3 ms·Matmul() using PyTorch's MPS back end is faster than Apple's MLX1 points by lonesword 2y agodrewbitt 2y agoI have noticed the same thing with PyTorch and MLX.