run.py

#!/usr/bin/env python3

"""init: data
iter:
        00.train
        01.model_devi
        02.vasp
        03.data.
"""

import argparse
import copy
import glob
import itertools
import json
import logging
import logging.handlers
import os
import queue
import random
import re
import shlex
import shutil
import sys
import warnings
from collections import Counter
from collections.abc import Iterable
from pathlib import Path
from typing import Optional

import dpdata
import numpy as np
import scipy.constants as pc
from numpy.linalg import norm
from packaging.version import Version

from dpgen import ROOT_PATH, SHORT_CMD, dlog
from dpgen.auto_test.lib.vasp import make_kspacing_kpoints
from dpgen.dispatcher.Dispatcher import make_submission
from dpgen.generator.lib.abacus_scf import (
    get_abacus_input_parameters,
    get_abacus_STRU,
    make_abacus_scf_input,
    make_abacus_scf_kpt,
    make_abacus_scf_stru,
)
from dpgen.generator.lib.cp2k import (
    make_cp2k_input,
    make_cp2k_input_from_external,
    make_cp2k_xyz,
)
from dpgen.generator.lib.ele_temp import NBandsEsti
from dpgen.generator.lib.gaussian import make_gaussian_input, take_cluster
from dpgen.generator.lib.lammps import (
    get_all_dumped_forces,
    get_dumped_forces,
    make_lammps_input,
)
from dpgen.generator.lib.make_calypso import (
    _make_model_devi_buffet,
    _make_model_devi_native_calypso,
)
from dpgen.generator.lib.parse_calypso import (
    _parse_calypso_dis_mtx,
    _parse_calypso_input,
)

# from dpgen.generator.lib.pwscf import cvt_1frame
from dpgen.generator.lib.pwmat import (
    make_pwmat_input_dict,
    make_pwmat_input_user_dict,
    write_input_dict,
)
from dpgen.generator.lib.pwscf import make_pwscf_input
from dpgen.generator.lib.run_calypso import (
    run_calypso_model_devi,
)
from dpgen.generator.lib.siesta import make_siesta_input
from dpgen.generator.lib.utils import (
    check_api_version,
    create_path,
    log_iter,
    log_task,
    make_iter_name,
    record_iter,
    symlink_user_forward_files,
)
from dpgen.generator.lib.vasp import (
    incar_upper,
    make_vasp_incar_user_dict,
    write_incar_dict,
)
from dpgen.remote.decide_machine import convert_mdata
from dpgen.util import (
    convert_training_data_to_hdf5,
    expand_sys_str,
    load_file,
    normalize,
    sepline,
    set_directory,
    setup_ele_temp,
)

from .arginfo import run_jdata_arginfo

template_name = "template"
train_name = "00.train"
train_task_fmt = "%03d"
train_tmpl_path = os.path.join(template_name, train_name)
default_train_input_file = "input.json"
data_system_fmt = "%03d"
model_devi_name = "01.model_devi"
model_devi_task_fmt = data_system_fmt + ".%06d"
model_devi_conf_fmt = data_system_fmt + ".%04d"
fp_name = "02.fp"
fp_task_fmt = data_system_fmt + ".%06d"
cvasp_file = os.path.join(ROOT_PATH, "generator/lib/cvasp.py")
# for calypso
calypso_run_opt_name = "gen_stru_analy"
calypso_model_devi_name = "model_devi_results"
calypso_run_model_devi_file = os.path.join(
    ROOT_PATH, "generator/lib/calypso_run_model_devi.py"
)
check_outcar_file = os.path.join(ROOT_PATH, "generator/lib/calypso_check_outcar.py")
run_opt_file = os.path.join(ROOT_PATH, "generator/lib/calypso_run_opt.py")


def _get_model_suffix(jdata) -> str:
    """Return the model suffix based on the backend."""
    mlp_engine = jdata.get("mlp_engine", "dp")
    if mlp_engine == "dp":
        suffix_map = {"tensorflow": ".pb", "pytorch": ".pth", "jax": ".savedmodel"}
        backend = jdata.get("train_backend", "tensorflow")
        if backend in suffix_map:
            suffix = suffix_map[backend]
        else:
            raise ValueError(
                f"The backend {backend} is not available. Supported backends are: 'tensorflow', 'pytorch', 'jax'."
            )
        return suffix
    else:
        raise ValueError(f"Unsupported engine: {mlp_engine}")


def get_job_names(jdata):
    jobkeys = []
    for ii in jdata.keys():
        if ii.split("_")[0] == "job":
            jobkeys.append(ii)
    jobkeys.sort()
    return jobkeys


def make_model_devi_task_name(sys_idx, task_idx):
    return "task." + model_devi_task_fmt % (sys_idx, task_idx)


def make_model_devi_conf_name(sys_idx, conf_idx):
    return model_devi_conf_fmt % (sys_idx, conf_idx)


def make_fp_task_name(sys_idx, counter):
    return "task." + fp_task_fmt % (sys_idx, counter)


def get_sys_index(task):
    task.sort()
    system_index = []
    for ii in task:
        system_index.append(os.path.basename(ii).split(".")[1])
    set_tmp = set(system_index)
    system_index = list(set_tmp)
    system_index.sort()
    return system_index


def _check_empty_iter(iter_index, max_v=0):
    fp_path = os.path.join(make_iter_name(iter_index), fp_name)
    # check the number of collected data
    sys_data = glob.glob(os.path.join(fp_path, "data.*"))
    empty_sys = []
    for ii in sys_data:
        nframe = 0
        sys_paths = expand_sys_str(ii)
        for single_sys in sys_paths:
            sys = dpdata.LabeledSystem(os.path.join(single_sys), fmt="deepmd/npy")
            nframe += len(sys)
        empty_sys.append(nframe < max_v)
    return all(empty_sys)


