What causes "{auto_class} is not a valid auto class." in huggingface/transformers?

Use a valid auto class name, e.g. "AutoConfig" for configs or "AutoModelForCausalLM" for models.

huggingface/transformers · error · ValueError

is not a valid auto class.

Error message

{auto_class} is not a valid auto class.

What it means

ValueError from PreTrainedConfig.register_for_auto_class when the given auto_class name (string or class __name__) does not exist as an attribute of transformers.models.auto. Only classes actually exported by the auto module (e.g. AutoConfig, AutoModelForCausalLM) can back a registration.

Solutions

Use a valid auto class name, e.g. "AutoConfig" for configs or "AutoModelForCausalLM" for models.
Check availability first: import transformers.models.auto as auto; hasattr(auto, name).
Verify the name is exported in your installed Transformers version's auto module.

Example fix

// before
MyConfig.register_for_auto_class("AutoModelForTextGeneration")  # ValueError

// after
MyConfig.register_for_auto_class("AutoModelForCausalLM")

Defensive patterns

Strategy: validation

Validate before calling

import transformers.models.auto as auto
valid = [n for n in dir(auto) if n.startswith("Auto")]
if auto_class_name not in valid:
    raise ValueError(f"{auto_class_name!r} not in {valid}")
MyConfig.register_for_auto_class(auto_class_name)

Type guard

def is_valid_auto_class(name: str) -> bool:
    import transformers.models.auto as auto
    return hasattr(auto, name)

Prevention

Reference auto class names from the transformers.models.auto module instead of typing them by hand.
Pin the Transformers version your custom model code was developed against.

When it happens

Trigger: MyConfig.register_for_auto_class("AutoModelForCausulLM") (typo), passing "AutoModel"-style names that are not in the auto module, or passing a custom class whose name differs from any auto export.

Common situations: Registering custom remote-code models for auto-loading; renamed or hypothetical auto classes; code written against a different Transformers version where an auto class does not exist.

Related errors

AI-assisted analysis of huggingface/transformers@a597f97485 (2026-08-14). Data as JSON: /api/errors/c26f84b9c62b6ba5. Report an issue: GitHub.

Appendix: source

Thrown at src/transformers/configuration_utils.py:1270

    @classmethod
    def register_for_auto_class(cls, auto_class="AutoConfig"):
        """
        Register this class with a given auto class. This should only be used for custom configurations as the ones in
        the library are already mapped with `AutoConfig`.



        Args:
            auto_class (`str` or `type`, *optional*, defaults to `"AutoConfig"`):
                The auto class to register this new configuration with.
        """
        if not isinstance(auto_class, str):
            auto_class = auto_class.__name__

        import transformers.models.auto as auto_module

        if not hasattr(auto_module, auto_class):
            raise ValueError(f"{auto_class} is not a valid auto class.")

        cls._auto_class = auto_class

    @classmethod
    def is_remote_code(cls) -> bool:
        """Return whether the current config is custom code, i.e. code loaded from the hub, or class that we just
        registered via `register_for_auto_class`."""
        return cls._auto_class is not None

    @classmethod
    def is_custom_code(cls) -> bool:
        """Return whether the current config is custom code, i.e. either code loaded from the hub, or defined in any
        user-specific module/session."""
        return cls.is_remote_code() or not cls.__module__.startswith("transformers.")

    def _get_generation_parameters(self) -> dict[str, Any]:
        """
        Checks if there are generation parameters in `PreTrainedConfig` instance. Note that

View on GitHub (pinned to a597f97485)