103  Downloads Updated  3 months ago

A custom merged model: the vision layers were borrowed from Mistral-3 and joined with Bielik-7B (SpeakLeash). The result is a compact multimodal model that reads text from images — printed pages, scans.