mlflow/mlflow · error · MlflowException

Record {i} is missing required 'inputs' field or it is empty

Error message

Record {i} is missing required 'inputs' field or it is empty

What it means

Raised during prompt optimization when a training record lacks the 'inputs' field or it is empty/None. Optimization needs per-record inputs to run the predict_fn and scorers, so MLflow validates the train_data (DataFrame) before proceeding and fails fast on malformed rows.

Source

Thrown at mlflow/genai/optimize/util.py:104

def validate_train_data(
    train_data: "pd.DataFrame",
    scorers: list[Scorer] | None,
    predict_fn: Callable[..., Any] | None = None,
) -> None:
    """
    Validate that training data has required fields for prompt optimization.

    Args:
        train_data: Training data as a pandas DataFrame.
        scorers: Scorers to validate the training data for. Can be None for zero-shot mode.
        predict_fn: The predict function to validate the training data for.

    Raises:
        MlflowException: If any record is missing required 'inputs' field or it is empty.
    """
    for i, record in enumerate(train_data.to_dict("records")):
        if "inputs" not in record or not record["inputs"]:
            raise MlflowException.invalid_parameter_value(
                f"Record {i} is missing required 'inputs' field or it is empty"
            )

    if scorers is not None:
        builtin_scorers = [scorer for scorer in scorers if isinstance(scorer, BuiltInScorer)]
        valid_data_for_builtin_scorers(train_data, builtin_scorers, predict_fn)


def infer_type_from_value(value: Any, model_name: str = "Output") -> type:
    """
    Infer the type from the value.
    Only supports primitive types, lists, and dict and Pydantic models.
    """
    if value is None:
        return type(None)
    elif isinstance(value, (bool, int, float, str)):
        return type(value)
    elif isinstance(value, list):

View on GitHub (pinned to 6a27f2decc)

Solutions

1. Rename/add the 'inputs' column to train_data so every row has it
2. Drop or fix rows where inputs is None/empty, e.g. df = df[df['inputs'].apply(lambda x: bool(x))]
3. Verify the DataFrame passed to optimize_prompts is the one produced by to_df()/the expected schema, not a differently-shaped frame

Example fix

// before
train_data = pd.DataFrame({"input": [{"question": "q"}], "expectations": [{}]})
// after
train_data = pd.DataFrame({"inputs": [{"question": "q"}], "expectations": [{}]})

import pandas as pd
missing = train_data[~train_data["inputs"].apply(lambda x: bool(x))]
if not missing.empty:
    raise ValueError(f"Rows missing/empty 'inputs': {missing.index.tolist()}")

def has_inputs(record: dict) -> bool:
    return bool(record.get("inputs"))

from mlflow.exceptions import MlflowException
try:
    result = mlflow.genai.optimize_prompts(...)
except MlflowException as e:
    if "missing required 'inputs'" in str(e):
        train_data = train_data[train_data["inputs"].apply(bool)]
    else:
        raise

When it happens

Trigger: Calling mlflow.genai.optimize_prompts (via validate_train_data) with a train_data DataFrame whose 'inputs' column is absent, or where a row's 'inputs' is None or an empty value (e.g. empty dict/string).

Common situations: Building training data with inconsistent column names (e.g. 'input' instead of 'inputs'); rows filtered down until some have no inputs; concatenating datasets where one batch lacks the column; constructing records programmatically with None placeholders.

Understand the failure class

Background: "Missing required field" and "field is required" errors: why libraries reject payloads that omit mandatory fields — this error's family across 20 libraries.

Related errors

• INVALID_PARAMETER_VALUE
• INVALID_PARAMETER_VALUE
• INVALID_PARAMETER_VALUE
• SerializedScorer cannot have multiple types of scorer fields
• Failed to deserialize InstructionsJudge scorer '{serialized.

AI-assisted analysis of mlflow/mlflow@6a27f2decc (2026-08-29). Data as JSON: /api/errors/773a6c483ae88107. Report an issue: GitHub.