apache/beam · error · TypeError

No SerializeToString method is detected on loaded model…

Error message

No SerializeToString method is detected on loaded model. Type of model: {type(model_proto)}

What it means

ONNXModelHandler.load_model expects the loaded model as bytes or an object exposing SerializeToString() (a protobuf message, e.g. onnx.ModelProto). If the object is neither bytes nor protobuf-like, a TypeError is raised because the handler cannot serialize it for onnxruntime's InferenceSession.

Solutions

1. Load with onnx.load(path) and pass the ModelProto (it has SerializeToString).
2. Or read the file yourself and pass bytes: open(path,'rb').read().
3. Ensure you pass onnxruntime-compatible ONNX protobuf, not a TorchScript/TF model.

Example fix

// before
handler = ONNXModelHandler(model_url='model.onnx')
session_input = onnx.load('model.onnx').graph  # graph has no SerializeToString
// after
model_proto = onnx.load('model.onnx')  # ModelProto: has SerializeToString

if not isinstance(model, bytes) and not (hasattr(model, 'SerializeToString') and callable(getattr(model, 'SerializeToString'))):
    raise TypeError('Pass bytes or an onnx.ModelProto to ONNXModelHandler')

def is_protobuf_like(obj) -> bool:
    return hasattr(obj, 'SerializeToString') and callable(obj.SerializeToString)

try:
    handler.load_model()
except TypeError as e:
    if 'SerializeToString' in str(e):
        model_bytes = open(model_path, 'rb').read()
        # retry with bytes input
    else:
        raise

When it happens

Trigger: onnx.load() returning a ModelProto is fine, but passing e.g. a session, a numpy object, or a str path into the handler's model slot leads to load_model hitting the else branch.

Common situations: Passing a file path string instead of loaded bytes; using tf/savedmodel exports without protobuf conversion; custom model wrappers lacking SerializeToString.

Understand the failure class

Background: Type mismatch errors: IllegalArgumentException, TypeError and type guards across 150 open-source libraries — this error's family across 150 libraries.

Related errors

• A cluster_identifier should be Optional[Union[str…
• Attempted to encode null for non-nullable field
• Cannot convert to a JSON value.
• Cannot get a type descriptor for
• Cannot interpret as Duration.

AI-assisted analysis of apache/beam@12126d8942 (2026-09-13). Data as JSON: /api/errors/bdadc19c598f58e7. Report an issue: GitHub.

Appendix: source

Thrown at sdks/python/apache_beam/ml/inference/onnx_inference.py:134

        **kwargs)
    self._model_uri = model_uri
    self._session_options = session_options
    self._providers = providers
    self._provider_options = provider_options
    self._model_inference_fn = inference_fn

  def load_model(self) -> ort.InferenceSession:
    """Loads and initializes an onnx inference session for processing."""
    # when path is remote, we should first load into memory then deserialize
    f = FileSystems.open(self._model_uri, "rb")
    model_proto = onnx.load(f)
    model_proto_bytes = model_proto
    if not isinstance(model_proto, bytes):
      if (hasattr(model_proto, "SerializeToString") and
          callable(model_proto.SerializeToString)):
        model_proto_bytes = model_proto.SerializeToString()
      else:
        raise TypeError(
            "No SerializeToString method is detected on loaded model. " +
            f"Type of model: {type(model_proto)}")
    ort_session = ort.InferenceSession(
        model_proto_bytes,
        sess_options=self._session_options,
        providers=self._providers,
        provider_options=self._provider_options)
    return ort_session

  def run_inference(
      self,
      batch: Sequence[numpy.ndarray],
      inference_session: ort.InferenceSession,
      inference_args: Optional[dict[str, Any]] = None
  ) -> Iterable[PredictionResult]:
    """Runs inferences on a batch of numpy arrays.

    Args:

View on GitHub (pinned to 12126d8942)