huggingface/pytorch-image-models · error · RuntimeError

features_only not implemented for ConvMixer models.

Error message

features_only not implemented for ConvMixer models.

What it means

ConvMixer factory functions reject features_only=True. ConvMixer is a stack of identical conv+patch-embed blocks without hierarchical stages, so it does not register the feature_locations needed by timm's feature extraction wrapper; _create_convmixer raises immediately.

Source

Thrown at timm/models/convmixer.py:114

            x = checkpoint_seq(self.blocks, x)
        else:
            x = self.blocks(x)
        return x

    def forward_head(self, x, pre_logits: bool = False):
        x = self.pooling(x)
        x = self.head_drop(x)
        return x if pre_logits else self.head(x)

    def forward(self, x):
        x = self.forward_features(x)
        x = self.forward_head(x)
        return x


def _create_convmixer(variant, pretrained=False, **kwargs):
    if kwargs.get('features_only', None):
        raise RuntimeError('features_only not implemented for ConvMixer models.')

    return build_model_with_cfg(ConvMixer, variant, pretrained, **kwargs)


def _cfg(url='', **kwargs):
    return {
        'url': url,
        'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None,
        'crop_pct': .96, 'interpolation': 'bicubic',
        'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD, 'classifier': 'head',
        'first_conv': 'stem.0', 'license': 'mit',
        **kwargs
    }


default_cfgs = generate_default_cfgs({
    'convmixer_1536_20.in1k': _cfg(hf_hub_id='timm/'),
    'convmixer_768_32.in1k': _cfg(hf_hub_id='timm/'),

View on GitHub (pinned to 9a5261e31b)

Solutions

1. Drop features_only and use forward_intermediates to obtain intermediate ConvMixer block outputs
2. If true multi-scale maps are needed, choose a hierarchical backbone (e.g. ResNet, ConvNeXt, EfficientNet)
3. Register a custom feature_cfgs via timm's build_model_with_cfg escape hatch only if you fully control the wrapper

Example fix

# before
model = timm.create_model('convmixer_1536_20', features_only=True)
# after
model = timm.create_model('convmixer_1536_20')
feats = model.forward_intermediates(x, indices=[5, 10, 15, 19])

import timm
m = timm.create_model('convmixer_1024_20_ks9_p14')
assert not hasattr(m, 'feature_info'), 'no feature_info; do not pass features_only'
model = timm.create_model('convmixer_1024_20_ks9_p14')

def supports_features_only(name: str) -> bool:
    import timm
    m = timm.create_model(name)
    ok = hasattr(m, 'feature_info')
    del m
    return ok

try:
    model = timm.create_model(variant, features_only=True)
except RuntimeError:
    model = timm.create_model(variant)

When it happens

Trigger: Calling timm.create_model('convmixer_1024_20_ks9_p14', features_only=True) (or the 1536/768 variants) with features_only truthy.

Common situations: Reuse of detection/segmentation scaffold code that unconditionally sets features_only=True; assuming feature-extraction support is uniform across timm models.

Related errors

• features_only not implemented for Vision Transformer models.
• features_only not implemented for Vision Transformer models.
• features_only not implemented for Vision Transformer models.
• Gemma4VitEncoder does not support classification use cases.
• MobileNetV5Encoder does not support classification use cases

AI-assisted analysis of huggingface/pytorch-image-models@9a5261e31b (2026-08-27). Data as JSON: /api/errors/a22f47aee51040b5. Report an issue: GitHub.