Google DeepMind recently released a paper that massively improves the viability of democratised AI training by showing that it's possible to train a large scale model on distributed and varied hardware.
We've built and improved upon it - Read more below