Nayana

Document-centric vision-language models for 22 languages, trained entirely on synthetic data