[tdehaene]

You can also do something similar but with seq2seq models instead of decoder-only models: https://blog.ml6.eu/ocr-correction-with-byt5-5994d1217c07