huggingface/pytorch-image-models · warning

Overwriting in registry with . . This is because the name…

Error message

Overwriting {model_name} in registry with {fn.__module__}.{model_name}. This is because the name being registered conflicts with an existing name. Please check if this is not expected.

What it means

timm's model registry warns whenever a model name is registered twice, overwriting the previous entrypoint. This happens when two @register_model-decorated functions share the same name (often across different modules). The registry silently keeps only the last one, which may not be what you want.

Solutions

1. Rename your custom model function (the registry key is the function name) so it does not collide with the built-in name
2. If overriding is intentional, filter/suppress the warning with warnings.filterwarnings around the import
3. Check timm.list_models() and the module of the registered entrypoint (timm.models._registry) to identify which module is overwriting which

Example fix

// before
@register_model
def vit_small_patch16_224(pretrained=False, **kwargs):  # collides with timm
    ...
// after
@register_model
def my_vit_small_patch16_224(pretrained=False, **kwargs):
    ...

import timm
from timm.models._registry import _model_entrypoints
name = 'vit_small_patch16_224'
if name in _model_entrypoints:
    print('already registered by', _model_entrypoints[name].__module__)  # rename yours if it's not yours

When it happens

Trigger: Two functions decorated with @register_model produce the same model_name (e.g. defining a custom 'vit_small_patch16_224' while timm already registers one); re-registering an existing timm model under its own name; plugin packages colliding with built-in names.

Common situations: Custom model forks that keep the original function name, duplicated modules both imported, refactors that copy a model file and import both copies, or downstream libraries overriding timm models unintentionally.

Related errors

• Importing from is deprecated, please import via timm.models
• A progressive schedule requires at least two choices.
• adamw_lr is deprecated, use fallback_lr_scale=adamw_lr/lr…
• All scheduled batch sizes must be positive integers.
• API has changed, `state_steps` argument must contain a list…

AI-assisted analysis of huggingface/pytorch-image-models@9a5261e31b (2026-08-27). Data as JSON: /api/errors/1c81f92d9a7cf4b3. Report an issue: GitHub.

Appendix: source

Thrown at timm/models/_registry.py:90

    return out


def register_model(fn: Callable[..., Any]) -> Callable[..., Any]:
    # lookup containing module
    mod = sys.modules[fn.__module__]
    module_name_split = fn.__module__.split('.')
    module_name = module_name_split[-1] if len(module_name_split) else ''

    # add model to __all__ in module
    model_name = fn.__name__
    if hasattr(mod, '__all__'):
        mod.__all__.append(model_name)
    else:
        mod.__all__ = [model_name]  # type: ignore

    # add entries to registry dict/sets
    if model_name in _model_entrypoints:
        warnings.warn(
            f'Overwriting {model_name} in registry with {fn.__module__}.{model_name}. This is because the name being '
            'registered conflicts with an existing name. Please check if this is not expected.',
            stacklevel=2,
        )
    _model_entrypoints[model_name] = fn
    _model_to_module[model_name] = module_name
    _module_to_models[module_name].add(model_name)
    if hasattr(mod, 'default_cfgs') and model_name in mod.default_cfgs:
        # this will catch all models that have entrypoint matching cfg key, but miss any aliasing
        # entrypoints or non-matching combos
        default_cfg = mod.default_cfgs[model_name]
        if not isinstance(default_cfg, DefaultCfg):
            # new style default cfg dataclass w/ multiple entries per model-arch
            assert isinstance(default_cfg, dict)
            # old style cfg dict per model-arch
            pretrained_cfg = PretrainedCfg(**default_cfg)
            default_cfg = DefaultCfg(tags=deque(['']), cfgs={'': pretrained_cfg})

View on GitHub (pinned to 9a5261e31b)