All of ComfyUI runs on pytorch, and pytorch barely touches that new silicon avaialbe in M5. Apple's MLX does: same clip, about twice as fast. So I wrote the missing MLX audio node. A four second…