Comment by andreinwald

It works on small Llama-3.2-1B model, specially for less powerfull GPU devices