tensorplay.distributed.distributed_c10d.all_to_all_single

tensorplay.distributed.distributed_c10d.all_to_all_single(output: TensorBase, input: TensorBase, output_split_sizes: List[int] | None = None, input_split_sizes: List[int] | None = None, group=None, async_op: bool = False)[source]

Splits input evenly (or by split sizes) and scatters the chunks.

Split sizes are in group-rank order, matching torch semantics.