Lightning-AI/pytorch-lightning · error · TypeError

f"An invalid dataloader was passed to `Trainer.{trainer_fn.v

Error message

f"An invalid dataloader was passed to `Trainer.{trainer_fn.value}({prefix}dataloaders=...)`." f" Found {dataloader}." f" Either pass the dataloader to the `.{trainer_fn.value}()` method OR implement" f" `def {source.name}(self):` in your LightningModule/LightningDataModule."

What it means

Same iterable check as 558 but for the case where the non-iterable dataloader comes from a hook that IS overridden on the LightningModule or LightningDataModule: the object returned by (train_/val_/test_/predict_)_dataloader is not iterable (e.g., a bare HF dataset or numpy array), and Lightning tells you to fix the method's return value.

Source

Thrown at src/lightning/pytorch/trainer/connectors/data_connector.py:405

    source: _DataLoaderSource,
    trainer_fn: TrainerFn,
) -> None:
    if isinstance(dataloader, DataLoader):
        # Fast path: `torch.utils.data.DataLoader` is always iterable, calling iter() would be expensive
        return

    try:
        iter(dataloader)  # type: ignore[call-overload]
    except TypeError:
        # A prefix in the message to disambiguate between the train- and (optional) val dataloader that .fit() accepts
        prefix = "train_" if trainer_fn == TrainerFn.FITTING else ""
        if not source.is_module():
            raise TypeError(
                f"An invalid dataloader was passed to `Trainer.{trainer_fn.value}({prefix}dataloaders=...)`."
                f" Found {dataloader}."
            )
        if not is_overridden(source.name, source.instance):
            raise TypeError(
                f"An invalid dataloader was passed to `Trainer.{trainer_fn.value}({prefix}dataloaders=...)`."
                f" Found {dataloader}."
                f" Either pass the dataloader to the `.{trainer_fn.value}()` method OR implement"
                f" `def {source.name}(self):` in your LightningModule/LightningDataModule."
            )
        raise TypeError(
            f"An invalid dataloader was returned from `{type(source.instance).__name__}.{source.name}()`."
            f" Found {dataloader}."
        )


def _worker_check(trainer: "pl.Trainer", dataloader: object, name: str) -> None:
    if not isinstance(dataloader, DataLoader):
        return

    upper_bound = suggested_max_num_workers(trainer.num_devices)
    start_method = (
        dataloader.multiprocessing_context.get_start_method()

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Return a torch DataLoader from the hook: return DataLoader(self.dataset, batch_size=self.bs)
2. For HF datasets: return ds.to_iterable_dataset() (optionally with .with_format("torch"))
3. Return an iterable-style object implementing __iter__ (and __len__ where possible)

Example fix

# before
class M(LightningModule):
    def train_dataloader(self):
        return self.hf_dataset  # datasets.Dataset, not iterable
# after
class M(LightningModule):
    def train_dataloader(self):
        return DataLoader(self.hf_dataset, batch_size=32)
# or: return self.hf_dataset.to_iterable_dataset()

dl = model.train_dataloader()
try:
    iter(dl)
except TypeError:
    raise TypeError("train_dataloader() must return a DataLoader/iterable, wrap the dataset")

def hook_returns_iterable(obj) -> bool:
    try:
        iter(obj)
        return True
    except TypeError:
        return False

When it happens

Trigger: def train_dataloader(self): return self.hf_dataset (a datasets.Dataset) or return np.array(...) / return SomeDatasetClass; the method is detected as overridden but its return value fails iter().

Common situations: Porting sklearn/HF pipelines into LightningModule hooks; returning a Dataset object where a DataLoader or IterableDataset is required; forgetting to instantiate/wrap.

Related errors

• f"An invalid dataloader was passed to `Trainer.{trainer_fn.v
• You seem to have configured a sampler in your DataLoader whi
• `val_dataloader` must be implemented to be used with the Lig
• `predict_dataloader` must be implemented to be used with the
• Couldn't infer the batch indices fetched from your dataloade

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/178fd47a9fc7e08f. Report an issue: GitHub.