fix: normalize legacy parser configuration (#17761)

api/utils/validation_utils.py

@@ -371,6 +371,35 @@ class RaptorConfig(Base):
     scope: Annotated[Literal["file", "dataset"], Field(default="file")]
     auto_disable_for_structured_data: Annotated[bool, Field(default=True)]
 
+    @model_validator(mode="before")
+    @classmethod
+    def normalize_legacy_fields(cls, value: Any) -> Any:
+        """Accept old RAPTOR fields but do not retain them in the config."""
+        if not isinstance(value, dict):
+            return value
+
+        normalized = dict(value)
+        changed_fields = []
+        legacy_ext = normalized.pop("ext", None)
+        if legacy_ext is not None:
+            changed_fields.append("ext")
+        if isinstance(legacy_ext, dict) and normalized.get("clustering_threshold") is None:
+            if "clustering_threshold" in legacy_ext:
+                normalized["clustering_threshold"] = legacy_ext["clustering_threshold"]
+                changed_fields.append("ext.clustering_threshold")
+
+        for field in ("threshold", "clustering_method", "tree_builder"):
+            if field in normalized:
+                normalized.pop(field)
+                changed_fields.append(field)
+        max_token = normalized.get("max_token")
+        if isinstance(max_token, (int, float)) and not isinstance(max_token, bool) and max_token < 512:
+            normalized["max_token"] = 512
+            changed_fields.append("max_token")
+        if changed_fields:
+            logging.debug("RaptorConfig normalized legacy fields: %s", sorted(changed_fields))
+        return normalized
+
 
 class GraphragConfig(Base):
     """Dataset parser configuration for GraphRAG generation."""