Researchers adapt Ai2's open olmOCR 2 for historical and archaeological archives

Ai2 releases olmOCR 2, a document-to-text model with open weights and code that researchers can run locally or adapt to their own libraries.

Historian Maurice Casey built ArtificeOCR to transcribe his photographs of early twentieth-century radical politics sources on his own computer.

TU Wien and the Austrian Archaeological Institute fine-tuned olmOCR 2 using LoRA, a method that trains a small set of additional parameters, on 57 records. On 450 test records of Carnuntum pottery, transcription error fell from 15.58% to 1.11%, according to the Ai2 report.