Collected sources and patterns will appear here. Add from search or the patterns library.
TokenSequence -> CachedKVBlocks
Store and match KV cache blocks of common prompt prefixes using a radix tree to share state across overlapping requests.
Problem it solves
Redundant computation of KV caches for identical system prompts or history contexts across sequential queries.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.