没有找到引用Characterization and prediction of deep learning workloads in large-scale gpu datacenters的文章。