model_name: DPA3-v1-OpenLAM model_key: dpa3-v1-openlam model_version: v0.1 # 2025-01-10 date_added: '2025-01-10' date_published: '2025-01-10' authors: - name: Duo Zhang affiliation: AI for Science Institute, Beijing orcid: https://orcid.org/0000-0001-9591-2659 - name: Anyang Peng affiliation: AI for Science Institute, Beijing orcid: https://orcid.org/0000-0002-0630-2187 - name: Chun Cai affiliation: AI for Science Institute, Beijing orcid: https://orcid.org/0000-0001-6242-0439 - name: Linfeng Zhang affiliation: AI for Science Institute, Beijing; DP Technology email: linfeng.zhang.zlf@gmail.com corresponding: true - name: Han Wang affiliation: Beijing Institute of Applied Physics and Computational Mathematics (IAPCM) email: wang_han@iapcm.ac.cn corresponding: true trained_by: - name: Anyang Peng affiliation: AI for Science Institute, Beijing orcid: https://orcid.org/0000-0002-0630-2187 repo: https://github.com/deepmodeling/deepmd-kit/tree/dpa3-alpha url: https://github.com/deepmodeling/deepmd-kit/tree/dpa3-alpha doi: https://github.com/deepmodeling/deepmd-kit/tree/dpa3-alpha # to be released soon paper: https://github.com/deepmodeling/deepmd-kit/tree/dpa3-alpha # to be released soon pr_url: https://github.com/janosh/matbench-discovery/pull/192 # checkpoints reported in https://github.com/deepmodeling/deepmd-kit/discussions/4682 checkpoint_url: https://bohrium-api.dp.tech/ds-dl/dpa3openlam-74ng-v3.zip license: code: LGPL-3.0 code_url: https://github.com/deepmodeling/deepmd-kit/blob/70bc6d89/LICENSE checkpoint: LGPL-3.0 checkpoint_url: https://github.com/deepmodeling/deepmd-kit/blob/70bc6d89/LICENSE openness: OSCD train_task: S2EFS test_task: IS2RE-SR targets: EFS_G model_type: UIP model_params: 8_184_608 n_estimators: 1 status: superseded superseded_by: dpa3-v2-openlam trained_for_benchmark: true hyperparams: max_force: 0.05 max_steps: 500 ase_optimizer: FIRE cell_filter: ExpCellFilter n_layers: 6 e_rcut: 6.0 a_rcut: 4.0 n_dim: 384 e_dim: 96 a_dim: 32 optimizer: Adam pretrain: loss: MSE loss_weights: energy: 0.02 -> 1 force: 1000 -> 100 virial: 0.02 -> 1 initial_learning_rate: 0.001 learning_rate_schedule: ExpLR - start_lr=0.001, decay_steps=5000, stop_lr=0.00001 training_steps: 1_600_000 batch_size: 960 # 120 (gpus) * 8 (batch per gpu) = 960 (total batch size) epochs: 9.4 finetune: loss: Huber loss_weights: energy: 15 force: 1 virial: 2.5 initial_learning_rate: 0.0001 learning_rate_schedule: ExpLR - start_lr=0.0001, decay_steps=5000, stop_lr=0.000006 training_steps: 400_000 batch_size: 256 # 64 (gpus) * 4 (batch per gpu) = 256 (total batch size) epochs: 9 loss_continue: Huber loss_weights_continue: energy: 30 force: 1 virial: 2.5 initial_learning_rate_continue: 0.0001 learning_rate_schedule_continue: ExpLR - start_lr=0.0001, decay_steps=5000, stop_lr=0.000000001 training_steps_continue: 600_000 batch_size_continue: 256 # 64 (gpus) * 4 (batch per gpu) = 256 (total batch size) epochs_continue: 13.5 graph_construction_radius: 6.0 # Å max_neighbors: 120 # from https://github.com/deepmodeling/deepmd-kit/discussions/4682#discussioncomment-12836651 requirements: torch: 2.3.1 torch-geometric: 2.5.2 ase: 3.23.0 pymatgen: 2024.6.10 numpy: 1.26.4 training_set: [OpenLAM] training_cost: missing notes: Description: | DPA3 is an advanced