Pytorch Lightning

zenml.integrations.pytorch_lightning special

Initialization of the PyTorch Lightning integration.

PytorchLightningIntegration (Integration)

Definition of PyTorch Lightning integration for ZenML.

Source code in zenml/integrations/pytorch_lightning/__init__.py

class PytorchLightningIntegration(Integration):
    """Definition of PyTorch Lightning integration for ZenML."""

    NAME = PYTORCH_L
    REQUIREMENTS = ["pytorch_lightning"]

    @classmethod
    def activate(cls) -> None:
        """Activates the integration."""
        from zenml.integrations.pytorch_lightning import materializers  # noqa

activate() classmethod

Activates the integration.

Source code in zenml/integrations/pytorch_lightning/__init__.py

@classmethod
def activate(cls) -> None:
    """Activates the integration."""
    from zenml.integrations.pytorch_lightning import materializers  # noqa

materializers special

Initialization of the PyTorch Lightning Materializer.

pytorch_lightning_materializer

Implementation of the PyTorch Lightning Materializer.

PyTorchLightningMaterializer (BaseMaterializer)

Materializer to read/write PyTorch models.

Source code in zenml/integrations/pytorch_lightning/materializers/pytorch_lightning_materializer.py

class PyTorchLightningMaterializer(BaseMaterializer):
    """Materializer to read/write PyTorch models."""

    ASSOCIATED_TYPES = (Trainer,)
    ASSOCIATED_ARTIFACT_TYPES = (ModelArtifact,)

    def handle_input(self, data_type: Type[Any]) -> Trainer:
        """Reads and returns a PyTorch Lightning trainer.

        Args:
            data_type: The type of the trainer to load.

        Returns:
            A PyTorch Lightning trainer object.
        """
        super().handle_input(data_type)
        return Trainer(
            resume_from_checkpoint=os.path.join(
                self.artifact.uri, CHECKPOINT_NAME
            )
        )

    def handle_return(self, trainer: Trainer) -> None:
        """Writes a PyTorch Lightning trainer.

        Args:
            trainer: A PyTorch Lightning trainer object.
        """
        super().handle_return(trainer)
        trainer.save_checkpoint(
            os.path.join(self.artifact.uri, CHECKPOINT_NAME)
        )

handle_input(self, data_type)

Reads and returns a PyTorch Lightning trainer.

Parameters:

Name	Type	Description	Default
data_type	Type[Any]	The type of the trainer to load.	required

Returns:

Type	Description
Trainer	A PyTorch Lightning trainer object.

Source code in zenml/integrations/pytorch_lightning/materializers/pytorch_lightning_materializer.py

def handle_input(self, data_type: Type[Any]) -> Trainer:
    """Reads and returns a PyTorch Lightning trainer.

    Args:
        data_type: The type of the trainer to load.

    Returns:
        A PyTorch Lightning trainer object.
    """
    super().handle_input(data_type)
    return Trainer(
        resume_from_checkpoint=os.path.join(
            self.artifact.uri, CHECKPOINT_NAME
        )
    )

handle_return(self, trainer)

Writes a PyTorch Lightning trainer.

Parameters:

Name	Type	Description	Default
trainer	Trainer	A PyTorch Lightning trainer object.	required

Source code in zenml/integrations/pytorch_lightning/materializers/pytorch_lightning_materializer.py

def handle_return(self, trainer: Trainer) -> None:
    """Writes a PyTorch Lightning trainer.

    Args:
        trainer: A PyTorch Lightning trainer object.
    """
    super().handle_return(trainer)
    trainer.save_checkpoint(
        os.path.join(self.artifact.uri, CHECKPOINT_NAME)
    )