新建
快速上手使用 DP-GEN + CALYPSO | 加速新材料发现
rainvibe
推荐镜像 :dp222-dpgen-calypso:v0.1
推荐机型 :c2_m4_cpu
赞 2
2
目录
数据集
dpgen-calypso-MgAl(v1)
dp-calypso-example(v1)
快速上手使用 DP-GEN + CALYPSO 加速新材料发现
代码
文本
©️ Copyright 2024 @ Authors
作者：王振雨(wzy@calypso.cn)
日期：2024-03-25
共享协议：本作品采用知识共享署名-非商业性使用-相同方式共享 4.0 国际许可协议进行许可。
快速开始：你可以点击界面上方蓝色按钮 开始连接 ，选择 `dp222-dpgen-calypso:v0.1` 镜像及`c2_m4_cpu`节点配置稍等片刻即可运行。
代码
文本
目标
使用DPGEN结合CALYPSO进行势能面的快速探索并构建势函数
在学习本教程后，你将能够：
- 学会安装dpgen和dpdata
- 使用dp结合CALYPSO进行MLP加速的结构预测
- 使用dpgen结合CALYPSO进行适用于结构预测的势函数的构建
阅读该教程【最多】约需 10 分钟(算例计算完成大概需要1个小时)，让我们开始吧！
代码
文本
代码
文本
代码
文本
[1]
import os
os.environ["PATH"]=os.getenv("PATH")+":/opt/deepmd/deepmd-kit-222/bin"
os.getenv("PATH")
'/root/.local/bin:/CALYPSO-Bohrium/utils:/opt/intel/oneapi/mkl/2023.2.0/bin/intel64:/opt/intel/oneapi/mpi/2021.7.1//libfabric/bin:/opt/intel/oneapi/mpi/2021.7.1//bin:/opt/intel/oneapi/debugger/2021.7.1/gdb/intel64/bin:/opt/intel/oneapi/compiler/2022.2.1/linux/bin/intel64:/opt/intel/oneapi/compiler/2022.2.1/linux/bin:/opt/root/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/opt/deepmd/deepmd-kit-222/bin'
代码
文本
[31]
!which dp
/opt/deepmd/deepmd-kit-222/bin/dp
代码
文本
[32]
!dpgen -h #查看是否安装成功
DeepModeling
------------
Version: 0.10.1.dev310+g4719ba1
Path: /opt/deepmd/deepmd-kit-222/lib/python3.10/site-packages/dpgen
Dependency
------------
numpy 1.25.0 /opt/deepmd/deepmd-kit-222/lib/python3.10/site-packages/numpy
dpdata 0.2.15 /opt/deepmd/deepmd-kit-222/lib/python3.10/site-packages/dpdata
pymatgen unknown version or path
monty 2024.2.26 /opt/deepmd/deepmd-kit-222/lib/python3.10/site-packages/monty
ase 3.22.1 /opt/deepmd/deepmd-kit-222/lib/python3.10/site-packages/ase
paramiko 3.4.0 /opt/deepmd/deepmd-kit-222/lib/python3.10/site-packages/paramiko
custodian 2024.3.12 /opt/deepmd/deepmd-kit-222/lib/python3.10/site-packages/custodian
Reference
------------
Please cite:
Yuzhi Zhang, Haidi Wang, Weijie Chen, Jinzhe Zeng, Linfeng Zhang, Han Wang, and Weinan E,
DP-GEN: A concurrent learning platform for the generation of reliable deep learning
based potential energy models, Computer Physics Communications, 2020, 107206.
------------
Description
------------
usage: dpgen [-h]
{init_surf,init_bulk,auto_gen_param,init_reaction,run,run/report,collect,simplify,autotest,db}
...
dpgen is a convenient script that uses DeepGenerator to prepare initial data,
drive DeepMDkit and analyze results. This script works based on several sub-
commands with their own options. To see the options for the sub-commands, type
"dpgen sub-command -h".
positional arguments:
{init_surf,init_bulk,auto_gen_param,init_reaction,run,run/report,collect,simplify,autotest,db}
init_surf Generating initial data for surface systems.
init_bulk Generating initial data for bulk systems.
auto_gen_param auto gen param.json
init_reaction Generating initial data for reactive systems.
run Main process of Deep Potential Generator.
run/report Report the systems and the thermodynamic conditions of
the labeled frames.
collect Collect data.
simplify Simplify data.
autotest Auto-test for Deep Potential.
db Collecting data from DP-GEN.
