G github.com DSpark: Speculative decoding accelerates LLM inference [pdf] 797 points aurenvale 4 months ago 361 comments
