IS it possible for someone to give us the lower bound on the cost of running a 70B model in the cloud? How much memory does Llamba-2 take? What would it cost to

IS it possible for someone to give us the lower bound on the cost of running a 70B model in the cloud? How much memory does Llamba-2 take? What would it cost to fine tune it?