Method __init__

rl_engine/alignment/model_wrappers.py:116–127

(
        self,
        model: torch.nn.Module,
        *,
        freeze: bool = True,
        eval_mode: bool = True,
    )

    """Standard adapter for the frozen reference model used by KL penalties."""

    def __init__(
        self,
        model: torch.nn.Module,
        *,
        freeze: bool = True,
        eval_mode: bool = True,
    ):
        super().__init__(model)
        if freeze:
            self.freeze()
        if eval_mode:
            self.eval()

    def freeze(self) -> "ReferenceModelWrapper":
        for parameter in self.model.parameters():