Update processing_markupdm.py

processing_markupdm.py

@@ -117,7 +117,8 @@ class MarkupDMProcessor(ProcessorMixin):  # type: ignore
             example = self.preprocess_images(example["images"])
 
         assert vision_model is not None, "Vision model must be provided."
-        image = example.pop("image").to(vision_model.device)
+        image = example.pop("image")
+        image = image.to(dtype=vision_model.dtype, device=vision_model.device)
         with torch.inference_mode():
             _, _, (_, _, image_ids) = vision_model.model.encode(image)
         example["image_ids"] = list(image_ids.view(image.size(0), -1).cpu())