Performance using Resnets, googlenet, alexnet on Julia vs Python [ Both when training time and testing time]?

I am curious, about
How much time did it take to train and also on testing for both Julia and Python languages.
In networks such as googlenet, resnets, or Stylegan . :slight_smile:

Thank you for your answer.
Excited to see results as a motivation. :slight_smile:

It is an interesting question.

A small benchmark information there are in https://github.com/dfdx/Avalon.jl and http://denizyuret.github.io/Knet.jl/latest/tutorial/#Benchmarks-1
but with not so complex networks.

However, taking in account the performance of PyTorch and TensorFlow (and how much its code is actually in C++) I would guess that it is going to be worse than these famous frameworks.

Anyway, if any one have times/comparison, it should be always interesting to see it.

1 Like

They are pretty much same speed. Although pytorch or tensorflow bounded by c++ so they are fast.
But why people are not adopting julia (fast and easy) instead of these frameworks ??? :slight_smile:

I was searching for Pytorch.jl (official) mainted by official community.

I had heard facebook converting php to Julia.
I hope facebook will also switch from pytorch python to pytorch julia. And it will be easy for us. :slight_smile: