Sync DeepSeek OCR HF job code

ds_batch_ocr/stages.py

@@ -165,6 +165,44 @@ def _push_dataset_records(
     for record in records:
         new_record = dict(record)
         new_record["figures"] = _normalize_figures(record.get("figures", []))
+
+        coerced_figures: List[Dict[str, str]] = []
+        for fig in new_record["figures"]:
+            if isinstance(fig, dict):
+                coerced_figures.append(
+                    {
+                        "figure_id": str(fig.get("figure_id", "") or ""),
+                        "image_path": str(fig.get("image_path", "") or ""),
+                        "description": str(fig.get("description", "") or ""),
+                    }
+                )
+                continue
+
+            if isinstance(fig, (list, tuple)):
+                coerced_figures.append(
+                    {
+                        "figure_id": str(fig[0]) if len(fig) > 0 and fig[0] is not None else "",
+                        "image_path": str(fig[1]) if len(fig) > 1 and fig[1] is not None else "",
+                        "description": str(fig[2]) if len(fig) > 2 and fig[2] is not None else "",
+                    }
+                )
+                LOGGER.debug("Coerced list/tuple figure entry to dict: %s", fig)
+                continue
+
+            LOGGER.warning(
+                "Unexpected figure entry type after normalization | type=%s | value=%r",
+                type(fig),
+                fig,
+            )
+            coerced_figures.append(
+                {
+                    "figure_id": str(fig),
+                    "image_path": "",
+                    "description": "",
+                }
+            )
+
+        new_record["figures"] = coerced_figures
         normalized_records.append(new_record)
 
     dataset = Dataset.from_list(normalized_records, features=_dataset_features())