DSpark: Speculative decoding accelerates LLM inference [pdf]

Posted 8 hours ago by aurenvale
645 points

https://github.com/deepseek-ai/DeepSpec/blob/main/DSpark_paper.pdf

25 comments

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..

Loading..