可以先流式输出，最后再执行post-process，这样用户体验更好

[zhouwenjun0820]

          Since we need to post-process the LLM output to match the citation number such as [6-8] with the context sentences, it currently does not support stream output.

Originally posted by @Neo-Zhangjiajie in #6 (comment)

[HanHenryOnlyFaceany]

不太理解，如何做呢