DSpark: Speculative decoding accelerates LLM inference [pdf]

 (github.com)

662 points | by aurenvale 9 hours ago

251 comments