PyTorch is 10x easier to debug than even TF2, and it's been that way all along. TF2 is no easier to debug than the previous releases if you're not using eager mode (which most people don't), and even in eager mode it sometimes errors out in ways that do not offer any suggestion as to _which op_ caused the error. This is nuts. Modern architectures have hundreds, sometimes thousands of ops. It basically boils down to flying blind and guessing and can easy take days of trial and error to figure out each issue. Plus, every time you start a TF program it just sort of sits there for a minute or so before it starts doing anything. This severely hampers productivity when debugging.

To all the folks who are just starting out: just go with PyTorch. It's downright intuitive compared to anything Google has been able to put out so far.

Disclosure: ex-Googler. Used TF while there (and DistBelief before it). Gave it up as soon as PyTorch came out. Couldn't be happier.