mindscience.distributed.mappings.all_to_all_from_sequence_to_hidden

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

Performs an all-to-all from sequence layout to hidden layout.

Parameters
  • x (Tensor) – Input tensor in sequence layout.

  • group (Union[CommGroup, CommGroupBase]) – Communication group for the operation.

Returns

Tensor transformed from sequence layout to hidden layout.