DP-GEN simplify
本文介绍如何在Bohrium平台进行DP-GEN simplify。
推荐
「Bohrium 帮助文档｜DP-GEN simplify」是 Notebook 版帮助文档，可直接运行，请点击查看。
步骤1 开启节点
如何开启节点参照"DPGEN教程"。
步骤2 准备输入文件
使用DP-GEN simplify需要准备初始数据、参数文件与计算文件。以使用钙钛矿为例：
(base) ➜ example tree
.
├── data
│ ├── init.000
│ │ ├── set.000
│ │ │ ├── box.npy
│ │ │ ├── coord.npy
│ │ │ ├── energy.npy
│ │ │ ├── force.npy
│ │ │ └── virial.npy
│ │ ├── type_map.raw
│ │ └── type.raw
│ ├── sys.000
│ │ ├── set.000
│ │ │ ├── box.npy
│ │ │ ├── coord.npy
│ │ │ ├── energy.npy
│ │ │ ├── force.npy
│ │ │ └── virial.npy
│ │ ├── type_map.raw
│ │ └── type.raw
│ └── sys.001
│ ├── set.000
│ │ ├── box.npy
│ │ ├── coord.npy
│ │ ├── energy.npy
│ │ ├── force.npy
│ │ └── virial.npy
│ ├── type_map.raw
│ └── type.raw
├── simplify_example
│ ├── INCAR
│ ├── machine_test_simplify.json
│ └── simplify.json
└── vasp_inputs
├── POTCAR_C
├── POTCAR_H
├── POTCAR_I
├── POTCAR_N
└── POTCAR_Pb_d
示例simplify.json文件
{
"type_map": ["I","Pb","C","N","H"],
"mass_map": [126.90447, 207.2, 12.0108, 14.0067, 1.00795],
"pick_data": "../data",
"init_data_prefix": "",
"init_data_sys": [],
"sys_configs": [null],
"sys_batch_size": [1,1,1],
"_comment": " 00.train ",
"numb_models": 4,
"model_devi_activation_func":[["tanh","tanh"],["tanh","gelu"],["gelu","gelu"],["gelu","tanh"]],
"train_param": "input.json",
"default_training_param": {
"model": {
"type_map": ["I","Pb","C","N","H"],
"descriptor": {
"type": "se_e2_a",
"sel":
[
20,
8,
8,
8,
48
],
"rcut_smth": 2.00,
"rcut": 6.00,
"neuron": [25, 50, 100],
"resnet_dt": false,
"type_one_side": true,
"trainable": true,
"axis_neuron": 12,
"seed": 0
},
"fitting_net": {
"neuron": [240, 240, 240],
"resnet_dt": true,
"trainable": [true, true, true, true],
"seed": 0
}
},
"loss": {
"start_pref_e": 0.02,
"limit_pref_e": 2,
"start_pref_f": 1000,
"limit_pref_f": 2,
"start_pref_v": 0.01,
"limit_pref_v": 1
},
"learning_rate": {
"type": "exp",
"start_lr": 0.001,
"decay_steps": 100000,
"decay_rate": 0.95
},
"training": {
"set_prefix": "set",
"stop_batch": 1000000,
"batch_size": "auto",
"seed": 1,
"_comment": "frequencies counted in batch",
"disp_file": "lcurve.out",
"disp_freq": 100000,
"numb_test": 4,
"save_freq": 100000,
"save_ckpt": "model.ckpt",
"disp_training": true,
"time_training": true,
"profiling": false,
"profiling_file": "timeline.json"
}
},
"_comment": "02.fp",
"fp_style": "vasp",
"fp_skip_bad_box": "length_ratio:5;height_ratio:5",
"fp_accurate_threshold": 0.95,
"fp_accurate_soft_threshold": 0.90,
"shuffle_poscar": false,
"fp_task_max": 20,
"fp_task_min": 5,
"ratio_failed": 0.30,
"fp_pp_path": "../vasp_inputs/",
"fp_pp_files": ["POTCAR_I","POTCAR_Pb_d","POTCAR_C","POTCAR_N","POTCAR_H"],
