keras-team/keras · error · ValueError

`call_function` argument is not supported with Sequential mo

Error message

`call_function` argument is not supported with Sequential models.  In a Sequential model, layers aren't called at model-construction time (they're merely listed). Use `call_function` with Functional models only. Received model of type '{model.__class__.__name__}', with call_function={clone_function}

What it means

call_function lets you replay a Functional model's call graph at clone time. Sequential models have no recorded call graph - layers are only listed, never called - so clone_model(model, call_function=fn) on a Sequential raises this ValueError. Use clone_function for Sequential instead.

Source

Thrown at keras/src/models/cloning.py:150

    In the case of a subclassed model, you cannot using a custom
    `clone_function`.
    """
    cache = kwargs.pop("cache", None)
    if kwargs:
        raise ValueError(
            f"Unexpected keyword argument(s): {tuple(kwargs.keys())}"
        )

    if isinstance(model, Sequential):
        # Wrap clone_function to handle recursiveness and layer sharing.
        clone_function = _wrap_clone_function(
            clone_function,
            call_function=call_function,
            recursive=recursive,
            cache=cache,
        )
        if call_function is not None:
            raise ValueError(
                "`call_function` argument is not supported with Sequential "
                "models.  In a Sequential model, layers aren't called "
                "at model-construction time (they're merely listed). "
                "Use `call_function` with Functional models only. "
                "Received model of "
                f"type '{model.__class__.__name__}', with "
                f"call_function={clone_function}"
            )
        return _clone_sequential_model(
            model,
            clone_function=clone_function,
            input_tensors=input_tensors,
        )
    if isinstance(model, Functional):
        # Wrap clone_function to handle recursiveness and layer sharing.
        clone_function = _wrap_clone_function(
            clone_function,
            call_function=call_function,

View on GitHub (pinned to 7a34a03db6)

Solutions

1. Pass clone_function(layer) -> layer instead; it customizes per-layer cloning for Sequential.
2. If you need call-graph customization, rebuild the model as Functional (built from keras.Input plus layer calls) first.
3. Simply drop call_function for Sequential models.

Example fix

# before
clone = keras.models.clone_model(seq_model, call_function=my_call_fn)

# after
clone = keras.models.clone_model(
    seq_model, clone_function=lambda l: l.__class__.from_config(l.get_config()))

kwargs = {'call_function': fn} if is_functional(model) else {}
clone = keras.models.clone_model(model, **kwargs)

def is_functional(model) -> bool:
    return getattr(model, '_functional_construction', False) or getattr(model, '_is_graph_network', False)

When it happens

Trigger: keras.models.clone_model(sequential_model, call_function=my_fn).

Common situations: Generic cloning utilities that always pass call_function regardless of model type; migrating a workflow from Functional to Sequential models.

Related errors

• Argument `call_function` is only supported for Functional mo
• Expected `model` argument to be a `Sequential` model instanc
• Argument `input_tensors` must contain a single tensor.
• Unexpected keyword argument(s): {tuple(kwargs.keys())}
• Arguments `clone_function` and `input_tensors` are only supp

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/014f4f4e0d6ac8f6. Report an issue: GitHub.