Exemplo n.º 1
0
def _bufferize_torch_parameter(worker: AbstractWorker,
                               param: torch.nn.Parameter) -> ParameterPB:
    protobuf_param = ParameterPB()
    set_protobuf_id(protobuf_param.id, param.id)
    protobuf_param.tensor.CopyFrom(
        syft.serde.protobuf.serde._bufferize(worker, param.data))
    protobuf_param.requires_grad = param.requires_grad
    if param.grad:
        protobuf_param.grad.CopyFrom(
            syft.serde.protobuf.serde._bufferize(worker, param.grad))
    return protobuf_param
Exemplo n.º 2
0
def _bufferize_torch_tensor(worker: AbstractWorker,
                            tensor: torch.Tensor) -> bin:
    """
    This function converts a Torch tensor into a serialized tensor
    using Protobuf. Depending on the worker's serializer, the tensor
    contents may be serialized to binary representations using Torch
    or Numpy, or to a generic inner Protobuf message for cross-platform
    communication.

    Args:
        tensor (torch.Tensor): an input tensor to be serialized

    Returns:
        protobuf_obj: Protobuf version of torch tensor.
    """
    serialized_tensor = _serialize_tensor(worker, tensor)

    if tensor.grad is not None:
        if hasattr(tensor, "child"):
            if isinstance(tensor.child, PointerTensor):
                grad_chain = None
            else:
                grad_chain = _bufferize_torch_tensor(worker, tensor.grad)
        else:
            grad_chain = _bufferize_torch_tensor(worker, tensor.grad)

    else:
        grad_chain = None

    chain = None
    if hasattr(tensor, "child"):
        chain = syft.serde.protobuf.serde._bufferize(worker, tensor.child)

    protobuf_tensor = TorchTensorPB()
    set_protobuf_id(protobuf_tensor.id, tensor.id)

    protobuf_tensor.serializer = SERIALIZERS_SYFT_TO_PROTOBUF[
        worker.serializer]
    if worker.serializer == TENSOR_SERIALIZATION.ALL:
        protobuf_tensor.contents_data.CopyFrom(serialized_tensor)
    else:
        protobuf_tensor.contents_bin = serialized_tensor

    if chain:
        protobuf_tensor.chain.CopyFrom(chain)
    if grad_chain:
        protobuf_tensor.grad_chain.CopyFrom(grad_chain)
    if tensor.description:
        protobuf_tensor.description = tensor.description

    protobuf_tensor.tags.extend(tensor.tags)

    return protobuf_tensor
Exemplo n.º 3
0
    def bufferize(worker: AbstractWorker, param: torch.nn.Parameter) -> ParameterPB:
        """
            This method converts a torch.nn.Parameter into a serialized parameter using ParameterPB.

            Args:
                param (torch.nn.Parameter): input nn.parameter to be serialized.

            Returns:
                protobuf_param: serialized torch.nn.Parameter.
        """
        protobuf_param = ParameterPB()
        set_protobuf_id(protobuf_param.id, param.id)
        protobuf_param.tensor.CopyFrom(syft.serde.protobuf.serde._bufferize(worker, param.data))
        protobuf_param.requires_grad = param.requires_grad
        if param.grad:
            protobuf_param.grad.CopyFrom(syft.serde.protobuf.serde._bufferize(worker, param.grad))
        return protobuf_param