It is correct that Deep Learning Toolbox training acceleration only works with NVIDIA GPUs.
Unfortunately your hardware and software is not compatible with training acceleration.
Last time I checked a few years ago, Mathworks was not actively working on making GPU work with Apple Silicon. They did investigate doing so, but unfortunately Apple does not provide detailed design documents for its GPUs, making it difficult to design efficient GPU operations. Furthermore, Apple has a distinct history of changing internal operations, so if Mathworks did manage to figure out how to use the GPU efficiently, then Apple could easily change GPU operations and say, "We never documented that it worked that way, so it is tough luck that you relied upon it."
