Update main.py

main.py

@@ -98,7 +98,7 @@ async def stream_response(tokens: Any) -> None:
                             'role': 'system',
                             'content': llm.detokenize(chat_chunk)
                         },
-                        'finish_reason': 'stop' if llm.detokenize(chat_chunk) == "[DONE]" else 'unknown'
+                        'finish_reason': 'stop' if llm.is_eos_token(chat_chunk) else 'unknown'
                     }
                 ]
             }