paper-conference

MegTaiChi: Dynamic tensor-based memory management optimization for DNN training