ReduceType¶

class paddle.distributed. ReduceType ¶

指定分布式 Tensor 在 Partial 状态下规约操作的类型，必须是下述值之一：

ReduceType.kRedSum

ReduceType.kRedMax

ReduceType.kRedMin

ReduceType.kRedProd

ReduceType.kRedAvg

ReduceType.kRedAny

ReduceType.kRedAll

代码示例¶

          >>> import paddle
>>> import paddle.distributed as dist
>>> mesh = dist.ProcessMesh([0, 1], dim_names=["x"])
>>> a = paddle.ones([10, 20])
>>> # distributed tensor
>>> d_tensor = dist.shard_tensor(a, mesh, [dist.Partial(dist.ReduceType.kRedSum)])

         