FineBooks Project Aims to Enhance OCR for Language Model Training

The FineBooks Project, a collaboration between Hugging Face and EleutherAI, is addressing the limitations of existing OCR technology in language model training. The top-performing OCR model, dots.mocr, achieves a character accuracy of 97.6 percent and costs under two dollars per thousand pages processed. More than 2,000 historical book pages have been tested, revealing that while the accuracy is sufficient for AI training data, it falls short for scholarly transcriptions, as noted by the FineBooks team. This initiative follows ongoing discussions about the quality of training data in AI, emphasizing the need for improved OCR capabilities to enhance model performance. Practitioners can expect advancements in the quality of training datasets as FineBooks works to refine OCR technology. For further details, see The Decoder.