options:
-h, --help show this help message and exit
代码
文本
如果是在自有集群上使用dpgen+calypso,可按照以下步骤安装：
- 使用conda安装deepmdkit：
conda create -n deepmd_222_cpu deepmd-kit=2.2.2=*cpu libdeepmd=2.2.2=*cpu lammps horovod -c https://conda.deepmodeling.com -c defaultsorconda create -n deepmd_222_gpu deepmd-kit=2.2.2=*gpu libdeepmd=2.2.2=*gpu lammps cudatoolkit=11.6 horovod -c https://conda.deepmodeling.com -c defaults - 激活环境：
conda activate deepmd_222_gpu - 下载dpgen：
pip install git+https://github.com/wangzyphysics/dpgen.git@devel - 获取CALYPSO：http://calypso.cn/getting-calypso/
至此环境准备完毕
代码
文本
[15]
#进入/personal目录，从数据集中下载算例
!cp -r /bohr/dpgen-calypso-ia74/v1 /personal/dpgen-calypso-example
sh: 0: getcwd() failed: No such file or directory
代码
文本
[2]
cd /personal/dpgen-calypso-example
/personal/dpgen-calypso-example
代码
文本
[17]
ls
dpgen-calypso.zip*
代码
文本
[19]
!unzip dpgen-calypso.zip
已隐藏输出
代码
文本
[34]
cd dpgen-calypso
/personal/dpgen-calypso-example/dpgen-calypso
代码
文本
[26]
ls *
MgAl.json machine.json run.sh calypso_input: calypso.x* input.dat input.dat.Al input.dat.Mg input.dat.MgAl data: trainingset/ vasp_input: INCAR POTCAR POTCAR.Al POTCAR.Mg
代码
文本
当前目录中共有六个文件(夹)，其中：
MgAl.json: dpgen的主要控制文件，包含了训练、采样和dft计算的参数machine.json: 计算资源的控制文件run.sh: 提交任务脚本calypso_input: 目录中包含了calypso的输入参数文件input.datdata: 初始训练集，由500个CALYPSO产生的随机结构的单点组成vasp_input: 第一性原理软件VASP的输入文件
代码
文本
2.1 dpgen输入文件：machine.json
machine.json为计算资源配置文件，参数的详细说明请参考dpdispatcher的文档
后续使用时需注意：
command为执行的命令，已经帮大家写好，需要注意，在model devi部分除command外，还需要提供calypso的绝对路径calypso_path和deepmd的python路径deepmdkit_python(使用DP对结构优化并计算model deviation)email/password为Bohrium平台的账号和密码program_id为相应的项目号scass_type为调用的机器型号image_name为调用的镜像名group_size为每个计算节点运行的任务数
代码
文本
[27]
cat machine.json
{
"api_version": "1.0",
"deepmd_version": "2.2.2",
"train" :[
{
"command": "/opt/deepmd/deepmd-kit-222/bin/dp",
"machine": {
"batch_type": "Bohrium",
"context_type": "BohriumContext",
"local_root" : "./",
"remote_profile":{
"email": "",
"password": "",
"program_id": 00000,
"keep_backup":true,
"input_data":{
"log_file": "00*/train.log",
"grouped":true,
"job_name": "dpgen_train_job",
"disk_size": 100,
"scass_type":"c8_m32_1 * NVIDIA V100",
"checkpoint_files":["00*/checkpoint","00*/model.ckpt*"],
"checkpoint_time":30,
"platform": "ali",
"job_type": "container",
"image_address":"registry.dp.tech/dptech/prod-11265/dp222-dpgen-calypso:v0.1",
"on_demand":0
}
}
},
"resources": {
"number_node": 1,
"cpu_per_node": 8,
"gpu_per_node": 1,
"queue_name": "V100_8_32",
"group_size": 1,
"custom_flags": [],
"strategy": {"if_cuda_multi_devices": true},
"para_deg": 3,
"source_list": []
}
}],
"model_devi":[
{
"_comment_1": "calypso_path is the local calypso path",
"calypso_path":"/usr/local/bin/",
"_comment_2": "deepmdkit_python is the remote machine python path",
"deepmdkit_python":"/opt/deepmd/deepmd-kit-222/bin/python",
"command": "",
"machine": {
"batch_type": "Bohrium",
"context_type": "BohriumContext",
"local_root" : "./",
"remote_profile":{
"email": "",
"password": "",
"program_id": 0000,
