PyTorch has become bloated, its dynamic computation graph was a nice idea 5 years ago but now it's just a liability, TensorFlow is where it's at for serious production deployments.