Lightning-AI/pytorch-lightning · error · RuntimeError

You provided only a single `{stage.dataloader_prefix}_datalo

Error message

You provided only a single `{stage.dataloader_prefix}_dataloader`, but have included `dataloader_idx` in `{type(pl_module).__name__}.{hook}()`. Either remove the argument or give it a default value i.e. `dataloader_idx=0`.

What it means

Raised by _verify_dataloader_idx_requirement when a single {train/val/test/predict}_dataloader was provided but the corresponding step hook (e.g. training_step) declares a required `dataloader_idx` parameter with no default. Lightning would have to pass an index that does not exist for a single loader, so the signature is rejected.

Source

Thrown at src/lightning/pytorch/loops/utilities.py:195

            context_manager = torch.no_grad
        with context_manager():
            return loop_run(self, *args, **kwargs)

    return _decorator


def _verify_dataloader_idx_requirement(
    hooks: tuple[str, ...], is_expected: bool, stage: RunningStage, pl_module: "pl.LightningModule"
) -> None:
    for hook in hooks:
        fx = getattr(pl_module, hook)
        # this validation only works if "dataloader_idx" is used, no other names such as "dl_idx"
        param_present = is_param_in_hook_signature(fx, "dataloader_idx")
        if not is_expected:
            if param_present:
                params = inspect.signature(fx).parameters
                if "dataloader_idx" in params and params["dataloader_idx"].default is inspect.Parameter.empty:
                    raise RuntimeError(
                        f"You provided only a single `{stage.dataloader_prefix}_dataloader`, but have included "
                        f"`dataloader_idx` in `{type(pl_module).__name__}.{hook}()`. Either remove the"
                        " argument or give it a default value i.e. `dataloader_idx=0`."
                    )
        elif not param_present:
            raise RuntimeError(
                f"You provided multiple `{stage.dataloader_prefix}_dataloader`, but no `dataloader_idx`"
                f" argument in `{type(pl_module).__name__}.{hook}()`. Try adding `dataloader_idx=0` to its"
                " signature."
            )

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Give the parameter a default: `def training_step(self, batch, batch_idx, dataloader_idx=0)`
2. Or remove `dataloader_idx` from the signature entirely for single-loader setups
3. Or pass a list of dataloaders so the index is meaningful

Example fix

# before
def training_step(self, batch, batch_idx, dataloader_idx):
    ...
trainer.fit(model, single_loader)

# after
def training_step(self, batch, batch_idx, dataloader_idx=0):
    ...
trainer.fit(model, single_loader)

import inspect

params = inspect.signature(model.training_step).parameters
n_loaders = 1 if not isinstance(train_dataloader, (list, tuple)) else len(train_dataloader)
if n_loaders == 1 and 'dataloader_idx' in params:
    assert params['dataloader_idx'].default is not inspect.Parameter.empty, 'give dataloader_idx a default'

def signature_ok_single_loader(fn) -> bool:
    p = inspect.signature(fn).parameters.get('dataloader_idx')
    return p is None or p.default is not inspect.Parameter.empty

When it happens

Trigger: `def training_step(self, batch, batch_idx, dataloader_idx):` while passing one dataloader (not a list) to fit; removing a second dataloader from a datamodule without simplifying the step signature.

Common situations: Downsizing an experiment from multi-dataloader to single dataloader; sharing a LightningModule between single- and multi-loader experiments.

Related errors

• You provided multiple `{stage.dataloader_prefix}_dataloader`
• Device should be CPU, got {device} instead.
• `devices` selected with `CPUAccelerator` should be an int >
• Device should be CUDA, got {device} instead.
• You requested to find {num_devices} devices but there are no

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/e07ec025927dd73e. Report an issue: GitHub.