Lightning-AI/pytorch-lightning · error · NotImplementedError

The `{type(self).__name__}` does not use the `CheckpointIO`

Error message

The `{type(self).__name__}` does not use the `CheckpointIO` plugin interface.

What it means

ModelParallelStrategy does not use the CheckpointIO plugin interface: both the checkpoint_io getter and setter raise NotImplementedError by design, because checkpointing for this strategy is handled through a different mechanism (state-dict based saving/loading via the strategy itself). Any generic code path that reads or assigns strategy.checkpoint_io will hit this.

Source

Thrown at src/lightning/fabric/strategies/model_parallel.py:126

        self._tensor_parallel_size = tensor_parallel_size
        self._num_nodes = 1
        self._save_distributed_checkpoint = save_distributed_checkpoint
        self._process_group_backend: Optional[str] = process_group_backend
        self._timeout: Optional[timedelta] = timeout
        self._backward_sync_control = _ParallelBackwardSyncControl()

        self._device_mesh: Optional[DeviceMesh] = None

    @property
    def device_mesh(self) -> "DeviceMesh":
        if self._device_mesh is None:
            raise RuntimeError("Accessing the device mesh before processes have initialized is not allowed.")
        return self._device_mesh

    @property
    @override
    def checkpoint_io(self) -> CheckpointIO:
        raise NotImplementedError(f"The `{type(self).__name__}` does not use the `CheckpointIO` plugin interface.")

    @checkpoint_io.setter
    @override
    def checkpoint_io(self, io: CheckpointIO) -> None:
        raise NotImplementedError(f"The `{type(self).__name__}` does not support setting a `CheckpointIO` plugin.")

    @property
    @override
    def root_device(self) -> torch.device:
        assert self.parallel_devices is not None
        return self.parallel_devices[self.local_rank]

    @property
    def num_nodes(self) -> int:
        return self._num_nodes

    @num_nodes.setter
    def num_nodes(self, num_nodes: int) -> None:

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Remove any code that sets or reads checkpoint_io when using ModelParallelStrategy; rely on fabric.save/load
2. For custom storage backends, serialize via the strategy's state_dict hooks instead of a CheckpointIO plugin
3. If you need a pluggable CheckpointIO, use a strategy that supports it (e.g. DDPStrategy/FSDPStrategy) or subclass and override the property knowingly

Example fix

# before
strategy = ModelParallelStrategy()
strategy.checkpoint_io = MyCheckpointIO()  # NotImplementedError

# after
strategy = ModelParallelStrategy()
fabric = Fabric(strategy=strategy)
fabric.save("ckpt.path", state)  # state-dict based, no CheckpointIO plugin

from lightning.fabric.strategies import ModelParallelStrategy
uses_checkpoint_io = not isinstance(strategy, ModelParallelStrategy)
if uses_checkpoint_io:
    strategy.checkpoint_io = my_io

def supports_checkpoint_io(strategy) -> bool:
    try:
        _ = strategy.checkpoint_io
        return True
    except NotImplementedError:
        return False

try:
    strategy.checkpoint_io = io
except NotImplementedError:
    pass  # strategy handles checkpointing via state_dict path

When it happens

Trigger: Calling strategy.checkpoint_io or assigning strategy.checkpoint_io = ... on ModelParallelStrategy; generic tooling (profiling, checkpoint wrappers, older Lightning abstractions) that unconditionally accesses the property on any Strategy instance.

Common situations: Porting code from DDP/FSDP strategies that set a custom CheckpointIO (e.g. fsspec or deepspeed-style IO plugins); shared utilities that iterate strategies and touch checkpoint_io; version upgrades where the base-class contract changed to require the property.

Related errors

• The `CSVLogger` does not yet support logging hyperparameters
• Loading a single optimizer object from a checkpoint is not s
• Accessing the device mesh before processes have initialized
• __setitem__ is not supported
• Accessing the device mesh before processes have initialized

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/6c38ed7943b19d12. Report an issue: GitHub.