Some frameworks have tried to do this automatically, but its extremely slow. You are better off just running whatever won't fit on the CPU (which is what llama.

Some frameworks have tried to do this automatically, but its extremely slow. You are better off just running whatever won't fit on the CPU (which is what llama.cpp does).