What causes "Model does not have the "pyfunc" flavor" in mlflow/mlflow?

Verify the target has an MLmodel file listing a `python_function` flavor; if not, re-log with `mlflow.pyfunc.log_model` or the flavor's log_model

mlflow/mlflow · error · MlflowException

RESOURCE_DOES_NOT_EXIST

RESOURCE_DOES_NOT_EXIST

Error message

Model does not have the "pyfunc" flavor

What it means

`mlflow.pyfunc.load_model` requires the model's MLmodel file to contain a `python_function` (pyfunc) flavor configuration. If `flavors.get('python_function')` is None, the model was never saved with pyfunc support, so pyfunc loading is impossible and MLflow throws RESOURCE_DOES_NOT_EXIST.

Source

Thrown at mlflow/pyfunc/__init__.py:1151

        artifact_uri=model_uri, output_path=dst_path, lineage_header_info=lineage_header_info
    )

    if not suppress_warnings:
        model_requirements = _get_pip_requirements_from_model_path(local_path)
        warn_dependency_requirement_mismatches(model_requirements)

    model_meta = Model.load(os.path.join(local_path, MLMODEL_FILE_NAME))

    if model_meta.metadata and model_meta.metadata.get(MLFLOW_MODEL_IS_EXTERNAL, False) is True:
        raise MlflowException(
            "This model's artifacts are external and are not stored in the model directory."
            " This model cannot be loaded with MLflow.",
            BAD_REQUEST,
        )

    conf = model_meta.flavors.get(FLAVOR_NAME)
    if conf is None:
        raise MlflowException(
            f'Model does not have the "{FLAVOR_NAME}" flavor',
            RESOURCE_DOES_NOT_EXIST,
        )
    model_py_version = conf.get(PY_VERSION)
    if not suppress_warnings:
        _warn_potentially_incompatible_py_version_if_necessary(model_py_version=model_py_version)

    _add_code_from_conf_to_system_path(local_path, conf, code_key=CODE)
    data_path = os.path.join(local_path, conf[DATA]) if (DATA in conf) else local_path

    if isinstance(model_config, str):
        model_config = _validate_and_get_model_config_from_file(model_config)

    model_config = _get_overridden_pyfunc_model_config(
        conf.get(MODEL_CONFIG, None), model_config, _logger
    )

    try:

View on GitHub (pinned to 6a27f2decc)

Solutions

Verify the target has an MLmodel file listing a `python_function` flavor; if not, re-log with `mlflow.pyfunc.log_model` or the flavor's log_model
Use the flavor-specific loader (e.g. `mlflow.sklearn.load_model`) if the model only has that flavor
Check the model_uri: ensure it points at the model directory itself, not a parent/subfolder
Re-save the model with a current MLflow version if produced by an old or third-party exporter

Example fix

// before
mlflow.pyfunc.load_model("runs:/abc/raw_artifacts")  # no MLmodel/pyfunc flavor
// after
mlflow.pyfunc.log_model("model", python_model=my_model)  # re-log with pyfunc
mlflow.pyfunc.load_model("runs:/abc/model")

Defensive patterns

Strategy: validation

Validate before calling

from mlflow.models import Model
meta = Model.load(os.path.join(local_path, "MLmodel"))
assert "python_function" in meta.flavors, f"flavors available: {list(meta.flavors)}"

Type guard

def has_pyfunc_flavor(model_path) -> bool:
    from mlflow.models import Model
    return "python_function" in Model.load(os.path.join(model_path, "MLmodel")).flavors

Try / catch

try:
    model = mlflow.pyfunc.load_model(uri)
except MlflowException as e:
    if 'does not have the "python_function" flavor' in str(e):
        model = mlflow.sklearn.load_model(uri)  # flavor-specific fallback
    else:
        raise

Prevention

Verify the MLmodel file exists and lists python_function before pyfunc loading
Always log models via log_model APIs which add the pyfunc flavor
Confirm model_uri points at the model directory, not a parent folder
Re-save old/third-party models with current MLflow

When it happens

Trigger: Loading a model_uri that points to a model saved only with non-pyfunc flavors (e.g. only `mlflow.sklearn` with no pyfunc flavor — rare, only flavor-specific loaders), a path that is not an MLflow model at all (missing/incorrect MLmodel), or a corrupt/truncated model directory.

Common situations: Passing a run artifact directory that contains raw files rather than a logged model; loading a model from a very old MLflow version or third-party tool that omitted the pyfunc flavor; typos in model_uri pointing at the wrong directory.

Related errors

AI-assisted analysis of mlflow/mlflow@6a27f2decc (2026-08-29). Data as JSON: /api/errors/63126dc7e582b1d3. Report an issue: GitHub.