Training a Foundation Model for Materials on a Budget

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Koker, Teddy, Kotak, Mit, Smidt, Tess
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866914083405561856
author Koker, Teddy
Kotak, Mit
Smidt, Tess
author_facet Koker, Teddy
Kotak, Mit
Smidt, Tess
contents Foundation models for materials modeling are advancing quickly, but their training remains expensive, often placing state-of-the-art methods out of reach for many research groups. We introduce Nequix, a compact E(3)-equivariant potential that pairs a simplified NequIP design with modern training practices, including equivariant root-mean-square layer normalization and the Muon optimizer, to retain accuracy while substantially reducing compute requirements. Nequix has 700K parameters and was trained in 100 A100 GPU-hours. On the Matbench-Discovery and MDR Phonon benchmarks, Nequix ranks third overall while requiring a 20 times lower training cost than most other methods, and it delivers two orders of magnitude faster inference speed than the current top-ranked model. We release model weights and fully reproducible codebase at https://github.com/atomicarchitects/nequix.
format Preprint
id arxiv_https___arxiv_org_abs_2508_16067
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Training a Foundation Model for Materials on a Budget
Koker, Teddy
Kotak, Mit
Smidt, Tess
Computational Physics
Machine Learning
Foundation models for materials modeling are advancing quickly, but their training remains expensive, often placing state-of-the-art methods out of reach for many research groups. We introduce Nequix, a compact E(3)-equivariant potential that pairs a simplified NequIP design with modern training practices, including equivariant root-mean-square layer normalization and the Muon optimizer, to retain accuracy while substantially reducing compute requirements. Nequix has 700K parameters and was trained in 100 A100 GPU-hours. On the Matbench-Discovery and MDR Phonon benchmarks, Nequix ranks third overall while requiring a 20 times lower training cost than most other methods, and it delivers two orders of magnitude faster inference speed than the current top-ranked model. We release model weights and fully reproducible codebase at https://github.com/atomicarchitects/nequix.
title Training a Foundation Model for Materials on a Budget
topic Computational Physics
Machine Learning
url https://arxiv.org/abs/2508.16067