Sequoia: Speculative decoding boosting LLM inference by 8-10x

3 points| fgfm | 1 year ago |infini-ai-lab.github.io