Acellera's AceFF-2-OMOL pushes the MLIP speed-accuracy Pareto frontier with NVIDIA-optimized TensorNet kernels
gdefabritiis · x · 2026-11-12
Acellera released AceFF-2-OMOL, a next-generation machine learning interatomic potential (MLIP) for small molecules, claiming a wide-margin advance of the speed-vs-accuracy Pareto frontier.
- Built on the TensorNet v2 architecture, implemented in TorchMD-Net, targeting drug-like compounds with support for all key elements and charged molecules
- Unlike AceFF-2.0 (trained on proprietary data), AceFF-2-OMOL is trained on the public OMol25 dataset computed at the wB97M-V/def2-tzvppd level of theory
- On the Wiggle150 benchmark it achieves 1.25 kcal/mol MAE and 1.64 RMSE, beating AceFF-2.0 (1.76/2.34) as well as AIMNet2 and ANI-2x
- With NVIDIA-optimized kernels for TensorNet, the Acellera-TensorNet version reaches 100 steps/second on a 2,661-atom water box while keeping SPICE MAE around 1.2 kcal/mol, clearly ahead of the prior frontier
The model is available on Hugging Face under CC BY-ND 4.0.
More from Research
- Academia is developing an 'AI caste system,' warns investor as models erase research moats — sebkrier · 2026-11-12
- PyTorch 2.15 regression slows LU solver up to 100x, warns researcher — ducha_aiki · 2026-11-12
- Turing Institute paper flags assurance gap in IMO's autonomous ship code — turinginst · 2026-11-12
- Caltech's Anima Anandkumar: AI-powered forecasting makes hurricane tracking faster and more reliable — AnimaAnandkumar · 2026-11-12
- Preprint: low-frequency EEG precedes pupil dynamics by ~764ms — AnnaCiaunica · 2026-11-12
- PACT: giving intermediate draft steps of diffusion language models informative RL training signals — burkov · 2026-11-12