Method load_state_dict

imperative/python/megengine/module/module.py:495–574

r"""Loads a given dictionary created by :func:`state_dict` into this module. If ``strict`` is ``True``, the keys of :func:`state_dict` must exactly match the keys returned by :func:`state_dict`. Users can also pass a closure: ``Function[key: str, var: Tensor] -> Optional[np.

(
        self,
        state_dict: Union[dict, Callable[[str, Tensor], Optional[np.ndarray]]],
        strict=True,
    )

        return rst

    def load_state_dict(
        self,
        state_dict: Union[dict, Callable[[str, Tensor], Optional[np.ndarray]]],
        strict=True,
    ):
        r"""Loads a given dictionary created by :func:`state_dict` into this module.
        If ``strict`` is ``True``, the keys of :func:`state_dict` must exactly match the keys
        returned by :func:`state_dict`.

        Users can also pass a closure: ``Function[key: str, var: Tensor] -> Optional[np.ndarray]``
        as a `state_dict`, in order to handle complex situations. For example, load everything
        except for the final linear classifier:

        .. code-block::

            state_dict = {...}  #  Dict[str, np.ndarray]
            model.load_state_dict({
                k: None if k.startswith('fc') else v
                for k, v in state_dict.items()
            }, strict=False)

        Here returning ``None`` means skipping parameter ``k``.

        To prevent shape mismatch (e.g. load PyTorch weights), we can reshape before loading:

        .. code-block::

            state_dict = {...}
            def reshape_accordingly(k, v):
                return state_dict[k].reshape(v.shape)
            model.load_state_dict(reshape_accordingly)

        We can also perform inplace re-initialization or pruning:

        .. code-block::

            def reinit_and_pruning(k, v):
                if 'bias' in k:
                    M.init.zero_(v)
                if 'conv' in k:

        """
        unused = []
        if isinstance(state_dict, dict):
            unused = state_dict.keys()

            def closure(k, _):  # var unused
                return state_dict[k] if k in state_dict else None

        elif callable(state_dict):
            closure = state_dict
        else:
            raise ValueError(
                "`state_dict` must load a dict or callable, got {}".format(
                    type(state_dict)
                )
            )