Extract, parse, and understand documents from images
Grounded 4B VLM that segments what it talks about