DoublePrecisionPlugin¶

class lightning.pytorch.plugins.precision.DoublePrecisionPlugin[source]¶

Bases: PrecisionPlugin

Plugin for training with double (torch.float64) precision.

connect(model, optimizers, lr_schedulers)[source]¶

Converts the model to double precision and wraps it in a LightningDoublePrecisionModule to convert incoming floating point data to double (torch.float64) precision.

Does not alter optimizers or lr_schedulers.

Return type: Tuple[Module, List[Optimizer], List[Any]]

forward_context()[source]¶

A context manager to change the default tensor type.

See: torch.set_default_tensor_type()

Return type: Generator[None, None, None]