sgl-project/sglang · error · ValueError

Incorrect type of image sizes. Got type: {type(images_spatia

Error message

Incorrect type of image sizes. Got type: {type(images_spatial_crop)}

What it means

Companion guard to 5637: images_spatial_crop must be a torch.Tensor or list; other types raise with the offending type in the message.

Source

Thrown at python/sglang/srt/models/unlimited_ocr.py:187

        images_spatial_crop = kwargs.pop("images_spatial_crop", None)
        images_crop = kwargs.pop("images_crop", None)
        has_images = kwargs.pop("has_images", None)

        if pixel_values is None:
            return None
        if has_images is not None:
            if not has_images:
                return None
        elif torch.sum(pixel_values).item() == 0:
            return None

        if pixel_values is not None:
            if not isinstance(pixel_values, (torch.Tensor, list)):
                raise ValueError(
                    "Incorrect type of pixel values. " f"Got type: {type(pixel_values)}"
                )
            if not isinstance(images_spatial_crop, (torch.Tensor, list)):
                raise ValueError(
                    "Incorrect type of image sizes. "
                    f"Got type: {type(images_spatial_crop)}"
                )
            if not isinstance(images_crop, (torch.Tensor, list)):
                raise ValueError(
                    "Incorrect type of image crop. " f"Got type: {type(images_crop)}"
                )
            return [pixel_values, images_crop, images_spatial_crop]

        raise AssertionError("This line should be unreachable.")

    def _pixel_values_to_embedding(
        self,
        pixel_values: torch.Tensor,
        images_crop: torch.Tensor,
        images_spatial_crop: torch.Tensor,
        has_local_crops: Optional[List[bool]] = None,
    ) -> NestedTensors:

View on GitHub (pinned to 0132848349)

Solutions

Ensure the processor supplies images_spatial_crop as tensor/list alongside pixel_values
Convert numpy inputs to torch tensors
Use the standard sglang multimodal processor path instead of manual payloads

Example fix

# before
spatial_crop=np.array([[2,2]])
# after
spatial_crop=torch.tensor([[2,2]])

Defensive patterns

Strategy: type-guard

Validate before calling

assert images_spatial_crop is None or isinstance(images_spatial_crop, (torch.Tensor, list))

Type guard

def valid_spatial_crop(s):
    return s is None or isinstance(s, (torch.Tensor, list))

Prevention

Build all image fields through one processor call
Type-check payloads in tests

When it happens

Trigger: Supplying images_spatial_crop as numpy array or omitted/wrong object while pixel_values is present.

Common situations: Custom request payloads lacking spatial crop metadata or providing numpy formats; processor version changes renaming the field.

Related errors

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/38a23ceb3d10f55a. Report an issue: GitHub.