In fact you can use both options you mentioned:
1- you use one relational model and split it up on in several subviews. In your case a fact per subview could be a good solution.
2- you use one relational model per fact. To prevent duplicate entities among relational models, you can link tables instead of making copies. You can make a link of a table by dragging the table into the subview of another relational model.
You can of course mix both models by using multiple relational models and have in a relational model multiple views.
I use one sub-view per fact. I also do the overall model in the main diagram for the overview. That is helpful if you have conformed (shared) dimensions.
I would only use a separate relational model to represent different versions of the same logical model 0 such as dimensional vs 3NF views of the data or Oracle vs DB2 or SQL Server.