"keep_backup":true,
"input_data":{
"log_file": "*/model_devi.log",
"grouped":true,
"job_name": "dpgen_model_devi_job",
"disk_size": 200,
"scass_type":"c8_m32_1 * NVIDIA V100",
"platform": "ali",
"job_type": "container",
"image_address":"registry.dp.tech/dptech/prod-11265/dp222-dpgen-calypso:v0.1",
"on_demand":0
}
}
},
"resources": {
"number_node": 1,
"cpu_per_node": 8,
"gpu_per_node": 1,
"queue_name": "V100_8_32",
"group_size": 50,
"source_list": []
}
}],
"fp":[
{
"command": "ulimit -s unlimited; mpirun -np 16 vasp_std",
"machine": {
"batch_type": "Bohrium",
"context_type": "BohriumContext",
"local_root" : "./",
"remote_profile":{
"email": "",
"password": "",
"program_id": 0000,
"input_data":{
"api_version":2,
"log_file": "task*/fp.log",
"grouped":true,
"job_name": "dpgen_fp_job",
"disk_size": 100,
"scass_type":"c16_m32_cpu",
"platform": "ali",
"job_type":"container",
"image_address":"registry.dp.tech/dptech/vasp:5.4.4-calypso",
"on_demand": 0
}
}
},
"resources": {
"number_node": 1,
"cpu_per_node": 8,
"gpu_per_node": 0,
"queue_name": "CPU",
"group_size": 10,
"source_list": ["/opt/intel/oneapi/setvars.sh"]
}
}
]
}
代码
文本
[29]
%%writefile machine.json
{
"api_version": "1.0",
"deepmd_version": "2.2.2",
"train" :[
{
"command": "/opt/deepmd/deepmd-kit-222/bin/dp",
"machine": {
"batch_type": "Bohrium",
"context_type": "BohriumContext",
"local_root" : "./",
"remote_profile":{
"email": "wzy@calypso.cn",
"password": "xxxx",
"program_id": xxxxx,
"keep_backup":true,
"input_data":{
"log_file": "00*/train.log",
"grouped":true,
"job_name": "dpgen_train_job",
"disk_size": 100,
"scass_type":"c8_m32_1 * NVIDIA V100",
"checkpoint_files":["00*/checkpoint","00*/model.ckpt*"],
"checkpoint_time":30,
"platform": "ali",
"job_type": "container",
"image_address":"registry.dp.tech/dptech/prod-11265/dp222-dpgen-calypso:v0.1",
"on_demand":0
}
}
},
"resources": {
"number_node": 1,
"cpu_per_node": 8,
"gpu_per_node": 1,
"queue_name": "V100_8_32",
"group_size": 1,
"custom_flags": [],
"strategy": {"if_cuda_multi_devices": true},
"para_deg": 3,
"source_list": []
}
}],
"model_devi":[
{
"_comment_1": "calypso_path is the local calypso path",
"calypso_path":"/usr/local/bin/",
"_comment_2": "deepmdkit_python is the remote machine python path",
"deepmdkit_python":"/opt/deepmd/deepmd-kit-222/bin/python",
"command": "",
"machine": {
"batch_type": "Bohrium",
"context_type": "BohriumContext",
"local_root" : "./",
"remote_profile":{
"email": "wzy@calypso.cn",
"password": "zhenyu1234!",
"program_id": 12361,
"keep_backup":true,
"input_data":{
"log_file": "*/model_devi.log",
"grouped":true,
"job_name": "dpgen_model_devi_job",
"disk_size": 200,
"scass_type":"c8_m32_1 * NVIDIA V100",
"platform": "ali",
"job_type": "container",
"image_address":"registry.dp.tech/dptech/prod-11265/dp222-dpgen-calypso:v0.1",
"on_demand":0
}
}
},
"resources": {
"number_node": 1,
"cpu_per_node": 8,
"gpu_per_node": 1,
"queue_name": "V100_8_32",
"group_size": 50,
"source_list": []
}
}],
"fp":[
{
"command": "ulimit -s unlimited; mpirun -np 16 vasp_std",