def copy_model(numb_model, prv_iter_index, cur_iter_index, suffix=".pb"):
    cwd = os.getcwd()
    prv_train_path = os.path.join(make_iter_name(prv_iter_index), train_name)
    cur_train_path = os.path.join(make_iter_name(cur_iter_index), train_name)
    prv_train_path = os.path.abspath(prv_train_path)
    cur_train_path = os.path.abspath(cur_train_path)
    create_path(cur_train_path)
    for ii in range(numb_model):
        prv_train_task = os.path.join(prv_train_path, train_task_fmt % ii)
        os.chdir(cur_train_path)
        os.symlink(os.path.relpath(prv_train_task), train_task_fmt % ii)
        os.symlink(
            os.path.join(train_task_fmt % ii, f"frozen_model{suffix}"),
            "graph.%03d%s" % (ii, suffix),  # noqa: UP031
        )
        os.chdir(cwd)
    with open(os.path.join(cur_train_path, "copied"), "w") as fp:
        None


def poscar_natoms(lines):
    numb_atoms = 0
    for ii in lines[6].split():
        numb_atoms += int(ii)
    return numb_atoms


def expand_idx(in_list):
    ret = []
    for ii in in_list:
        if isinstance(ii, int):
            ret.append(ii)
        elif isinstance(ii, str):
            step_str = ii.split(":")
            if len(step_str) > 1:
                step = int(step_str[1])
            else:
                step = 1
            range_str = step_str[0].split("-")
            assert (len(range_str)) == 2
            ret += range(int(range_str[0]), int(range_str[1]), step)
    return ret


def _check_skip_train(job):
    try:
        skip = _get_param_alias(job, ["s_t", "sk_tr", "skip_train", "skip_training"])
    except ValueError:
        skip = False
    return skip


def poscar_to_conf(poscar, conf):
    sys = dpdata.System(poscar, fmt="vasp/poscar")
    sys.to_lammps_lmp(conf)


# def dump_to_poscar(dump, poscar, type_map, fmt = "lammps/dump") :
#    sys = dpdata.System(dump, fmt = fmt, type_map = type_map)
#    sys.to_vasp_poscar(poscar)


def dump_to_deepmd_raw(dump, deepmd_raw, type_map, fmt="gromacs/gro", charge=None):
    system = dpdata.System(dump, fmt=fmt, type_map=type_map)
    system.to_deepmd_raw(deepmd_raw)
    if charge is not None:
        with open(os.path.join(deepmd_raw, "charge"), "w") as f:
            f.write(str(charge))


def make_train(iter_index, jdata, mdata):
    mlp_engine = jdata.get("mlp_engine", "dp")
    if mlp_engine == "dp":
        return make_train_dp(iter_index, jdata, mdata)
    else:
        raise ValueError(f"Unsupported engine: {mlp_engine}")


def make_train_dp(iter_index, jdata, mdata):
    # load json param
    # train_param = jdata['train_param']
    train_input_file = default_train_input_file
    numb_models = jdata["numb_models"]
    init_data_prefix = os.path.abspath(jdata["init_data_prefix"])
    init_data_sys_ = jdata["init_data_sys"]
    fp_task_min = jdata["fp_task_min"]
    model_devi_jobs = jdata["model_devi_jobs"]
    use_ele_temp = jdata.get("use_ele_temp", 0)
    training_iter0_model = jdata.get("training_iter0_model_path", [])
    training_init_model = jdata.get("training_init_model", False)
    training_reuse_iter = jdata.get("training_reuse_iter")
    training_reuse_old_ratio = jdata.get("training_reuse_old_ratio", "auto")

    # if you want to use DP-ZBL potential , you have to give the path of your energy potential file
    if "srtab_file_path" in jdata.keys():
        srtab_file_path = os.path.abspath(jdata.get("srtab_file_path", None))

    if "training_reuse_stop_batch" in jdata.keys():
        training_reuse_stop_batch = jdata["training_reuse_stop_batch"]
    elif "training_reuse_numb_steps" in jdata.keys():
        training_reuse_stop_batch = jdata["training_reuse_numb_steps"]
    else:
        training_reuse_stop_batch = None

    training_reuse_start_lr = jdata.get("training_reuse_start_lr")
    training_reuse_start_pref_e = jdata.get("training_reuse_start_pref_e")
    training_reuse_start_pref_f = jdata.get("training_reuse_start_pref_f")
    model_devi_activation_func = jdata.get("model_devi_activation_func", None)
    training_init_frozen_model = (
        jdata.get("training_init_frozen_model") if iter_index == 0 else None
    )
    training_finetune_model = (
        jdata.get("training_finetune_model") if iter_index == 0 else None
    )

    auto_ratio = False
    if (
        training_reuse_iter is not None
        and isinstance(training_reuse_old_ratio, str)
        and training_reuse_old_ratio.startswith("auto")
    ):
        s = training_reuse_old_ratio.split(":")
        if len(s) == 1:
            new_to_old_ratio = 10.0
        elif len(s) == 2:
            new_to_old_ratio = float(s[1])
        else:
            raise ValueError(
                f"training_reuse_old_ratio is not correct, got {training_reuse_old_ratio}"
            )
        dlog.info(
            "Use automatic training_reuse_old_ratio to make new-to-old ratio close to %d times of the default value.",
            new_to_old_ratio,
        )
        auto_ratio = True
        number_old_frames = 0
        number_new_frames = 0

    suffix = _get_model_suffix(jdata)
    model_devi_engine = jdata.get("model_devi_engine", "lammps")
    if iter_index > 0 and _check_empty_iter(iter_index - 1, fp_task_min):
        log_task("prev data is empty, copy prev model")
        copy_model(numb_models, iter_index - 1, iter_index, suffix)
        return
    elif (
        model_devi_engine != "calypso"
        and iter_index > 0
        and _check_skip_train(model_devi_jobs[iter_index - 1])
    ):
        log_task("skip training at step %d " % (iter_index - 1))  # noqa: UP031
        copy_model(numb_models, iter_index - 1, iter_index, suffix)
        return
    else:
        iter_name = make_iter_name(iter_index)
        work_path = os.path.join(iter_name, train_name)
        copy_flag = os.path.join(work_path, "copied")
        if os.path.isfile(copy_flag):
            os.remove(copy_flag)

