he ``torch.futures.Future``.
        This is the preferred way of constructing Work objects when writing a custom ProcessGroup
        in python.
        Example::
            >>> class SingleRankProcessGroup(torch.distributed.ProcessGroup):
            >>>     def broadcast(self, tensor_list, opts):
            >>>         fut = torch.futures.Future()
            >>>         fut.set_result(tensor_list)
            >>>         return torch._C._distributed_c10d._create_work_from_future(fut)
        .. warning ::
            This API is experimental and subject to change.
            The returned Work object has multiple limitations:
            - synchronize() does nothing. Use ``torch.futures.Future`` based synchronization.
            - wait() ignored timeout argument.
            - sourceRank() raises.
            - abort() raises.
            The provided Future object result must be a Tensor or a list of Tensors.