Open Models Almanac
Methodsinference & servingKV cache management

single mention · inference & serving

KV-cache reset on weight synchronization

Resets inference KV cache after weight broadcasts to prevent mixing states encoded under different weight versions.

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 KV-cache reset on weight sync.

Filed under inference & serving › KV cache management