Add request_config support to cot_reflection

The cot_reflection function now accepts a request_config parameter, allowing dynamic configuration of temperature and max_tokens for API calls. Defaults are set to temperature=0.6 and max_tokens=4096 if not provided.

Author: codelion

optillm.py

@@ -331,7 +331,7 @@ def execute_single_approach(approach, system_prompt, initial_query, client, mode
                           c=server_config['rstar_c'])
             return rstar.solve(initial_query)
         elif approach == "cot_reflection":
-            return cot_reflection(system_prompt, initial_query, client, model, return_full_response=server_config['return_full_response'])
+            return cot_reflection(system_prompt, initial_query, client, model, return_full_response=server_config['return_full_response'], request_config=request_config)
         elif approach == 'plansearch':
             return plansearch(system_prompt, initial_query, client, model, n=server_config['n'])
         elif approach == 'leap':

optillm/cot_reflection.py

@@ -3,8 +3,16 @@
 
 logger = logging.getLogger(__name__)
 
-def cot_reflection(system_prompt, initial_query, client, model: str, return_full_response: bool=False):
+def cot_reflection(system_prompt, initial_query, client, model: str, return_full_response: bool=False, request_config: dict = None):
     cot_completion_tokens = 0
+    
+    # Extract temperature and max_tokens from request_config with defaults
+    temperature = 0.6  # Default to 0.6 as requested
+    max_tokens = 4096  # Default to 4096 as requested
+    
+    if request_config:
+        temperature = request_config.get('temperature', temperature)
+        max_tokens = request_config.get('max_tokens', max_tokens)
     cot_prompt = f"""
         {system_prompt}
 
@@ -32,15 +40,15 @@ def cot_reflection(system_prompt, initial_query, client, model: str, return_full
         </output>
         """
 
-    # Make the API call
+    # Make the API call using user-provided or default parameters
     response = client.chat.completions.create(
         model=model,
         messages=[
             {"role": "system", "content": cot_prompt},
             {"role": "user", "content": initial_query}
         ],
-        temperature=0.7,
-        max_tokens=4096
+        temperature=temperature,
+        max_tokens=max_tokens
     )
 
     # Extract the full response