opendatalab/MinerU · error · ValueError

architecture {file_name} is not in arch_config.yaml

Error message

architecture {file_name} is not in arch_config.yaml

What it means

Raised by get_arch_config when the stem (filename without extension) of the model path is not a key in arch_config.yaml (loaded from DEFAULT_CFG_PATH). This mapping is how the inference utility knows which network architecture corresponds to a given weights file.

Source

Thrown at mineru/model/utils/tools/infer/pytorchocr_utility.py:230

    return s_len - math.ceil(en_dg_count / 2)


def base64_to_cv2(b64str):
    import base64
    data = base64.b64decode(b64str.encode('utf8'))
    data = np.fromstring(data, np.uint8)
    data = cv2.imdecode(data, cv2.IMREAD_COLOR)
    return data


def get_arch_config(model_path):
    import yaml
    with open(DEFAULT_CFG_PATH, encoding='utf-8') as f:
        all_arch_config = yaml.safe_load(f)
    path = Path(model_path)
    file_name = path.stem
    if file_name not in all_arch_config:
        raise ValueError(f"architecture {file_name} is not in arch_config.yaml")

    arch_config = all_arch_config[file_name]
    return arch_config

View on GitHub (pinned to 4fe4bde114)

Solutions

1. Restore the original weights filename so its stem matches a known architecture name.
2. Pass yaml_path explicitly to AnalysisConfig so arch inference from the filename is skipped.
3. Update mineru so the bundled arch_config.yaml matches your models version.

Example fix

# before
get_arch_config("/models/my_renamed_rec.pth")

# after
# keep the official filename, or bypass filename inference:
AnalysisConfig(weights_path="/models/my_renamed_rec.pth", yaml_path="/models/rec_config.yml")

import yaml
from pathlib import Path
with open(DEFAULT_CFG_PATH, encoding="utf-8") as f:
    known = yaml.safe_load(f)
if Path(model_path).stem not in known:
    # fall back to explicit config instead of filename inference
    return AnalysisConfig(model_path, yaml_path=explicit_yaml)

try:
    arch = get_arch_config(model_path)
except ValueError:
    arch = read_network_config_from_yaml(explicit_yaml_path)

When it happens

Trigger: Renaming a weights file (e.g. my_model.pth) so it no longer matches known names like ch_PP-OCRv4_rec_infer; using custom/foreign weights whose name is absent from the bundled arch_config.yaml; passing a path with an unexpected extension so path.stem differs.

Common situations: Users renaming downloaded models; side-loading third-party OCR weights; version mismatch between the weights bundle and the arch_config.yaml shipped with the installed mineru version.

Related errors

• {} is not existed.
• Unsupported activation: {name}
• PPLCNetV4 only supports 3 input channels, got {in_channels}.
• PPLCNetV4 {mode} model_size must be one of {list(config_dict
• {} has no Architecture

AI-assisted analysis of opendatalab/MinerU@4fe4bde114 (2026-08-14). Data as JSON: /api/errors/42cbb342c248376c. Report an issue: GitHub.