model_name: Allegro-MP-L # required (this must match the model's label which is the 3rd arg in the matbench_discovery.preds.Model enum) model_key: allegro-MP-L-0.1 # this should match the name of the YAML file and determines the URL /models/ on which details of the model are displayed on the website model_version: '0.1' date_added: '2025-09-08' date_published: '2025-08-28' authors: - name: Seán R. Kavanagh affiliation: Center for the Environment, Harvard University & MIR Group, Harvard University email: skavanagh@seas.harvard.edu orcid: https://orcid.org/0000-0003-4577-9647 url: https://seankavanagh.com corresponding: true - name: Chuin Wei Tan affiliation: MIR Group, Harvard University - name: Albert Musaelian affiliation: MIR Group, Harvard University & Mirian Technologies - name: William C. Witt affiliation: MIR Group, Harvard University - name: Gabriel de Miranda Nascimento affiliation: MIR Group, Harvard University & MIT - name: Ulrik Unneberg affiliation: MIR Group, Harvard University & MIT - name: Marc L. Descoteaux affiliation: MIR Group, Harvard University - name: Boris Kozinsky affiliation: MIR Group, Harvard University trained_by: - name: Seán R. Kavanagh affiliation: Center for the Environment, Harvard University & MIR Group, Harvard University email: skavanagh@seas.harvard.edu orcid: https://orcid.org/0000-0003-4577-9647 url: https://seankavanagh.com corresponding: true repo: https://github.com/mir-group/allegro url: https://allegro.readthedocs.io/en/latest/ doi: https://doi.org/10.5281/zenodo.16980200 paper: https://arxiv.org/abs/2504.16068 # To be updated... checkpoint_url: https://www.nequip.net/models/mir-group/Allegro-MP-L:0.1 pr_url: https://github.com/janosh/matbench-discovery/pull/282 openness: OSOD # see `Open` enum in matbench_discovery/enums.py train_task: S2EFS # see `Task` enum in matbench_discovery/enums.py test_task: IS2RE-SR # see `Task` enum in matbench_discovery/enums.py targets: EFS_G # see `Targets` enum in matbench_discovery/enums.py model_type: UIP # see `ModelType` enum in matbench_discovery/enums.py model_params: 18_700_000 trained_for_benchmark: true n_estimators: 1 license: code: MIT code_url: https://github.com/mir-group/nequip/blob/main/LICENSE checkpoint: CC-BY-4.0 checkpoint_url: https://creativecommons.org/licenses/by/4.0/legalcode hyperparams: max_force: 0.05 max_steps: 500 ase_optimizer: GOQN # faster than FIRE with same results; see SI of https://doi.org/10.1088/2515-7655/ade916 cell_filter: FrechetCellFilter optimizer: AdamW weight_decay: 1e-3 graph_construction_radius: 6.0 sph_harmonics_l_max: 3 n_layers: 5 n_features: 96 parity: false zbl_potential: false allegro_mlp_depth: 3 allegro_mlp_width: 1024 tensor_path_channel_coupling: true polynomial_cutoff: 8 n_radial_bessel_basis: 12 loss: Huber - delta=0.01 for energy, delta=0.1 for stress, stratified delta (0.01, 0.007, 0.004, 0.001) for force loss_weights: energy: 1.0 force: 5.0 stress: 0.01 batch_size: 520 # 8 (gpus) * 65 (batch per gpu) = 520 (total batch size) initial_learning_rate: 0.01 gradient_clip_val: 0.015 learning_rate_schedule: