# MAD-1.5 Massive Atomic Diversity r2SCAN Training Split (102 Elements), ColabFit Conversion

## Facts

- Provider: Cesare Malosso, Michele Ceriotti, Arslan Mazitov and coauthors (EPFL COSMO); ColabFit conversion
- License: CC-BY-4.0 (https://archive.materialscloud.org/api/records/18tke-tt476)
- Size: 6 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/colabfit/Massive_Atomic_Diversity_MAD-1.5_r2SCAN_Train
- Access: public, readable without an account
- DOI: https://doi.org/10.24435/materialscloud:jc-9f
- Category: Chemistry & Materials
- Source: https://doi.org/10.48550/arXiv.2603.02089
- Page: https://scigantic.com/archive/e1ca04a9-83c2-5789-ad03-1ac95462282b
- Keywords: Chemistry & Materials, interatomic potentials, Parquet, Hugging Face

## About

Training split of MAD-1.5, the second version of the Massive Atomic Diversity dataset from EPFL used to train the PET-MAD universal interatomic potentials. All structures are computed with one all-electron workflow: the r2SCAN meta-GGA functional in FHI-aims (tight basis, 8 per Angstrom k-point density, 0.05 eV Gaussian smearing). It spans 102 elements and 14 subsets including bulk crystals from MC3D, rattled and random structures, surfaces, nanoclusters, 2D materials, molecular crystals and molecular fragments, with monomers, dimers and trimers all placed in training; outliers were removed by force thresholds and LLPR uncertainty filtering. ColabFit counts 180,174 configurations (about 2.6 million atoms) with energy, atomization energy, forces and stress. This is the ColabFit Exchange conversion: configurations sit in Parquet files under co/ (positions, cell, species, 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 schema at materials.colabfit.org/docs. Published on Materials Cloud under CC-BY-4.0; cite Malosso et al., arXiv:2603.02089 (2026).

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("colabfit/Massive_Atomic_Diversity_MAD-1.5_r2SCAN_Train")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `co/co_0.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/e1ca04a9-83c2-5789-ad03-1ac95462282b
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/e1ca04a9-83c2-5789-ad03-1ac95462282b
- Over MCP: `get_schema_card` with id `e1ca04a9-83c2-5789-ad03-1ac95462282b` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Chemistry & Materials datasets](https://scigantic.com/topics/chemistry-and-materials)
- [Interatomic potentials datasets](https://scigantic.com/topics/interatomic-potentials)
