sAlex: Subsampled Alexandria Training Set for OMat24 Fine-Tuning, ColabFit Conversion

Provider
Meta FAIR Chemistry (Barroso-Luque et al.), from Alexandria (Schmidt, Marques et al.); ColabFit conversion
License
CC-BY-4.0 (source)
Size
96 files
Format
Parquet (in sampled files)
Stored at
Hugging Face (https://huggingface.co/datasets/colabfit/sAlex_train)
Access
Free and public. No account with the data provider is needed.
DOI
10.60732/efbb7935
Category
Chemistry & Materials
Source
https://arxiv.org/abs/2410.12771

About this dataset

sAlex is the subsampled, Matbench-Discovery-compliant version of the Alexandria inorganic crystal database that Meta FAIR Chemistry used to fine-tune its OMat24 (Open Materials 2024) interatomic potentials such as eqV2 and eSEN. It was built from Alexandria PBE relaxation trajectories by removing any structure matched to the WBM test set of Matbench Discovery, and by keeping only frames along a trajectory that differ in energy by more than 10 meV/atom from the previously kept frame. Calculations are VASP PBE and PBE+U, labelled with total energy, atomic forces and stress. ColabFit's card counts 10,345,613 configurations, about 107 million atoms and 89 elements; the repository holds 93 Parquet shards, about 6.9 GB (measured 2026-11-15). Use it to fine-tune or pre-train universal MLIPs for crystal stability prediction without leaking the WBM benchmark. This is the ColabFit Exchange conversion: configurations are Parquet files under co/ (structure, positions, cell, computed properties and metadata per row) with dataset-level information in ds.parquet; read them with pyarrow, pandas, polars or Hugging Face datasets, following the ColabFit configuration schema at materials.colabfit.org/docs. Released by Meta under CC-BY-4.0 with OMat24; cite Barroso-Luque et al., arXiv:2410.12771 (2024), and Schmidt et al. for Alexandria.

How to load this dataset in Python

from datasets import load_dataset
ds = load_dataset("colabfit/sAlex_train")
print(ds)

Data structure

File types in sampled files: Parquet

Files

Example file: co/co_0.parquet

Analyze this dataset without downloading it

On Scigantic, sAlex: Subsampled Alexandria Training Set for OMat24 Fine-Tuning, ColabFit Conversion opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.

This page is also available as Markdown and as JSON.

Related Chemistry & Materials datasets