    # establish work path
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, train_name)
    create_path(work_path)
    # link init data
    cwd = os.getcwd()
    os.chdir(work_path)
    os.symlink(os.path.abspath(init_data_prefix), "data.init")
    # link iter data
    os.mkdir("data.iters")
    os.chdir("data.iters")
    for ii in range(iter_index):
        os.symlink(
            os.path.relpath(os.path.join(cwd, make_iter_name(ii))), make_iter_name(ii)
        )
    os.chdir(cwd)

    init_data_sys = []
    init_batch_size = []
    if "init_batch_size" in jdata:
        init_batch_size_ = list(jdata["init_batch_size"])
        if len(init_data_sys_) > len(init_batch_size_):
            warnings.warn(
                "The batch sizes are not enough. Assume auto for those not spefified."
            )
            init_batch_size.extend(
                ["auto" for aa in range(len(init_data_sys_) - len(init_batch_size))]
            )
    else:
        init_batch_size_ = ["auto" for aa in range(len(jdata["init_data_sys"]))]
    if "sys_batch_size" in jdata:
        sys_batch_size = jdata["sys_batch_size"]
    else:
        sys_batch_size = ["auto" for aa in range(len(jdata["sys_configs"]))]

    # make sure all init_data_sys has the batch size -- for the following `zip`
    assert len(init_data_sys_) <= len(init_batch_size_)
    for ii, ss in zip(init_data_sys_, init_batch_size_):
        sys_paths = expand_sys_str(os.path.join(init_data_prefix, ii))
        for single_sys in sys_paths:
            init_data_sys.append(
                Path(
                    os.path.normpath(
                        os.path.join(
                            "../data.init",
                            ii,
                            os.path.relpath(
                                single_sys, os.path.join(init_data_prefix, ii)
                            ),
                        )
                    )
                ).as_posix()
            )
            init_batch_size.append(detect_batch_size(ss, single_sys))
            if auto_ratio:
                number_old_frames += get_nframes(single_sys)
    old_range = None
    if iter_index > 0:
        for ii in range(iter_index):
            if ii == iter_index - 1:
                old_range = len(init_data_sys)
            fp_path = os.path.join(make_iter_name(ii), fp_name)
            fp_data_sys = glob.glob(os.path.join(fp_path, "data.*"))
            if model_devi_engine == "calypso":
                _modd_path = os.path.join(
                    make_iter_name(ii), model_devi_name, calypso_model_devi_name
                )
                sys_list = glob.glob(os.path.join(_modd_path, "*.structures"))
                sys_batch_size = ["auto" for aa in range(len(sys_list))]
            for jj in fp_data_sys:
                sys_idx = int(jj.split(".")[-1])
                sys_paths = expand_sys_str(jj)
                nframes = 0
                for sys_single in sys_paths:
                    nframes += dpdata.LabeledSystem(
                        sys_single, fmt="deepmd/npy"
                    ).get_nframes()
                if auto_ratio:
                    if ii == iter_index - 1:
                        number_new_frames += nframes
                    else:
                        number_old_frames += nframes
                if nframes < fp_task_min:
                    log_task(
                        "nframes (%d) in data sys %s is too small, skip" % (nframes, jj)  # noqa: UP031
                    )
                    continue
                for sys_single in sys_paths:
                    init_data_sys.append(
                        Path(
                            os.path.normpath(os.path.join("../data.iters", sys_single))
                        ).as_posix()
                    )
                    batch_size = (
                        sys_batch_size[sys_idx]
                        if sys_idx < len(sys_batch_size)
                        else "auto"
                    )
                    init_batch_size.append(detect_batch_size(batch_size, sys_single))
    # establish tasks
    jinput = jdata["default_training_param"]
    try:
        mdata["deepmd_version"]
    except KeyError:
        mdata = set_version(mdata)
    # setup data systems
    if Version(mdata["deepmd_version"]) >= Version("1") and Version(
        mdata["deepmd_version"]
    ) < Version("2"):
        # 1.x
        jinput["training"]["systems"] = init_data_sys
        jinput["training"]["batch_size"] = init_batch_size
        jinput["model"]["type_map"] = jdata["type_map"]
        # electron temperature
        if use_ele_temp == 0:
            pass
        elif use_ele_temp == 1:
            jinput["model"]["fitting_net"]["numb_fparam"] = 1
            jinput["model"]["fitting_net"].pop("numb_aparam", None)
        elif use_ele_temp == 2:
            jinput["model"]["fitting_net"]["numb_aparam"] = 1
            jinput["model"]["fitting_net"].pop("numb_fparam", None)
        else:
            raise RuntimeError("invalid setting for use_ele_temp " + str(use_ele_temp))
    elif Version(mdata["deepmd_version"]) >= Version("2") and Version(
        mdata["deepmd_version"]
    ) < Version("4"):
        # 2.x
        jinput["training"].setdefault("training_data", {})
        jinput["training"]["training_data"]["systems"] = init_data_sys
        old_batch_size = jinput["training"]["training_data"].get("batch_size", "")
        if not (
            isinstance(old_batch_size, str) and old_batch_size.startswith("mixed:")
        ):
            jinput["training"]["training_data"]["batch_size"] = init_batch_size
        jinput["model"]["type_map"] = jdata["type_map"]
        # electron temperature
        if use_ele_temp == 0:
            pass
        elif use_ele_temp == 1:
            jinput["model"]["fitting_net"]["numb_fparam"] = 1
            jinput["model"]["fitting_net"].pop("numb_aparam", None)
        elif use_ele_temp == 2:
            jinput["model"]["fitting_net"]["numb_aparam"] = 1
            jinput["model"]["fitting_net"].pop("numb_fparam", None)
        else:
            raise RuntimeError("invalid setting for use_ele_temp " + str(use_ele_temp))
    else:
        raise RuntimeError(
            "DP-GEN currently only supports for DeePMD-kit 1.x to 3.x version!"
        )
    # set training reuse model
    if auto_ratio:
        training_reuse_old_ratio = number_old_frames / (
            number_old_frames + number_new_frames * new_to_old_ratio
        )
    if training_reuse_iter is not None and iter_index >= training_reuse_iter:
        if "numb_steps" in jinput["training"] and training_reuse_stop_batch is not None:
            jinput["training"]["numb_steps"] = training_reuse_stop_batch
        elif (
            "stop_batch" in jinput["training"] and training_reuse_stop_batch is not None
        ):
            jinput["training"]["stop_batch"] = training_reuse_stop_batch
        if Version("1") <= Version(mdata["deepmd_version"]) < Version("2"):
            jinput["training"]["auto_prob_style"] = (
                "prob_sys_size; 0:%d:%f; %d:%d:%f"  # noqa: UP031
                % (
                    old_range,
                    training_reuse_old_ratio,
                    old_range,
                    len(init_data_sys),
                    1.0 - training_reuse_old_ratio,
                )
            )
        elif Version("2") <= Version(mdata["deepmd_version"]) < Version("4"):
            jinput["training"]["training_data"]["auto_prob"] = (
                "prob_sys_size; 0:%d:%f; %d:%d:%f"  # noqa: UP031
                % (
                    old_range,
                    training_reuse_old_ratio,
                    old_range,
                    len(init_data_sys),
                    1.0 - training_reuse_old_ratio,
                )
            )
        else:
            raise RuntimeError(
                "Unsupported DeePMD-kit version: {}".format(mdata["deepmd_version"])
            )
        if (
            jinput["loss"].get("start_pref_e") is not None
            and training_reuse_start_pref_e is not None
        ):
            jinput["loss"]["start_pref_e"] = training_reuse_start_pref_e
        if (
            jinput["loss"].get("start_pref_f") is not None
            and training_reuse_start_pref_f is not None
        ):
            jinput["loss"]["start_pref_f"] = training_reuse_start_pref_f
        if training_reuse_start_lr is not None:
            jinput["learning_rate"]["start_lr"] = training_reuse_start_lr

