I'm calling it - the current hype around PyTorch is overhyped. The syntax is clunky, the is fragmented, and don't even get me started on the autograd system. Give me TensorFlow's simplicity any day of the week.