Open Models Almanac
Methodsmodel architecturetoken mixer

single mention · model architecture

Bidirectional self-attention with VLM cross-attention

Action-expert blocks combine bidirectional self-attention across an action chunk with cross-attention to VLM context and an MLP.

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

Filed under model architecture › token mixer