Comment by mingus88

And an LLM is just generating a token stream from a set of model weights