Merge pull request #13072 from grandtele/fix/openai_o_series_handler

* fix: Refactor openai handler to support all o-series models universally

backend/open_webui/routers/openai.py

@@ -94,20 +94,19 @@ async def cleanup_response(
         await session.close()
 
 
-def openai_o1_o3_handler(payload):
+def openai_o_series_handler(payload):
     """
-    Handle o1, o3 specific parameters
+    Handle "o" series specific parameters
     """
     if "max_tokens" in payload:
-        # Remove "max_tokens" from the payload
+        # Convert "max_tokens" to "max_completion_tokens" for all o-series models
         payload["max_completion_tokens"] = payload["max_tokens"]
         del payload["max_tokens"]
 
-    # Fix: o1 and o3 do not support the "system" role directly.
-    # For older models like "o1-mini" or "o1-preview", use role "user".
-    # For newer o1/o3 models, replace "system" with "developer".
+    # Handle system role conversion based on model type
     if payload["messages"][0]["role"] == "system":
         model_lower = payload["model"].lower()
+        # Legacy models use "user" role instead of "system"
         if model_lower.startswith("o1-mini") or model_lower.startswith("o1-preview"):
             payload["messages"][0]["role"] = "user"
         else:
@@ -670,10 +669,10 @@ async def generate_chat_completion(
     url = request.app.state.config.OPENAI_API_BASE_URLS[idx]
     key = request.app.state.config.OPENAI_API_KEYS[idx]
 
-    # Fix: o1,o3 does not support the "max_tokens" parameter, Modify "max_tokens" to "max_completion_tokens"
-    is_o1_o3 = payload["model"].lower().startswith(("o1", "o3-"))
-    if is_o1_o3:
-        payload = openai_o1_o3_handler(payload)
+    # Check if model is from "o" series
+    is_o_series = payload["model"].lower().startswith(("o1", "o3", "o4"))
+    if is_o_series:
+        payload = openai_o_series_handler(payload)
     elif "api.openai.com" not in url:
         # Remove "max_completion_tokens" from the payload for backward compatibility
         if "max_completion_tokens" in payload: