Collected sources and patterns will appear here. Add from search, explore, or the patterns library.
PromptTokens -> KVCache
Retrieve matching KV-cache prefixes from a tiered GPU-CPU-Disk storage hierarchy to skip redundant prefill computations.
Problem it solves
Long-context prompts cause redundant prefill overhead if the active KV-cache is evicted from GPU VRAM.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.