https://hackerpaper.kaipfister.de/p/2402.01528/Collected HN comments and submissions linking this paper.enMon, 02 Nov 2026 11:16:22 GMThttps://news.ycombinator.com/item?id=41676369urn:hackerpapers:mention:2402.01528:41676369Fri, 27 Sep 2024 23:01:55 GMT<p>Orders of magnitude seems a bit ambitious. The implementation from the DeepMind paper achieved a 2-2.5X https://arxiv.org/pdf/2302.01318 and most of the tests I&#x27;ve seen [1][2] have been similar, but there are different variations (Medusa, Ouroboros, etc) that can do better/be combined. Recently Together.ai published SpecExec, a SD variant which did claim to get a 10-18X speedups: https://www.together.ai/blog/specexec [1] https://www.reddit.com/r/LocalLLaMA/comments/17h4rqz/specula... [2] https://arxiv.org/pdf/2402.01528v3</p><p><a href="https://hackerpaper.kaipfister.de/p/2402.01528/">Paper and discussion history</a>. A mention is not an endorsement.</p>cs.LG