Efficient Parallelization Layouts for Large-Scale Distributed Model Training