Sort:  

HuggingFace is like a Github for Machine Learning... You can fork many LLMs and Image/Voice generating AIs over there. (Some AI tools come with Licenses, though. You have to provide your contact data to Meta if you want to download Llama.)

It comes with different tiers of compute power. The Free tier has 2 Cours and 2GB RAM which might be enough to run small LLMs... I'm sure they get a LOT of paying costumers who want more compute power and RAM, don't you think?

I would imagine. There is a business model that I am sure they are profiting from.

But it is good that those who know how to set up small models can run them.