FineBooks

FineBooks is a collaboration between Hugging Face and EleutherAI with two goals: to test whether current open OCR models are good enough to unlock historical knowledge at scale, and to re-process public domain book collections using leading models, publishing the improved text as open datasets.

The first collection we are working on is the Biodiversity Heritage Library (BHL).

Current releases

• BHL OCR Leaderboard — 14 open OCR models scored against expert-transcribed historical book pages. New models are added as they are released.
• bhl-impact-gt — the ground-truth dataset behind the leaderboard: 2,165 pages in English, French, German and Latin, transcribed by the EU IMPACT project and BHL-Europe.
• bhl-ocr-eval — the evaluation harness, so anyone can verify a score or run the same measurement on their own material.

Read the announcement: FineBooks: are open OCR models good enough to unlock historical knowledge?