ml-atomistic

ML Atomistic

ML atomistic interoperability banner

Foundational libraries enabling ML models and simulation engines to communicate

ML atomistic interoperability architecture

Context

Atomistic ML lives in an O(M x E) world: M models (PyTorch, JAX, MACE, PET, …) times E simulation engines (LAMMPS, GROMACS, ASE, eOn). Every model-engine pair needs its own glue, and a model trained under one framework cannot be dropped into another without rewriting that glue.

The approach

Metatensor provides a shared data format (TensorMap) for atomistic ML tensors - descriptors, predictions, training labels - with metadata about which atoms and properties each block describes (Bigi et al. 2025). The format is framework-agnostic, sparse, and block-structured. Metatomic wraps trained models behind one interface that simulation engines call directly. Integration cost drops from O(M x E) to O(M + E).

My contributions

At EPFL labCOSMO I work on the systems-level pieces that make metatensor practical: DLPack support for zero-copy tensor interchange between PyTorch, NumPy, and the Rust core; device-aware execution that routes tensors to the right hardware; and performance tuning on the Rust side. On the simulation side, I maintain the integration with eOn for saddle point searches and contribute to the GROMACS integration for scalable MD.

Code

Open directions

  • Hardware-aware execution paths in metatomic: automatically selecting GPU vs CPU kernels based on system size and available devices.
  • Extending metatensor to handle long-range electrostatics and periodic boundary conditions natively in the tensor format.

References

Bigi, Filippo, Joseph W. Abbott, Philip Loche, Arslan Mazitov, Davide Tisi, Marcel F. Langer, Alexander Goscinski, et al. 2025. “Metatensor and Metatomic: Foundational Libraries for Interoperable Atomistic Machine Learning.” arXiv. https://doi.org/10.48550/arXiv.2508.15704.

← All research threads