Lightning-AI/pytorch-lightning · error · NotImplementedError

Unsupported {cls}

Error message

Unsupported {cls}

What it means

NotImplementedError raised at the end of _load_from_checkpoint when the class being loaded is neither a pl.LightningModule nor a pl.LightningModule subclass handled by the earlier branches. The function only knows how to instantiate and hydrate Lightning modules; anything else falls through to this raise.

Source

Thrown at src/lightning/pytorch/core/saving.py:109

    # for past checkpoint need to add the new key
    checkpoint.setdefault(cls.CHECKPOINT_HYPER_PARAMS_KEY, {})
    # override the hparams with values that were passed in
    checkpoint[cls.CHECKPOINT_HYPER_PARAMS_KEY].update(kwargs)

    if issubclass(cls, pl.LightningDataModule):
        return _load_state(cls, checkpoint, **kwargs)
    if issubclass(cls, pl.LightningModule):
        model = _load_state(cls, checkpoint, strict=strict, **kwargs)
        state_dict = checkpoint["state_dict"]
        if not state_dict:
            rank_zero_warn(f"The state dict in {checkpoint_path!r} contains no parameters.")
            return model

        device = next((t for t in state_dict.values() if isinstance(t, torch.Tensor)), torch.tensor(0)).device
        assert isinstance(model, pl.LightningModule)
        return model.to(device)

    raise NotImplementedError(f"Unsupported {cls}")


def _default_map_location(storage: "UntypedStorage", location: str) -> Optional["UntypedStorage"]:
    if (
        location.startswith("mps")
        and not MPSAccelerator.is_available()
        or location.startswith("cuda")
        and not CUDAAccelerator.is_available()
        or location.startswith("xla")
        and not XLAAccelerator.is_available()
    ):
        return storage.cpu()
    return None  # default behavior by `torch.load()`


def _load_state(
    cls: Union[type["pl.LightningModule"], type["pl.LightningDataModule"]],
    checkpoint: dict[str, Any],

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Make the class inherit from lightning.pytorch.LightningModule
2. Or load manually with torch.load(ckpt) and model.load_state_dict(state_dict)
3. Check that cls is the class you intended (not accidentally the metaclass or a factory)

Example fix

// before
class MyModel(nn.Module): ...
model = MyModel.load_from_checkpoint(ckpt)
// after
class MyModel(pl.LightningModule): ...
model = MyModel.load_from_checkpoint(ckpt)

import lightning.pytorch as pl
assert issubclass(cls, pl.LightningModule), "load_from_checkpoint requires a LightningModule"

def is_lightning_module(cls) -> bool:
    import lightning.pytorch as pl
    return isinstance(cls, type) and issubclass(cls, pl.LightningModule)

try:
    model = cls.load_from_checkpoint(ckpt)
except NotImplementedError:
    state = torch.load(ckpt, map_location="cpu")["state_dict"]
    model = cls(); model.load_state_dict(state)

When it happens

Trigger: Calling load_from_checkpoint where cls is a plain nn.Module, a Callback, or a class that doesn't inherit from LightningModule; also if a custom __init__ returns an object of a different type via a patched constructor.

Common situations: Trying to reuse the checkpoint-loading convenience API on non-Lightning classes, or refactoring a LightningModule into a plain module while old loader code remains.

Related errors

• The `{type(self).__name__}` does not use the `CheckpointIO`
• The `{type(self).__name__}` does not support setting a `Chec
• Loading a single optimizer object from a checkpoint is not s
• Could not find a distributed model in the provided checkpoin
• Found multiple distributed models in the given state. Loadin

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/3d4d0e1d46a86cdb. Report an issue: GitHub.