DPA-3.1-3M-FT

Version: v0.3 Added: 2025-06-05 Published: 2025-06-05 3.27M parameters Missing preds: 0
Repo Paper Docs DOI Files PR Checkpoint
Predictions
κSRMEGeometry OptimizationDiscovery

Convex hull distance prediction errors projected onto elements

1 H 0.128
2 He 0
3 Li 0.0286
4 Be 0.0304
5 B 0.0641
6 C 0.0556
7 N 0.0841
8 O 0.11
9 F 0.111
10 Ne 0
11 Na 0.0301
12 Mg 0.0386
13 Al 0.0502
14 Si 0.0612
15 P 0.0532
16 S 0.069
17 Cl 0.0949
18 Ar 0
19 K 0.0355
20 Ca 0.0417
21 Sc 0.033
22 Ti 0.0385
23 V 0.0592
24 Cr 0.0826
25 Mn 0.11
26 Fe 0.0924
27 Co 0.0476
28 Ni 0.0428
29 Cu 0.0423
30 Zn 0.0399
31 Ga 0.0481
32 Ge 0.0509
33 As 0.0525
34 Se 0.0884
35 Br 0.0786
36 Kr 0
37 Rb 0.0374
38 Sr 0.0378
39 Y 0.0416
40 Zr 0.0402
41 Nb 0.047
42 Mo 0.0506
43 Tc 0.0373
44 Ru 0.0537
45 Rh 0.0499
46 Pd 0.0471
47 Ag 0.0355
48 Cd 0.0326
49 In 0.0553
50 Sn 0.0477
51 Sb 0.054
52 Te 0.113
53 I 0.0654
54 Xe 0.032
55 Cs 0.0362
56 Ba 0.038
57 La 0.0343
58 Ce 0.0355
59 Pr 0.0356
60 Nd 0.0331
61 Pm 0.0318
62 Sm 0.0346
63 Eu 0.0636
64 Gd 0.0447
65 Tb 0.0339
66 Dy 0.0364
67 Ho 0.0331
68 Er 0.0334
69 Tm 0.0336
70 Yb 0.0532
71 Lu 0.0351
72 Hf 0.0404
73 Ta 0.07
74 W 0.0459
75 Re 0.0428
76 Os 0.0538
77 Ir 0.0583
78 Pt 0.0578
79 Au 0.0613
80 Hg 0.0312
81 Tl 0.0348
82 Pb 0.056
83 Bi 0.0408
84 Po 0
85 At 0
86 Rn 0
87 Fr 0
88 Ra 0
89 Ac 0.0317
90 Th 0.0466
91 Pa 0.0473
92 U 0.0579
93 Np 0.0917
94 Pu 0.197
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. Duo Zhang  AI for Science Institute, Beijing  AI for Science Institute, Beijing logo  
  2. Anyang Peng  AI for Science Institute, Beijing  AI for Science Institute, Beijing logo  
  3. Chun Cai  AI for Science Institute, Beijing  AI for Science Institute, Beijing logo  
  4. Linfeng Zhang  AI for Science Institute, Beijing; DP Technology  
  5. Han Wang  Beijing Institute of Applied Physics and Computational Mathematics (IAPCM)  

Trained By

  1. Anyang Peng (AI for Science Institute, Beijing)

Model Info

  • Model Version v0.3
  • Model Type UIP
  • Targets EFSG
  • Openness OSCD
  • Train Task S2EFS
  • Test Task IS2RE-SR
  • Trained for Benchmark Yes

Training Set

OpenLAM dataset v1: 163M structures

Description

DPA3 is an advanced interatomic potential leveraging the message passing architecture, implemented within the DeePMD-kit framework, available on GitHub. 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.

Hyperparameters

  • max_force: 0.05
  • max_steps: 500
  • ase_optimizer: "FIRE"
  • cell_filter: "ExpCellFilter"
  • n_layers: 16
  • e_rcut: 6
  • a_rcut: 4
  • n_dim: 128
  • e_dim: 64
  • 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":4000000,"batch_size":960,"epochs":23.5}
  • finetune: {"loss":"Huber","loss_weights":{"energy":30,"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":2000000,"batch_size":256,"epochs":45}
  • graph_construction_radius: 6
  • max_neighbors: null

Dependencies