Quick question: what would you estimate the running cost of Llama 2 70b to be? (On GPU, and assuming maximum utilization)?