Open Models Almanac
Methodsmodel architecturemultimodal architecture

specific method · filed under model architecture

Visual token reduction

Reduces visual-token count, in one described implementation by using a global image and pixel shuffle without tiling.

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

Also called Reduce the number of visual tokens, Using fewer visual tokens, using a minimal number of visual tokens.

sources
2
model
1
lab adopt it
1
strongest
core

Documented in

Hugging Face

Evidence

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

How sources treat it
used 1core 1

SmolVLA

Hugging Faceadopted2 sources

we do not use tiling. We use the global image only, in addition to a pixel shuffle operation, limiting the visual tokens to 64 per frame.

usedmodel architecturein SmolVLAHugging Face
1 more from this model

Using fewer visual tokens

coremodel architecturein SmolVLAHugging Face
↑