    input_files = []
    for ii in range(numb_models):
        task_path = os.path.join(work_path, train_task_fmt % ii)
        create_path(task_path)
        os.chdir(task_path)

        if "srtab_file_path" in jdata.keys():
            shutil.copyfile(srtab_file_path, os.path.basename(srtab_file_path))

        for jj in init_data_sys:
            # HDF5 path contains #
            if not (
                os.path.isdir(jj) if "#" not in jj else os.path.isfile(jj.split("#")[0])
            ):
                raise RuntimeError(
                    f"data sys {jj} does not exists, cwd is {os.getcwd()}"
                )
        os.chdir(cwd)
        # set random seed for each model
        if Version(mdata["deepmd_version"]) >= Version("1") and Version(
            mdata["deepmd_version"]
        ) < Version("4"):
            # 1.x
            if "descriptor" not in jinput["model"]:
                pass
            elif jinput["model"]["descriptor"]["type"] == "hybrid":
                for desc in jinput["model"]["descriptor"]["list"]:
                    desc["seed"] = random.randrange(sys.maxsize) % (2**32)
            elif jinput["model"]["descriptor"]["type"] == "loc_frame":
                pass
            else:
                jinput["model"]["descriptor"]["seed"] = random.randrange(
                    sys.maxsize
                ) % (2**32)
            if "fitting_net" in jinput["model"]:
                jinput["model"]["fitting_net"]["seed"] = random.randrange(
                    sys.maxsize
                ) % (2**32)
            if "type_embedding" in jinput["model"]:
                jinput["model"]["type_embedding"]["seed"] = random.randrange(
                    sys.maxsize
                ) % (2**32)
            jinput["training"]["seed"] = random.randrange(sys.maxsize) % (2**32)
        else:
            raise RuntimeError(
                "DP-GEN currently only supports for DeePMD-kit 1.x to 3.x version!"
            )
        # set model activation function
        if model_devi_activation_func is not None:
            if Version(mdata["deepmd_version"]) < Version("1"):
                raise RuntimeError(
                    "model_devi_activation_func does not suppport deepmd version",
                    mdata["deepmd_version"],
                )
            assert (
                isinstance(model_devi_activation_func, list)
                and len(model_devi_activation_func) == numb_models
            )
            if (
                len(np.array(model_devi_activation_func).shape) == 2
            ):  # 2-dim list for emd/fitting net-resolved assignment of actF
                jinput["model"]["descriptor"]["activation_function"] = (
                    model_devi_activation_func[ii][0]
                )
                jinput["model"]["fitting_net"]["activation_function"] = (
                    model_devi_activation_func[ii][1]
                )
            if (
                len(np.array(model_devi_activation_func).shape) == 1
            ):  # for backward compatibility, 1-dim list, not net-resolved
                jinput["model"]["descriptor"]["activation_function"] = (
                    model_devi_activation_func[ii]
                )
                jinput["model"]["fitting_net"]["activation_function"] = (
                    model_devi_activation_func[ii]
                )
        # dump the input.json
        with open(os.path.join(task_path, train_input_file), "w") as outfile:
            json.dump(jinput, outfile, indent=4)
        input_files.append(os.path.join(task_path, train_input_file))

    # link old models
    if iter_index > 0:
        prev_iter_name = make_iter_name(iter_index - 1)
        prev_work_path = os.path.join(prev_iter_name, train_name)
        for ii in range(numb_models):
            prev_task_path = os.path.join(prev_work_path, train_task_fmt % ii)
            old_model_files = glob.glob(os.path.join(prev_task_path, "model.ckpt*"))
            _link_old_models(work_path, old_model_files, ii)
    else:
        if isinstance(training_iter0_model, str):
            training_iter0_model = [training_iter0_model]
        iter0_models = []
        for ii in training_iter0_model:
            model_is = glob.glob(ii)
            model_is.sort()
            iter0_models += [os.path.abspath(ii) for ii in model_is]
        if training_init_model:
            assert numb_models == len(iter0_models), (
                "training_iter0_model should be provided, and the number of models should be equal to %d"  # noqa: UP031
                % numb_models
            )
        for ii in range(len(iter0_models)):
            old_model_path = os.path.join(iter0_models[ii], "model.ckpt*")
            old_model_files = glob.glob(old_model_path)
            if not len(old_model_files):
                raise FileNotFoundError(f"{old_model_path} not found!")
            _link_old_models(work_path, old_model_files, ii)
    copied_models = next(
        (
            item
            for item in (training_init_frozen_model, training_finetune_model)
            if item is not None
        ),
        None,
    )
    if copied_models is not None:
        for ii in range(len(copied_models)):
            _link_old_models(
                work_path, [copied_models[ii]], ii, basename=f"init{suffix}"
            )
    # Copy user defined forward files
    symlink_user_forward_files(mdata=mdata, task_type="train", work_path=work_path)
    # HDF5 format for training data
    if jdata.get("one_h5", False):
        convert_training_data_to_hdf5(input_files, os.path.join(work_path, "data.hdf5"))