"fp_incar": "INCAR",
"use_clusters": false,
"labeled": false,
"init_pick_number":20,
"iter_pick_number":20,
"model_devi_f_trust_lo":0.30,
"model_devi_f_trust_hi":100.00,
"cvasp": false
}
示例machine.json文件
{
"api_version": "1.0",
"train" :[
{
"command": "dp",
"machine": {
"batch_type": "Lebesgue",
"context_type": "LebesgueContext",
"local_root" : "./",
"remote_profile":{
"email": "",
"password": "",
"project_id": 0000,
"input_data":{
"api_version":2,
"job_type": "indicate",
"log_file": "00*/lcurve.out",
"grouped": true,
"job_name": "simplify_MAPbI3-scan_train_job",
"disk_size": 100,
"scass_type":"c8_m32_1 * NVIDIA V100",
"platform": "ali",
"checkpoint_files": ["00*/model.ckpt*", "00*/checkpoint"],
"checkpoint_time": 15,
"job_type": "container",
"image_name": "registry.dp.tech/dptech/deepmd-kit:2.1.5-cuda11.6" ,
"on_demand":0
}
}
},
"resources": {
"number_node": 15,
"cpu_per_node": 4,
"gpu_per_node": 1,
"queue_name": "V100_8_32",
"group_size": 1
}
}],
"model_devi":
[{
"command": "dp",
"machine": {
"batch_type": "Lebesgue",
"context_type": "LebesgueContext",
"local_root" : "./",
"remote_profile":{
"email": "",
"password": "",
"project_id": 0000,
"input_data":{
"api_version":2,
"job_type": "indicate",
"log_file": "cubic-*/*/md.log",
"grouped": true,
"job_name": "MAPbI3-scan_md_job",
"disk_size": 100,
"scass_type":"c8_m32_1 * NVIDIA V100",
"platform": "ali",
"checkpoint_files": "sync_files",
"checkpoint_time": 15,
"job_type": "container",
"image_name": "registry.dp.tech/dptech/deepmd-kit:2.1.5-cuda11.6" ,
"on_demand":0
}
}
},
"resources": {
"number_node": 15,
"cpu_per_node": 4,
"gpu_per_node": 1,
"queue_name": "V100_8_32",
"group_size": 1
}
}],
"fp":
[{
"command": "ulimit -m unlimited; ulimit -s unlimited; mpirun -n 32 vasp_std",
"machine": {
"batch_type": "Lebesgue",
"context_type": "LebesgueContext",
"local_root" : "./",
"remote_profile":{
"email": "",
"password": "",
"project_id": 0000,
"input_data":{
"api_version":2,
"job_type": "indicate",
"log_file": "**/fp.log",
"grouped": true,
"job_name": "simplify_MAPbI3-scan_fp_job",
"disk_size": 100,
"scass_type":"c32_m256_cpu",
"platform": "ali",
"checkpoint_files": "sync_files",
"checkpoint_time": 15,
"job_type": "container",
"image_name":"您需提供授权凭证到 bohrium@dp.tech 邮箱获取VASP镜像名称",
"on_demand":0
}
}
},
"resources": {
"number_node": 15,
"cpu_per_node": 32,
"gpu_per_node": 0,
"queue_name": "CPU",
"group_size": 1,
"source_list": ["/opt/intel/oneapi/setvars.sh"]
}
}
]
}
注意：所有 "project_id": 后的 0000 均需替换为您自己的项目ID，可在“项目管理”页查看。
步骤3 提交任务
在/simplify_example目录下执行：
nohup dpgen simplify simplify.json machine.json 1>log 2>err&
监测任务状态与运行结果获取参考 DP-GEN 文档。