Check out our updated website, publish your data, or leverage the 600 TB of data we have on hand at
www.materialsdatafacility.org

- First, a new section showing how to use MDF's Foundry datasets. Load ML-ready datasets with just a few lines of code
- Second, a clean hero, with updated Navbar and clear Calls-to-Action
- Third, a more compact and intuitive view of Foundry dataset contents, browsable from the search interface

Within Cursor, Gemini 3, and Claude missed simple things that caused component errors. Sonnet fixed these immediately in CC, showing the importance of the harness.

There were lots of little updates, but the attached images show a few specifically

2. The model harness matters...a LOT. That means that you might see vastly different performance e.g., in Cursor vs Antigravity (Google's IDE) vs Droid.
3. The cost was low. This update took 2h, $0.90 Cursor credits, 2 free Antigravity prompts and 1 Claude Code (CC) prompt.

So glad you are getting some relief from all of this! 💜

Wait until you try to do a hire. Welcome…

Excited to help build the TorchSim community. Reach out to me and join us:

🔸TorchSim GitHub: github.com/torchSim/tor...

🔹TorchSim Slack: join.slack.com/t/torchsim/s...

🔸Hugging Face Science Discord: discord.gg/jDz2UyzQxW

Thanks to the many community contributors already pushing this forward, including Thomas Loux, Ryan Liu, J Kian Pu, Filippo Bigi, Stefan Bringuier, Ph.D. , Myles Stapelberg, Yutack Park, John Gardner, Guillame Fraux, Chuin Wei Tan, and Timo Reents.

But, for more success we now need your help to!
🔷 Build examples, tutorials, and benchmark your workflows
🔶 Integrate new MLIP architectures and optimize GPU utilization
🔷 Implement integrators, optimizers, and simulation methods
🔶 Help document and build this ecosystem.

The original development team, including Abhijeet Gangan, Orion Archer Cohen, Janosh Riebesell, Rhys Goodall, Adeesh Kolluru, Stefano Falletta, and Curtis Chong, Joseph Krause and Jorge Colindres built something special, and we want to ensure their work flourishes.

TorchSim offers faster batched inference, full GPU utilization, and perhaps most importantly, a unified interface across model architectures enabling rapid prototyping and model swapping.

This paper won Best Paper Award at EScience 2025, congrats to the entire team team: Marcus Schwarting, @WardLT2, @NCHudson95, Xiaoli Yan, me, Santanu Chaudhuri, Eliu Huerta, @ianfoster
@argonne, @UChicago, @thisisUIC

Applied to MOF discovery, this queue prioritization:
🔶 Doubled discovery of high-performing candidates
🔷 Required <1% extra compute for the RL loop
🔶 Discovered more stable, varied materials

📘 Read the full paper: arxiv.org/pdf/2509.25538

Thanks to support from NIST and James Warren for making the MDF vision of vast troves of open data to fuel discovery possible.

We can't wait to see what you build and discover!

Access Details: github.com/facebookrese...

Eagle provides researchers access to high-volume data pioneered by @ianfoster42.bsky.social, Rachana Ananthakrishnan, Kyle Chard, Michael Papka, Rick Stevens, and others. @globusorg.bsky.social provides platform capabilities (auth, data transfer, automation, and compute) to over 600k researchers.

For this first release, the data are quite raw, and as-created by the Meta team. There's an opportunity for the community to build tools that simplify access to these data, allow data query and browsing, create databases of calculated properties and descriptors, and much more.

The Materials Data Facility is proud to make these data available via the Eagle cluster at ALCF through a high-performance Globus endpoint. Given the dataset's scale, we're first releasing data for a 4M random OMol25 split, with the multi-petabyte dataset following based on community engagement.

This dataset includes DFT outputs, electronic densities, wavefunctions, and molecular orbital information for over 4M high-accuracy calculations. We see this as an opportunity to develop higher quality partial charges, partial spins, and advanced electronic features to unlock next generation models.

Data to train next-generation AI models. Data to push the boundaries of what's computationally possible in molecular chemistry and lead the world in AI for science. Data that captures the full complexity of chemical systems, from small organic molecules to massive biomolecular complexes.