"more papers use it" != "won"

I mean PyTorch is nicer and all that, and so much of the cutting edge stuff is built with it but TensorFlow is still used a lot in many companies.

Do you see any new challengers to PyTorch?