Diff Coverage

                    "RowwiseParallel expects a DTensor from Linear outputs; "
                    f"got {type(outputs)}. If this is an unsupported module, extend I/O hooks."
                )
        if reduce_dtype is not None and tuple(outputs.placements) != tuple(output_layouts):
            local_output = _cast_fp_tensor(reduce_dtype, outputs.to_local())
            outputs = DTensor.from_local(local_output, outputs.device_mesh, outputs.placements)
        if tuple(outputs.placements) != tuple(output_layouts):
            outputs = outputs.redistribute(device_mesh, output_layouts)
        if use_local_output:

                    self.reduce_dtype is not None
                    and has_partial_output
                    and out_layout != desired_out_layout
                ):
                    local_out = _cast_fp_tensor(self.reduce_dtype, dt_out.to_local())
                    dt_out = DTensor.from_local(local_out, dt_out.device_mesh, dt_out.placements)
                if out_layout != desired_out_layout:
                    dt_out = dt_out.redistribute(device_mesh, (desired_out_layout,))
                prepared_outputs.append(