Nequix MP PFT

Version: 0.1.0 Added: 2026-01-08 Published: 2026-01-08 708k parameters Missing preds: 0
Repo Paper Docs DOI Files PR Checkpoint
Predictions
κSRMEGeometry OptimizationDiscovery

Convex hull distance prediction errors projected onto elements

1 H 0.28
2 He 0.00
3 Li 0.05
4 Be 0.11
5 B 0.14
6 C 0.11
7 N 0.15
8 O 0.16
9 F 0.17
10 Ne 0.00
11 Na 0.06
12 Mg 0.07
13 Al 0.11
14 Si 0.12
15 P 0.12
16 S 0.14
17 Cl 0.16
18 Ar 0.00
19 K 0.07
20 Ca 0.07
21 Sc 0.08
22 Ti 0.09
23 V 0.11
24 Cr 0.14
25 Mn 0.16
26 Fe 0.15
27 Co 0.09
28 Ni 0.09
29 Cu 0.07
30 Zn 0.08
31 Ga 0.10
32 Ge 0.11
33 As 0.10
34 Se 0.14
35 Br 0.15
36 Kr 0.00
37 Rb 0.07
38 Sr 0.07
39 Y 0.08
40 Zr 0.09
41 Nb 0.12
42 Mo 0.11
43 Tc 0.10
44 Ru 0.13
45 Rh 0.11
46 Pd 0.09
47 Ag 0.06
48 Cd 0.06
49 In 0.10
50 Sn 0.10
51 Sb 0.09
52 Te 0.16
53 I 0.13
54 Xe 0.11
55 Cs 0.07
56 Ba 0.07
57 La 0.07
58 Ce 0.07
59 Pr 0.07
60 Nd 0.07
61 Pm 0.06
62 Sm 0.07
63 Eu 0.08
64 Gd 0.07
65 Tb 0.07
66 Dy 0.07
67 Ho 0.07
68 Er 0.07
69 Tm 0.07
70 Yb 0.07
71 Lu 0.07
72 Hf 0.10
73 Ta 0.15
74 W 0.10
75 Re 0.11
76 Os 0.12
77 Ir 0.13
78 Pt 0.11
79 Au 0.11
80 Hg 0.06
81 Tl 0.08
82 Pb 0.10
83 Bi 0.09
84 Po 0.00
85 At 0.00
86 Rn 0.00
87 Fr 0.00
88 Ra 0.00
89 Ac 0.07
90 Th 0.10
91 Pa 0.12
92 U 0.10
93 Np 0.16
94 Pu 0.36
95 Am 0.00
96 Cm 0.00
97 Bk 0.00
98 Cf 0.00
99 Es 0.00
100 Fm 0.00
101 Md 0.00
102 No 0.00
103 Lr 0.00
104 Rf 0.00
105 Db 0.00
106 Sg 0.00
107 Bh 0.00
108 Hs 0.00
109 Mt 0.00
110 Ds 0.00
111 Rg 0.00
112 Cn 0.00
113 Nh 0.00
114 Fl 0.00
115 Mc 0.00
116 Lv 0.00
117 Ts 0.00
118 Og 0.00
57-71 La-Lu Lanthanides
89-103 Ac-Lr Actinides

Model Authors

  1. Teddy Koker  Massachusetts Institute of Technology      
  2. Abhijeet Gangan  University of California, Los Angeles      
  3. Mit Kotak  Massachusetts Institute of Technology      
  4. Jaime Marian  University of California, Los Angeles      
  5. Tess Smidt  Massachusetts Institute of Technology      

Trained By

  1. Teddy Koker (Massachusetts Institute of Technology)

Model Info

  • Model Version 0.1.0
  • Model Type UIP
  • Targets EFSHG
  • Openness OSOD
  • Train Task S2EFS
  • Test Task IS2RE-SR
  • Trained for Benchmark Yes

Training Set

MPtrj: 1.58M structures from 146k materials

MDR PBE Phonons in MPtrj: 8.51k structures

Description

Nequix MP PFT is a fine-tuned version of Nequix MP 1, which was fine-tuned using phonon fine-tuning (PFT) on a subset of the MDR Phonon PBE dataset that is contained within MPtrj.

Reproduction

git clone https://github.com/atomicarchitects/nequix.git && cd nequix uv sync --extra pft bash data/download_pbe_mdr_preprocessed.sh uv run nequix/pft/train.py configs/nequix-mp-1-pft.yml

Hyperparameters

  • max_force: 0.02
  • max_steps: 500
  • ase_optimizer: "FIRE"
  • cell_filter: "FrechetCellFilter"
  • hidden_irreps: "128x0e+64x1o+32x2e+32x3o"
  • graph_construction_radius: 6
  • max_neighbors: null
  • lmax: 3
  • radial_basis_size: 8
  • radial_mlp_size: 64
  • radial_mlp_layers: 2
  • radial_polynomial_p: 6
  • layer_norm: true
  • optimizer: "adamw"
  • learning_rate: 0.0001
  • weight_decay: 0.001
  • grad_clip_norm: 100
  • batch_size: 16
  • n_epochs: 100
  • loss: {"energy":"mae","force":"l2","stress":"mae","hessian":"mae"}
  • loss_weights: {"energy":0,"force":20,"stress":5,"hessian":100}
  • cotrain_loss_weights: {"energy":500,"force":200,"stress":50}
  • ema_decay: 0.999

Dependencies