KVBoost Enables Non-Contiguous KV Cache Reuse via Dual-Hashing

KVBoost is a chunk-level KV cache reuse system for HuggingFace-compatible models that works regardless of content position. It uses a dual-hash scheme (prefix and content) and selective recomputation to manage attention boundary errors during cache reuse.