feat: Do not pass autorater_config for predefined metrics in SDK

PiperOrigin-RevId: 910847685

Author: vertex-sdk-bot

vertexai/_genai/_evals_metric_handlers.py

@@ -1026,11 +1026,9 @@ def _build_request_payload(
             "instance": instance_payload,
         }
 
-        autorater_config = _get_autorater_config(self.metric)
-        if autorater_config:
-            request_payload["autorater_config"] = genai_types.AutoraterConfig(
-                **autorater_config
-            )
+        # Note: autorater_config is intentionally not passed for predefined
+        # metrics. The server uses its own model configuration for predefined
+        # metrics and ignores the autorater_config field.
         return request_payload
 
     @override
@@ -1045,7 +1043,6 @@ def get_metric_result(
                 lambda: self.module._evaluate_instances(
                     metrics=[self.metric],
                     instance=payload.get("instance"),
-                    autorater_config=payload.get("autorater_config"),
                 ),
                 metric_name,
             )

vertexai/_genai/types/common.py

@@ -2392,7 +2392,7 @@ class EvaluationRunConfig(_common.BaseModel):
         default=None, description="""The output config for the evaluation run."""
     )
     autorater_config: Optional[genai_types.AutoraterConfig] = Field(
-        default=None, description="""The autorater config for the evaluation run."""
+        default=None, description="""The autorater config for the evaluation run. Not applicable for predefined metrics, the server uses its own model configuration for predefined metrics and this field is ignored."""
     )
     prompt_template: Optional[EvaluationRunPromptTemplate] = Field(
         default=None, description="""The prompt template used for inference."""
@@ -2413,7 +2413,7 @@ class EvaluationRunConfigDict(TypedDict, total=False):
     """The output config for the evaluation run."""
 
     autorater_config: Optional[genai_types.AutoraterConfigDict]
-    """The autorater config for the evaluation run."""
+    """The autorater config for the evaluation run. Not applicable for predefined metrics, the server uses its own model configuration for predefined metrics and this field is ignored."""
 
     prompt_template: Optional[EvaluationRunPromptTemplateDict]
     """The prompt template used for inference."""
@@ -4533,7 +4533,7 @@ class _EvaluateInstancesRequestParameters(_common.BaseModel):
         default=None, description=""""""
     )
     autorater_config: Optional[genai_types.AutoraterConfig] = Field(
-        default=None, description=""""""
+        default=None, description="""The autorater config for the evaluation. Not applicable for predefined metrics, the server uses its own model configuration for predefined metrics and this field is ignored."""
     )
     metrics: Optional[list[Metric]] = Field(
         default=None,