I am sincerely hoping that anyone working with LLMs has already figured out that they need lot of memory at the very least and perhaps the fastest cores they can get their hands on.