Dataset card

README.md

@@ -1,4 +1,10 @@
 ---
+language:
+- en
+license: cc-by-4.0
+size_categories:
+- 1M<n<10M
+pretty_name: LeMaterial
 dataset_info:
   features:
   - name: Database
@@ -23,4 +29,60 @@ configs:
     path: data/train-*
   - split: validation
     path: data/validation-*
+tags:
+- materials-science
+- crystallography
+- cif
 ---
+
+# Dataset Card for LeMaterial
+
+Unlabelled inorganic crystal structures from LeMat-BulkUnique, used to pre-train CrystaLLM-π.
+
+## Dataset Details
+
+- **Curated by:** Bone et al. (University College London)
+- **Source data:** [LeMat-BulkUnique](https://huggingface.co/datasets/LeMaterial/LeMat-BulkUnique)
+- **Used to train:** [c-bone/CrystaLLM-pi_base](https://huggingface.co/c-bone/CrystaLLM-pi_base)
+- **Paper:** [Discovery and recovery of crystalline materials with property-conditioned transformers (arXiv:2511.21299)](https://arxiv.org/abs/2511.21299)
+- **Repository:** [GitHub: CrystaLLM-pi](https://github.com/C-Bone-UCL/CrystaLLM-pi)
+- **License:** CC BY 4.0, following the source data
+
+## Dataset Structure
+
+| Column | Description |
+|---|---|
+| `Database` | Material identifier in LeMat-BulkUnique |
+| `Reduced Formula` | Reduced chemical formula |
+| `CIF` | Crystal structure in CrystaLLM-π augmented CIF format |
+
+| Split | Rows |
+|---|---|
+| train | 4,034,556 |
+| validation | 212,346 |
+
+## Dataset Creation
+
+The structures come from the PBE subset of LeMat-BulkUnique, accessed in April 2025. Within each group of structures with identical cell composition and space group, the structure with the lowest volume per formula unit was retained. The dataset uses a random 95:5 train:validation split.
+
+## Citation
+
+```bibtex
+@misc{bone2025discoveryrecoverycrystallinematerials,
+      title={Discovery and recovery of crystalline materials with property-conditioned transformers},
+      author={Cyprien Bone and Matthew Walker and Bradley A. A. Martin and Kuangdai Leng and Luis M. Antunes and Ricardo Grau-Crespo and Amil Aligayev and Javier Dominguez and Keith T. Butler},
+      year={2025},
+      eprint={2511.21299},
+      archivePrefix={arXiv},
+      primaryClass={cond-mat.mtrl-sci},
+      url={https://arxiv.org/abs/2511.21299},
+}
+
+@misc{lematerial_2024,
+      title={LeMat-BulkUnique Dataset},
+      author={Martin Siron and Inel Djafar and Lucile Ritchie and Etienne Du-Fayet and Amandine Rossello and Ali Ramlaoui and Leandro von Werra and Thomas Wolf and Alexandre Duval},
+      year={2024},
+      publisher={Hugging Face},
+      url={https://huggingface.co/datasets/LeMaterial/LeMat-BulkUnique},
+}
+```