Train Everything, Everywhere, All at Once: A Version Control Convention for Training Multiple Deep Learning Models