ReduceLROnPlateau - factor=0.5, patience=250, min_lr=1e-6 epochs: 250 max_neighbors: .inf training_cost: Nvidia H100 GPUs: { amount: 8, hours: 50, cost: 400 } requirements: torch: '2' nequip: '>=0.7.0' # >=0.14.0 recommended allegro: '>=0.5.0' # >=0.7.1 recommended training_set: [MPtrj] notes: Description: Large 'compliant' Allegro foundation potential; see for details and for model/training infrastructure. Steps: Single training run on MPtrj with specified parameters. metrics: phonons: kappa_103: pred_file: models/allegro/Allegro-MP-L-0.1/Allegro-MP-L-0.1-2025-08-27-kappa-103-FIRE-dist_0_03-fmax_0_0001-symprec_1e-05.json.gz pred_file_url: https://figshare.com/files/57576301 κ_SRME: 0.5039 κ_SRE: 0.2309 geo_opt: pred_file: models/allegro/Allegro-MP-L-0.1/Allegro-MP-L-0.1-2025-08-27-wbm-IS2RE-GOQN.jsonl.gz pred_file_url: https://figshare.com/files/57731014 struct_col: nequip_structure # same for NequIP/Allegro symprec=1e-5: rmsd: 0.0816 # unitless n_sym_ops_mae: 1.902 # unitless symmetry_decrease: 0.0712 # fraction symmetry_match: 0.7211 # fraction symmetry_increase: 0.2015 # fraction n_structures: 256963 # count analysis_file: models/allegro/Allegro-MP-L-0.1/Allegro-MP-L-0.1-2025-08-27-wbm-IS2RE-GOQN-symprec=1e-5-moyo=0.4.4.csv.gz analysis_file_url: https://figshare.com/files/57731050 symprec=1e-2: rmsd: 0.0816 # unitless n_sym_ops_mae: 1.7918 # unitless symmetry_decrease: 0.0572 # fraction symmetry_match: 0.8125 # fraction symmetry_increase: 0.123 # fraction n_structures: 256963 # count analysis_file: models/allegro/Allegro-MP-L-0.1/Allegro-MP-L-0.1-2025-08-27-wbm-IS2RE-GOQN-symprec=1e-2-moyo=0.4.4.csv.gz analysis_file_url: https://figshare.com/files/57731053 discovery: pred_file: models/allegro/Allegro-MP-L-0.1/Allegro-MP-L-0.1-2025-08-27-wbm-IS2RE.csv.gz pred_file_url: https://figshare.com/files/57575095 pred_col: e_form_per_atom_nequip # same for NequIP/Allegro full_test_set: F1: 0.742 # fraction DAF: 3.998 # dimensionless Precision: 0.686 # fraction Recall: 0.809 # fraction Accuracy: 0.904 # fraction TPR: 0.809 # fraction FPR: 0.077 # fraction TNR: 0.923 # fraction FNR: 0.191 # fraction TP: 35671.0 # count FP: 16330.0 # count TN: 196541.0 # count FN: 8421.0 # count MAE: 0.042 # eV/atom RMSE: 0.085 # eV/atom R2: 0.775 # dimensionless missing_preds: 2 # count unique_prototypes: F1: 0.751 # fraction DAF: 4.516 # dimensionless Precision: 0.69 # fraction Recall: 0.823 # fraction Accuracy: 0.915 # fraction TPR: 0.823 # fraction FPR: 0.068 # fraction TNR: 0.932 # fraction FNR: 0.177 # fraction TP: 27483.0 # count FP: 12329.0 # count TN: 169785.0 # count FN: 5891.0 # count MAE: 0.044 # eV/atom RMSE: 0.087 # eV/atom R2: 0.778 # dimensionless missing_preds: 0 # count most_stable_10k: F1: 0.917 # fraction DAF: 5.536 # dimensionless Precision: 0.846 # fraction Recall: 1.0 # fraction Accuracy: 0.846 # fraction TPR: 1.0 # fraction FPR: 1.0 # fraction TNR: 0.0 # fraction FNR: 0.0 # fraction TP: 8463.0 # count FP: 1537.0 # count TN: 0.0 # count FN: 0.0 # count MAE: 0.064 # eV/atom RMSE: 0.124 # eV/atom R2: 0.706 # dimensionless missing_preds: 0 # count