Lightning-AI/pytorch-lightning · error · MisconfigurationException

`{self.__class__.__name__}` should have been `setup` with a

Error message

`{self.__class__.__name__}` should have been `setup` with a `CombinedLoader`.

What it means

DataFetcher.combined_loader raises MisconfigurationException when its _combined_loader attribute is None, i.e. setup(combined_loader) was never called (or reset) before the property was accessed. The fetcher must be set up by the training/eval loop before iteration.

Source

Thrown at src/lightning/pytorch/loops/fetchers.py:42

def _profile_nothing() -> None:
    pass


class _DataFetcher(Iterator):
    def __init__(self) -> None:
        self._combined_loader: Optional[CombinedLoader] = None
        self.iterator: Optional[Iterator] = None
        self.fetched: int = 0
        self.done: bool = False
        self.length: Optional[int] = None
        self._start_profiler = _profile_nothing
        self._stop_profiler = _profile_nothing

    @property
    def combined_loader(self) -> CombinedLoader:
        if self._combined_loader is None:
            raise MisconfigurationException(
                f"`{self.__class__.__name__}` should have been `setup` with a `CombinedLoader`."
            )
        return self._combined_loader

    def setup(self, combined_loader: CombinedLoader) -> None:
        self._combined_loader = combined_loader

    @override
    def __iter__(self) -> "_DataFetcher":
        self.iterator = iter(self.combined_loader)
        self.reset()
        return self

    @override
    def __next__(self) -> _ITERATOR_RETURN:
        assert self.iterator is not None
        self._start_profiler()
        try:

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Call fetcher.setup(CombinedLoader(dataloaders)) before accessing combined_loader
2. Let the Trainer own fetcher setup — don't access it before trainer.state is RUNNING
3. If writing a custom loop, mirror the setup order used in lightning's fit_loop

Example fix

# before
fetcher = _DataFetcher()
cl = fetcher.combined_loader  # raises
# after
from lightning.pytorch.loops.utilities import _DataFetcher
from lightning.pytorch.overrides.distributed import CombinedLoader  # illustrative
fetcher.setup(CombinedLoader(dataloader))
cl = fetcher.combined_loader

if fetcher._combined_loader is None:
    fetcher.setup(CombinedLoader(train_dataloader))

When it happens

Trigger: Accessing fetcher.combined_loader on a freshly constructed _DataFetcher (or after resetting) without calling fetcher.setup(loader); typically from custom loop code that subclasses or drives the fetcher manually.

Common situations: Custom training loops or tests that instantiate a fetcher and skip the Trainer's setup sequence; also calling internal loop APIs outside a running fit/validate.

Related errors

• No models were set up for backward. Did you forget to call `
• You need to set up the model first before you can call `fabr
• f"`{hook_name}` is not a shared hook within `LightningModule
• Received multiple values for {', '.join(duplicated_plugin_ke
• Received both `precision={precision_input}` and `plugins={se

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/b2e7cb0575bca496. Report an issue: GitHub.