Add lightweight visual-token spatial adapters (7 VLMs, frozen base)

031dd80

verified

quin210 commited on 16 days ago