def _link_old_models(work_path, old_model_files, ii, basename: Optional[str] = None):
    """Link the `ii`th old model given by `old_model_files` to
    the `ii`th training task in `work_path`.
    """
    task_path = os.path.join(work_path, train_task_fmt % ii)
    task_old_path = os.path.join(task_path, "old")
    create_path(task_old_path)
    cwd = os.getcwd()
    for jj in old_model_files:
        absjj = os.path.abspath(jj)
        if basename is None:
            basejj = os.path.basename(jj)
        else:
            basejj = basename
        os.chdir(task_old_path)
        os.symlink(os.path.relpath(absjj), basejj)
        os.chdir(cwd)


def detect_batch_size(batch_size, system=None):
    if isinstance(batch_size, int):
        return batch_size
    elif batch_size == "auto":
        # automaticcaly set batch size, batch_size = 32 // atom_numb (>=1, <=fram_numb)
        # check if h5 file
        format = "deepmd/npy" if "#" not in system else "deepmd/hdf5"
        s = dpdata.LabeledSystem(system, fmt=format)
        return int(
            min(np.ceil(32.0 / float(s["coords"].shape[1])), s["coords"].shape[0])
        )
    else:
        raise RuntimeError("Unsupported batch size")


def get_nframes(system):
    format = "deepmd/npy" if "#" not in system else "deepmd/hdf5"
    s = dpdata.LabeledSystem(system, fmt=format)
    return s.get_nframes()


def run_train(iter_index, jdata, mdata):
    mlp_engine = jdata.get("mlp_engine", "dp")
    if mlp_engine == "dp":
        return run_train_dp(iter_index, jdata, mdata)
    else:
        raise ValueError(f"Unsupported engine: {mlp_engine}")


def run_train_dp(iter_index, jdata, mdata):
    # print("debug:run_train:mdata", mdata)
    # load json param
    numb_models = jdata["numb_models"]
    suffix = _get_model_suffix(jdata)
    # train_param = jdata['train_param']
    train_input_file = default_train_input_file
    training_reuse_iter = jdata.get("training_reuse_iter")
    training_init_model = jdata.get("training_init_model", False)
    training_init_frozen_model = (
        jdata.get("training_init_frozen_model") if iter_index == 0 else None
    )
    training_finetune_model = (
        jdata.get("training_finetune_model") if iter_index == 0 else None
    )

    if "srtab_file_path" in jdata.keys():
        zbl_file = os.path.basename(jdata.get("srtab_file_path", None))

    if training_reuse_iter is not None and iter_index >= training_reuse_iter:
        training_init_model = True
    try:
        mdata["deepmd_version"]
    except KeyError:
        mdata = set_version(mdata)

    if (
        training_init_model
        + (training_init_frozen_model is not None)
        + (training_finetune_model is not None)
        > 1
    ):
        raise RuntimeError(
            "training_init_model, training_init_frozen_model, and training_finetune_model are mutually exclusive."
        )

    train_command = mdata.get("train_command", "dp").strip()
    # assert train_command == "dp", "The 'train_command' should be 'dp'"     # the tests should be updated to run this command
    if suffix == ".pth":
        train_command += " --pt"
    elif suffix == ".savedmodel":
        train_command += " --jax"

    # paths
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, train_name)
    # check if is copied
    copy_flag = os.path.join(work_path, "copied")
    if os.path.isfile(copy_flag):
        log_task("copied model, do not train")
        return
    # make tasks
    all_task = []
    for ii in range(numb_models):
        task_path = os.path.join(work_path, train_task_fmt % ii)
        all_task.append(task_path)
    commands = []
    if Version(mdata["deepmd_version"]) >= Version("1") and Version(
        mdata["deepmd_version"]
    ) < Version("4"):
        # 1.x
        ## Commands are like `dp train` and `dp freeze`
        ## train_command should not be None
        assert train_command
        extra_flags = ""
        init_flag = ""
        if jdata.get("dp_train_skip_neighbor_stat", False):
            extra_flags += " --skip-neighbor-stat"
        if training_init_model:
            init_flag = " --init-model old/model.ckpt"
        elif training_init_frozen_model is not None:
            init_flag = f" --init-frz-model old/init{suffix}"
        elif training_finetune_model is not None:
            init_flag = f" --finetune old/init{suffix}"
        command = f"{train_command} train {train_input_file}{extra_flags}"
        if suffix == ".pb":
            ckpt_suffix = ".index"
        elif suffix == ".pth":
            ckpt_suffix = ".pt"
        elif suffix == ".savedmodel":
            ckpt_suffix = ".jax"
        else:
            raise RuntimeError(f"Unknown suffix {suffix}")
        command = f"{{ if [ ! -f model.ckpt{ckpt_suffix} ]; then {command}{init_flag}; else {command} --restart model.ckpt; fi }}"
        command = f"/bin/sh -c {shlex.quote(command)}"
        commands.append(command)
        command = f"{train_command} freeze"
        commands.append(command)
        if jdata.get("dp_compress", False):
            commands.append(f"{train_command} compress")
    else:
        raise RuntimeError(
            "DP-GEN currently only supports for DeePMD-kit 1.x to 3.x version!"
        )

