huggingface/pytorch-image-models · error · RuntimeError

Model architecture ({arch_name}) has no pretrained cfg regis

Error message

Model architecture ({arch_name}) has no pretrained cfg registered.

What it means

timm's pretrained-configuration lookup could not find any pretrained cfg registered for the given architecture name. Every model registered via register_model has a default pretrained cfg (usually tagged 'default'); if the name is not in _model_default_cfgs at all and allow_unregistered is False, this RuntimeError is raised. It typically means the model name is misspelled, the model was never registered, or timm was imported in a way that skipped model registration.

Source

Thrown at timm/models/_registry.py:336

    assert isinstance(module_names, (tuple, list, set))
    return any(arch_name in _module_to_models[n] for n in module_names)


def is_model_pretrained(model_name: str) -> bool:
    return model_name in _model_has_pretrained


def get_pretrained_cfg(model_name: str, allow_unregistered: bool = True) -> Optional[PretrainedCfg]:
    if model_name in _model_pretrained_cfgs:
        return deepcopy(_model_pretrained_cfgs[model_name])
    arch_name, tag = split_model_name_tag(model_name)
    if arch_name in _model_default_cfgs:
        # if model arch exists, but the tag is wrong, error out
        raise RuntimeError(f'Invalid pretrained tag ({tag}) for {arch_name}.')
    if allow_unregistered:
        # if model arch doesn't exist, it has no pretrained_cfg registered, allow a default to be created
        return None
    raise RuntimeError(f'Model architecture ({arch_name}) has no pretrained cfg registered.')


def get_pretrained_cfg_value(model_name: str, cfg_key: str) -> Optional[Any]:
    """ Get a specific model default_cfg value by key. None if key doesn't exist.
    """
    cfg = get_pretrained_cfg(model_name, allow_unregistered=False)
    return getattr(cfg, cfg_key, None)


def get_arch_pretrained_cfgs(model_name: str) -> Dict[str, PretrainedCfg]:
    """ Get all pretrained cfgs for a given architecture.
    """
    arch_name, _ = split_model_name_tag(model_name)
    model_names = _model_with_tags[arch_name]
    cfgs = {m: _model_pretrained_cfgs[m] for m in model_names}
    return cfgs

View on GitHub (pinned to 9a5261e31b)

Solutions

1. Verify the exact name via timm.list_models() (optionally with a filter like timm.list_models('*coat*')) and correct the string
2. If it is a custom model, pass pretrained_cfgs (or a default cfg) when calling register_model
3. Pass allow_unregistered=True if you intentionally want a cfg-less model (only valid through APIs that expose the flag)
4. Pin/upgrade timm to a version that contains the model you expect

Example fix

# before
model = timm.create_model('resnet50v', pretrained=True)
# after
import timm
print([m for m in timm.list_models('*resnet50*')])
model = timm.create_model('resnet50s', pretrained=True)

import timm

valid = set(timm.list_models(pretrained=True))
if name not in valid:
    close = timm.list_models(f'{name[:6]}*')
    raise KeyError(f'{name} unknown; candidates: {close}')
model = timm.create_model(name, pretrained=True)

def is_registered_timm_model(name: str) -> bool:
    import timm
    return name in set(timm.list_models())

try:
    cfg = timm.resolve_pretrained_cfg(name)
except RuntimeError as e:
    # fall back to offline/custom weights or corrected name
    suggestions = timm.list_models(name[:5] + '*')
    name = suggestions[0]
    cfg = timm.resolve_pretrained_cfg(name)

When it happens

Trigger: Calling timm.create_model('nonexistent_or_typo_arch', pretrained=True), timm.resolve_pretrained_cfg('bad_name'), or timm.get_pretrained_cfg_value('bad_name', 'input_size') where the arch string does not match any registered model.

Common situations: Typos in model names ('resnet50' vs 'resnet_50'), referencing a model removed/renamed in a newer timm release, custom models registered with register_model but no pretrained_cfgs argument, or partial imports that bypass timm.models register decorators.

Related errors

• batch_sizes must contain at least one value.
• Cannot initialize position embeddings without grid_size.Plea
• Unknown rope_type: {cfg.rope_type}
• Invalid net configuration
• Unrecognized union:

AI-assisted analysis of huggingface/pytorch-image-models@9a5261e31b (2026-08-27). Data as JSON: /api/errors/68cf0d6c092e08ea. Report an issue: GitHub.