[adastra22]

There is nothing magical about GPU memory though. It’s just faster. But people have been doing CPU inference since the first llama code came out.