Make minimal monkey-patch

Signed-off-by: Christoph Auer <cau@zurich.ibm.com>

Author: cau-git

docling/models/stages/ocr/nemotron_ocr_model.py

@@ -29,17 +29,22 @@ class _GridSamplerStorageWorkaround:
     # Temporary upstream workaround:
     # `nemotron_ocr_cpp.indirect_grid_sample_forward(...)` intermittently
     # raises "Tensor doesn't have storage" for tensors that appear valid from
-    # Python. Force fresh owned contiguous storage at the wrapper boundary.
+    # Python. Retry the failed call once with fresh owned contiguous tensors.
     def __init__(self, original_sampler: Any):
         self._original_sampler = original_sampler
 
     def __call__(self, input_tensor: Any, grid: Any, input_indices: Any) -> Any:
         # Workaround call site for the upstream custom-op storage bug.
-        return self._original_sampler(
-            input_tensor.contiguous().clone(),
-            grid.contiguous().clone(),
-            input_indices.contiguous().clone(),
-        )
+        try:
+            return self._original_sampler(input_tensor, grid, input_indices)
+        except RuntimeError as exc:
+            if "doesn't have storage" not in str(exc):
+                raise
+            return self._original_sampler(
+                input_tensor.contiguous().clone(),
+                grid.contiguous().clone(),
+                input_indices.contiguous().clone(),
+            )
 
 
 class NemotronOcrPrediction(TypedDict):