PaddlePaddle/PaddleOCR · error · ValueError

{value} is not a valid {cls.__name__}, please select one of

Error message

{value} is not a valid {cls.__name__}, please select one of {list(cls._value2member_map_.keys())}

What it means

ExplicitEnum is the base class (copied from HuggingFace transformers) for tokenizer strategy enums such as TruncationStrategy. Python calls _missing_ when a value cannot be matched to a member; this override replaces the default cryptic message with an explicit list of valid values.

Source

Thrown at ppocr/data/imaug/label_ops.py:1898

            for idx, seq in enumerate(process_seq):
                l = len(seq)
                labels[idx][:l] = seq
            topk[k] = labels
        return (
            np.array(topk["input_ids"]).astype(np.int64),
            np.array(topk["attention_mask"]).astype(np.int64),
            max_length,
        )


class ExplicitEnum(str, Enum):
    """
    Enum with more explicit error message for missing values.
    """

    @classmethod
    def _missing_(cls, value):
        raise ValueError(
            f"{value} is not a valid {cls.__name__}, please select one of {list(cls._value2member_map_.keys())}"
        )


class TruncationStrategy(ExplicitEnum):
    """
    Possible values for the `truncation` argument in [`PreTrainedTokenizerBase.__call__`]. Useful for tab-completion in
    an IDE.
    """

    ONLY_FIRST = "only_first"
    ONLY_SECOND = "only_second"
    LONGEST_FIRST = "longest_first"
    DO_NOT_TRUNCATE = "do_not_truncate"


class PaddingStrategy(ExplicitEnum):
    """

View on GitHub (pinned to 2661c7c0ef)

Solutions

1. Read the valid values from the error message itself and use one of them exactly, e.g. truncation: only_first or longest_only equivalent ('longest' for padding)
2. If the setting comes from YAML, quote the string and avoid boolean keywords: truncation: 'only_first', not truncation: yes
3. Check for boolean leakage: use str(value).lower() comparisons or normalize config values before passing them to the tokenizer
4. Align the transformers version with the one PaddleOCR pins if you also import HF tokenizers

Example fix

# before
tokenizer(text, truncation='True')
# after
from ppocr.data.imaug.label_ops import TruncationStrategy
tokenizer(text, truncation=TruncationStrategy.ONLY_FIRST)

from ppocr.data.imaug.label_ops import TruncationStrategy
if isinstance(truncation, str):
    if truncation not in TruncationStrategy._value2member_map_:
        raise SystemExit(f'truncation must be one of {list(TruncationStrategy._value2member_map_)}')
    truncation = TruncationStrategy(truncation)

def is_valid_truncation(v) -> bool:
    from ppocr.data.imaug.label_ops import TruncationStrategy
    return v is None or v in TruncationStrategy._value2member_map_ or isinstance(v, TruncationStrategy)

try:
    tokenizer(texts, truncation=truncation)
except ValueError as e:
    if 'is not a valid' in str(e):
        # fall back to a safe default instead of crashing the loader
        tokenizer(texts, truncation='only_first')
    else:
        raise

When it happens

Trigger: Passing a 'truncation' (or padding) argument to the tokenizer used by the recognition label pipeline whose value is not one of the enum members, e.g. truncation='True' (string), truncation='both', or a boolean True.

Common situations: Reading truncation/padding settings from a yml config where YAML renders booleans as strings; passing Python True instead of 'longest'/'only_first'; version drift between transformers and the vendored tokenizer code in label_ops.py.

Related errors

• Found a {token.__class__} in the saved `added_tokens_decoder
• RecResizeImg.image_shape is required in rec inference.yml
• Unexpected recognition channels: ${String(channels)}
• OCR pipeline config text must decode to an object.
• OCR pipeline config must be an object or YAML text.

AI-assisted analysis of PaddlePaddle/PaddleOCR@2661c7c0ef (2026-08-14). Data as JSON: /api/errors/2304e340714e6801. Report an issue: GitHub.