I hope the output as token by token during the inference rather than the whole output after the inference finished, if ipex-llm support or not?