1. X
  2. Roy Frostig
Log inSign up
Roy Frostig
123 posts
Roy Frostig profile banner
user avatar

Roy Frostig

@froystig
research scientist at @googledeepmind. co-author of JAX (github.com/jax-ml/jax)
sfba
cs.stanford.edu/~rfrostig/
Joined April 2008
788
Following
1,564
Followers
RepliesRepliesMediaMedia
  • user avatar
    Roy Frostig
    @froystig
    Mar 7, 2025
    A nice and concise R1 inference jax:tpu port by @rdyro128523. Good for both reading and running. Watch the repo for more.
    user avatar
    rdyro
    @rdyro128523
    Mar 6, 2025
    Deepseek R1 inference in pure JAX! Currently on TPU, with GPU and distilled models in-progress. Features MLA-style attention, expert/tensor parallelism & int8 quantization. Contributions welcome!
    Image
  • user avatar
    Roy Frostig
    @froystig
    Feb 4, 2025
    Our online book on systems principles of LLM scaling is live. We hope that it helps you make the most of your computing resources. Enjoy!
    user avatar
    Jacob Austin
    @jacobaustin132
    Feb 4, 2025
    Making LLMs run efficiently can feel scary, but scaling isn’t magic, it’s math! We wanted to demystify the “systems view” of LLMs and wrote a little textbook called “How To Scale Your Model” which we’re releasing today. 1/n
    Image
  • user avatar
    Roy Frostig
    @froystig
    Apr 28, 2022
    In JAX and Dex, we do automatic differentiation (AD) in a distinctive way: by "linearizing" and then "transposing" programs. We wrote up what this looks like in a model language: arxiv.org/abs/2204.10923 with Alexey Radul, @apaszke, @SingularMattrix, @DougalMaclaurin
    arXiv logo
    arxiv.org
    You Only Linearize Once: Tangents Transpose to Gradients
    Automatic differentiation (AD) is conventionally understood as a family of distinct algorithms, rooted in two "modes" -- forward and reverse -- which are typically presented (and implemented)...
  • user avatar
    Roy Frostig
    @froystig
    Oct 21, 2020
    "... small, generalizable operations and compiler infrastructure are more reliable than hand tuned kernels."
    user avatar
    Gautam Kamath
    @thegautamkamath
    Oct 20, 2020
    Feeling blue that differentially private machine learning is slow?😢 Answer is simple: learn JAX! 😮 50x speedups vs TF Privacy & Opacus! 🏎️ Project carried by @PranavSubramani and @nicvadivelu. 📝Paper: arxiv.org/abs/2010.09063 💻Code: github.com/TheSalon/fast-… 🧵Thread ⬇️ 1/8
    Image
  • user avatar
    Roy Frostig
    @froystig
    Sep 8, 2020
    rebooting. had a nice seven years away. what'd i miss?

Log in or sign up for X

See what’s happening and join the conversation

Continue with phone
or
Log in with username or email
Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
Advertisement
Advertisement