Nội dung này: SMSG: Profiling-Free Parallelism Modeling for Distributed Training of DNN