With the growing model size of deep neural networks (DNN), deep learning training is
increasingly relying on handcrafted search spaces to find efficient parallelization execution …