huggingface/pytorch-image-models · error · RuntimeError

No node names found matching {names}.

Error message

No node names found matching {names}.

What it means

AttentionExtractor with graph-based extraction (hook_type='graph') matches the provided node names (fnmatch globs or regexes) against the traced FX graph node names. If nothing matches, RuntimeError is raised because extraction would be a no-op — usually the names don't correspond to any module/functional node in the model's graph.

Source

Thrown at timm/utils/attention_extract.py:53

        if mode == 'train':
            model = model.train()
        else:
            model = model.eval()

        assert method in ('fx', 'hook')
        if method == 'fx':
            # names are activation node names
            from timm.models._features_fx import get_graph_node_names, GraphExtractNet

            node_names = get_graph_node_names(model)[0 if mode == 'train' else 1]
            names = names or self.default_node_names
            if use_regex:
                regexes = [re.compile(r) for r in names]
                matched = [g for g in node_names if any([r.match(g) for r in regexes])]
            else:
                matched = [g for g in node_names if any([fnmatch.fnmatch(g, n) for n in names])]
            if not matched:
                raise RuntimeError(f'No node names found matching {names}.')

            self.model = GraphExtractNet(model, matched, return_dict=True)
            self.hooks = None
        else:
            # names are module names
            assert hook_type in ('forward', 'forward_pre')
            from timm.models._features import FeatureHooks

            module_names = [n for n, m in model.named_modules()]
            names = names or self.default_module_names
            if use_regex:
                regexes = [re.compile(r) for r in names]
                matched = [m for m in module_names if any([r.match(m) for r in regexes])]
            else:
                matched = [m for m in module_names if any([fnmatch.fnmatch(m, n) for n in names])]
            if not matched:
                raise RuntimeError(f'No module names found matching {names}.')

View on GitHub (pinned to 9a5261e31b)

Solutions

1. Inspect node names first: print([n for n in torch.fx.symbolic_trace(model).graph.nodes]) and adjust patterns
2. Use correct glob/regex (remember re.match anchors at the string start; use '.*' prefix if needed)
3. Verify hook_type: if names are module names, drop hook_type='graph' so module matching is used
4. Check the model class actually contains the attention modules you named

Example fix

# before
ext = AttentionExtractor(model, names=['attn_drop'], hook_type='graph')
# after
import torch.fx as fx
print([n.name for n in fx.symbolic_trace(model).graph.nodes])
ext = AttentionExtractor(model, names=['.*attn.*'], use_regex=True, hook_type='graph')

import torch.fx as fx
node_names = [n.name for n in fx.symbolic_trace(model).graph.nodes]
assert any(fnmatch.fnmatch(n, pat) for n in node_names for pat in names), 'no graph node matches'

try:\n    ext = AttentionExtractor(model, names, hook_type='graph')\nexcept RuntimeError as e:\n    if 'No node names' in str(e):\n        print([n for n in node_names])\n        raise\n    raise

When it happens

Trigger: Calling AttentionExtractor(model, names=['attn.softmax'], hook_type='graph') where no graph node has that name; using a wildcard pattern that doesn't match any node; using module-name patterns when graph node names differ (e.g. missing trailing call suffix).

Common situations: Porting hook names from one backbone to another; assuming names look like attribute paths when FX graph node names differ; typos in regex patterns; regex mismatch because re.match anchors at start.

Related errors

• No module names found matching {names}.
• Input image must have positive dimensions, got H={height}, W
• Invalid class map file, expected a dict ({class_map_path}).
• Dataset length is unknown, please pass `num_samples` explici
• Found 0 images in subfolders of {root}. Supported image exte

AI-assisted analysis of huggingface/pytorch-image-models@9a5261e31b (2026-08-27). Data as JSON: /api/errors/47ec765cba2bede3. Report an issue: GitHub.