SevenNet-0 (11July2024) is an interatomic potential pre-trained on the MPTrj(CHGNet) dataset.
Warning: This is NOT the potential referred to in our paper. Please check 'SevenNet/pretrained_potentials/SevenNet_0__22May2024' for the referenced there.
SevenNet-0 (11July2024) has the same architecture and number of parameters as SevenNet-0 (22May2024), but the training process and dataset differ.
We found that this model performs better than the previous SevenNet-0 (22May2024).
It can be directly applied to any system without training and fine-tuned with another dataset if accuracy is unsatisfactory.
- checkpoint_sevennet_0.pth: checkpoint of SevenNet-0 model.
- fine_tune.yaml: example input.yaml for fine-tuning.
- pre_train.yaml: example input.yaml file used in pre-training.
MAE (Mean absolute error), for SevenNet-0 (11July2024), we did not split the dataset.
| Energy (eV/atom) | Force (eV/Å) | Stress (GPa) | |
|---|---|---|---|
| Train | 0.011 | 0.040 | 0.28 |
You can obtain models for LAMMPS by running the following command (-p for parallel)
$ sevenn_get_model {-p} 7net-0_11July2024Refer to example_inputs/md_{serial/parallel}_example/ for their usage in LAMMPS. Both serial and parallel model gives the same results but the parallel model enables multi-GPU MD simulation.