fix

Author: mayinghan

eval_protocol/models.py

@@ -258,7 +258,6 @@ def __iter__(self):
 class Message(BaseModel):
     """Chat message model with trajectory evaluation support."""
 
-    model_config = ConfigDict(extra="allow")
     role: str  # assistant, user, system, tool
     content: Optional[Union[str, List[ChatCompletionContentPartTextParam]]] = Field(
         default="", description="The content of the message."
@@ -271,13 +270,11 @@ class Message(BaseModel):
     tool_calls: Optional[List[ChatCompletionMessageToolCall]] = None
     function_call: Optional[FunctionCall] = None
     control_plane_step: Optional[Dict[str, Any]] = None
+    weight: Optional[int] = None
 
     def dump_mdoel_for_chat_completion_request(self):
         """Only keep chat completion accepted fields"""
-        exclude_fields = (
-            {"control_plane_step", "reasoning_content"} | set(self.model_extra.keys()) if self.model_extra else set()
-        )
-        return self.model_dump(exclude_none=True, exclude=exclude_fields)
+        return self.model_dump(exclude_none=True, exclude={"control_plane_step", "reasoning_content", "weight"})
 
     @classmethod
     def model_validate(cls, obj, *args, **kwargs):

tests/adapters/test_openai_responses_adapter.py

@@ -22,7 +22,7 @@ def test_openai_responses_adapter_with_real_response_simple(snapshot: SnapshotAs
     assert len(eval_rows) == 1
 
     # Convert to dict for snapshot testing
-    eval_rows_dict = [row.model_dump(exclude={"created_at", "execution_metadata"}) for row in eval_rows]
+    eval_rows_dict = [row.model_dump(exclude={"created_at": True, "execution_metadata": True, "messages": {"__all__": {"weight"}}}) for row in eval_rows]
 
     # Assert against snapshot
     assert eval_rows_dict == snapshot
@@ -42,7 +42,7 @@ def test_openai_responses_adapter_with_real_response_parallel_tool_calls(snapsho
     assert len(eval_rows) == 1
 
     # Convert to dict for snapshot testing
-    eval_rows_dict = [row.model_dump(exclude={"created_at", "execution_metadata"}) for row in eval_rows]
+    eval_rows_dict = [row.model_dump(exclude={"created_at": True, "execution_metadata": True, "messages": {"__all__": {"weight"}}}) for row in eval_rows]
 
     # Assert against snapshot
     assert eval_rows_dict == snapshot