Looking for advice about my training tool based on PyTorch

Hi all,
It has been a while since I switched from Keras and Tensorflow to PyTorch. During these time, I’ve received much help from many developers and researchers from community, and I also want to contribute to the PyTorch community.

I’ve completed serveral projects based on PyTorch, and I think PyTorch is a much more pythonic and user-friendly tool than Tensorflow. However, I need to repeat myself every time to write codes to fetch batches of data, forward prop, backward prop, compute loss, optimize model, it’s kind of annoying. As we all know, Keras provides a nice API to users, so I try to build a Keras-like tool on top of PyTorch to ease the pain. So here comes torchtools.

torchtools is a High-Level training API on top of PyTorch with many useful features to simplifiy the traing process for users. It was developed based on ideas from tnt and Keras. Although It works pretty good for me, I have limited resources and skills to expand it. I want to share it with community and get some advice about how to make it better.