I can't comment on the dominance of either framework on production-level deployment, but when it comes to academic work, Pytorch is unequivocaly winning. There

I can't comment on the dominance of either framework on production-level deployment, but when it comes to academic work, Pytorch is unequivocaly winning. There were some studies showing the number of mentions of deep learning frameworks in the top ML conferences (ICLR, NAACL, ..etc) that demonstrated this. I'd expect that to trickle down to the industry in the coming years. There's Andrej Karpathy's (chief of AI at Tesla) endorsement too: I've been using PyTorch a few months now and I've never felt better. I have more energy. My skin is clearer. My eye sight has improved.