Hey @rafa-br34

It depends on what you are going for 😃 Let me give you options.

• If you just need some AI Services (ChatGPT, Claude, ...) you can use their APIs. (like that one). No special managing of anything, as these are pretty much just function calls.
• If you need one or several models that are already on Huggingface: I recommend using classes and inheritance to abstract the different implementations away, providing only a simple interface like "load()", "chat(...)", "save()".
• If you develop your own model and then use it in your code: Put the model on Huggingface, then apply the above. Note that individual Github repositories typically do not have enough storage for big AI models. (You shouldn't really go much beyond 1 GB.) So Huggingface is probably the place for the AI model.

If you want a real code example using option 2, you can have a look at my code here. It provides several different options to analyze PDF documents, using different Huggingface models for each option. A baseclass provides basic functionality and defines the interface.