    # _tasks = [os.path.basename(ii) for ii in all_task]
    # run_tasks = []
    # for ii in all_task:
    #     check_pb = os.path.join(ii, "frozen_model.pb")
    #     check_lcurve = os.path.join(ii, "lcurve.out")
    #     if os.path.isfile(check_pb) and os.path.isfile(check_lcurve):
    #         pass
    #     else:
    #         run_tasks.append(ii)
    run_tasks = [os.path.basename(ii) for ii in all_task]

    forward_files = [train_input_file]
    if "srtab_file_path" in jdata.keys():
        forward_files.append(zbl_file)
    if training_init_model:
        if suffix == ".pb":
            forward_files += [
                os.path.join("old", "model.ckpt.meta"),
                os.path.join("old", "model.ckpt.index"),
                os.path.join("old", "model.ckpt.data-00000-of-00001"),
            ]
        elif suffix == ".pth":
            forward_files += [os.path.join("old", "model.ckpt.pt")]
        elif suffix == ".savedmodel":
            forward_files += [os.path.join("old", "model.ckpt.jax")]
        else:
            raise RuntimeError(f"Unknown suffix {suffix}")
    elif training_init_frozen_model is not None or training_finetune_model is not None:
        forward_files.append(os.path.join("old", f"init{suffix}"))

    backward_files = [
        f"frozen_model{suffix}",
        "lcurve.out",
        "train.log",
        "checkpoint",
    ]
    if jdata.get("dp_compress", False):
        backward_files.append(f"frozen_model_compressed{suffix}")

    if suffix == ".pb":
        backward_files += [
            "model.ckpt.meta",
            "model.ckpt.index",
            "model.ckpt.data-00000-of-00001",
        ]
    elif suffix == ".pth":
        backward_files += ["model.ckpt.pt"]
    elif suffix == ".savedmodel":
        backward_files += ["model.ckpt.jax"]
    else:
        raise RuntimeError(f"Unknown suffix {suffix}")

    if not jdata.get("one_h5", False):
        init_data_sys_ = jdata["init_data_sys"]
        init_data_sys = []
        for ii in init_data_sys_:
            init_data_sys.append(os.path.join("data.init", ii))
        trans_comm_data = []
        cwd = os.getcwd()
        os.chdir(work_path)
        fp_data = glob.glob(os.path.join("data.iters", "iter.*", "02.fp", "data.*"))
        for ii in itertools.chain(init_data_sys, fp_data):
            sys_paths = expand_sys_str(ii)
            for single_sys in sys_paths:
                if "#" not in single_sys:
                    trans_comm_data += glob.glob(os.path.join(single_sys, "set.*"))
                    trans_comm_data += glob.glob(os.path.join(single_sys, "type*.raw"))
                    trans_comm_data += glob.glob(os.path.join(single_sys, "nopbc"))
                else:
                    # H5 file
                    trans_comm_data.append(single_sys.split("#")[0])
    else:
        cwd = os.getcwd()
        trans_comm_data = ["data.hdf5"]
    # remove duplicated files
    trans_comm_data = list(set(trans_comm_data))
    os.chdir(cwd)

    try:
        train_group_size = mdata["train_group_size"]
    except Exception:
        train_group_size = 1

    user_forward_files = mdata.get("train" + "_user_forward_files", [])
    forward_files += [os.path.basename(file) for file in user_forward_files]
    backward_files += mdata.get("train" + "_user_backward_files", [])

    ### Submit jobs
    check_api_version(mdata)

    submission = make_submission(
        mdata["train_machine"],
        mdata["train_resources"],
        commands=commands,
        work_path=work_path,
        run_tasks=run_tasks,
        group_size=train_group_size,
        forward_common_files=trans_comm_data,
        forward_files=forward_files,
        backward_files=backward_files,
        outlog="train.log",
        errlog="train.log",
    )
    submission.run_submission()


def post_train(iter_index, jdata, mdata):
    mlp_engine = jdata.get("mlp_engine", "dp")
    if mlp_engine == "dp":
        return post_train_dp(iter_index, jdata, mdata)
    else:
        raise ValueError(f"Unsupported engine: {mlp_engine}")


def post_train_dp(iter_index, jdata, mdata):
    # load json param
    numb_models = jdata["numb_models"]
    # paths
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, train_name)
    # check if is copied
    copy_flag = os.path.join(work_path, "copied")
    if os.path.isfile(copy_flag):
        log_task("copied model, do not post train")
        return
    # symlink models
    suffix = _get_model_suffix(jdata)
    for ii in range(numb_models):
        model_name = f"frozen_model{suffix}"
        if jdata.get("dp_compress", False):
            model_name = f"frozen_model_compressed{suffix}"

        ofile = os.path.join(work_path, "graph.%03d%s" % (ii, suffix))  # noqa: UP031
        task_file = os.path.join(train_task_fmt % ii, model_name)
        if os.path.isfile(ofile):
            os.remove(ofile)
        os.symlink(task_file, ofile)


def _get_param_alias(jdata, names):
    for ii in names:
        if ii in jdata:
            return jdata[ii]
    raise ValueError(
        f"one of the keys {str(names)} should be in jdata {json.dumps(jdata, indent=4)}"
    )


def parse_cur_job(cur_job):
    ensemble = _get_param_alias(cur_job, ["ens", "ensemble"])
    temps = [-1]
    press = [-1]
    if "npt" in ensemble:
        temps = _get_param_alias(cur_job, ["Ts", "temps"])
        press = _get_param_alias(cur_job, ["Ps", "press"])
    elif "nvt" == ensemble or "nve" == ensemble:
        temps = _get_param_alias(cur_job, ["Ts", "temps"])
    nsteps = _get_param_alias(cur_job, ["nsteps"])
    trj_freq = _get_param_alias(cur_job, ["t_freq", "trj_freq", "traj_freq"])
    if "pka_e" in cur_job:
        pka_e = _get_param_alias(cur_job, ["pka_e"])
    else:
        pka_e = None
    if "dt" in cur_job:
        dt = _get_param_alias(cur_job, ["dt"])
    else:
        dt = None
    if "nbeads" in cur_job:
        nbeads = _get_param_alias(cur_job, ["nbeads"])
    else:
        nbeads = None
    return ensemble, nsteps, trj_freq, temps, press, pka_e, dt, nbeads


