可以先流式输出，最后再执行post-process，这样用户体验更好 #15
Description
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)