Breaking the Training Barrier of Billion-Parameter Universal Machine Learning Interatomic Potentials
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866908973755531264 |
|---|---|
| author | Zhou, Yuanchang Wang, Hongyu Du, Yiming Wang, Yan Li, Mingzhen Hu, Siyu Zhang, Xiangyu Liu, Weijian Wang, Chen Guo, Zhuoqiang Wang, Long Bu, Jingde Lu, Yutong Tan, Guangming Jia, Weile |
| author_facet | Zhou, Yuanchang Wang, Hongyu Du, Yiming Wang, Yan Li, Mingzhen Hu, Siyu Zhang, Xiangyu Liu, Weijian Wang, Chen Guo, Zhuoqiang Wang, Long Bu, Jingde Lu, Yutong Tan, Guangming Jia, Weile |
| contents | Universal Machine Learning Interatomic Potentials (uMLIPs), pre-trained on massively diverse datasets encompassing inorganic materials and organic molecules across the entire periodic table, serve as foundational models for quantum-accurate physical simulations. However, uMLIP training requires second-order derivatives, which lack corresponding parallel training frameworks; moreover, scaling to the billion-parameter regime causes explosive growth in computation and communication overhead, making its training a tremendous challenge. We introduce MatRIS-MoE, a billion-parameter Mixture-of-Experts model built upon invariant architecture, and {Janus}, a pioneering high-dimensional distributed training framework for uMLIPs with hardware-aware optimizations. Deployed across two Exascale supercomputers, our code attains a peak performance of 1.2/1.0 EFLOPS (24\%/{35.5\%} of theoretical peak) in single precision at over 90\% parallel efficiency, compressing the training of billion-parameter uMLIPs from weeks to hours. This work establishes a new high-water mark for AI-for-Science (AI4S) foundation models at Exascale and provides essential infrastructure for rapid scientific discovery. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_15821 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Breaking the Training Barrier of Billion-Parameter Universal Machine Learning Interatomic Potentials Zhou, Yuanchang Wang, Hongyu Du, Yiming Wang, Yan Li, Mingzhen Hu, Siyu Zhang, Xiangyu Liu, Weijian Wang, Chen Guo, Zhuoqiang Wang, Long Bu, Jingde Lu, Yutong Tan, Guangming Jia, Weile Distributed, Parallel, and Cluster Computing Machine Learning Universal Machine Learning Interatomic Potentials (uMLIPs), pre-trained on massively diverse datasets encompassing inorganic materials and organic molecules across the entire periodic table, serve as foundational models for quantum-accurate physical simulations. However, uMLIP training requires second-order derivatives, which lack corresponding parallel training frameworks; moreover, scaling to the billion-parameter regime causes explosive growth in computation and communication overhead, making its training a tremendous challenge. We introduce MatRIS-MoE, a billion-parameter Mixture-of-Experts model built upon invariant architecture, and {Janus}, a pioneering high-dimensional distributed training framework for uMLIPs with hardware-aware optimizations. Deployed across two Exascale supercomputers, our code attains a peak performance of 1.2/1.0 EFLOPS (24\%/{35.5\%} of theoretical peak) in single precision at over 90\% parallel efficiency, compressing the training of billion-parameter uMLIPs from weeks to hours. This work establishes a new high-water mark for AI-for-Science (AI4S) foundation models at Exascale and provides essential infrastructure for rapid scientific discovery. |
| title | Breaking the Training Barrier of Billion-Parameter Universal Machine Learning Interatomic Potentials |
| topic | Distributed, Parallel, and Cluster Computing Machine Learning |
| url | https://arxiv.org/abs/2604.15821 |