mlflow/mlflow · error · MlflowException

INTERNAL_ERROR

INTERNAL_ERROR

Error message

Unrecognized serialization format: {serialization_format}

What it means

mlflow.sklearn's save path dispatches on the `serialization_format` argument and only understands 'skops' and 'cloudpickle'. Any other value falls into the final else branch and raises this MlflowException with error code INTERNAL_ERROR. The output directory is not written with a usable model.

Source

Thrown at mlflow/sklearn/__init__.py:712

            shutil.rmtree(output_path, ignore_errors=True)
            raise MlflowException(
                "The sklearn model could not be serialized in the skops serialization format. "
                "skops does not support custom functions or classes that are not defined at the "
                "top level. To work around this limitation, you can set the serialization_format "
                "'cloudpickle', while exercising caution due to the possible arbitrary "
                "code during model deserialization using CloudPickle."
            ) from e
        return

    with open(output_path, "wb") as out:
        if serialization_format == SERIALIZATION_FORMAT_PICKLE:
            _dump_model(pickle, sk_model, out)
        elif serialization_format == SERIALIZATION_FORMAT_CLOUDPICKLE:
            import cloudpickle

            _dump_model(cloudpickle, sk_model, out)
        else:
            raise MlflowException(
                message=f"Unrecognized serialization format: {serialization_format}",
                error_code=INTERNAL_ERROR,
            )


def load_model(model_uri, dst_path=None):
    """
    Load a scikit-learn model from a local file or a run.

    Args:
        model_uri: The location, in URI format, of the MLflow model, for example:

            - ``/Users/me/path/to/local/model``
            - ``relative/path/to/local/model``
            - ``s3://my_bucket/path/to/model``
            - ``runs:/<mlflow_run_id>/run-relative/path/to/model``
            - ``models:/<model_name>/<model_version>``
            - ``models:/<model_name>/<stage>``

View on GitHub (pinned to 6a27f2decc)

Solutions

1. Set serialization_format to exactly 'skops' or 'cloudpickle' (lowercase).
2. Import the constants instead of hard-coding strings: from mlflow.sklearn import SERIALIZATION_FORMAT_SKOPS, SERIALIZATION_FORMAT_CLOUDPICKLE.
3. Check for accidental whitespace or case differences in config/env-driven values (e.g. .strip().lower() before passing).

Example fix

// before
mlflow.sklearn.save_model(model, path, serialization_format='pickle')
// after
mlflow.sklearn.save_model(model, path, serialization_format='cloudpickle')

VALID = {'skops', 'cloudpickle'}
assert serialization_format in VALID, f'serialization_format must be one of {VALID}, got {serialization_format!r}'

try:
    mlflow.sklearn.save_model(model, path, serialization_format=fmt)
except MlflowException as e:
    if 'Unrecognized serialization format' in str(e):
        raise ValueError(f'Bad format {fmt!r}; use skops or cloudpickle') from e

When it happens

Trigger: Passing serialization_format to mlflow.sklearn.save_model/log_model (or via MLFLOW_SKLEARN_DEFAULT_SERIALIZATION_FORMAT-style config paths that ultimately reach _save_model) with a typo or unsupported value, e.g. 'pickle', 'joblib', or 'CloudPickle'.

Common situations: Typos in the format string; case-sensitivity mistakes ('CloudPickle' vs 'cloudpickle'); copying config from older MLflow versions or blog posts that predate the skops format.

Understand the failure class

Background: Invalid enum value errors: "Unknown type", "Invalid scope", "must be one of" — when a string is not on the library's allowed list — this error's family across 23 libraries.

Related errors

• Databricks host not found. Please either: 1. Set the DATAB
• @kubernetes/client-node is not installed. It is required for
• Invalid response format: missing credential_info
• An MLflow Tracking URI is required, please provide the track
• An MLflow experiment ID is required, please provide the expe

AI-assisted analysis of mlflow/mlflow@6a27f2decc (2026-08-29). Data as JSON: /api/errors/b570055e2cc32ff9. Report an issue: GitHub.