Google’s New AI Just Broke My Brain

❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambda.ai/papers

📝 The TurboQuant paper is available here:
https://arxiv.org/abs/2504.19874

Reproductions:
https://github.com/tonbistudio/turboquant-pytorch
https://www.reddit.com/r/Local....LLM/comments/1s6edoi
https://www.reddit.com/r/Local....LLaMA/comments/1s73y
https://x.com/AlicanKiraz0/sta....tus/2038245538865275
I'll note that I have found several reproductions and benchmarks with all kinds of results, so take each of these as just one data point, not definitive yet (thanks for the feedback Alon Torres). Some later tests even better at KV-cache compression!

KV-cache source: https://huggingface.co/blog/not-lain/kv-caching

Reviews and criticisms of the paper:
https://openreview.net/forum?id=tO3ASKZlok
https://x.com/gaoj0017/status/2037532673812443214

Our Patreon if you wish to support us: https://www.patreon.com/TwoMinutePapers

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Adam Bridges, Benji Rabhan, B Shang, Cameron Navor, Charles Ian Norman Venn, Christian Ahlin, Eric T, Fred R, Gordon Child, Juan Benet, Michael Tedder, Owen Skarpness, Richard Sundvall, Ryan Stankye, Shawn Becker, Steef, Taras Bobrovytsky, Tazaur Sagenclaw, Tybie Fitzhugh, Ueli Gallizzi

My research: https://cg.tuwien.ac.at/~zsolnai/
Thumbnail design: https://felicia.hu