What causes "Model {cls.__name__} has no config class or model type" in huggingface/transformers?

Set cls.config_class = MyConfig on the model class and define MyConfig.model_type (e.g. 'my-model')

huggingface/transformers · error · ValueError

Model {cls.name} has no config class or model type

Error message

Model {cls.__name__} has no config class or model type

What it means

In the module-fusion machinery, after discovering fusable modules and registering patch mappings, the code requires cls.config_class with a model_type attribute to key checkpoint conversion mappings. Model classes lacking a proper config class (or whose config class has no model_type) cannot participate in fusion.

Source

Thrown at src/transformers/fusion_mapping.py:211

    """Register one fusion family for `cls`.

    This function updates the two global registries used by fused loading:
    - the monkey-patching registry, so compatible module classes are replaced before initialization
    - the checkpoint conversion mapping, so fused runtime modules still load from the original checkpoint layout

    Notes:
    - conflicting checkpoint transforms fail fast
    """

    fusable_classes = _discover_fusable_modules(cls, config, fusion_name=fusion_name, spec=spec)
    if not fusable_classes:
        logger.info(spec.get_empty_log(cls.__name__))
        return

    register_patch_mapping(fusable_classes, overwrite=True)

    if not hasattr(cls, "config_class") or not hasattr(cls.config_class, "model_type"):
        raise ValueError(f"Model {cls.__name__} has no config class or model type")
    model_type = cls.config_class.model_type
    converters = spec.make_transforms(config)

    existing_converters = get_checkpoint_conversion_mapping(model_type)
    if existing_converters is not None:
        # WeightConverter matching stops at the first matching source pattern, so
        # conflicting converters must fail fast instead of being appended.
        existing_converter_sources = {tuple(existing.source_patterns): existing for existing in existing_converters}
        for converter in converters:
            source_patterns = tuple(converter.source_patterns)
            existing_converter = existing_converter_sources.get(source_patterns)
            if existing_converter is not None:
                raise ValueError(
                    f"Fusion {fusion_name} for model type {model_type} conflicts with an existing conversion mapping "
                    f"for source patterns {source_patterns}."
                )

        # TODO: allow compatible fusions mentioned https://github.com/huggingface/transformers/pull/45041#discussion_r3028989716

View on GitHub (pinned to a597f97485)

Solutions

Set cls.config_class = MyConfig on the model class and define MyConfig.model_type (e.g. 'my-model')
Verify fusable modules were actually discovered first — the error only fires when fusable_classes is non-empty
Do not enable fusion_config for model types that do not support fusion

Example fix

# before
class MyModel(PreTrainedModel):
    config_class = None  # fusion then fails
# after
class MyConfig(PretrainedConfig):
    model_type = "my-model"
class MyModel(PreTrainedModel):
    config_class = MyConfig

Defensive patterns

Strategy: validation

Validate before calling

def supports_fusion(cls) -> bool:
    return hasattr(cls, "config_class") and hasattr(cls.config_class, "model_type")

Type guard

from transformers import PreTrainedModel, PretrainedConfig

def has_model_type(cls: type[PreTrainedModel]) -> bool:
    cfg = getattr(cls, "config_class", None)
    return isinstance(cfg, type) and issubclass(cfg, PretrainedConfig) and bool(getattr(cfg, "model_type", None))

Prevention

Always define config_class and model_type on custom PreTrainedModel subclasses
Only enable fusion_config on models documented to support fusion

When it happens

Trigger: Calling register_fusion_patches (directly or via model loading with a fusion_config) on a custom PreTrainedModel subclass whose config_class is missing or whose config class does not define model_type.

Common situations: Experimental/custom model implementations that skip config_class, or configs built from a plain dict without setting model_type before fusion is requested.

Related errors

AI-assisted analysis of huggingface/transformers@a597f97485 (2026-08-14). Data as JSON: /api/errors/8c8c22bf2c878d2e. Report an issue: GitHub.