We have said it before, and we will say it again right here: If you can make a matrix math engine that runs the PyTorch framework and the Llama large