Making large transformer based models like Jukebox [1] efficient enough so that they can be trained on a single 8x GPU machine to the same level of quality as t

Making large transformer based models like Jukebox [1] efficient enough so that they can be trained on a single 8x GPU machine to the same level of quality as the original model, in the same amount of time. [1] https://openai.com/blog/jukebox/ https://openai.com/blog/jukebox/