Comment by febed

Would this technique work with inference optimization techniques like DFlash and MTP? In other words does it slow down things to the point where its just better to retry?