Module bastionlab.torch.client

Classes

BastionLabTorch()

BastionLab Torch RPC Handle

Methods

RemoteDataset(self, *args, **kwargs) ‑> bastionlab.torch.RemoteDataset

Returns a RemoteDataset object encapsulating a training and testing dataloaders on the remote server that uses this client to communicate with the server.

Args: *args: all arguments are forwarded to the bastionlab.torch.RemoteDataset constructor. **kwargs: all keyword arguments are forwarded to the bastionlab.torch.RemoteDataset constructor.

RemoteLearner(self, *args, **kwargs) ‑> bastionlab.torch.RemoteLearner

Returns a bastionlab.torch.RemoteLearner object encapsulating a model and hyperparameters for training and testing on the remote server and that uses this client to communicate with the server.

Args: *args: all arguments are forwarded to the bastionlab.torch.RemoteLearner constructor. **kwargs: all keyword arguments are forwarded to the bastionlab.torch.RemoteLearner constructor.

RemoteTensor(self, tensor: torch.Tensor) ‑> bastionlab.torch.RemoteTensor

Returns a RemoteTensor which represents a reference to the uploaded tensor.

Args: tensor: The tensor to be uploaded.

delete_dataset(self, ref: Union[ForwardRef('bastionlab.torch.RemoteDataset'), bastionlab_pb2.Reference])

Deletes the dataset correponding to the given ref reference on the BastionLab Torch server.

Args: ref: BastionLab Torch gRPC protocol reference of the dataset to be deleted.

delete_module(self, ref: bastionlab_pb2.Reference) ‑> None

Deletes the module correponding to the given ref reference on the BastionLab Torch server.

Args: ref: BastionLab Torch gRPC protocol reference of the module to be deleted.

fetch_dataset(self, ref: Union[ForwardRef('bastionlab.torch.RemoteDataset'), bastionlab_pb2.Reference]) ‑> bastionlab.torch.utils.TensorDataset

Fetches the distant dataset with a BastionLab Torch gRPC protocol reference.

Args: ref: BastionLab Torch gRPC protocol reference object corresponding to the distant dataset.

Returns: A dataset instance built from received data.

fetch_model_weights(self, model: torch.nn.modules.module.Module, ref: bastionlab_pb2.Reference) ‑> None

Fetches the weights of a distant trained model with a BastionLab Torch gRPC protocol reference and loads the weights into the passed model instance.

Args: model: The Pytorch's nn.Module whose weights will be replaced by the fetched weights. ref: BastionLab Torch gRPC protocol reference object corresponding to the distant trained model.

get_available_datasets(self) ‑> List[bastionlab_pb2.Reference]

Returns the list of BastionLab Torch gRPC protocol references of all datasets on the server.

get_available_devices(self) ‑> List[str]

Returns the list of devices available on the server.

get_available_models(self) ‑> List[bastionlab_pb2.Reference]

Returns the list of BastionLab Torch gRPC protocol references of all available models on the server.

get_available_optimizers(self) ‑> List[str]

Returns the list of optimizers supported by the server.

get_metric(self, run: bastionlab_pb2.Reference) ‑> bastionlab_torch_pb2.Metric

Returns the value of the metric associated with the given run reference.

Args: run: BastionLab Torch gRPC protocol reference of the run whose metric is read.

send_dataset(self, dataset: torch.utils.data.dataset.Dataset, name: str, description: str = '', privacy_limit: Optional[float] = None, chunk_size: int = 4194285, batch_size: int = 1024, train_dataset: Optional[bastionlab_pb2.Reference] = None, progress: bool = False) ‑> bastionlab_pb2.Reference

Uploads a Pytorch Dataset to the BastionLab Torch server.

Args: model: The Pytorch Dataset to upload. name: A name for the dataset being uploaded. description: A string description of the dataset being uploaded. chunk_size: Size of a chunk in the BastionLab Torch gRPC protocol in bytes. batch_size: Size of a unit of serialization in number of samples, increasing this value may increase serialization throughput at the price of a higher memory consumption. train_dataset: metadata, True means this dataset is suited for training, False that it should be used for testing/validating only

Returns: BastionLab Torch gRPC protocol's reference object.

send_model(self, model: torch.nn.modules.module.Module, name: str, description: str = '', chunk_size: int = 4194285, progress: bool = False) ‑> bastionlab_pb2.Reference

Uploads a Pytorch module to the BastionLab Torch server.

This endpoint transforms Pytorch modules into TorchScript modules and sends them to the BastionLab Torch server over gRPC.

Args: model: The Pytorch nn.Module to upload. name: A name for the module being uploaded. description: A string description of the module being uploaded. chunk_size: Size of a chunk in the BastionLab Torch gRPC protocol in bytes.

Returns: BastionLab Torch gRPC protocol's reference object.