I have an AMD GPU and I'm running on Windows 10 with Python 3.8. Tensorflow didn't provide any support for AMD users in terms of GPU acceleration. I've found a way on StackOverflow describing how to run Keras on AMD GPU using plaid-ml that works perfectly on my machine (link below) but does not make the trick since Keras only take numpy array as input of models and not Tensors, particularly tf.data.Dataset as input.
Anyway to work with Keras in Mac with AMD GPU?
How can we feed a tf.data.Dataset as input of a Keras model with plaid-ml backend ? Or is there any way to build Tensorflow 2.x so that it uses AMD GPU on Windows 10 (and then no need to use Keras as a stand-alone) ? I have seen people forking Tensorflow repository for AMD users but the version was always Tensorflow 1.x.
Has someone ever overcome this issue ? Deep learning on CPU is pain !
