huggingface/pytorch-image-models · error · ValueError

Invalid net configuration

Error message

Invalid net configuration 

What it means

_create_selecsls builds a config from a hardcoded table keyed by variant string; an unknown variant falls through to the generic 'Invalid net configuration' error. The public factories (selecsls42, selecsls42b, selecsls60, selecsls60b, selecsls84) always pass valid names, so this fires only on custom/misspelled variants.

Source

Thrown at timm/models/selecsls.py:316

        feature_info.extend([
            dict(num_chs=144, reduction=4, module='features.1'),
            dict(num_chs=304, reduction=8, module='features.6'),
            dict(num_chs=512, reduction=16, module='features.12'),
        ])
        # Head can be replaced with alternative configurations depending on the problem
        cfg['head'] = [
            (512, 960, 3, 2),
            (960, 1024, 3, 1),
            (1024, 1024, 3, 2),
            (1024, 1280, 3, 1),
        ]
        cfg['num_features'] = 1280
        feature_info.extend([
            dict(num_chs=1024, reduction=32, module='head.1'),
            dict(num_chs=1280, reduction=64, module='head.3')
        ])
    else:
        raise ValueError('Invalid net configuration ' + variant + ' !!!')
    cfg['feature_info'] = feature_info

    # this model can do 6 feature levels by default, unlike most others, leave as 0-4 to avoid surprises?
    return build_model_with_cfg(
        SelecSls,
        variant,
        pretrained,
        model_cfg=cfg,
        feature_cfg=dict(out_indices=(0, 1, 2, 3, 4), flatten_sequential=True),
        **kwargs,
    )


def _cfg(url='', **kwargs):
    return {
        'url': url,
        'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': (4, 4),
        'crop_pct': 0.875, 'interpolation': 'bilinear',

View on GitHub (pinned to 9a5261e31b)

Solutions

1. Use one of the registered variants: selecsls42, selecsls42b, selecsls60, selecsls60b, selecsls84
2. If adding a new variant, add a matching branch in _create_selecsls that populates cfg and feature_info

Example fix

# before
model = timm.create_model('selecsls42c', pretrained=False)
# after
model = timm.create_model('selecsls42b', pretrained=False)

from timm.models.selecsls import selecsls_variants if False else None
valid = {'selecsls42','selecsls42b','selecsls60','selecsls60b','selecsls84'}
assert variant in valid

When it happens

Trigger: Calling build_model_with_cfg(SelecSls, 'selecsls42c', ...) or a custom factory with a variant not in the if/elif chain of _create_selecsls.

Common situations: Adding a new SelecSls width/depth variant by copying a factory without extending _create_selecsls, or renaming variants inconsistently.

Related errors

• batch_sizes must contain at least one value.
• Model architecture ({arch_name}) has no pretrained cfg regis
• Cannot initialize position embeddings without grid_size.Plea
• Unknown rope_type: {cfg.rope_type}
• Unrecognized union:

AI-assisted analysis of huggingface/pytorch-image-models@9a5261e31b (2026-08-27). Data as JSON: /api/errors/7792ce9dcfa27b9a. Report an issue: GitHub.