sgl-project/sglang · error · ValueError

original_sr must be a positive number, but got {self.audio[1

Error message

original_sr must be a positive number, but got {self.audio[1]}

What it means

The sample-rate element of the (waveform, sr) tuple must be a positive int/float. Zero or negative values are rejected because resampling would be meaningless.

Source

Thrown at python/sglang/srt/multimodal/processors/mimo_audio.py:67

        if not isinstance(self.audio, (str, bytes, tuple, torch.Tensor, np.ndarray)):
            raise ValueError(
                f"audio must be a str, bytes, tuple, torch.Tensor, or np.ndarray, but got {type(self.audio)}"
            )
        if isinstance(self.audio, tuple):
            if (
                len(self.audio) != 2
                or not isinstance(self.audio[0], torch.Tensor)
                or not isinstance(self.audio[1], (int, float))
            ):
                raise ValueError(
                    f"audio must be a tuple of (waveform-T, original_sr-int/float), but got {len(self.audio)} elements and {type(self.audio[0])} and {type(self.audio[1])}"
                )
            if self.audio[0].ndim != 1:
                raise ValueError(
                    f"waveform must be a 1D tensor, but got {self.audio[0].ndim}D tensor"
                )
            if self.audio[1] <= 0:
                raise ValueError(
                    f"original_sr must be a positive number, but got {self.audio[1]}"
                )
        if isinstance(self.audio, torch.Tensor) and self.audio.ndim != 2:
            raise ValueError(
                f"audio must be a 2D tensor, but got {self.audio.ndim}D tensor"
            )


class MiMoAudioPipeline:
    """Stateful audio preprocessing pipeline.

    Composable: held by both MiMoProcessor (multimodal) and MiMoV2ASRProcessor.
    Owns the mel spectrogram, resampler cache, http session, and the special
    token ids for ``<|sosp|> <|empty|>* <|eosp|>`` placeholders.
    """

    def __init__(
        self,

View on GitHub (pinned to 0132848349)

Solutions

1. Pass the true original sample rate, e.g. 16000 or 44100
2. Fix the upstream metadata read that yields 0/-1
3. Validate sr > 0 before constructing the input

Example fix

# before
sr = int(meta.get("sample_rate", 0))
audio = (waveform, sr)
# after
sr = int(meta["sample_rate"])  # 16000
audio = (waveform, sr)

assert isinstance(audio[1], (int, float)) and audio[1] > 0, f"bad sr {audio[1]!r}"

def is_valid_sr(sr) -> bool:
    return isinstance(sr, (int, float)) and sr > 0

When it happens

Trigger: Passing sr = 0, a negative number, or a value that became 0 through a parsing/default bug in the (waveform, sr) tuple.

Common situations: Sample rate read from a metadata field that defaults to 0, or computed by a buggy division; passing -1 as a sentinel meaning 'unknown'.

Related errors

• audio_sr must be positive, got {audio_sr}
• audio must be a str, bytes, tuple, torch.Tensor, or np.ndarr
• audio must be a tuple of (waveform-T, original_sr-int/float)
• waveform must be a 1D tensor, but got {self.audio[0].ndim}D
• audio must be a 2D tensor, but got {self.audio.ndim}D tensor

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/d852d029c6e74208. Report an issue: GitHub.