Multi-GPU Programming

This paper extended Mosaic Flows to scale training and inference to distributed GPUs. This greatly improved the training time.