Model split into two vision_model and text_decoder. a6340a4 verified johnpaultaken commited on Dec 28, 2025