Open Models Almanac
Methodsinference & servingdecoding strategy

single mention · inference & serving

Throughput-based draft-block sizing

Selects speculative draft-block size using end-to-end throughput rather than acceptance rate alone.

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

Filed under inference & serving › decoding strategy