Currently mlflow only supports the following engine types: {

Error message

Currently mlflow only supports the following engine types: {SUPPORTED_ENGINES}. {engine_type} is not supported, so please use one of the above types.

What it means

When saving a LlamaIndex index via the legacy serialization path, MLflow validates the engine_type argument against SUPPORTED_ENGINES = {"chat", "query", "retriever"} (mlflow/llama_index/pyfunc_wrapper.py:14). Any other string raises ValueError. MLflow can only wrap and serve these three engine types at prediction time.

Source

Thrown at mlflow/llama_index/model.py:83

        that, at a minimum, contains these requirements.
    """
    return [_get_pinned_requirement("llama-index")]


def get_default_conda_env():
    """
    Returns:
        The default Conda environment for MLflow Models produced by calls to
        :func:`save_model()` and :func:`log_model()`.
    """
    return _mlflow_conda_env(additional_pip_deps=get_default_pip_requirements())


def _validate_engine_type(engine_type: str):
    from mlflow.llama_index.pyfunc_wrapper import SUPPORTED_ENGINES

    if engine_type not in SUPPORTED_ENGINES:
        raise ValueError(
            f"Currently mlflow only supports the following engine types: "
            f"{SUPPORTED_ENGINES}. {engine_type} is not supported, so please "
            "use one of the above types."
        )


def _get_llama_index_version() -> str:
    try:
        import llama_index.core

        return llama_index.core.__version__
    except ImportError:
        raise MlflowException(
            "The llama_index module is not installed. "
            "Please install it via `pip install llama-index`."
        )

View on GitHub (pinned to 6a27f2decc)

Solutions

Set engine_type to one of "chat", "query", or "retriever".
If you need another engine type, save the model in Model-from-Code mode (pass a script path) where engine_type constraints differ.
Check the value for typos/case — the set match is exact.
Upgrade MLflow if a newer engine type may have been added to SUPPORTED_ENGINES.

Example fix

// before
mlflow.llama_index.save_model(index, name="model", engine_type="query_engine")
// after
mlflow.llama_index.save_model(index, name="model", engine_type="query")

Defensive patterns

Strategy: validation

Validate before calling

from mlflow.llama_index.pyfunc_wrapper import SUPPORTED_ENGINES

if engine_type not in SUPPORTED_ENGINES:
    raise ValueError(f"engine_type must be one of {SUPPORTED_ENGINES}, got {engine_type!r}")

Type guard

from typing import Literal
EngineType = Literal["chat", "query", "retriever"]

def is_valid_engine_type(t: str) -> bool:
    return t in {"chat", "query", "retriever"}

Try / catch

try:
    mlflow.llama_index.save_model(index, name="model", engine_type=engine_type)
except ValueError as e:
    if "not supported" in str(e):
        mlflow.llama_index.save_model(index, name="model", engine_type="query")
    else:
        raise

Prevention

Use a Literal/enum type for engine_type in your code so typos fail at type-check time
Import SUPPORTED_ENGINES and validate against it rather than hardcoding strings
Verify exact lowercase spelling: chat, query, retriever

When it happens

Trigger: Calling mlflow.llama_index.save_model(model=index, engine_type="<other>") with an unsupported engine_type string such as "node_postprocessor", "query_engine", a typo like "qurey", or a custom engine name.

Common situations: Copy-pasting engine_type from LlamaIndex terminology rather than MLflow's supported set; typos or casing issues; trying to log a raw LLM or custom pipeline via the legacy path.

Related errors

AI-assisted analysis of mlflow/mlflow@6a27f2decc (2026-08-29). Data as JSON: /api/errors/3787d19f87d182b8. Report an issue: GitHub.