wang +5
The linear growth of Key-Value (KV) cache remains a bottleneck for multi-turn LLM deployment. Existing KV cache compression methods often fail to account for the structural propert…