The Pytorch Toolkit is a library of functions and classes I wrote to provide a Keras-like interface to train & evaluate Pytorch models and make predictions. It works with Numpy data as well as torchvision datasets.
You can download it from my Github repository - https://github.com/mjbhobe/dl-pytorch (it includes examples as well)
I also include a Step-by-Step Tutorial to get you started quickly.
Hope you love using this as much as I loved writing it.
Thanks for posting your work here! 
Could you describe some advantages or use cases of your library compared to e.g. Lightning or Ignite?
To be honest, I have not used Lightening or Ignite, so I can’t really comment.
I came to Pytorch from a primary Keras over Tensorflow background. I hated using pure Tensorflow as you had to write a lot of verbose non-pythonic code. Keras was a welcome relief. Then I heard that Pytorch was more Pythonic in it’s approach, so though I’d give it a try. But to my dismay, I saw that yet again I had to write a lot of boilerplate code to train/test my DL models.
So to alleviate my problem, I wrote some re-useable code, primarily to avoid writing repeated boilerplate code. Though I’d share my work with the community in the hope that someone else also finds it useful. This is a very simple library (just 1 file of code), so I don’t doubt that Ignite or Lightening would be far more capable.
Ah OK, sounds like a reasonable approach
Maybe your library (or the other mentioned ones) could benefit from the design decisions made, so feel free if you think some kind of collaboration would be useful.
Well, there is also the fast.ai thing! 