Configuration class for logging settings. Attributes: enabled (bool): Whether logging is enabled. file_output (bool): Whethe

A dataclass representing a pair of contrasting strings. Attributes: positive: The positive string in the pair. negative: The

A wrapper layer that implements conditional activation steering for language models. This layer can be applied to existing model layers to e

Singleton class for global configuration settings. Class Attributes: console (Console): Rich console instance for pretty printing.

MalleableModel implements conditional activation steering for language models. This class wraps a pre-trained language model and provides me

Create a formatted dataset for steering a language model. This class takes a list of examples (either contrastive messages or contrastive te

A dataclass representing a steering vector used for guiding the language model. Attributes: model_type: The type of the model this v