Unlimited OCR Works 2

Baidu has released Unlimited OCR 2, an open vision-language model (about 3 billion parameters) that turns scanned documents into text. Its new attention mechanism keeps memory use and speed nearly constant whatever the document length, so documents of dozens or even hundreds of pages can be parsed in one request with cross-page context. The model is MIT-licensed on Hugging Face and ModelScope, with an online demo and access through Baidu Cloud.

See original publication↗Baidu (Hugging Face officiel) · huggingface.co