There’s no need to run an LLM on the same system it was trained on. Once the model is built it contains all the information already. If you want a model to live on long term you would just release the file(s) publicly, like hugging face does with theirs, then anyone could use it or host an interface for it.