The goal is to train YOLO with multi-GPU. According to Darknet AlexeyAB, we should train YOLO with single GPU for 1000 iterations first, and then continue it with multi-GPU from
