DATASET: Fine-Tuning Unifies Foundational Machine-learned Interatomic Potential Architectures at ab initio Accuracy

From foundation to fine-tuning: Advancing bulk-phase molecular dynamics at ab initio accuracy This dataset accompanies the publication “Fine-Tuning Unifies Foundational Machine-learned Interatomic Potential Architectures at ab initio Accuracy.”It contains fine-tuned and foundation machine learning interatomic potentials (MLIPs) across five major frameworks MACE, GRACE, SevenNet, MatterSim, and ORB for seven representative chemical systems (including MoS₂, KOH, Li₁₃Si₄, and others). The repository includes model weights, configuration files, training logs, and reference data used for benchmarking energy, force, and diffusion accuracy. These data support the study’s findings that fine-tuned foundation models can achieve near-_ab initio_ accuracy in molecular dynamics while maintaining nanosecond-scale efficiency.