diff --git a/src/vila/predictors.py b/src/vila/predictors.py
index c96e46f..dbc52e5 100644
--- a/src/vila/predictors.py
+++ b/src/vila/predictors.py
@@ -177,7 +177,8 @@ def predict_page(
         visual_group_detector: Optional[Any] = None,
         page_size=None,
         batch_size=None,
-    ) -> lp.Layout:
+        return_type: Optional[str] = "layout",
+    ) -> Union[lp.Layout, List]:
         """The predict_page function is used for running the model on a single page
         in the vila page_token objects.
 
@@ -187,8 +188,8 @@ def predict_page(
             visual_group_detector:
                 The visual group model to use for detecting the required visual groups.
             page_size (Tuple):
-                A tuple of (width, height) for this page. By default it will use the 
-                page_size from the page_tokens directly unless the page_size is explicitly 
+                A tuple of (width, height) for this page. By default it will use the
+                page_size from the page_tokens directly unless the page_size is explicitly
                 specified.
             batch_size (Optional[int]):
                 Specifying the maximum number of batches for each model run.
@@ -198,7 +199,7 @@ def predict_page(
         required_agg_level = self.preprocessor.config.agg_level
         required_group = AGG_LEVEL_TO_GROUP_NAME[required_agg_level]
 
-        if not getattr(page_tokens, required_group + "s"): # either none or empty
+        if not getattr(page_tokens, required_group + "s"):  # either none or empty
             if page_image is not None and visual_group_detector is not None:
                 warnings.warn(
                     f"The required_group {required_group} is missing in page_tokens."
@@ -216,7 +217,7 @@ def predict_page(
             page_data=pdf_data,
             page_size=page_tokens.page_size if page_size is None else page_size,
             batch_size=batch_size,
-            return_type="layout",
+            return_type=return_type,
         )
 
         return predicted_tokens