📄 中文摘要
Zatom-1 是首个统一三维分子和材料生成与预测学习的基础模型，旨在克服现有 AI 方法在单一领域和单一任务上的局限性。该模型采用 Transformer 架构，结合多模态流匹配目标，能够同时建模离散原子类型和连续三维几何形状。Zatom-1 支持可扩展的预训练，随着模型容量的增加，性能提升可预测，同时实现快速稳定的采样。该研究为化学建模提供了一种新的解决方案，推动了分子与材料科学的交叉应用。
Zatom-1：用于三维分子和材料的多模态流基础模型
📄 English Summary
Zatom-1: A Multimodal Flow Foundation Model for 3D Molecules and Materials
Zatom-1 is the first foundation model that unifies generative and predictive learning for 3D molecules and materials, addressing the limitations of existing AI approaches that focus on a single domain and task. This model employs a Transformer architecture with a multimodal flow matching objective, enabling the joint modeling of discrete atom types and continuous 3D geometries. Zatom-1 supports scalable pretraining with predictable performance gains as model capacity increases, while facilitating fast and stable sampling. This research offers a novel solution for chemical modeling, advancing the intersection of molecular and materials science.