arXiv Survey Maps KV Cache Optimization Landscape: 5 Strategies for Million-Token LLM Inference

A comprehensive arXiv review categorizes five principal KV cache optimization techniques—eviction, compression, hybrid memory, novel attention, and co