An event flowchart is a way to visualize the different steps in a process or event. As illustrated in this event flowchart, alpa automatically creates model-parallel training plans that can scale out complex deep learning models on distributed computing devices. It views parallelisms as two hierarchical levels: inter-operator and intra-operator parallelisms, and uses this information to create an optimized hierarchical space for massive model-parallel execution plans. This event flowchart provides an efficient method of understanding and tracking the progress of model-parallel training across multiple devices.
