Hi, I was told that pytorch is for research purpose and tensorflow for developme

https://www.facebook.com/groups/ComputerVisionGroup/permalink/2045159488961575/

John Pokhrel

Hi,
I was told that pytorch is for research purpose and tensorflow for development. Need some clarrification because I have seen people using pytorch too in development. Can't we do all tasks in pytorch which can be done in tensorflow? Give your views

When it comes to deployment in embedded devices, you need to convert pytorch model to tensorflow using third party software.
So, you can do all tasks in pytorch but tensorflow is needed when it comes to deployment in devices like android, raspberry pie, etc

This used to be more true than it is now.

Tensorflow didn't have "eager mode" which made it hard to interactively debug research (it does have an eager mode now).

Pytorch didn't have static graphs or a c++ or mobile api which made it harder to deploy in production in some cases (it does now have static "jit" graphs and c++ and mobile APIs).

There are still some differences, eg some researchers prefer the numpy-like api of pytorch. I think Tensorflow has a Javascript api.

The differences are much smaller today and a lot of it comes down to personal preference.