No longer the upstart nipping at TensorFlow’s heels, PyTorch is a major force in the deep learning world today, perhaps primarily for research, but also in production applications more and more.

TensorFlow is your ally for scalability and production. PyTorch is your friend for research flexibility and ease of use. The choice depends on your project needs, expertise, and long-term goals.

If this is what matters most for you, then your choice is probably TensorFlow. A network written in PyTorch is a Dynamic Computational Graph (DCG). It allows you to do any crazy thing you want to do.

PyTorch introduced "Torchscript" and a JIT compiler, whereas TensorFlow announced that it would be moving to an "eager mode" of execution starting from version 2.0.