We announce a survey about efficient generative LLM serving on arXiv. :mega: