mindscience.distributed.mappings.scatter_to_hidden

mindscience.distributed.mappings.scatter_to_hidden(x, group)[source]

Scatters tensors into hidden partitions along the last dimension.

Parameters

x (Tensor) – Input tensor to be scattered into hidden partitions.
group (Union[CommGroup, CommGroupBase]) – Communication group for the operation.

Returns

Tensor partition corresponding to the current rank.