"machine": {
"batch_type": "Bohrium",
"context_type": "BohriumContext",
"local_root" : "./",
"remote_profile":{
"email": "wzy@calypso.cn",
"password": "zhenyu1234!",
"program_id": 12361,
"input_data":{
"api_version":2,
"log_file": "task*/fp.log",
"grouped":true,
"job_name": "dpgen_fp_job",
"disk_size": 100,
"scass_type":"c16_m32_cpu",
"platform": "ali",
"job_type":"container",
"image_address":"registry.dp.tech/dptech/vasp:5.4.4-calypso",
"on_demand": 0
}
}
},
"resources": {
"number_node": 1,
"cpu_per_node": 8,
"gpu_per_node": 0,
"queue_name": "CPU",
"group_size": 10,
"source_list": ["/opt/intel/oneapi/setvars.sh"]
}
}
]
}
Overwriting machine.json
代码
文本
2.2 dpgen输入文件：MgAl.json
”MgAljson“为dpgen run的主要参数设置文件。
基于CALYPSO构建势函数需要注意以下几个参数：
model_devi_engine: str, 默认为lmp，需要替换为calypso.calypso_input_path: str, CALYPSO的输入参数文件绝对路径，当提供该路径后，model_devi_jobs不起作用model_devi_max_iter: int, 最大迭代次数，超过最大迭代次数后程序会停止，当calypso_input_path存在时起作用model_devi_jobs: List[dict], 类似lmp控制采样的形式，可控制参数包括：model_devi_jobs["times"], List[int], 列表中的数字代表iteration的index，该列表中的index会使用相同的calypso输入文件
model_devi_jobs["NameOfAtoms"]，List[str], 目标结构的元素列表
model_devi_jobs["NumberOfAtoms"]，List[int]，一倍分子式中每种元素的原子数
model_devi_jobs["NumberOfFormula"]，List[int]，产生结构的分子式范围
model_devi_jobs["Volume"]，List[float]，一倍分子式的体积，A^3
model_devi_jobs["DistanceOfIon"]，List[List[float]]，不同元素之间允许的最短距离 Angstrom
model_devi_jobs["PsoRatio"]，float，使用演化算法产生结构的比例
model_devi_jobs["PopSize"]，int，每一代产生结构的数目
model_devi_jobs["MaxStep"]，int，CALYPSO运行需要的最大代数
model_devi_jobs["ICode"]，int, 局域优化软件的选择，15为使用DP优化结构
model_devi_jobs["Split"]，str，Split模式表示将产生结构与优化结构分离，dpgen中必须为"T"
model_devi_jobs["PSTRESS"]，List[float]，压力，单位KBar
model_devi_jobs["fmax"]，float，DP结构优化收敛标准，结构中原子所受力的最大值小于fmax时优化停止
model_devi_jobs["VSC"]，str，是否进行变组分结构产生
model_devi_jobs["MaxNumAtom"]，int，变组分产生的结构最大允许原子数
model_devi_jobs["CtrlRange"]，List[List[int]]，变组分结构预测时，每种元素变化的范围[[1,2].[1,3]]
建议大家可以直接使用
calypso_input_path直接指定input.dat来运行
代码
文本
双击即可修改
代码
文本
[35]
!cat MgAl.json
{
"model_devi_engine":"calypso",
"calypso_input_path":"./calypso_input",
"model_devi_max_iter": 2,
"vsc":false,
"ratio_failed":0.2,
"type_map": [
"Mg",
"Al"
],
"mass_map": [
27,
64
],
"init_data_prefix":"./data/trainingset",
"init_data_sys": [
"Mg1Al1",
"Mg1Al11",
"Mg1Al13",
"Mg1Al14",
"Mg1Al16",
"Mg1Al17",
"Mg1Al18",
"Mg1Al19",
"Mg1Al2",
"Mg1Al20",
"Mg1Al3",
"Mg1Al4",
"Mg1Al6",
"Mg1Al7",
"Mg1Al8",
"Mg1Al9",
"Mg2Al10",
"Mg2Al12",
"Mg2Al13",
"Mg2Al15",
"Mg2Al16",
"Mg2Al17",
"Mg2Al18",
"Mg2Al2",
"Mg2Al20",
"Mg2Al3",
"Mg2Al4",
"Mg2Al5",
"Mg2Al6",
"Mg2Al7",
"Mg2Al8",
"Mg2Al9",
"Mg3Al1",
"Mg3Al10",
"Mg3Al11",
"Mg3Al12",
"Mg3Al13",
"Mg3Al14",
"Mg3Al15",
"Mg3Al16",
"Mg3Al17",
"Mg3Al2",
"Mg3Al20",
"Mg3Al3",
"Mg3Al4",
"Mg3Al5",
"Mg3Al6",
"Mg3Al7",
"Mg3Al8",
"Mg3Al9",
"Mg4Al11",
"Mg4Al13",
"Mg4Al14",
"Mg4Al15",
"Mg4Al16",
"Mg4Al17",
"Mg4Al18",
"Mg4Al19",
"Mg4Al2",
"Mg4Al20",
"Mg4Al3",
"Mg4Al6",
"Mg5Al10",
"Mg5Al11",
"Mg5Al12",
"Mg5Al16",
"Mg5Al18",
"Mg5Al3",
"Mg5Al4",
"Mg5Al5",
"Mg5Al7",
"Mg5Al8",
"Mg6Al1",
"Mg6Al11",
