Open Models Almanac
Methodsinference & servingdecoding strategy

implementation detail · filed under inference & serving

Three-token speculative draft depth

Sets the speculative draft or verification depth to a block of three tokens.

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 Three-token block speculative decoding, propose a block of 3 tokens, speculative depth is fixed at 3.

sources
2
model
1
lab adopt it
1
strongest
core

Documented in

Mistral AI

Evidence

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

How sources treat it
default 1core 1
↑