Scalable and Robust Speculative Decoding

As the usage of large language models (LLMs) grows, it becomes increasingly important to serve them quickly and efficiently.