About

This package is designed to make it easy to organize your machine learning projects and quickly implement a project idea. It is similar to something like PyTorch Lightning or Keras, but it is a bit more opinionated, making it easier to get a project off the ground.

The closest similar thing would be something like this, but I wanted to avoid using PyTorch Lightning or Hydra directly, since they have been a bit unweildy in my experience. I tried to keep the amount of magic to a minimum; I use OmegaConf for dealing with config files, and the framework expects a specific project structure so that it can do some caching for associating the objects referenced in the config file with their implementations, but other than that it’s just a bunch of standard PyTorch code.

I built this almost entirely for my own purposes, to address some of the pain points I had while using PyTorch Lightning and Fairseq. If you’re reading this then I hope you find it useful as well!