apply torch int div to layoutlmv2 (#15457)

* apply torch int div

* black linting fixup

* update path to torch_int_div

* clarify imports

src/transformers/models/layoutlmv2/modeling_layoutlmv2.py

@@ -32,7 +32,7 @@ from ...modeling_outputs import (
     TokenClassifierOutput,
 )
 from ...modeling_utils import PreTrainedModel
-from ...pytorch_utils import apply_chunking_to_forward
+from ...pytorch_utils import apply_chunking_to_forward, torch_int_div
 from ...utils import (
     add_start_docstrings,
     add_start_docstrings_to_model_forward,
@@ -770,25 +770,25 @@ class LayoutLMv2Model(LayoutLMv2PreTrainedModel):
         return embeddings
 
     def _calc_visual_bbox(self, image_feature_pool_shape, bbox, device, final_shape):
-        visual_bbox_x = (
+        visual_bbox_x = torch_int_div(
             torch.arange(
                 0,
                 1000 * (image_feature_pool_shape[1] + 1),
                 1000,
                 device=device,
                 dtype=bbox.dtype,
-            )
-            // self.config.image_feature_pool_shape[1]
+            ),
+            self.config.image_feature_pool_shape[1],
         )
-        visual_bbox_y = (
+        visual_bbox_y = torch_int_div(
             torch.arange(
                 0,
                 1000 * (self.config.image_feature_pool_shape[0] + 1),
                 1000,
                 device=device,
                 dtype=bbox.dtype,
-            )
-            // self.config.image_feature_pool_shape[0]
+            ),
+            self.config.image_feature_pool_shape[0],
         )
         visual_bbox = torch.stack(
             [