I found vllm support streaming in api server. But I want to know is vllm support stream chat in offline inference. I didn't found it in example.