用图神经网络加速材料发现，1秒筛10亿结构
Exascale Multi-Task Graph Foundation Models for Imbalanced, Multi-Fidelity Atomistic Data
- 构建多任务图模型，联合训练54400万原子结构数据
- 1.1亿结构仅用50秒完成筛选，效率提升数年量级
- 适合材料设计、高通量筛选和小样本场景的科研人员
我们提出一种面向材料发现的百亿级原子结构计算流程，基于HydraGNN构建图基础模型，在Frontier超级计算机上联合训练16个开源第一性原理数据集（共5.44亿以上结构，覆盖85种以上元素），采用每数据集独立输出头的多任务架构，并通过可扩展的ADIOS2/DDStore数据管道支持大规模训练。在FP64下执行六次大型DeepHyper超参数优化，将表现最优的消息传递模型迁移至2048节点持续训练，最终获得以PaiNN为基础的领先模型。该模型实现十亿级结构快速筛查，50秒内完成11亿结构评估，相当于将原本需数年完成的第一性原理计算压缩至分钟级。同时支持数据稀缺条件下的下游任务微调，量化了BF16/FP32/FP64精度-性能权衡，验证了在12类化学差异显著任务中的跨域迁移能力，并在Frontier、Aurora和Perlmutter系统上实现无缝强弱扩展。本工作使此前无法触及的庞大化学设计空间得以快速可靠探索。
原文摘要 · Abstract (English)
We present an exascale workflow for materials discovery using atomistic graph foundation models built on HydraGNN. We jointly train on 16 open first-principles datasets (544+ million structures covering 85+ elements) using a multi-task architecture with per-dataset heads and a scalable ADIOS2/DDStore data pipeline. On Frontier, we execute six large-scale DeepHyper hyperparameter optimization campaigns in FP64 and promote the top-performing message-passing models to sustained 2,048-node training, yielding a PaiNN-based lead model. The resulting model enables billion-scale screening, evaluating 1.1 billion atomistic structures in 50 seconds, compressing a workload that would require years of first-principles computation, and supports data-scarce fine-tuning across diverse downstream tasks. We quantify precision-performance tradeoffs (BF16/FP32/FP64), demonstrate transfer across twelve chemically diverse downstream tasks, and establish seamless strong- and weak-scaling across Frontier, Aurora, and Perlmutter. This work allows fast and reliable exploration of vast chemical design spaces that are otherwise inaccessible to first-principles methods.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。