Enhancing cost efficiency in LLM serving through an edge-assisted speculative decoding framework.