A 2.48% CER VLM-OCR Pipeline for a Heavily Accented 1544 Florentine Treatise

 Pierfrancesco Giambullari’s Del sito, forma, & misure, dello Inferno di Dante (Florence: Neri Dortelata, 1544) is a representative case of what we call primary-source typographic idiosyncrasy: a single pseudonymous 16th-century press with an experimental dense stress-accent orthography, long-s ß ligatures, and scribal tilde abbreviations, none of which appears in any modern OCR training distribution. On this text we report a character error rate of 2.48%, measured on a 7-page hand-corrected leakage-free held-out set (7,660 characters, 190 edits), using a LoRA fine-tune of LightOnOCR-2-1B on 16 training pages as the OCR engine. Training took ∼ 15 minutes on a single H100; total compute cost for all experiments, including full-book inference on 151 pages, was US$6.96. For context, a zero-shot Claude Sonnet 4.5 baseline reaches 8.12% CER on the same gold standard, and the author’s first-pass Transkribus PyLaia custom HTR model (20 manually annotated training pages) reaches 15.66% CER; both numbers are reported as reference points, not as a head-to-head ranking, and the Transkribus baseline would likely improve with more effort.

The headline of this note is not that one OCR stack “wins”, but that off-the-shelf VLM-OCR with a small LoRA fine-tune, produced by a single person on an iPhone-scanned book in one day, reaches a CER suitable for critical-edition workflows on exactly the kind of idiosyncratic primary-source typography where zero-shot frontier models still degrade. A complementary finding concerns stress-accent coverage: the fine-tuned LightOn model preserves +26% more accented characters than the Transkribus baseline corpus-wide, a difference that matters for a Dortelata edition where stress accents are lexically load-bearing.

We release this note ahead of the forthcoming critical edition to document the pipeline and the accent-coverage finding.