apache/beam · error · RuntimeError

Please specify either torch_script_model_path or…

Error message

Please specify either torch_script_model_path or (state_dict_path, model_class) to successfully load the model.

What it means

Raised by _validate_constructor_args in PytorchModelHandlerKeyedModel/PytorchModelHandler when the constructor arguments are inconsistent. Loading a PyTorch model requires either a TorchScript serialized model path alone, or a state_dict path paired with the model's class. Passing torch_script_model_path together with state_dict_path is ambiguous, so the handler refuses to construct.

Solutions

1. Remove torch_script_model_path if you intend to load via (state_dict_path, model_class)
2. Remove state_dict_path and model_class if you intend to load a TorchScript model via torch_script_model_path
3. Ensure your config/kwargs builder emits exactly one of the two loading styles

Example fix

// before
handler = PytorchModelHandlerKeyedModel(
    state_dict_path='gs://bucket/model.pt',
    model_class=MyNet,
    torch_script_path='gs://bucket/model_scripted.pt')
// after
handler = PytorchModelHandlerKeyedModel(
    state_dict_path='gs://bucket/model.pt',
    model_class=MyNet)

def check_pytorch_handler_kwargs(kwargs):
    has_ts = bool(kwargs.get('torch_script_model_path'))
    has_sd = bool(kwargs.get('state_dict_path') and kwargs.get('model_class'))
    if has_ts and has_sd:
        raise ValueError('Pass either torch_script_model_path or (state_dict_path, model_class), not both.')
    if not has_ts and not has_sd:
        raise ValueError('Provide torch_script_model_path or (state_dict_path, model_class).')

def is_valid_loading_style(kwargs: dict) -> bool:
    has_ts = kwargs.get('torch_script_model_path') is not None
    has_sd = kwargs.get('state_dict_path') is not None and kwargs.get('model_class') is not None
    return has_ts ^ has_sd

try:
    handler = PytorchModelHandlerKeyedModel(**kwargs)
except RuntimeError as e:
    if 'torch_script_model_path' in str(e):
        kwargs.pop('torch_script_model_path')
        handler = PytorchModelHandlerKeyedModel(**kwargs)
    else:
        raise

When it happens

Trigger: Constructing a ModelHandler (e.g. PytorchModelHandlerKeyedModel(...)) with both torch_script_model_path and state_dict_path set to non-None values.

Common situations: Migrating a handler from TorchScript loading to state_dict loading (or vice versa) and forgetting to remove the old path argument; building handler kwargs from a config that contains both keys; copy-pasting example code that mixes the two loading styles.

Related errors

• Callable create_model_fn must be passedwith…
• A has been supplied to the model handler, but the required…
• Batch does not have expected dtype
• Batch does not have expected shape
• Batch is not an instance of torch.Tensor

AI-assisted analysis of apache/beam@12126d8942 (2026-09-13). Data as JSON: /api/errors/f263a8e7f4237815. Report an issue: GitHub.

Appendix: source

Thrown at sdks/python/apache_beam/ml/inference/pytorch_inference.py:91

def _validate_constructor_args(
    state_dict_path, model_class, torch_script_model_path):
  message = (
      "A {param1} has been supplied to the model "
      "handler, but the required {param2} is missing. "
      "Please provide the {param2} in order to "
      "successfully load the {param1}.")
  # state_dict_path and model_class are coupled with each other
  # raise RuntimeError if user forgets to pass any one of them.
  if state_dict_path and not model_class:
    raise RuntimeError(
        message.format(param1="state_dict_path", param2="model_class"))

  if not state_dict_path and model_class:
    raise RuntimeError(
        message.format(param1="model_class", param2="state_dict_path"))

  if torch_script_model_path and state_dict_path:
    raise RuntimeError(
        "Please specify either torch_script_model_path or "
        "(state_dict_path, model_class) to successfully load the model.")


def _load_model(
    model_class: Optional[Callable[..., torch.nn.Module]],
    state_dict_path: Optional[str],
    device: torch.device,
    model_params: Optional[dict[str, Any]],
    torch_script_model_path: Optional[str],
    load_model_args: Optional[dict[str, Any]]):
  if device == torch.device('cuda') and not _cuda_device_is_usable():
    logging.warning(
        "Model handler specified a 'GPU' device, but GPUs are not available. "
        "Switching to CPU.")
    device = torch.device('cpu')

  try:

View on GitHub (pinned to 12126d8942)