Lstm architecture

help me understand how to make lstm architecture.
My English is not so good as to understand the documentation correctly))
I understand that it should not be so difficult, but I can not understand the principle.

I prepared a file in which I drew my diagram and questions.

I can not understand what is hidden_size. Maybe someone can draw on a simple example?

Help someone make the right LSTM architecture
I don’t understand how to make hidden_size, out