Eastsheng's Wiki

GPUMD NEP训练例子

2024-09-24 12:18:28

[toc]

输入文件准备

  • nep.in，指定用于训练NEP的超参数
1
2
3
4
5
6
7
8
9
10
11
12
13
type          2 Te Pb # number of atom types and list of chemical species
version       4       # select the NEP version
cutoff        8 4     # 径向（rcR）和角向（rcA）截断
n_max         4 4     # 径向（nmaxR）和角向（nmaxA）基的尺寸
basis_size    8 8     # 径向（NbasR）和角度（NbasA）基函数的数目
l_max         4 2 0   # expansion order for angular terms
neuron        30      # number of neurons in the hidden layer 
lambda_e      1.0     # weight of energy loss term
lambda_f      1.0     # weight of force loss term
lambda_v      0.1     # weight of virial loss term
batch         1000    # batch size for training
population    50      # population size used in the SNES algorithm
generation    100000  # number of generations used by the SNES algorithm

训练与结果

1
nep < nep.in

.
├── energy_train.out
├── force_train.out
├── loss.out
├── nep.in
├── nep.restart
├── nep.txt
├── stress_train.out
├── train.xyz
└── virial_train.out

0 directories, 9 files

1
2
gpumdkit.sh -plt -h
gpumdkit.sh -plt train

train

参考

[1] https://gpumd.org/dev/gpumd/output_files/index.html

Tags: GPUMD NEP
← GBDT 和 GBR 的区别 GPUMD安装与测试 →