I'm trying to apply TensorFlow-macOS to accelerate the fine-tune of hugging face transformer program on my Radeon Pro 5300M. However, there shows that "Cannot assign a device for operation", would it be possible to tell me why and how to solve this problem? issues: Version of TensorFlow-macOS is 2.5.0; I've tried to update into 2.6.0, and it've worked. However, it's extremely slower than the 2.5.0 one. It would be a better way to fix 2.5.0 according to the 2.6.0, but I don't know how to do.

Would it be possible to use GPU acceleration when training a Tensorflow model on macOS? How’s the performance when we training the same model on an Apple-chip platform?