PyTorch is a deep learning framework by Facebook, similar to TensorFlow by Google.
Being backed by Google, thousands of dollars have been spent on TensorFlow's marketing, development, and documentation. It also got to a stable 1.0 release almost a year ago, while PyTorch has only recently gotten to 0.4.1. This means that it's usually easier to find a TensorFlow solution to your problem and that you can copy and paste code off the internet.
On the other hand, PyTorch is programmer-friendly. It is semantically similar to NumPy and deep learning operations in one. This means that I can use the Python debugging tools that I am already familiar with.
Pythonic: TensorFlow worked like a C program in the sense that the code was all written ...