This article explores distinguishing details of Markov Models and Transformer-based models like GPT, focusing on how they predict the next character in a sequence. It explores the fundamental differences between these models, with a particular emphasis on how the self-attention mechanism in Transformer models makes a difference compared to the fixed context length in Markov models.

Sometimes input for document processing tasks such as OCR, table detection or text segmentation can be scanned or photo taken from hand that do not have ideal perspective - is rotated or spatially distorted in some way (warped document). If you are looking for my recommendations go straight to the last section of this article "Summary and recommendations".