A simple framework that takes inspiration from all the best outside. From an idea of Marco Cannici PyTorch trainer Mostly inspired from PyTorch Lightning, but better suited for our purposes.
We developed this library to help our students from Polimi (and ourselvelves) to stay organized when writing code and designing experiments. The goal is to have a non-monolithic code base that is easily debuggable and scalable. We have a seemless integration with Neptune AI to track and compare experiments.
- Distributed and Multi-gpu training
This is experimental code and it should be used for research purposes only. We don't provide any support for this repo. We strongly suggest you to consider the awesome PyTorch Lightning.