keras-team/keras · error · ValueError

Arguments `clone_function` and `input_tensors` are only supp

Error message

Arguments `clone_function` and `input_tensors` are only supported for Sequential models or Functional models. Received model of type '{model.__class__.__name__}', with clone_function={clone_function} and input_tensors={input_tensors}

What it means

clone_function and input_tensors are only implemented for Sequential and Functional models. For custom subclassed models, clone_model falls back to a config round-trip and raises this ValueError if either argument was supplied, since there is no layer list or graph to clone from.

Source

Thrown at keras/src/models/cloning.py:191

        # If the get_config() method is the same as a regular Functional
        # model, we're safe to use _clone_functional_model (which relies
        # on a Functional constructor). In the case where the get_config
        # is custom, this may not necessarily work, but if clone_function
        # or input_tensors are passed, we attempt it anyway
        # in order to preserve backwards compatibility.
        if utils.is_default(model.get_config) or (
            clone_function or input_tensors
        ):
            return _clone_functional_model(
                model,
                clone_function=clone_function,
                call_function=call_function,
                input_tensors=input_tensors,
            )

    # Case of a custom model class
    if clone_function or input_tensors:
        raise ValueError(
            "Arguments `clone_function` and `input_tensors` "
            "are only supported for Sequential models "
            "or Functional models. Received model of "
            f"type '{model.__class__.__name__}', with "
            f"clone_function={clone_function} and "
            f"input_tensors={input_tensors}"
        )
    if call_function is not None:
        raise ValueError(
            "Argument `call_function` is only supported "
            "for Functional models. Received model of "
            f"type '{model.__class__.__name__}', with "
            f"call_function={clone_function}"
        )
    config = serialization_lib.serialize_keras_object(model)
    return serialization_lib.deserialize_keras_object(
        config, custom_objects={model.__class__.__name__: model.__class__}
    )

View on GitHub (pinned to 7a34a03db6)

Solutions

1. For subclassed models call clone_model(model) with no extra args - it rebuilds via get_config()/from_config().
2. Ensure the subclass implements get_config()/from_config() correctly so the round trip works.
3. Only pass clone_function/input_tensors after confirming the model is Sequential or Functional.

Example fix

# before
clone = keras.models.clone_model(subclassed_model, clone_function=fn)

# after
clone = keras.models.clone_model(subclassed_model)  # config round-trip

if not isinstance(model, keras.Sequential) and not is_functional(model):
    assert not clone_function and not input_tensors, 'clone_function/input_tensors only for Sequential/Functional'

def supports_clone_args(model) -> bool:
    import keras
    return isinstance(model, keras.Sequential) or getattr(model, '_is_graph_network', False)

When it happens

Trigger: keras.models.clone_model(my_subclassed_model, clone_function=fn) or input_tensors=tensor where the model is neither Sequential nor Functional.

Common situations: Model-agnostic cloning utilities; subclassed models embedding custom training logic, common in research code.

Related errors

• Argument `call_function` is only supported for Functional mo
• Unexpected keyword argument(s): {tuple(kwargs.keys())}
• `call_function` argument is not supported with Sequential mo
• Expected `model` argument to be a `Sequential` model instanc
• Expected `clone_function` argument to be a callable. Receive

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/38c43166cd578675. Report an issue: GitHub.