interatomic potential leveraging the message passing architecture, implemented within the DeePMD-kit framework, available on [GitHub](https://github.com/deepmodeling/deepmd-kit/tree/dpa3-alpha). Designed as a large atomic model (LAM), DPA3 is tailored to integrate and simultaneously train on datasets from various disciplines, encompassing diverse chemical and materials systems across different research domains. Its model design ensures exceptional fitting accuracy and robust generalization both within and beyond the training domain. Furthermore, DPA3 maintains energy conservation and respects the physical symmetries of the potential energy surface, making it a dependable tool for a wide range of scientific applications. metrics: phonons: kappa_103: κ_SRME: 0.741 pred_file: models/deepmd/dpa3-v1-openlam/2025-01-10-kappa-103-FIRE-dist=0.01-fmax=1e-4-symprec=1e-5.json.gz pred_file_url: https://figshare.com/files/52134863 geo_opt: pred_file: models/deepmd/dpa3-v1-openlam/2025-01-10-wbm-geo-opt.jsonl.gz struct_col: dp_structure pred_file_url: https://figshare.com/files/57754516 symprec=1e-5: analysis_file: models/deepmd/dpa3-v1-openlam/2025-01-10-wbm-geo-opt-symprec=1e-5-moyo=0.3.3.csv.gz analysis_file_url: https://figshare.com/files/52291973 # deleted from Figshare (model superseded by dpa3-v2) rmsd: 0.0128 # unitless n_sym_ops_mae: 2.1477 # unitless symmetry_decrease: 0.0657 # fraction symmetry_match: 0.7188 # fraction symmetry_increase: 0.2094 # fraction n_structures: 256963 # count symprec=1e-2: analysis_file: models/deepmd/dpa3-v1-openlam/2025-01-10-wbm-geo-opt-symprec=1e-2-moyo=0.3.3.csv.gz analysis_file_url: https://figshare.com/files/52291976 # deleted from Figshare (model superseded by dpa3-v2) rmsd: 0.0128 # unitless n_sym_ops_mae: 1.8912 # unitless symmetry_decrease: 0.0515 # fraction symmetry_match: 0.8097 # fraction symmetry_increase: 0.1314 # fraction n_structures: 256963 # count discovery: pred_file: models/deepmd/dpa3-v1-openlam/2025-01-10-wbm-IS2RE.csv.gz pred_file_url: https://figshare.com/files/52057532 pred_col: e_form_per_atom_dp full_test_set: F1: 0.862 # fraction DAF: 5.013 # dimensionless Precision: 0.86 # fraction Recall: 0.865 # fraction Accuracy: 0.953 # fraction TPR: 0.865 # fraction FPR: 0.029 # fraction TNR: 0.971 # fraction FNR: 0.135 # fraction TP: 38130.0 # count FP: 6197.0 # count TN: 206674.0 # count FN: 5962.0 # count MAE: 0.023 # eV/atom RMSE: 0.067 # eV/atom R2: 0.863 # dimensionless missing_preds: 2 # count most_stable_10k: F1: 0.987 # fraction DAF: 6.371 # dimensionless Precision: 0.974 # fraction Recall: 1.0 # fraction Accuracy: 0.974 # fraction TPR: 1.0 # fraction FPR: 1.0 # fraction TNR: 0.0 # fraction FNR: 0.0 # fraction TP: 9740.0 # count FP: 260.0 # count TN: 0.0 # count FN: 0.0 # count MAE: 0.019 # eV/atom RMSE: 0.066 # eV/atom R2: 0.905 # dimensionless missing_preds: 0 # count unique_prototypes: F1: 0.883 # fraction DAF: 5.754 # dimensionless Precision: 0.88 # fraction Recall: 0.885 # fraction Accuracy: 0.963 # fraction TPR: 0.885 # fraction FPR: 0.022 # fraction TNR: 0.978 # fraction FNR: 0.115 # fraction TP: 29549.0 # count FP: 4042.0 # count TN: 178072.0 # count FN: 3825.0 # count MAE: 0.023 # eV/atom RMSE: 0.067 # eV/atom R2: 0.869 # dimensionless missing_preds: 0 # count