"Mg6Al12",
"Mg6Al13",
"Mg6Al14",
"Mg6Al17",
"Mg6Al18",
"Mg6Al19",
"Mg6Al2",
"Mg6Al20",
"Mg6Al3",
"Mg6Al6",
"Mg6Al7",
"Mg6Al8",
"Mg6Al9",
"Mg7Al10",
"Mg7Al12",
"Mg7Al14",
"Mg7Al15",
"Mg7Al19",
"Mg7Al2",
"Mg7Al20",
"Mg7Al3",
"Mg7Al4",
"Mg7Al5",
"Mg7Al6",
"Mg8Al11",
"Mg8Al12",
"Mg8Al13",
"Mg8Al15",
"Mg8Al16",
"Mg8Al17",
"Mg8Al18",
"Mg8Al19",
"Mg8Al2",
"Mg8Al3",
"Mg8Al4",
"Mg8Al5",
"Mg8Al6",
"Mg8Al7",
"Mg8Al8",
"Mg8Al9",
"Mg9Al12",
"Mg9Al14",
"Mg9Al15",
"Mg9Al17",
"Mg9Al18",
"Mg9Al19",
"Mg9Al2",
"Mg9Al20",
"Mg9Al4",
"Mg9Al5",
"Mg9Al6",
"Mg9Al7",
"Mg9Al8",
"Mg9Al9",
"Mg10Al1",
"Mg10Al10",
"Mg10Al11",
"Mg10Al12",
"Mg10Al13",
"Mg10Al14",
"Mg10Al15",
"Mg10Al16",
"Mg10Al2",
"Mg10Al20",
"Mg10Al3",
"Mg10Al4",
"Mg10Al6",
"Mg10Al7",
"Mg10Al8",
"Mg10Al9",
"Mg11Al1",
"Mg11Al10",
"Mg11Al12",
"Mg11Al13",
"Mg11Al14",
"Mg11Al15",
"Mg11Al16",
"Mg11Al17",
"Mg11Al19",
"Mg11Al2",
"Mg11Al20",
"Mg11Al3",
"Mg11Al4",
"Mg11Al5",
"Mg11Al6",
"Mg11Al7",
"Mg11Al8",
"Mg11Al9",
"Mg12Al1",
"Mg12Al10",
"Mg12Al11",
"Mg12Al12",
"Mg12Al13",
"Mg12Al14",
"Mg12Al15",
"Mg12Al16",
"Mg12Al17",
"Mg12Al19",
"Mg12Al20",
"Mg12Al3",
"Mg12Al5",
"Mg12Al6",
"Mg12Al8",
"Mg13Al11",
"Mg13Al12",
"Mg13Al13",
"Mg13Al14",
"Mg13Al15",
"Mg13Al16",
"Mg13Al17",
"Mg13Al18",
"Mg13Al19",
"Mg13Al2",
"Mg13Al20",
"Mg13Al3",
"Mg13Al5",
"Mg13Al6",
"Mg13Al7",
"Mg13Al8",
"Mg13Al9",
"Mg14Al12",
"Mg14Al13",
"Mg14Al16",
"Mg14Al17",
"Mg14Al18",
"Mg14Al19",
"Mg14Al3",
"Mg14Al4",
"Mg14Al5",
"Mg14Al8",
"Mg14Al9",
"Mg15Al10",
"Mg15Al11",
"Mg15Al16",
"Mg15Al17",
"Mg15Al18",
"Mg15Al19",
"Mg15Al2",
"Mg15Al20",
"Mg15Al3",
"Mg15Al4",
"Mg15Al5",
"Mg15Al7",
"Mg15Al9",
"Mg16Al1",
"Mg16Al12",
"Mg16Al13",
"Mg16Al14",
"Mg16Al15",
"Mg16Al16",
"Mg16Al17",
"Mg16Al19",
"Mg16Al2",
"Mg16Al3",
"Mg16Al4",
"Mg16Al6",
"Mg16Al7",
"Mg17Al1",
"Mg17Al11",
"Mg17Al12",
"Mg17Al13",
"Mg17Al14",
"Mg17Al15",
"Mg17Al16",
"Mg17Al17",
"Mg17Al18",
"Mg17Al19",
"Mg17Al2",
"Mg17Al20",
"Mg17Al3",
"Mg17Al4",
"Mg17Al5",
"Mg17Al6",
"Mg17Al7",
"Mg17Al8",
"Mg17Al9",
"Mg18Al11",
"Mg18Al12",
"Mg18Al13",
"Mg18Al14",
"Mg18Al15",
"Mg18Al16",
"Mg18Al17",
"Mg18Al19",
"Mg18Al2",
"Mg18Al20",
"Mg18Al3",
"Mg18Al5",
"Mg18Al6",
"Mg18Al7",
"Mg18Al8",
"Mg18Al9",
"Mg19Al1",
"Mg19Al10",
"Mg19Al11",
"Mg19Al15",
"Mg19Al16",
"Mg19Al17",
"Mg19Al18",
"Mg19Al2",
"Mg19Al20",
"Mg19Al4",
"Mg19Al5",
"Mg19Al6",
"Mg19Al8",
"Mg19Al9",
"Mg20Al11",
"Mg20Al12",
"Mg20Al17",
"Mg20Al4",
"Mg20Al5",
"Mg20Al6",
"Mg20Al7",
"Mg20Al9"
],
"_comment": " that's all ",
"sys_configs":[""],
"training_init_model": false,
"numb_models": 4,
"default_training_param": {
"model": {
"descriptor": {
"type": "se_e2_a",
"sel": [
500,
1500
],
"rcut_smth": 2.0,
"rcut": 6.0,
"neuron": [
25,
50,
100
],
"resnet_dt": false,
"axis_neuron": 12,
"type_one_side": true,
"seed": 1801819940,
"_activation_function": "tanh"
},
"fitting_net": {
"neuron": [
240,
240,
240
],
"resnet_dt": true,
"seed": 2375417769
},
"type_map": [
"Mg",
"Al"
]
},
"learning_rate": {
"type": "exp",
"start_lr": 0.001,
"decay_steps": 2000
},
"loss": {
"start_pref_e": 0.02,
