From the above, which of the following have been used by Alpa to generate parallel execution plans $?($A$)$ Intra$-$operator parallelism may incur device idle time, but is placed to devices with high$-$rate interconnection bandwidth; $($B$)$ It partitions the compu$-$ tation graph of the model into stages, which are assigned to device meshes that a cluster is partitioned into; $($C$)$ Inter$-$operator parallelism leads to better device utilization at distant devices; $($D$)$ The required communication volume in intra$-$operator parallelism can be much less than the collective communication in inter$-$operator parallelism.