# Alex-MP-20 Training Set for Crystal Generative Models (MatterGen), ColabFit Conversion

## Facts

- Provider: Microsoft Research AI for Science (Zeni et al.), from Alexandria and Materials Project; ColabFit conversion
- License: CC-BY-4.0 (https://raw.githubusercontent.com/microsoft/mattergen/main/data-release/alex-mp/README.md)
- Size: 5 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/colabfit/Alex_MP-20_train
- Access: public, readable without an account
- Category: Chemistry & Materials
- Source: https://doi.org/10.1038/s41586-025-08628-5
- Page: https://scigantic.com/archive/5652cf1f-745b-5cba-b91d-ded862592ff0
- Keywords: Chemistry & Materials, DFT, Parquet, Hugging Face

## About

The training split of Alex-MP-20, the crystal structure dataset Microsoft Research used to train the MatterGen generative model for inorganic materials (Zeni et al., Nature 2025). It merges structures from the Alexandria database and MP-20 (Materials Project), relaxed with PBE DFT for consistent energies; structures containing Tc, Pm or elements with atomic number 84 or above were excluded, and the training set keeps only structures with at most 20 atoms per unit cell and energy above hull up to 0.1 eV/atom. Labels include energy above hull and band gap. ColabFit's card counts 540,162 configurations; the repository is about 0.9 GB (measured 2026-11-14). The validation and test splits are separate ColabFit repositories. 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. The MatterGen data release states that Alexandria and MP-20 are both CC-BY-4.0.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("colabfit/Alex_MP-20_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/5652cf1f-745b-5cba-b91d-ded862592ff0
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/5652cf1f-745b-5cba-b91d-ded862592ff0
- Over MCP: `get_schema_card` with id `5652cf1f-745b-5cba-b91d-ded862592ff0` 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)
- [DFT datasets](https://scigantic.com/topics/dft)
