Lightning-AI/pytorch-lightning · error · MisconfigurationException

Providing log_model={log_model} and offline={offline} is an

Error message

Providing log_model={log_model} and offline={offline} is an invalid configuration since model checkpoints cannot be uploaded in offline mode.
Hint: Set `offline=False` to log your model.

What it means

MisconfigurationException from WandbLogger.__init__: offline=True (no network run) combined with log_model set to a truthy value is rejected because model checkpoints cannot be uploaded to W&B without an online run.

Source

Thrown at src/lightning/pytorch/loggers/wandb.py:315

        save_dir: _PATH = ".",
        version: Optional[str] = None,
        offline: bool = False,
        dir: Optional[_PATH] = None,
        id: Optional[str] = None,
        anonymous: Optional[bool] = None,
        project: Optional[str] = None,
        log_model: Union[Literal["all"], bool] = False,
        experiment: Union["Run", "RunDisabled", None] = None,
        prefix: str = "",
        checkpoint_name: Optional[str] = None,
        add_file_policy: Literal["mutable", "immutable"] = "mutable",
        **kwargs: Any,
    ) -> None:
        if not _WANDB_AVAILABLE:
            raise ModuleNotFoundError(str(_WANDB_AVAILABLE))

        if offline and log_model:
            raise MisconfigurationException(
                f"Providing log_model={log_model} and offline={offline} is an invalid configuration"
                " since model checkpoints cannot be uploaded in offline mode.\n"
                "Hint: Set `offline=False` to log your model."
            )

        super().__init__()
        self._offline = offline
        self._log_model = log_model
        self._prefix = prefix
        self._experiment = experiment
        self._logged_model_time: dict[str, float] = {}
        self._checkpoint_callbacks: dict[int, ModelCheckpoint] = {}
        self.add_file_policy = add_file_policy

        # paths are processed as strings
        if save_dir is not None:
            save_dir = os.fspath(save_dir)
        elif dir is not None:

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Set offline=False (with WANDB_API_KEY configured) if you need model uploads
2. Keep offline=True but pass log_model=False / omit log_model
3. Log checkpoints manually to a local dir and upload later with wandb sync when online

Example fix

# before
logger = WandbLogger(offline=True, log_model='all')
# after (choose one)
logger = WandbLogger(offline=True)                        # no ckpt upload
logger = WandbLogger(offline=False, log_model='all')      # online upload

if cfg.wandb.offline and cfg.wandb.log_model:
    raise ValueError("offline=True cannot be combined with log_model")

When it happens

Trigger: WandbLogger(save_dir='./off', offline=True, log_model='all') (or log_model=True/every/interval) — any truthy log_model with offline truthy.

Common situations: Air-gapped or CI environments set offline=True while the code (copied from an online example) enables checkpoint uploading via log_model.

Related errors

• You have set `accumulate_grad_batches` and are using the `Gr
• You set `.load_from_checkpoint(..., strict={strict!r})` whic
• Experiment is not initialized
• `synchronous` requires mlflow>=2.8.0
• NeptuneLogger is no longer supported. Neptune has been sunse

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/65b247992e4c3277. Report an issue: GitHub.