colabfit/Massive_Atomic_Diversity_MAD_bench_mad

Massive Atomic Diversity MAD bench mad数据集是一个包含多种原子配置的集合，由MAD DFT设置计算得出，旨在通过扭曲稳定结构配置来达到巨大的原子多样性。该数据集包含少于100k的结构，但已经能够支持训练与传统数据集相比具有竞争力的通用原子间势能。数据集包含了能量、原子力、柯西应力等属性。The Massive Atomic Diversity MAD bench mad dataset is a collection of various atomic configurations computed with MAD DFT settings, aiming to achieve massive atomic diversity through aggressive distortion of stable structural configurations. Despite containing fewer than 100k structures, the dataset has already enabled the training of universal interatomic potentials that are competitive with models trained on traditional datasets with two to three orders of magnitude more structures. Properties included are energy, atomic forces, and Cauchy stress.