Training a model to add HTML formatting to a web article?

Hello Everyone, I want to train a seq2seq model that would take as an input a web/blog article in plain text and manage to add basic HTML formatting to it (such as < h3 > or < h2 > on some titles, < strong > on some keywords or phrases, < i > the same, etc…).

I have a dataset with thousands of articles containing those HTML tags and for each article I have its plain text version.
Am I supposed to fine-tune an existing seq2seq model like BERT or MT5, should I make one from scratch? How would I go about doing such thing? All I found on the internet is tutorials on fine-tuning pre existing models for sentiment analysis and translation but I cant find anything that gets close to what I want and I don’t know if it’s even possible to achieve.

I will appreciate enourmously any kind of help or pointers. Thank you very much.