Fix UnboundLocalError for stop_sequences in reasoning logic

coderai

@@ -5561,24 +5561,8 @@ async def chat_completions(request: ChatCompletionRequest, http_request: Request
                     break
             if not system_found:
                 messages = [ChatMessage(role="system", content=system_content)] + list(messages)
-        
-        # Handle stop tokens - add to stop_sequences for generation
-        if reasoning_action in ("stop", "both"):
-            _, _, additional_stops = get_reasoning_stop_tokens(model_family)
-            # Add model-specific stop tokens to the existing stop sequences
-            for stop_token in additional_stops:
-                if stop_token not in stop_sequences:
-                    stop_sequences.append(stop_token)
-            print(f"DEBUG: Added reasoning stop tokens for model family '{model_family}': {additional_stops}")
-    
-    # Format messages with tools if provided
-    if request.tools:
-        messages = format_tools_for_prompt(request.tools, messages)
-    
-    # Get the tool_parser from the current manager
-    tool_parser = current_manager.tool_parser if hasattr(current_manager, 'tool_parser') else ModelParserAdapter()
     
-    # Prepare stop sequences
+    # Prepare stop sequences (before reasoning block to avoid UnboundLocalError)
     stop_sequences = []
     if request.stop:
         if isinstance(request.stop, str):
@@ -5586,6 +5570,22 @@ async def chat_completions(request: ChatCompletionRequest, http_request: Request
         else:
             stop_sequences = request.stop
     
+    # Handle stop tokens - add to stop_sequences for generation
+    if reasoning_enabled and reasoning_action in ("stop", "both"):
+        _, _, additional_stops = get_reasoning_stop_tokens(model_family)
+        # Add model-specific stop tokens to the existing stop sequences
+        for stop_token in additional_stops:
+            if stop_token not in stop_sequences:
+                stop_sequences.append(stop_token)
+        print(f"DEBUG: Added reasoning stop tokens for model family '{model_family}': {additional_stops}")
+    
+    # Format messages with tools if provided
+    if request.tools:
+        messages = format_tools_for_prompt(request.tools, messages)
+    
+    # Get the tool_parser from the current manager
+    tool_parser = current_manager.tool_parser if hasattr(current_manager, 'tool_parser') else ModelParserAdapter()
+    
     # Convert messages to dict format for chat completion
     messages_dict = []
     for msg in messages: