huggingface/transformers · error · ValueError

Asking to pad but the feature_extractor does not have a padd

Error message

Asking to pad but the feature_extractor does not have a padding value. Please select a value to use as `padding_value`. For example: `feature_extractor.padding_value = 0.0`.

What it means

Padding writes self.padding_value into the padded region. If the extractor's padding_value is None (unset on the instance or subclass), numpy's np.pad would fail ambiguously, so the code raises a clear ValueError up front and suggests a value such as 0.0.

Source

Thrown at src/transformers/feature_extraction_sequence_utils.py:364

            if padding is True:
                padding_strategy = PaddingStrategy.LONGEST  # Default to pad to the longest sequence in the batch
            elif not isinstance(padding, PaddingStrategy):
                padding_strategy = PaddingStrategy(padding)
            elif isinstance(padding, PaddingStrategy):
                padding_strategy = padding
        else:
            padding_strategy = PaddingStrategy.DO_NOT_PAD

        # Set max length if needed
        if max_length is None:
            if padding_strategy == PaddingStrategy.MAX_LENGTH:
                raise ValueError(
                    f"When setting ``padding={PaddingStrategy.MAX_LENGTH}``, make sure that max_length is defined"
                )

        # Test if we have a padding value
        if padding_strategy != PaddingStrategy.DO_NOT_PAD and (self.padding_value is None):
            raise ValueError(
                "Asking to pad but the feature_extractor does not have a padding value. Please select a value to use"
                " as `padding_value`. For example: `feature_extractor.padding_value = 0.0`."
            )

        return padding_strategy

    def fetch_audio(self, audio_url_or_urls: str | list[str] | list[list[str]], sampling_rate: int | None = None):
        """
        Convert a single or a list of urls into the corresponding `np.ndarray` objects.

        If a single url is passed, the return value will be a single object. If a list is passed a list of objects is
        returned.
        """
        # Accepted input types for `raw_audio`: "np.ndarray | list[float] | list[np.ndarray] | list[list[float]]"
        sampling_rate = sampling_rate if sampling_rate else self.sampling_rate
        if isinstance(audio_url_or_urls, list) and not isinstance(audio_url_or_urls[0], float):
            return [self.fetch_audio(x, sampling_rate=sampling_rate) for x in audio_url_or_urls]
        elif isinstance(audio_url_or_urls, str):

View on GitHub (pinned to a597f97485)

Solutions

1. Set feature_extractor.padding_value = 0.0 (or the correct pad value for your modality) before padding
2. In a subclass, set self.padding_value in __init__
3. Persist the value by saving the updated preprocessor config

Example fix

# before
fe = MyFeatureExtractor()
fe(audio, padding=True)  # padding_value is None

# after
fe.padding_value = 0.0
fe(audio, padding=True)

if fe.padding_value is None:
    fe.padding_value = 0.0

def is_paddable(fe) -> bool:
    return fe.padding_value is not None

try:
    fe.pad(batch, padding=True)
except ValueError as e:
    if "padding value" in str(e):
        fe.padding_value = 0.0
        fe.pad(batch, padding=True)
    else:
        raise

When it happens

Trigger: Calling any padding path on a feature extractor whose padding_value attribute is None — typically a custom subclass that did not set it, or an instance where it was explicitly cleared.

Common situations: Writing a custom sequence feature extractor and forgetting padding_value; loading a config where padding_value is null; models whose correct pad value is nonzero (e.g. attention-style masks) and was never configured.

Related errors

• Invalid padding strategy:{padding_side}
• You should supply an instance of `transformers.BatchFeature`
• Some items in the output dictionary have a different batch s
• When setting ``padding={PaddingStrategy.MAX_LENGTH}``, make
• Found 'model.config.return_loss=True'. Loss computation is n

AI-assisted analysis of huggingface/transformers@a597f97485 (2026-08-14). Data as JSON: /api/errors/e704c024c8b72d6b. Report an issue: GitHub.