Mlforecast trains the models globally. For example, I have the following examples, as shown: When I remove a couple of models, the output of the remaining model changes. Is it normal for the output to always change when multiple models are removed?

How big are the differences?