Newer Models, Same Advantage

DharmaOCR surpassed Mistral OCR4 and Unlimited-OCR in Brazilian Portuguese due to domain specialization and targeted training, despite newer architectures. This advantage stems from how models are trained for specific tasks, as all generative model-based OCR systems are probabilistic and prone to transcription errors. Supervised fine-tuning, while concentrating resources on the target domain, can lead to degeneration and incoherent sequences if early token predictions diverge, as each subsequent prediction is conditioned on that divergent state.