Network graph view - tensorboard is only partly functional?

I wanted to understand StyleGAN2, and I do pytorch, so I grabbed a pytorch implementation and ran tensorboard - but I didn’t like what I saw; So I did the same for the original tensorflow repo.

  • Are there ways to get the cleanliness that one gets with tensorflow, using pytorch?
  • StyleGAN2 has a discriminator and generator, can I see both graphs in one environment?
  • is there an alternative to tensorboard that will make good graphs in pytorch, for more complex networks?
    I forked both repos, so this is all reproducible

StyleGAN2 Tensorflow Tensorboard


As you see the graphs are not nearly as neat - I only forked this implementation, I suppose I should build my own, as it has a few other problems that I havent yet sussed out, however if I do I would like to end up building a graph that looks as neat as what i see w tf. With tensorboard or other setup.
To be fair, I dont think that the tensorboard representation is clean enough either. I would like to see a functional graph - so I can evaluate or read the code of any of the nodes. I dont want to forgo the cleanliness of code, but to be able to view and encapsulate the code in a funcional graph would be grand.