5 ms·Matmul() using PyTorch's MPS back end is faster than Apple's MLX1 points by lonesword 1y agodrewbitt 1y agoI have noticed the same thing with PyTorch and MLX.