Christopher De Sa

[Dan Fu @realDanFu]

See everyone at MLSys 2025 in Santa Clara next week!

Super excited to organize the Young Professional Symposium program on day one (May 12) and our invited speakers @soumithchintala (keynote), @Tim_Dettmers, @infwinston, @simran_s_arora, @BeidiChen!

[Lester Mackey @LesterMackey]

@annabelle_cs @GongAlbert @AShettyV @raazdwivedi Inspired by the groundbreaking work of @insu_han @daliri__majid @aminkarbasi @WentaoGuo7 @chrismdesa @afedercooper @cdomingoenrich

[Matei Zaharia @matei_zaharia]

We're thrilled to announce the keynote speakers for #MLSys2025: @AnimaAnandkumar, @soumithchintala, Ling Liu and @istoica05! Registration is open to attend the conference in Santa Clara.

[Tianqi Chen @tqchenml]

Learn more about the latest advances in AI and systems, including LLM serving, efficient attentions, structured outputs, scaling up training, and more topics. Check out #MLSys2025. Accepted papers at https://t.co/sTsbrxWHlw and register today at https://t.co/2iRbuiDirc

[Katherine Lee is at ICML! @katherine1ee]

Machine unlearning is taking off! There is a ton of interest in getting generative AI models to “unlearn” targeted undesirable information, often to meet policy aims in privacy, copyright, safety, etc.

Our new cross-disciplinary collaboration shows why this isn’t simple.

[Albert Tseng @tsengalb99]

🦙QTIP 2, 3 and 4 bit Llama 3.3 models are now up on HF: https://t.co/mIkLRdbw76. Almost no zeroshot degradation at all sizes. 2 bit QTIP 3.3 70B fits on a single 4090 and gives pretty high quality generations (📹).

[Albert Tseng @tsengalb99]

🧵

🏎️ Want faster, better quantized LLMs? Introducing QTIP, a new LLM quantization method that achieves a SOTA combination of quality and speed – outperforming methods like QuIP#!

🧑‍💻+🦙(w/ 2 bit 405B!): https://t.co/o0m95o0WWo

📜https://t.co/uwdhZB5OOB

[Christopher De Sa @chrismdesa]

MLSys 2024 is next week. Looking forward to seeing you all in Santa Clara!

https://t.co/yagYbd1nyn

[SambaNova @SambaNovaAI]

🚀🌟🚀Excited to announce Samba-CoE v0.2, which outperforms DBRX by @DbrxMosaicAI and @databricks, Mixtral-8x7B from @MistralAI, and Grok-1 by @grok at a breakneck speed of 330 tokens/s.
These breakthrough speeds were achieved without sacrificing precision and only on 8 sockets, showcasing the true capabilities of dataflow! Why would you buy 576 sockets and go to 8 bits when you can run using 16 bits and just 8 sockets. Try out the model and check out the speed here - https://t.co/A3lB3RTEo2.

We are also providing a sneak peak of our next model, Samba-CoE v0.3, available soon with our partners at @LeptonAI. Read more about this announcement at https://t.co/DIiuJkPlKg

[Christopher De Sa @chrismdesa]

New this year, there will be a Young Professionals Symposium on Monday, which provides a forum for young professionals in industry and academia, to discuss important research & career questions/challenges: abstract submissions are now open (https://t.co/CEFUYSWRKu).

[Christopher De Sa @chrismdesa]

We are excited to announce the technical program for MLSys 2024! The provisional set of accepted papers is now available on the website at https://t.co/JPl5HbENr0.

Register for MLSys now at https://t.co/4yNtiwQkUQ

[Christopher De Sa @chrismdesa]

This year’s conference will be held at the Santa Clara Convention Center, from Mon May 13th through Thu the 16th, and features keynote speakers: Yejin Choi (University of Washington/Allen Institute for AI), Jeff Dean (Google), and Zico Kolter (CMU/Bosch Center for AI).

[Katherine Lee is at ICML! @katherine1ee]

We have a fun attack that lets you extract the last-layer embedding weights of an LM via public APIs. It's really simple & uses SVDs!

https://t.co/GZnCrJUSNd

We discovered this for ChatGPT + PaLM-2. We privately disclosed, they fixed, now, we release :)

[Christopher De Sa @chrismdesa]

We are excited to announce this year’s keynote speakers for #MLSys2024: Jeff Dean @JeffDean, Zico Kolter @zicokolter, and Yejin Choi @YejinChoinka! MLSys this year will be held in Santa Clara on May 13–16. More details at https://t.co/HlUiTXLJTK.

[Albert Tseng @tsengalb99]

🧵 (1/n)

👉 Introducing QuIP#, a new SOTA LLM quantization method that uses incoherence processing from QuIP & lattices to achieve 2 bit LLMs with near-fp16 performance! Now you can run LLaMA 2 70B on a 24G GPU w/out offloading!

💻 https://t.co/eM8Br2CqL3

[Christopher De Sa @chrismdesa]

MLSys 2024 call for papers is out. The submission deadline is Friday, October 27, 2023. https://t.co/vw3DjA0DRf

[Cornell Computer Science @Cornell_CS]

Christopher De Sa and Yucheng Lu won an Outstanding Paper Award Honorable Mention at the 2021 International Conference on Machine Learning (ICML) for "Optimal Complexity in Decentralized Training.” @chrismdesa @yuchengthekid

https://t.co/w3X1aEjJT4

[Cornell University @Cornell]

Cornell computer science professor Kavita Bala, a leading expert in computer graphics and computer vision, has been named dean of the Faculty of Computing and Information Science. @CornellCIS https://t.co/6LmZVRpbsl

[Brandon Amos @brandondamos]

Differentiating through the Fréchet Mean
A Lou, @isaykatsman, Q. Jiang, @SergeBelongie, S. Lim, @chrismdesa
https://t.co/G7WYgQ9b7v

[Malcolm Harris @BigMeanInternet]

Schools are shutting down today, but childcare is still a collective social responsibility. I've spent the last few days working with a crack team of coders to build this (incredibly cool) tool for scheduling MICRO childcare co-ops https://t.co/AuXwrzPZul