def expand_matrix_values(target_list, cur_idx=0):
    nvar = len(target_list)
    if cur_idx == nvar:
        return [[]]
    else:
        res = []
        prev = expand_matrix_values(target_list, cur_idx + 1)
        for ii in target_list[cur_idx]:
            tmp = copy.deepcopy(prev)
            for jj in tmp:
                jj.insert(0, ii)
                res.append(jj)
        return res


def parse_cur_job_revmat(cur_job, use_plm=False):
    templates = [cur_job["template"]["lmp"]]
    if use_plm:
        templates.append(cur_job["template"]["plm"])
    revise_keys = []
    revise_values = []
    if "rev_mat" not in cur_job.keys():
        cur_job["rev_mat"] = {}
    if "lmp" not in cur_job["rev_mat"].keys():
        cur_job["rev_mat"]["lmp"] = {}
    for ii in cur_job["rev_mat"]["lmp"].keys():
        revise_keys.append(ii)
        revise_values.append(cur_job["rev_mat"]["lmp"][ii])
    n_lmp_keys = len(revise_keys)
    if use_plm:
        if "plm" not in cur_job["rev_mat"].keys():
            cur_job["rev_mat"]["plm"] = {}
        for ii in cur_job["rev_mat"]["plm"].keys():
            revise_keys.append(ii)
            revise_values.append(cur_job["rev_mat"]["plm"][ii])
    revise_matrix = expand_matrix_values(revise_values)
    return revise_keys, revise_matrix, n_lmp_keys


def parse_cur_job_sys_revmat(cur_job, sys_idx, use_plm=False):
    templates = [cur_job["template"]["lmp"]]
    if use_plm:
        templates.append(cur_job["template"]["plm"])
    sys_revise_keys = []
    sys_revise_values = []
    if "sys_rev_mat" not in cur_job.keys():
        cur_job["sys_rev_mat"] = {}
    local_rev = cur_job["sys_rev_mat"].get(str(sys_idx), {})
    if "lmp" not in local_rev.keys():
        local_rev["lmp"] = {}
    for ii in local_rev["lmp"].keys():
        sys_revise_keys.append(ii)
        sys_revise_values.append(local_rev["lmp"][ii])
    n_sys_lmp_keys = len(sys_revise_keys)
    if use_plm:
        if "plm" not in local_rev.keys():
            local_rev["plm"] = {}
        for ii in local_rev["plm"].keys():
            sys_revise_keys.append(ii)
            sys_revise_values.append(local_rev["plm"][ii])
    sys_revise_matrix = expand_matrix_values(sys_revise_values)
    return sys_revise_keys, sys_revise_matrix, n_sys_lmp_keys


def find_only_one_key(lmp_lines, key):
    found = []
    for idx in range(len(lmp_lines)):
        words = lmp_lines[idx].split()
        nkey = len(key)
        if len(words) >= nkey and words[:nkey] == key:
            found.append(idx)
    if len(found) > 1:
        raise RuntimeError("found %d keywords %s" % (len(found), key))  # noqa: UP031
    if len(found) == 0:
        raise RuntimeError(f"failed to find keyword {key}")
    return found[0]


def revise_lmp_input_model(
    lmp_lines, task_model_list, trj_freq, deepmd_version="1", use_ele_temp=0, jdata=None
):
    idx = find_only_one_key(lmp_lines, ["pair_style", "deepmd"])
    graph_list = " ".join(task_model_list)

    # Check if D3 dispersion is configured
    lmp_d3 = jdata.get("lmp_d3", {}) if jdata else {}
    d3_enabled = lmp_d3.get("enable", False) if lmp_d3 else False

    if Version(deepmd_version) < Version("1"):
        if d3_enabled:
            d3_params = f"{lmp_d3['damping_function']} {lmp_d3['functional']} {lmp_d3['cutoff']} {lmp_d3['cn_cutoff']}"
            lmp_lines[idx] = (
                f"pair_style      hybrid/overlay deepmd {graph_list} {trj_freq} model_devi.out dispersion/d3 {d3_params}\n"
            )
        else:
            lmp_lines[idx] = "pair_style      deepmd %s %d model_devi.out\n" % (  # noqa: UP031
                graph_list,
                trj_freq,
            )
    else:
        # Build keywords string like in make_lammps_input
        keywords = ""
        if jdata is not None:
            if jdata.get("use_clusters", False):
                keywords += "atomic "
            if jdata.get("use_relative", False):
                keywords += "relative {} ".format(jdata["epsilon"])
            if jdata.get("use_relative_v", False):
                keywords += "relative_v {} ".format(jdata["epsilon_v"])

        if use_ele_temp == 1:
            keywords += "fparam ${ELE_TEMP}"

        if d3_enabled:
            d3_params = f"{lmp_d3['damping_function']} {lmp_d3['functional']} {lmp_d3['cutoff']} {lmp_d3['cn_cutoff']}"
            lmp_lines[idx] = (
                "pair_style      hybrid/overlay deepmd %s out_freq %d out_file model_devi.out %s dispersion/d3 %s\n"  # noqa: UP031
                % (
                    graph_list,
                    trj_freq,
                    keywords.rstrip(),
                    d3_params,
                )
            )
        else:
            lmp_lines[idx] = (
                "pair_style      deepmd %s out_freq %d out_file model_devi.out %s\n"  # noqa: UP031
                % (
                    graph_list,
                    trj_freq,
                    keywords.rstrip(),
                )
            )
    return lmp_lines


def revise_lmp_input_pair_coeff(lmp_lines, jdata=None):
    """Update pair_coeff lines for D3 support."""
    if jdata is None:
        return lmp_lines

    lmp_d3 = jdata.get("lmp_d3", {})
    d3_enabled = lmp_d3.get("enable", False) if lmp_d3 else False

    if not d3_enabled:
        return lmp_lines

    # D3 requires type maps (element symbols)
    type_map = jdata.get("type_map", [])
    type_map_str = " ".join(type_map)

    # Find pair_coeff line
    pair_coeff_idx = None
    for idx, line in enumerate(lmp_lines):
        if line.strip().startswith("pair_coeff") and "* *" in line:
            pair_coeff_idx = idx
            break

