The RevCity Model: HTR Recognition Model trained on 18th century English handwriting

RevCity Handwritten English Base Model This is a text recognition model trained on the Revolutionary City dataset of handwritten Latin-script text in English available here in its state of 2025-12-18. The material encompasses English text (~95k lines) in a variety of hands, with varying degrees of damage and noise. A negligible amount (<500 lines) of French material was also in this corpus. The model was trained from scratch using the python HTR package kraken using a learning rate of 0.0001 and with data augmentation enabled. Work on this model was sponsored by the Center for Digital Scholarship at the American Philosophical Society.