Based on a model like LayoutLM, how does it work to label the front part of a document and have the rest automatically annotated with components like headers, footers, etc?

Is this a form of finetuning, just adjusting the coordinates of the text locations to the context of the current document? It's not clear to me why this already works for a single new document. Thought that even fine-tuning requires a batch sample. However, maybe each page counts here?

adjusting components context current etc finetuning languagetechnology layoutlm locations part rest text work