h

Author: codelion

optillm/server.py

@@ -409,6 +409,7 @@ def execute_single_approach(approach, system_prompt, initial_query, client, mode
             if initial_query:
                 messages.append({"role": "user", "content": initial_query})
 
+            logger.debug(f"none_approach kwargs: {kwargs}")
             response = none_approach(original_messages=messages, client=client, model=model, request_id=request_id, **kwargs)
             # For none approach, we return the response and a token count of 0
             # since the full token count is already in the response

tests/test_api_compatibility.py

@@ -80,10 +80,10 @@ def test_extra_body_approach(client):
         messages=[
             {"role": "user", "content": "What is 2+2?"}
         ],
-        extra_body={"optillm_approach": "bon"},
-        max_tokens=10
+        extra_body={"optillm_approach": "re2"},  # Use re2 instead of bon (simpler, avoids role ordering issues with some models)
+        max_tokens=50
     )
-    
+
     assert hasattr(response, 'choices')
     assert len(response.choices) > 0