"limit_pref_e": 1,
"start_pref_f": 1000,
"limit_pref_f": 1,
"start_pref_v": 0,
"limit_pref_v": 0
},
"training": {
"numb_steps": 2000,
"seed": 3982377700,
"disp_file": "lcurve.out",
"disp_freq": 20,
"numb_test": 4,
"save_freq": 2000,
"save_ckpt": "model.ckpt",
"disp_training": true,
"time_training": true,
"profiling": false,
"profiling_file": "timeline.json",
"_set_prefix": "set"
}
},
"model_devi_dt": 0.002,
"model_devi_skip": 0,
"model_devi_f_trust_lo": 0.1,
"model_devi_f_trust_hi": 0.3,
"model_devi_e_trust_lo": 10000000000.0,
"model_devi_e_trust_hi": 10000000000.0,
"model_devi_clean_traj": true,
"model_devi_jobs": [
],
"fp_style": "vasp",
"shuffle_poscar": false,
"fp_task_max": 10,
"fp_task_min": 1,
"fp_pp_path": "./vasp_input",
"fp_pp_files": [
"POTCAR.Mg",
"POTCAR.Al"
],
"fp_incar":"./vasp_input/INCAR"
}
代码
文本
2.3 calypso输入文件：input.dat
input.dat文件是CALYPSO运行时的主要控制文件，结构的产生方式需要在文件中说明，详细参数解释可以参考：https://iccms-calypso.github.io/CALYPSO-Fortran/
代码
文本
[38]
!cat ./calypso_input/input.dat
################################ The Basic Parameters of CALYPSO ################################ # A string of one or several words contain a descriptive name of the system (max. 40 characters). SystemName =MgAl # Number of different atomic species in the simulation. NumberOfSpecies = 2 # Element symbols of the different chemical species. NameOfAtoms =Mg Al # Atomic Number of each chemical species. AtomicNumber = 12 13 # Number of atoms for each chemical species in one formula unit. NumberOfAtoms = 1 1 # The range of formula unit per cell in your simulation. NumberOfFormula = 1 1 # The volume per formula unit. Unit is in angstrom^3. # Volume=20 # Minimal distance between atoms of each chemical species. Unit is in angstrom. @DistanceOfIon 1.48 1.44 1.44 1.41 @End # It determines which algorithm should be adopted in the simulation. Ialgo = 2 # Ialgo = 1 for Global PSO # Ialgo = 2 for Local PSO (default value) # The proportion of the structures generated by PSO. PsoRatio = 0.0 # The popu2ation size. Normally, it has a larger number for larger systems. PopSize = 50 #It determines which method should be adopted in generation the random structure. GenType= 1 # 1 under symmetric constraints # 2 grid method for large system # 3 and 4 core grow method # 0 combination of all method # If GenType=3 or 4, it determined the small unit to grow the whole structure # It determines which local optimization method should be interfaced in the simulation. ICode= 15 # ICode= 1 interfaced with VASP # ICode= 2 interfaced with SIESTA # ICode= 3 interfaced with GULP # The number of lbest for local PSO NumberOfLbest=4 # The Number of local optimization for each structure. NumberOfLocalOptim= 3 # The command to perform local optimiztion calculation (e.g., VASP, SIESTA) on your computer. Command = sh submit.sh MaxTime = 9000 # The Max step for iteration MaxStep = 5 # If True, a previous calculation will be continued. PickUp=F # At which step will the previous calculation be picked up. PickStep = 1 # If True, the