sgl-project/sglang · error · ValueError

An exception occurred while loading {modality.name} data at

Error message

An exception occurred while loading {modality.name} data at index {idx}: {e}

What it means

In fast_load_mm_data, per-item load failures classified as invalid input (ValueError-family) are re-raised as ValueError with the modality name, index, and cause — preserving 'bad request' semantics for callers to map to a 4xx-style error.

Source

Thrown at python/sglang/srt/multimodal/processors/base_processor.py:1244

            )

        logger.debug("[load_mm_data(simple)] total futures submitted: %d", len(futures))

        images: List[Any] = [None] * len(image_data) if image_data else []
        videos: List[Any] = [None] * len(video_data) if video_data else []
        audios: List[Any] = [None] * len(audio_data) if audio_data else []

        for modality, idx, future in futures:
            try:
                result = await asyncio.wrap_future(future)
            except ValueError as e:
                logger.info(
                    "[load_mm_data(simple)] invalid %s data at index=%d: %s",
                    modality.name,
                    idx,
                    e,
                )
                raise ValueError(
                    f"An exception occurred while loading {modality.name} data "
                    f"at index {idx}: {e}"
                ) from e
            except Exception as e:
                logger.exception(
                    "[load_mm_data(simple)] error loading %s data at index=%d",
                    modality.name,
                    idx,
                )
                raise RuntimeError(
                    f"An exception occurred while loading {modality.name} data at index {idx}: {e}"
                )

            if modality == Modality.IMAGE:
                images[idx] = result
            elif modality == Modality.VIDEO:
                videos[idx] = result
            elif modality == Modality.AUDIO:

View on GitHub (pinned to 0132848349)

Solutions

1. Use the index in the message to identify and fix/remove the offending item
2. Pre-validate each item (scheme, base64 header, file signature) before batching
3. Catch ValueError at the API layer and return a per-item error to the client rather than failing the batch

Example fix

// before
await processor.fast_load_mm_data(mm_data, ...)  # one bad item kills batch
// after
bad = [i for i, it in enumerate(items) if not looks_valid(it)]
items = [it for i, it in enumerate(items) if i not in bad]
await processor.fast_load_mm_data(mm_data, ...)

def looks_valid(item):
    if isinstance(item, str):
        return item.startswith(('http://','https://','data:')) or os.path.exists(item)
    return item is not None

try:
    await processor.fast_load_mm_data(mm_data, ...)
except ValueError as e:
    if 'at index' in str(e):
        idx = int(re.search(r'index (\d+)', str(e)).group(1))
        return per_item_error(idx, str(e))
    raise

When it happens

Trigger: fast_load_mm_data encountering an item at index idx that fails validation/decoding with a ValueError-type error, e.g. malformed base64, wrong content type, unsupported scheme.

Common situations: Batch inference over user-submitted media where one item is malformed; frontend sending data URLs with typos ('data:image/png;base64!...').

Related errors

• Error while loading data {data_str}: {e}
• Unsupported image type: {type(image)}
• When using multiple prompts with multiple input images, prov
• {key}.position_ids is required
• {key}.{field} is required

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/f2cb30fbc7bd2960. Report an issue: GitHub.