Preview - rendered architecture diagrams; full validated atlas + dataset coming.
Back to models

V-JEPA2 ViT-G/16 384 (action-conditioned latent video predictor)

history v jepa2 validation pending

Architecture diagram

Rendered with TorchLens / Graphviz

Open SVG
Unrolled - full
V-JEPA2 ViT-G/16 384 (action-conditioned latent video predictor) architecture diagram