Customization guideline for beginner

Hi, there. I am new to here.
I’ve followed some contents in Transfomers docs.
And I am now reading the Philosophy chapter on Conceptual Guides.

On the following image, I have few questions.

  1. Developers said they give a single API to access and look into attention weights. What is the name of the API?

  2. I’d like to know what is the simple and consistent way to add new tokens to the vocabulary. Also the ways to change Transfomer haeds.

  3. Is there the other way of loading the newly-trained custom model and tokenizer from local directory?
    I only see the way of using push-to-hub method to load custom model while I was following tutorials.

Overall, I need some resources to study how to customize my own model and tokenizer.
I’d really appreciate any kind of related links.
FYI, I am torch user. :grinning:

Thx! :+1: