sgl-project/sglang · error · ValueError

Incorrect type of pixel values. Got type: {type(pixel_values

Error message

Incorrect type of pixel values. Got type: {type(pixel_values)}

What it means

_parse_and_validate_image_input requires pixel_values to be a torch.Tensor or list; anything else (numpy array, None-with-flag, string) is rejected before encoding.

Source

Thrown at python/sglang/srt/models/unlimited_ocr.py:183

    def _parse_and_validate_image_input(self, **kwargs: object):
        """Parse and validate pixel values, spatial crops, and image crops."""
        pixel_values = kwargs.pop("pixel_values", None)
        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,

View on GitHub (pinned to 0132848349)

Solutions

Convert numpy arrays with torch.from_numpy(...).to(device) before calling
Pass None when there are no images so the earlier branch skips encoding
Ensure the input dict keys match what the scheduler provides

Example fix

# before
model.get_multimodal_embeddings(pixel_values=np_imgs, ...)
# after
model.get_multimodal_embeddings(pixel_values=torch.from_numpy(np_imgs), ...)

Defensive patterns

Strategy: type-guard

Validate before calling

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

Type guard

def valid_pixels(p):
    return p is None or isinstance(p, (torch.Tensor, list))

Prevention

Convert numpy to torch at the preprocessing boundary
Use the built-in multimodal processor

When it happens

Trigger: Passing pixel_values as np.ndarray or another type into get_multimodal_embeddings of UnlimitedOCR.

Common situations: Feeders/preprocessors producing numpy arrays; intermediate layers forwarding the wrong field; None handled by earlier branches so unexpected types slip through.

Related errors

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/8d5bfe6790e493dd. Report an issue: GitHub.