Lightning-AI/pytorch-lightning · error · MisconfigurationException

You are trying to `self.log()` but the loop's result collect

Error message

You are trying to `self.log()` but the loop's result collection is not registered yet. This is most likely because you are trying to log in a `predict` hook, but it doesn't support logging

What it means

self.log requires trainer._results (the loop's result collection) to be registered, which only happens inside Trainer-run hooks. The predict loop does not register a result collection, so logging inside predict_step (or any hook outside managed training/validation flow) raises MisconfigurationException.

Source

Thrown at src/lightning/pytorch/core/module.py:465

        )

        trainer = self._trainer
        if trainer is None:
            # not an error to support testing the `*_step` methods without a `Trainer` reference
            rank_zero_warn(
                "You are trying to `self.log()` but the `self.trainer` reference is not registered on the model yet."
                " This is most likely because the model hasn't been passed to the `Trainer`"
            )
            return
        if trainer.barebones:
            rank_zero_warn(
                "You are trying to `self.log()` but `Trainer(barebones=True)` is configured."
                " Logging can impact raw speed so it is disabled under this setting."
            )
            return
        results = trainer._results
        if results is None:
            raise MisconfigurationException(
                "You are trying to `self.log()` but the loop's result collection is not registered"
                " yet. This is most likely because you are trying to log in a `predict` hook,"
                " but it doesn't support logging"
            )
        if self._current_fx_name is None:
            raise MisconfigurationException(
                "You are trying to `self.log()` but it is not managed by the `Trainer` control flow"
            )

        on_step, on_epoch = _FxValidator.check_logging_and_get_default_levels(
            self._current_fx_name, on_step=on_step, on_epoch=on_epoch
        )

        # make sure user doesn't introduce logic for multi-dataloaders
        if add_dataloader_idx and "/dataloader_idx_" in name:
            raise MisconfigurationException(
                f"You called `self.log` with the key `{name}`"
                " but it should not contain information about `dataloader_idx` when `add_dataloader_idx=True`"

View on GitHub (pinned to 9fed5c27d2)

Solutions

Remove self.log calls from predict_step; collect outputs from predict_step and compute metrics afterwards from trainer.predict return value
If logging during training/validation, ensure self.log is called within those hooks under Trainer control
Return predictions from predict_step and log via a Logger explicitly (e.g. fabric/loggers or trainer.logger) after inference

Example fix

# before
def predict_step(self, batch, batch_idx):
    preds = self(batch)
    self.log('acc', acc)  # raises
    return preds

# after
def predict_step(self, batch, batch_idx):
    return self(batch)
# afterwards:
outs = trainer.predict(model)
acc = compute_accuracy(outs)
trainer.logger.log_metrics({'acc': acc})

Defensive patterns

Strategy: validation

Validate before calling

def safe_log(model, name, value, **kw):
    if model.trainer is not None and model.trainer._results is not None:
        model.log(name, value, **kw)
    else:
        print(f'[unlogged] {name}={value}')

Type guard

def logging_supported(model) -> bool:
    t = getattr(model, '_trainer', None)
    return t is not None and getattr(t, '_results', None) is not None

Try / catch

from lightning.pytorch.utilities.exceptions import MisconfigurationException
try:
    self.log('m', v)
except MisconfigurationException:
    pass  # e.g. inside predict_step: collect and log later

Prevention

Never call self.log inside predict_step
Accumulate predictions and compute/log metrics after trainer.predict returns

When it happens

Trigger: Calling self.log(...) inside predict_step, or calling self.log manually outside the Trainer loop (e.g. in __init__ or a plain function before training starts).

Common situations: User copied a validation_step containing self.log into predict_step; tried logging metrics during inference; called model.log in a callback before the loop registered results.

Related errors

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/19c50cbde4572db0. Report an issue: GitHub.