local optimizations performed by parallel Parallel= F # The number node for parallel NumberOfParallel=4 #LMC = F Split = T ##### The Parameters For Variational Stoichiometry ############## ## If True, Variational Stoichiometry structure prediction is performed VSC=T ## VSCEnergy= 9.68045866 1.00102117 ## The Max Number of Atoms in unit cell MaxNumAtom=40 ## The Variation Range for each type atom @CtrlRange 1 20 1 20 @end ###################End Parameters for VSC ##########################
代码
文本
代码
文本
[39]
!bash run.sh #完成计算大概需要1-2个小时
代码
文本
[40]
ls
MgAl.json data/ iter.000000/ out vasp_input/ calypso_input/ dpgen.log machine.json run.sh
代码
文本
[42]
cat dpgen.log
2024-03-29 10:27:54,232 - INFO : start running 2024-03-29 10:27:54,246 - INFO : =============================iter.000000============================== 2024-03-29 10:27:54,246 - INFO : -------------------------iter.000000 task 00-------------------------- 2024-03-29 10:28:11,350 - INFO : -------------------------iter.000000 task 01--------------------------
代码
文本
3. 基于DP加速的CALYPSO结构预测
在DPGEN+CALYPSO迭代收敛后，我们会针对迭代后的模型长训，利用长训后的模型进行DP加速的CALYPSO结构预测。
基于DP加速的CALYPSO结构预测分为两种方式：
- 单节点模式：正常的CALYPSO运行模式，此模式下结构产生和优化是串行的，可能速度会较多节点split模式慢一些，适合(cpu/gpu)资源不充足的情况。
- 多节点split模式：在开启split模式后，CALYPSO的产生结构和优化结构部分分离，在资源充足的情况下，我们可以每个结构单独占据一个节点，大大加速了结构预测流程，由于bohrium平台资源充足，因此也更推荐使用这种方式。
以上两种方式同样适用于自有集群，本节例子的目的是为了给大家提供在自有集群运行任务的模板，不在实际运行。
关于如何更方便的在borhium平台使用DP加速的CALYPSO结构预测可在notebook1和notebook2中进一步学习。
代码
文本
[5]
cp -r /bohr/dp-calypso-example-ivbp/v1/dp-calypso-example/ /personal
代码
文本
[6]
cd /personal/dp-calypso-example
/personal/dp-calypso-example
代码
文本
[7]
ls *
nosplit_calypso_dp_example: calypso_check_outcar.py* graph.pb* run.sh* calypso_run_opt.py* input.dat* submit.sh* split_calypso_dp_example: calypso_check_outcar.py* evo.dispatcher.py* input.dat* resources.json* calypso_run_opt.py* graph.pb* machine.json* run.sh*
代码
文本
[10]
cat nosplit_calypso_dp_example/run.sh
#!/bin/bash calypso.x > caly.log 2>&1
代码
文本
[11]
cat nosplit_calypso_dp_example/submit.sh
#!/bin/bash /opt/deepmd/deepmd-kit-222/bin/python calypso_run_opt.py || /opt/deepmd/deepmd-kit-222/bin/python calypso_check_outcar.py > log 2>&1
代码
文本
在单节点模式中，需要以下文件：
- calypso_run_opt.py: 调用ASE结合DP进行结构优化的控制脚本
- calypso_check_outcar.py: 处理优化失败情况的脚本
- graph.pb: 训练好的模型
- input.dat: calypso的输入参数
- submit.sh: 优化结构的命令，内容一般为：python calypso_run_opt.py || python calypso_check_outcar.py
- run.sh: 任务提交脚本，对于本地机器，可直接运行；对于任务调度系统，需要按照队列系统对应的脚本进行修改，其中运行的命令不变：calypso.x > caly.log
代码
文本
运行命令为：bash run.sh
代码
文本
代码
文本
[12]
ls *
nosplit_calypso_dp_example: calypso_check_outcar.py* graph.pb* run.sh* calypso_run_opt.py* input.dat* submit.sh* split_calypso_dp_example: calypso_check_outcar.py* evo.dispatcher.py* input.dat* resources.json* calypso_run_opt.py* graph.pb* machine.json* run.sh*
代码
文本
在多节点split模式中，需要以下文件：
- calypso_run_opt.py: 调用ASE结合DP进行结构优化的控制脚本
- calypso_check_outcar.py: 处理优化失败情况的脚本
- graph.pb: 训练好的模型
- input.dat: calypso的输入参数
