Open Models Almanac
Methodsinference & servingserving parallelism

single mention · inference & serving

KDA state and KV-cache layout pinning

Requires matching KDA convolution-state and KV-cache layouts across prefill and decode pools.

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

Filed under inference & serving › serving parallelism