Open Models Almanac
Methodsmodel architecturemultimodal architecture

implementation detail · filed under model architecture

Reduced action-expert hidden size

Uses a smaller hidden size for the action expert to reduce computational cost.

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 its hidden size to 75% of the VLM’s, Reduced hidden size for the action expert, reduced hidden size.

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

reduce its hidden size to 75% of the VLM’s, keeping the model lightweight for deployment.

coremodel architecturein SmolVLAHugging Face
1 more from this model

To improve on inference’s efficiency, we use a reduced hidden size

usedmodel architecturein SmolVLAHugging Face
↑