Checkout UniMat -- a unified representation of materials that enables scaling of diffusion models to millions of stable crystal structures. Website: unified-materials.github.io Paper: arxiv.org/abs/2311.09235

Now out in Angewandte Chemie doi.org/10.1002/anie.2…

Here is an implementation of Mamba 🐍 in pure PyTorch using a parallel scan. It offers a nice tradeoff between performance and readability, being only composed of 2 files. >> github.com/alxndrTL/mamba… << Albert Gu Tri Dao

Proud to announce a huge QM data dump for ML training, related to our ANI-2x paper pubs.acs.org/doi/10.1021/ac… 1/n

2023 Small Molecule Approvals! @drughuntersite A lot of fascinating firsts in 2023 including: -The first FDA-approved oral ER degrader -The first FDA-approved reversible BTK inhibitor -The first FDA-approved gamma-secretase inhibitor (not in Alzheimer’s!) -A reversible

I wrote about the effort to decipher sperm whale language with artificial intelligence. I reached out to philosophers, linguists, animal rights lawyers, marine biologists, field scientists who specialize in whales, and paleontologists. Assume that this works, I told them.

I'll give this #PhD opportunity in computational materials chemistry another push. Apply or get in touch if interested. findaphd.com/phds/project/g…

Tried to capture why I’m so excited about NNPs. I honestly think they are a profoundly useful tool.

This is exactly what I hate with all big frameworks. TF is terrible. PyTorch used to be straightforward but turned terrible too. Torch7 was very direct. JAX/Flax still ok, but I pray every day that it doesn’t end up with the same fate over time.

scalable molecular simulation: github.com/molmod/psiflow scientific: ML potentials, DFT and post-HF calculations, (path-integral) MD, replica exchange, alchemical ΔF , hessians, ... technical: automated job submission, simple Python, scales to >100 nodes, containerized!

Data generation through simulation is one reason why the whole idea of ML for science is so exciting.

For too long, users have lived under the software lottery tyranny of fused attention implementations. No longer. Introducing FlexAttention, a new PyTorch API allowing for many attention variants to enjoy fused kernels in a few lines of PyTorch. pytorch.org/blog/flexatten… 1/10

A thread on our new paper Thermodynamic Bayesian Inference 250 years later, Bayes’s theorem is still the gold standard for probabilistic reasoning. But for complicated models it’s too hard to implement exactly, so approximations are used. For example, the complexity of Bayesian

New paper - Transformers, but without normalization layers (1/n)

New blog post: let's talk about latents! sander.ai/2025/04/15/lat…

btw, i wrote a post about "how to scale" based on what i've learned over the past few months. it covers muP, HP scaling laws, and some stuffs. would be happy to get any feedback or discussion. (it's pretty verbose and no TL;DR, sorry lol) howtoscalenn.github.io