I'm know the TensorFlow API really well considering I've been using it for about 2 years now.

I've used PyTorch before as well, its really not hard to learn the other once you're good at one of them.

PyTorch is a little more verbose than TensorFlow , e.g. TensorFlow has a default training loop that can be started by invoking the fit method where as in PyTorch you write the training loop yourself regardless, you always have to explicitly call methods for AD/Back propagation as well in PyTorch.