MLIP Training Datasets

This document consolidates publicly available datasets for training and fine-tuning machine-learned interatomic potentials. Please feel free to modify, update, or add any datasets you think are missing, as well as any additional details you believe would be helpful.
Inorganic Materials
Dataset
Dataset Size
Level of Theory
Dataset Link
Reference
OMat24
100M / 1M / 5M
PBE / PBE+U
facebook/OMAT24 · Datasets at Hugging Face
Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models
sAlex
10M
PBE / PBE+U
facebook/OMAT24 · Datasets at Hugging Face
Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models
sAlex25
1.3M
PBE
Alexandria Materials Database
AI-Driven expansion and application of the Alexandria database
HighP Alex
32M
PBE
Alexandria Materials Database
Universal Machine Learning Potentials under Pressure
WBM
215K
PBE / PBE+U
matbench-discovery/data/wbm at main · janosh/matbench-discovery
Predicting stable crystalline compounds using chemical similarity
MPtrj
1.58M
PBE/ PBE+U
https://doi.org/10.6084/m9.figshare.23713842
CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling
MATPES
434K
PBE
MatPES
A Foundational Potential Energy Surface Dataset for Materials
MATPES
434K
r2SCAN
MatPES
A Foundational Potential Energy Surface Dataset for Materials
MP-ALOE
1M
r2SCAN
https://doi.org/10.6084/m9.figshare.29452190
https://www.nature.com/articles/s41524-025-01834-9
MAD
217K
r2SCAN
High-quality, high-information datasets for universal atomistic machine learning
High-quality, high-information datasets for universal atomistic...
SMAX
1.69M
PBE
Exploring the extremes: atomic basis for multi-elemental materials science under complex thermodynamic conditions
ELEMENTA
210M
PBE
kairosmaterial/ELEMENTA · Datasets at Hugging Face
https://www.kairosmaterials.com/papers/ELEMENTA.pdf
Be extremely careful about the sign convention on the stresses, as VASP and ASE convention differ! In MatPES, for instance, you must convert the stresses from kbar to eV/A^3 and negate each value.
Metal-Organic Frameworks
Dataset
Dataset Size
Level of Theory
Dataset Link
Reference
ODAC25
70M
PBE+D3(BJ)
facebook/ODAC25 · Hugging Face
The Open DAC 2025 Dataset for Sorbent Discovery in Direct Air Capture
MACE-MP-MOF-0
4.7k
PBE+D3(BJ)
data/mace-mof-0 at main · ddmms/data
Machine learned potential for high-throughput phonon calculations of metal—organic frameworks
MOF-OFF
TBA
PBE+D3(BJ)
MOF-OFF
TBA
r2SCAN+D4
Molecules
Dataset
Dataset Size
Level of Theory
Dataset Link
Reference
OMol25+OPoly26
100M+
wB97M-V
facebook/OMol25 · Hugging Face
The Open Molecules 2025 (OMol25) Dataset, Evaluations, and Models
Molecular Crystals
Dataset
Dataset Size
Level of Theory
Dataset Link
Reference
OMC25
27M+
PBE+D3
facebook/OMC25 · Datasets at Hugging Face
Open Molecular Crystals 2025 (OMC25) dataset and models
Surfaces and Catalysts
Dataset
Dataset Size
Level of Theory
Dataset Link
Reference
OC20/22/25
A lot
Mixed
Datasets - FAIR Chemistry Documentation
Datasets - FAIR Chemistry Documentation
AQCat25
13.5M
RPBE
SandboxAQ/aqcat25-dataset · Datasets at Hugging Face
AQCat25: unlocking spin-aware, high-fidelity machine learning potentials for heterogeneous catalysis
﻿