If you followed "https://developer.apple.com/metal/tensorflow-plugin/" to install TF, they you should be already taking advantage of the GPU power. The fact that your TF can see your GPU (device Metal) also confirmed that. How slow is your training process? For your reference, training the CIFAR-100 example in the above link takes 90s per epoch on my M1 Pro GPU. In my benchmarks, M1 Pro GPU is 5 - 7 times faster than M1 Pro CPU and should be around the level of an RTX 3050 Mobile for deep learning. It is not going to be as fast as a good desktop GPU.

Thanks. The problem is not the privacy setting which controls app access to a recognized device. The problem is that the VM doesn't recognize the external sound device (see attached screenshot below from the VM, no external microphone). The external microphone can be recognized by the host and used without a problem (host can see and use both built-in and external microphones). The built-in microphone, despite being recognized in the Sound Setting, doesn't record sound at all. I tested it without external monitor so this is not a result of the clamshell mode.

I don't think I understand the error message better. But here is what I did to make tensorflow working on my Macbook Pro M1 Pro and hopefully it helps. First I removed all previous Anaconda installation I had. See tutorial here (https://docs.anaconda.com/anaconda/install/uninstall/). Then I followed every single step in the link you attached (https://developer.apple.com/metal/tensorflow-plugin/). Make sure you use miniconda instead of anaconda, but bash or graphical installer should not matter. I tried anaconda and it didn't work well due to package conflicts. I was able to run the sample model in Step 4 without a problem.