You don’t seem to be very familiar with Keras/Tensorflow and their relationship — especially in TF2 where one is the high level and the later the low level. TF2 and PyT are practically the same thing now at the API layer and much of what you wrote as PyTorch is essentially the same thing you’d write in Keras — so much so I thought maybe you had misquoted your PyTorch example.

That PyTorch has gained momentum is true, but that’s moreso due to the ability to more directly transfer papers to practice code. Otherwise they are effectively siblings and behave very similarly.