Comment by alainrk

13 hours ago

Yeah but energy is still a cost, and local inference without batch and multiplexing for many users (like an office would be) is even less optimal. I'm not sure I think it's just pushing the problem forward