Inference is what you do with the model. It's not an executable, so you don't run a model directly, instead a separate program runs and that program reads from the model. Also, hosting can be used to mean running inference on a model, but it could also be used to mean just storing the files and possibly making them available for download, so it's a bit more vague.