# MatterGen Alex-MP-20 Crystal Structures with Simulated XRD Peaks for Diffraction-Conditioned Generation (CrystaLLM-pi, UCL)

## Facts

- Provider: Cyprien Bone, Keith T. Butler et al. (University College London)
- License: CC-BY-4.0 (https://huggingface.co/datasets/c-bone/mattergen_XRD)
- Size: 5 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/c-bone/mattergen_XRD
- Access: public, readable without an account
- Category: Chemistry & Materials
- Source: https://arxiv.org/abs/2511.21299
- Page: https://scigantic.com/archive/6b66ed04-9d7b-5785-9a9e-b517b4c61ea0
- Keywords: Chemistry & Materials, DFT, Parquet, Hugging Face

## About

Training data from University College London for CrystaLLM-pi, a property-conditioned transformer that generates crystal structures. It pairs the Alex-MP-20 inorganic crystal structures released with Microsoft's MatterGen (drawn from the Alexandria DFT database and Materials Project MP-20) with up to 20 of their most intense simulated powder X-ray diffraction (XRD) peaks. Peaks were computed with pymatgen's XRDCalculator at the Cu K-alpha wavelength and stored as a condition vector of 2-theta positions followed by intensities, min-max normalized and padded with -100. Each row gives the source database, reduced formula and the structure in CrystaLLM-pi augmented CIF text, with a random 90:10 split of 591,964 training and 65,774 validation rows in Parquet. Use it to train or test models that recover a crystal structure from its diffraction pattern, or for XRD-conditioned materials discovery; load with datasets or pandas. License CC BY 4.0, following the Alexandria and Materials Project source data. Cite Bone et al., Discovery and recovery of crystalline materials with property-conditioned transformers, arXiv:2511.21299.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("c-bone/mattergen_XRD")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `data/train-00000-of-00002.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/6b66ed04-9d7b-5785-9a9e-b517b4c61ea0
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/6b66ed04-9d7b-5785-9a9e-b517b4c61ea0
- Over MCP: `get_schema_card` with id `6b66ed04-9d7b-5785-9a9e-b517b4c61ea0` 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)