- run.sh: 任务提交脚本，对于本地机器，可直接运行；对于任务调度系统，需要修改machine.json和resoureces.json定义需要的机器和资源信息，该脚本可以提交到计算节点（前提是计算节点可以继续向队列系统提交任务），也可以在主节点运行。
- machine.json: 控制使用的机器信息，详细参数解释见dpdispatcher的文档
- resources.json: 具体使用的资源信息，详细参数解释见dpdispatcher的文档
- evo.dispatcher.py: 流程控制脚本，需要对里面的参数做一些修改
代码
文本
[15]
cat split_calypso_dp_example/evo.dispatcher.py
#/usr/bin/env python
from dpdispatcher import Machine, Resources, Task, Submission
import os, sys, shutil, glob
from pathlib import Path
# os.environ["CUDA_VISIBLE_DEVICES"]="0"
MaxStep = 50
PopSize = 200
# N_INCAR = 1
# dp
python = "/opt/deepmd/deepmd-kit-222/bin/python"
forward_files = ['calypso_run_opt.py', 'calypso_check_outcar.py', 'input.dat', 'POSCAR']
backward_files = ['OUTCAR', 'CONTCAR', 'traj.traj', 'model_devi.out']
forward_common_files=['graph.pb']
command = f'{python} calypso_run_opt.py || {python} calypso_check_outcar.py > model_devi.out 2>&1'
machine = Machine.load_from_json('./machine.json')
resources = Resources.load_from_json('./resources.json')
os.system("./calypso.x")
for step in range(1, MaxStep+1):
task_list = []
for pop in range(1, PopSize + 1):
task_dir = "data/step%03d.pop%03d"% (step,pop)
Path(task_dir).mkdir(parents=True, exist_ok=True)
shutil.copyfile("POSCAR_%d"%pop, os.path.join(task_dir, "POSCAR"))
shutil.copyfile("POSCAR_%d"%pop, os.path.join(task_dir, "POSCAR.ori"))
# dp
for file_name in forward_files[:3]:
shutil.copyfile(file_name , os.path.join(task_dir, file_name))
task = Task(
command=command,
task_work_path=task_dir,
forward_files=forward_files,
backward_files=backward_files,
)
task_list.append(task)
submission = Submission(
work_base = os.getcwd(),
machine= machine,
resources =resources,
task_list = task_list,
forward_common_files=forward_common_files
)
submission.run_submission()
for pop in range(1, PopSize + 1):
task_dir = "data/step%03d.pop%03d"% (step,pop)
shutil.copyfile(os.path.join(task_dir, "CONTCAR"), "CONTCAR_%d"%pop )
shutil.copyfile(os.path.join(task_dir, "OUTCAR"), "OUTCAR_%d"%pop)
os.system("./calypso.x")
代码
文本
在evo.dispatcher.py脚本中，需要修改MaxStep， PopSize，python变量:
MaxStep和PopSize需要和input.dat中的参数保持一致python需要指定deepmd-kit的python路径
代码
文本
[14]
cat split_calypso_dp_example/run.sh
#!/bin/bash #SBATCH --job-name=split-mode #SBATCH --mem-per-cpu=2gb #SBATCH --nodes=1 #SBATCH --ntasks=2 #SBATCH --ntasks-per-node=2 #SBATCH --cpus-per-task=1 #SBATCH --output=%j.log #SBATCH --partition=hebhcnormal01 # load the environment module purge # module load apps/PyTorch/chatglm2_py38/pytorch1.13-dtk23.04-py38 source /public/software/profile.d/compiler_intel-compiler-2017.5.239.sh source /public/software/profile.d/mpi_intelmpi-2018.4.274.sh # ./calypso.x > caly.log 2>&1 /public/share/acsp3lax5q/miniconda3/envs/workshop2023/bin/python evo.dispatcher.py > caly.log 2>&1
代码
文本
可通过提交run.sh脚本运行，也可以通过直接在命令行运行evo.dispatcher.py脚本运行。
代码
文本
已赞2
本文被以下合集收录
MD
bohr61096f
更新于 2024-08-27
47 篇0 人关注
DPMD
RamonMi
更新于 2024-07-29
14 篇0 人关注
推荐阅读
公开
快速上手使用ABACUS + DP-GEN | init_bulk/runAISI 胡钰铂
更新于 2024-06-18
2 转存文件1 评论
公开
手搓RHF其实不难只是慢（1）：Gaussian型轨道和计算重叠积分hanyanbo
发布于 2023-09-25
3 赞3 转存文件