[ashirviskas]

I wonder whether this could help optimize AI inference and training engines or are they already squeezed to the max. llama.cpp could potentially be a great target.