Open Models Almanac
Methodsinference & servingdecoding strategy

implementation detail · filed under inference & serving

Five-layer SWA MTP drafter

Uses a five-layer sliding-window-attention MTP drafter that predicts seven subsequent tokens per forward pass for parallel verification.

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.

Also called 5-layer SWA MTP drafter (DFlash-style).

sources
2
models
2
lab adopt it
1
strongest
core

Documented in

Evidence

2 quotes from 2 sources, one card per model, each opening with its strongest.

How sources treat it
core 2
↑