Open Models Almanac
Methodsinference & servingKV cache management

single mention · inference & serving

Projected-input caching for speculative KDA rollback

Caches projected inputs, rebuilds accepted-token states on-chip, and writes verified and bonus-token states back.

The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See KV cache management for the methods filed beside it.

Also called cache only these projected inputs.

Filed under inference & serving › KV cache management