feat: add backend

main.py

@@ -58,8 +58,9 @@ RETRIEVER = DB.as_retriever()
 class MyCustomSyncHandler(BaseCallbackHandler):
     def __init__(self):
         self.end = False
+        self.token = ""
 
-    async def on_llm_start(
+    def on_llm_start(
         self, serialized: Dict[str, Any], prompts: List[str], **kwargs: Any
     ) -> None:
         self.end = False
@@ -68,7 +69,7 @@ class MyCustomSyncHandler(BaseCallbackHandler):
         self.end = True
 
     def on_llm_new_token(self, token: str, **kwargs) -> Any:
-        return token
+        self.token += token
 
 
 handlerToken = MyCustomSyncHandler()
@@ -257,7 +258,7 @@ async def websocket_endpoint(websocket: WebSocket):
         finish = False
         while finish == False:
             finish = handlerToken.end
-            token = handlerToken.on_llm_new_token()
+            token = handlerToken.token
             await websocket.send_text(f"result: {token}")