Kelvin (Keqiang) Yan
@KeqiangY

RS at Bytedance Seed, Seattle. LLMs for Scientific Discovery. Ex @MSFTResearch @Princeton @PKU1898, etc. Opinions are my own.

sites.google.com/view/keqian…
Joined September 2020
32 Photos and videos
Replying to @zhu_hanqing666
Looking forward to your posts!
Come work with Yuanqi!
Yuanqi Du
@YuanqiD
Oct 14
We are hiring research interns in the broad area of machine learning and statistics @MSRNE @MSFTResearch! Come work with us! Personally I have very broad interests and I’d love to learn about your favorite problems!
10
Replying to @EdwardSun0909
congratulations!
2
Interesting RL env analysis
Vals AI
@ValsAI
Oct 7
Xiaomi open-sourced the RL environments behind MiMo v2.6. In two-thirds of the coding tasks, the answer is still in the task's Git history, and MiMo finds it.
Will the 25T tokens be released? ☺️
Will Held@WilliamBarrHeld
Sep 23
There’s an enormous amount of open training data on Hugging Face. What does it take to make it work together 🤗? For Marin’s 535B run, we built on 25T tokens from 152 datasets with licenses permitting training. Here’s the work between downloading those and training a model 🧵
Flow diagram of the Marin data pipeline. Starting with 25.25 trillion tokens, deduplication removes 2.13 trillion and benchmark decontamination removes another 13.66 billion, leaving 23.11 trillion. Documents are organized into 40 topics and five quality bands, yielding 200 sampling groups. Results from 872 small-model experiments feed a regression used to choose sampling weights.

ALT Flow diagram of the Marin data pipeline. Starting with 25.25 trillion tokens, deduplication removes 2.13 trillion and benchmark decontamination removes another 13.66 billion, leaving 23.11 trillion. Documents are organized into 40 topics and five quality bands, yielding 200 sampling groups. Results from 872 small-model experiments feed a regression used to choose sampling weights.
3
Replying to @EdwardSun0909 @alexandr_wang @natfriedman @finkd
Great vision, congratulations!
2
Replying to @xiangfu_ml
makes perfect sense...
1
Kelvin (Keqiang) Yan retweeted
Jiaqi Zhang
@JiaqiZhangVic
Sep 9
I’m excited to share that I’ll be joining @Columbia in the CS Department as an Assistant Professor in January 2027, in New York City. My lab will work on advancing causality, AI, experimental design, and biomedical applications. I’ll be taking on PhD students and recruiting postdocs. If you’re interested in working with me, consider applying to the CS PhD program this fall or reach out at jz4255@columbia.edu.
55
88
964
Replying to @weijie444
To benefit the whole humanity!
1
Will cancer be cured in the near future? 🥰
Weijie Su
@weijie444
Sep 8
Last July, when LLMs won gold medals at the IMO, I felt that someday AI might be able to solve some of the hardest problems in math. Now, an internal OpenAI model (still improving!) has resolved a Millennium Prize Problem. It feels surreal that that day came so, so soon!
1
8
Replying to @zdhnarsil @xai
🐮 Ding!
1
Replying to @dr_k_choudhary
Many congrats!!
Replying to @gliu0329 @Meng_CS
Congratulations and all the best!
2
Replying to @xie_yaqi @UofIllinois
Congrats!
Replying to @weijie444 @OpenAI
Huge congratulations!
1
Kelvin (Keqiang) Yan retweeted
Dr. Kamal Choudhary@dr_k_choudhary
Jul 22
One year as an assistant professor at JHU this month and delighted to share #AtomGPTLab's work has been awarded through both the NSF PCL Test Bed and DOE Genesis programs! Huge thanks to the agencies, colleagues, and our incredible team. More ahead on AtomGPT.org 🚀
2
6
Replying to @zdhnarsil
Teach me how to post-train 😍🥰
1
Replying to @yingheng_wang
又帅了
1
1
Replying to @PimDeWitte @khoslaventures @generalcatalyst @JeffBezos @ericschmidt @NicoRosberg
Congratulations!
Replying to @demisama_
Doing research v.s. Publishing papers
1
1
Load more