This is wrong in many ways. First off you don't necessarily need a high-end consumer GPU, look at llama.cpp. Or even better look at something like phi-2. It's

This is wrong in many ways. First off you don't necessarily need a high-end consumer GPU, look at llama.cpp. Or even better look at something like phi-2. It's likely to go even lower. I'm sure there are people here who can detail more specifics. Its 1/24 and we're already there more or less.