Collected sources and patterns will appear here. Add from search, explore, or the patterns library.
KVCache -> HierarchicalStorageState
Save computed KV-cache prefixes across a tiered GPU-CPU-Disk storage hierarchy based on access frequency.
Problem it solves
GPU memory limits require evicting KV-caches without losing them entirely, which would require re-computation.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.