Skip to main content

model_reference

References to custom models.

Classes​

BitfountModelReference​

class BitfountModelReference(    model_ref: Union[Path, str],    datastructure: Optional[DataStructure] = None,    model_version: Optional[int] = None,    schema: Optional[BitfountSchema] = None,    username: Optional[str] = None,    hub: Optional[BitfountHub] = None,    hyperparameters: Optional[_StrAnyDict] = None,    private: bool = False,    new_version: bool = False,    secrets: Optional[Union[APIKeys, ExternallyManagedJWT]] = None,    weights: Optional[Union[Path, str]] = None,    model_description: Optional[str] = None,):

Describes a local or remote reference to a model implementing ModelProtocol.

tip

To use another user's custom model, simply provide that user's username instead of your own (along with the name of the model as the model_ref argument).

Arguments

  • datastructure: DataStructure to be passed to the model when initialised. This is an optional argument as it is only required for get_model to perform validation on the model before uploading it to the hub. Ensure that you provide this argument if you want to use get_model to upload your model.
  • hub: Required for upload/download of model. This attribute is set after initialisation on the worker side as the hub is not serialized. Defaults to None.
  • hyperparameters: Hyperparameters to be passed to the model constructor after it has been loaded from file or hub. Defaults to None.
  • model_ref: Either path to model file or name of model on hub.
  • model_version: The version of the model you wish to use. Defaults to the latest version.
  • new_version: Whether to upload a new version of the model to the hub. Defaults to False.
  • private: Boolean flag to set the model to be private to control useage or publicly accessible to all users.
  • schema: The BitfountSchema object associated with the datasource on which the model will be trained on.
  • secrets: The secrets to use when creating a BitfountHub instance. Defaults to None.
  • username: The username of the model owner. Defaults to bitfount session username if not provided.

Attributes

  • fields_dict: A dictionary mapping all attributes that will be serialized in the class to their marshmallow field type. (e.g. fields_dict = {"class_name": fields.Str()}).
  • nested_fields: A dictionary mapping all nested attributes to a registry that contains class names mapped to the respective classes. (e.g. nested_fields = {"datastructure": datastructure.registry})

Raises

  • ValueError: If username is not provided and hub is not provided.

Ancestors​

  • bitfount.models.base_models._BaseModelRegistryMixIn
  • bitfount.types._BaseSerializableObjectMixIn

Variables​

  • model_id : str - The full model ID (or model slug) for a model.

    This is of the form "owner_name/model_name:model_version".

  • model_name : str - The name of the model being referenced.

    If the model referenced is one already hosted in the hub then this will be the model name it was registered with.

    If the model referenced is a local file, to be uploaded to the hub, then this is the file stem (e.g. MyModel.py has name MyModel).

Static methods​

from_model_ref_config​

def from_model_ref_config(    config: BitfountModelReferenceConfig,    *,    hub: Optional[BitfountHub] = None,    secrets: Optional[Union[APIKeys, ExternallyManagedJWT]] = None,    new_version: bool = False,    private: bool = False,    datastructure: Optional[DataStructure] = None,    schema: Optional[BitfountSchema] = None,    hyperparameters: Optional[_StrAnyDict] = None,) ‑> BitfountModelReference:

Builds a BitfountModelReference instance from config.

Methods​

get_model_from_hub​

def get_model_from_hub(    self, project_id: Optional[str] = None,) ‑> type[ModelProtocol]:

Gets the model referenced.

If the model is a Path to a ModelProtocol, it will upload it to BitfountHub and return the model class. If it is a name of a model on the hub, it will download the model from the hub and return the model class.

If the model class is a v1 PyTorchBitfountModel it will be automatically converted to v2.

Returns The model class (auto converted to pytorch v2 if necessary).

Raises

  • TypeError: If the model is not a Path or a string.
  • TypeError: If the model does not implement DistributedModelProtocol, InferrableModelProtocol or EvaluableModelProtocol.
  • ValueError: If a BitfountHub instance has not been provided or if there was a communication error with the hub.
  • ValueError: If a datastructure has not been provided.

get_weights​

def get_weights(self, project_id: Optional[str] = None) ‑> Optional[bytes]:

Gets weights file uploaded for the model if one exists.

Returns The weights file as a byte stream.

send_weights​

def send_weights(self, pretrained_file: Union[Path, str]) ‑> None:

Sends the model weights from a pretrained file to Hub.

Arguments

  • pretrained_file: The path to the pretrained model file.

Raises

  • ValueError: If model_version has not been set on BitfountModelReference instance.

upload_model_and_weights​

def upload_model_and_weights(    self, project_id: Optional[str] = None,) ‑> type[ModelProtocol]:

Uploads model and weights to the hub.

Should be used by the modeller to upload the model and weights to the hub at the beginning of a task.

Arguments

  • project_id: The project ID to upload the model to. Defaults to None.

Returns The model class (auto converted to PyTorchBitfountModelv2 if needed).