Many of you asked for code & weights for π₀, we are happy to announce that we are releasing π₀ and pre-trained checkpoints in our new openpi repository! We tested the model on a few public robots, and we include code for you to fine-tune it yourself.

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

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!

Training our most capable Gemini models relies heavily on our JAX software stack + Google's TPU hardware platforms. If you want to learn more, see this awesome book "How to Scale Your Model": jax-ml.github.io/scaling-book/ It was put together by my Google DeepMind colleagues

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!

A nice and concise R1 inference jax:tpu port by rdyro. Good for both reading and running. Watch the repo for more.

Llama 4 inference in pure JAX! Expert/tensor parallelism with int8 quantization. Contributions welcome!

A thread of thoughts on radiance fields, from my keynote at 3DV: Radiance fields have had 3 distinct generations. First was NeRF: just posenc and a tiny MLP. This was slow to train but worked really well, and it was unusually compressed --- The NeRF was smaller than the images.

Pretty awesome result from the new version of Gemini 2.5 I changed one line of War and Peace, inserting a sentence into Book 14, Chapter 10 (halfway through), where Princess Mary "spoke to Crab Man the superhero" Gemini 2.5 consistently found this reference among 860,000 tokens

Excited to introduce PyRoki ("Python Robot Kinematics"): easier IK, trajectory optimization, motion retargeting... with an open-source toolkit on both CPU and GPU

What would truly open-source AI look like? Not just open weights, open code/data, but *open development*, where the entire research and development process is public *and* anyone can contribute. We built Marin, an open lab, to fulfill this vision:

For a rare look into how LLMs are really built, check out David Hall's retrospective on how we trained the Marin 8B model from scratch (and outperformed Llama 3.1 8B base). It’s an honest account with all the revelations and mistakes we made along our journey. Papers are forced to

Strong recommend for this book and the JAX/TPU docs, even if you are using Torch / GPUs. Clean notation and mental model for some challenging ideas. github.com/jax-ml/scaling… github.com/jax-ml/scaling… docs.jax.dev/en/latest/note…

So about a month ago, Percy posted a version of this plot of our Marin 32B pretraining run. We got a lot of feedback, both public and private, that the spikes were bad. (This is a thread about how we fixed the spikes. Bear with me. )

Today we're putting out an update to the JAX TPU book, this time on GPUs. How do GPUs work, especially compared to TPUs? How are they networked? And how does this affect LLM training? 1/n

Kudos to Terry Tao for this: newsletter.ofthebrave.org/p/im-an-award-…

Jeremy Howard Luckily we have alternatives :) github.com/jax-ml/jax/blo… Just 100 lines without leaving Python and SOTA performance

Curious how to write SOTA performance Blackwell matmul kernels using MGPU? We just published a short step-by-step tutorial: docs.jax.dev/en/latest/pall… At each step, we show exactly what (small) changes are necessary to refine the kernel and the final kernel is just under 150 lines.

Want to improve GPU compute/comms overlap? We just published a new short tutorial for you! A few small changes to the Pallas:MGPU matmul kernel is all it takes to turn it into an all-gather collective matmul that overlaps NVLINK comms with local compute: docs.jax.dev/en/latest/pall…

TPU-style collective matmuls on GPU!