Allegro-OAM-L

Version: 0.1 Added: 2025-09-08 Published: 2025-08-28 9.7M parameters Missing preds: 0
Repo Paper Docs DOI Files PR Checkpoint
Predictions
κSRMEGeometry OptimizationDiscovery

Convex hull distance prediction errors projected onto elements

1 H 0.135
2 He 0
3 Li 0.0259
4 Be 0.0244
5 B 0.057
6 C 0.0571
7 N 0.0755
8 O 0.123
9 F 0.118
10 Ne 0
11 Na 0.0297
12 Mg 0.0342
13 Al 0.0421
14 Si 0.0551
15 P 0.051
16 S 0.0753
17 Cl 0.0941
18 Ar 0
19 K 0.0349
20 Ca 0.0366
21 Sc 0.0294
22 Ti 0.036
23 V 0.0587
24 Cr 0.081
25 Mn 0.108
26 Fe 0.0876
27 Co 0.0447
28 Ni 0.0387
29 Cu 0.0365
30 Zn 0.0349
31 Ga 0.0426
32 Ge 0.0467
33 As 0.0493
34 Se 0.0927
35 Br 0.0856
36 Kr 0
37 Rb 0.0373
38 Sr 0.0337
39 Y 0.0376
40 Zr 0.0366
41 Nb 0.0462
42 Mo 0.0511
43 Tc 0.0373
44 Ru 0.0489
45 Rh 0.0443
46 Pd 0.0426
47 Ag 0.0331
48 Cd 0.0278
49 In 0.0481
50 Sn 0.0427
51 Sb 0.0516
52 Te 0.114
53 I 0.0744
54 Xe 0.03
55 Cs 0.0364
56 Ba 0.0348
57 La 0.0321
58 Ce 0.0313
59 Pr 0.0332
60 Nd 0.0302
61 Pm 0.0308
62 Sm 0.0315
63 Eu 0.061
64 Gd 0.0435
65 Tb 0.0303
66 Dy 0.0328
67 Ho 0.0297
68 Er 0.0297
69 Tm 0.0297
70 Yb 0.0479
71 Lu 0.0309
72 Hf 0.0374
73 Ta 0.0671
74 W 0.0387
75 Re 0.0397
76 Os 0.0487
77 Ir 0.0566
78 Pt 0.0534
79 Au 0.0554
80 Hg 0.0302
81 Tl 0.0317
82 Pb 0.0524
83 Bi 0.0371
84 Po 0
85 At 0
86 Rn 0
87 Fr 0
88 Ra 0
89 Ac 0.0286
90 Th 0.0429
91 Pa 0.0455
92 U 0.0511
93 Np 0.0882
94 Pu 0.196
95 Am 0
96 Cm 0
97 Bk 0
98 Cf 0
99 Es 0
100 Fm 0
101 Md 0
102 No 0
103 Lr 0
104 Rf 0
105 Db 0
106 Sg 0
107 Bh 0
108 Hs 0
109 Mt 0
110 Ds 0
111 Rg 0
112 Cn 0
113 Nh 0
114 Fl 0
115 Mc 0
116 Lv 0
117 Ts 0
118 Og 0
57-71 La-Lu Lanthanides
89-103 Ac-Lr Actinides

Model Authors

  1. Seán R. Kavanagh  Center for the Environment, Harvard University & MIR Group, Harvard University    
  2. Chuin Wei Tan  MIR Group, Harvard University  MIR Group, Harvard University logo
  3. Albert Musaelian  MIR Group, Harvard University & Mirian Technologies
  4. William C. Witt  MIR Group, Harvard University  MIR Group, Harvard University logo
  5. Gabriel de Miranda Nascimento  MIR Group, Harvard University & MIT
  6. Ulrik Unneberg  MIR Group, Harvard University & MIT
  7. Marc L. Descoteaux  MIR Group, Harvard University  MIR Group, Harvard University logo
  8. Boris Kozinsky  MIR Group, Harvard University  MIR Group, Harvard University logo

Trained By

  1. Seán R. Kavanagh (Center for the Environment, Harvard University & MIR Group, Harvard University)

Model Info

  • Model Version 0.1
  • Model Type UIP
  • Targets EFSG
  • Openness OSOD
  • Train Task S2EFS
  • Test Task IS2RE-SR
  • Trained for Benchmark Yes

Training Set

OMat24: 101M structures from 3.23M materials

Subsampled Alexandria: 10.4M structures from 3.23M materials

MPtrj: 1.58M structures from 146k materials

Description

Large Allegro foundation potential; see https://www.nequip.net/models/mir-group/Allegro-OAM-L:0.1 for details and https://arxiv.org/abs/2504.16068 for model/training infrastructure.

Steps

Training performed by: (1) pre-training on OMat24; (2) fine-tuning on MPtrj+sAlex, with a reduced learning rate (1e-4), energy-loss-upweighting (1:1:0.01 instead of 1:5:0.01) and StochasticWeightAveraging (SWA).

Hyperparameters

  • max_force: 0.05
  • max_steps: 500
  • ase_optimizer: "GOQN"
  • cell_filter: "FrechetCellFilter"
  • optimizer: "AdamW"
  • weight_decay: 1e-8
  • graph_construction_radius: 7
  • sph_harmonics_l_max: 3
  • n_layers: 4
  • n_features: 64
  • parity: false
  • zbl_potential: false
  • allegro_mlp_depth: 2
  • 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,"force":5,"stress":0.01}
  • batch_size: 480
  • initial_learning_rate: 0.005
  • gradient_clip_val: 0.0001
  • learning_rate_schedule: "ReduceLROnPlateau - factor=0.1, patience=10, min_lr=1e-6"
  • epochs: 15
  • max_neighbors: null

Dependencies