materials-toolkits/materials-project
Dataset Materials project (2019 dump) This dataset contains 133420 materials with formation energy per atom. Processed from mp.2019.04.01.json Download Download link: materials-project.tar.gz MD5 checksum c132f3781f32cd17f3a92aa6501b9531 Content Bundled in materials-project.tar.gz. Index (index.json) list of dict: index (int) => index of the structure in data file. id (str) => id of Materials Project. formula (str) => formula. natoms… See the full description on the dataset page: https://huggingface.co/datasets/materials-toolkits/materials-project.
Dataset
Materials project (2019 dump)
This dataset contains 133420 materials with formation energy per atom.
Processed from mp.2019.04.01.json
Download
Download link: materials-project.tar.gz
MD5 checksum c132f3781f32cd17f3a92aa6501b9531
Content
Bundled in materials-project.tar.gz.
Index (index.json)
list of dict:
index(int) => index of the structure in data file.id(str) => id of Materials Project.formula(str) => formula.natoms(int) => number of atoms.energy_pa(float) => formation energy per atom.
Data (data.hdf5)
fields:
structures=> a group containing structure information.structures/cell(float32) => lattice of the material.structures/natoms(int32) => number of atoms.structures/energy_pa(float32) => formation energy per atom.structures/atoms_ptr(int64) => position of the first atom of the structures in theatomsgroup.atoms=> a group containing information about atoms.atoms/positions(float32) => the positions of the atoms.atoms/atomic_number(uint8) => the atomic number of the atoms.
2026 release (all.hdf5)
all.hdf5 is a newer, separate export of the Materials Project, preprocessed with pycrystalgen. It does not replace materials-project.tar.gz above and uses a different layout.
It contains 271,325 structures (6,967,262 atoms, 1 to 444 atoms per structure): every material of the summary collection of database version 2026.09.28, except the 8,834 materials flagged as deprecated. Nothing else is filtered.
Labels
Four materials have no labels in the source and are kept with NaN. Material ids are not stored: structures are in the order of the source collection, deprecated materials removed.
Format
Flat concatenated arrays:
The lattice is rebuilt from cell lengths and angles, so it is in the standard orientation (first vector along x). The batch array is an artefact of the writer and is not used when reading.
Usage
from pycrystalgen.data import HDF5Dataset
dataset = HDF5Dataset.from_hub("materials-toolkits/materials-project", "all.hdf5")
structures = dataset[:32] # batched StructuresSource, processing and licence
- Source: the public open data bucket of the Materials Project,
materialsproject-build/collections/summary/version=2026-09-28/. - Built with pycrystalgen,
scripts/build_materials_project.py: collection to CIF, then CIF to HDF5. - Every structure was checked against the source: number of atoms, elements, fractional coordinates, lattice and the four labels.
Licence of `all.hdf5`: the license: mit of this repository describes the 2019 files, not this one. In the source, each material carries its own licence label: 154,377 of the structures kept here are labelled BY-C and 116,948 are labelled BY-NC (non-commercial). This file does not record which structure has which label, so treat the whole file as non-commercial unless you check against the source, and refer to the Materials Project for the terms of use.