    if pair_coeff_idx is None:
        # If no pair_coeff found, add them after pair_style
        pair_style_idx = find_only_one_key(lmp_lines, ["pair_style"])
        lmp_lines.insert(pair_style_idx + 1, "pair_coeff      * * deepmd\n")
        lmp_lines.insert(
            pair_style_idx + 2, f"pair_coeff      * * dispersion/d3 {type_map_str}\n"
        )
    else:
        # Replace existing pair_coeff with D3 version
        lmp_lines[pair_coeff_idx] = "pair_coeff      * * deepmd\n"
        lmp_lines.insert(
            pair_coeff_idx + 1, f"pair_coeff      * * dispersion/d3 {type_map_str}\n"
        )

    return lmp_lines


def revise_lmp_input_neigh_modify(lmp_lines, jdata=None):
    """Add neigh_modify one N if requested."""
    if jdata is None:
        return lmp_lines

    neigh_modify_one = jdata.get("lmp_neigh_modify_one")
    if neigh_modify_one is None:
        return lmp_lines

    # Find where to insert neigh_modify one N
    # Look for existing neigh_modify lines or insert after neighbor command
    neigh_modify_found = False
    neighbor_idx = None

    for idx, line in enumerate(lmp_lines):
        if line.strip().startswith("neigh_modify") and " one " in line:
            neigh_modify_found = True
            break
        elif line.strip().startswith("neighbor"):
            neighbor_idx = idx

    if not neigh_modify_found:
        if neighbor_idx is not None:
            lmp_lines.insert(
                neighbor_idx + 1, f"neigh_modify    one {neigh_modify_one}\n"
            )
        else:
            # Insert after units command if neighbor not found
            units_idx = None
            for idx, line in enumerate(lmp_lines):
                if line.strip().startswith("units"):
                    units_idx = idx
                    break
            if units_idx is not None:
                lmp_lines.insert(
                    units_idx + 1, f"neigh_modify    one {neigh_modify_one}\n"
                )

    return lmp_lines


def revise_lmp_input_dump(lmp_lines, trj_freq, model_devi_merge_traj=False):
    idx = find_only_one_key(lmp_lines, ["dump", "dpgen_dump"])
    if model_devi_merge_traj:
        lmp_lines[idx] = (
            "dump            dpgen_dump all custom %d    all.lammpstrj id type x y z\n"  # noqa: UP031
            % trj_freq
        )
    else:
        lmp_lines[idx] = (
            "dump            dpgen_dump all custom %d traj/*.lammpstrj id type x y z\n"  # noqa: UP031
            % trj_freq
        )

    return lmp_lines


def revise_lmp_input_plm(lmp_lines, in_plm, out_plm="output.plumed"):
    idx = find_only_one_key(lmp_lines, ["fix", "dpgen_plm"])
    lmp_lines[idx] = (
        f"fix            dpgen_plm all plumed plumedfile {in_plm} outfile {out_plm}\n"
    )
    return lmp_lines


def revise_by_keys(lmp_lines, keys, values):
    for kk, vv in zip(keys, values):
        for ii in range(len(lmp_lines)):
            lmp_lines[ii] = lmp_lines[ii].replace(kk, str(vv))
    return lmp_lines


def make_model_devi(iter_index, jdata, mdata):
    # The MD engine to perform model deviation
    # Default is lammps
    model_devi_engine = jdata.get("model_devi_engine", "lammps")

    model_devi_jobs = jdata["model_devi_jobs"]
    if model_devi_engine != "calypso":
        if iter_index >= len(model_devi_jobs):
            return False
    else:
        # mode 1: generate structures according to the user-provided input.dat file, so calypso_input_path and model_devi_max_iter are needed
        run_mode = 1
        if "calypso_input_path" in jdata:
            try:
                maxiter = jdata.get("model_devi_max_iter")
            except KeyError:
                raise KeyError(
                    "calypso_input_path key exists so you should provide model_devi_max_iter key to control the max iter number"
                )
        # mode 2: control each iteration to generate structures in specific way by providing model_devi_jobs key
        else:
            try:
                maxiter = max(model_devi_jobs[-1].get("times"))
                run_mode = 2
            except KeyError:
                raise KeyError('did not find model_devi_jobs["times"] key')
        if iter_index > maxiter:
            dlog.info(f"iter_index is {iter_index} and maxiter is {maxiter}")
            return False

    if "sys_configs_prefix" in jdata:
        sys_configs = []
        for sys_list in jdata["sys_configs"]:
            # assert (isinstance(sys_list, list) ), "Currently only support type list for sys in 'sys_conifgs' "
            temp_sys_list = [
                os.path.join(jdata["sys_configs_prefix"], sys) for sys in sys_list
            ]
            sys_configs.append(temp_sys_list)
    else:
        sys_configs = jdata["sys_configs"]
    shuffle_poscar = jdata.get("shuffle_poscar", False)

    if model_devi_engine != "calypso":
        cur_job = model_devi_jobs[iter_index]
        sys_idx = expand_idx(cur_job["sys_idx"])
    else:
        cur_job = {"model_devi_engine": "calypso", "input.dat": "user_provided"}
        sys_idx = []

    if len(sys_idx) != len(list(set(sys_idx))):
        raise RuntimeError("system index should be uniq")
    conf_systems = []
    for idx in sys_idx:
        cur_systems = []
        ss = sys_configs[idx]
        for ii in ss:
            ii_systems = sorted(glob.glob(ii))
            if ii_systems == []:
                warnings.warn(
                    f"There is no system in the path {ii}. Please check if the path is correct."
                )
            cur_systems += ii_systems
        # cur_systems should not be sorted, as we may add specific constrict to the similutions
        # cur_systems.sort()
        cur_systems = [os.path.abspath(ii) for ii in cur_systems]
        conf_systems.append(cur_systems)

    iter_name = make_iter_name(iter_index)
    train_path = os.path.join(iter_name, train_name)
    train_path = os.path.abspath(train_path)
    suffix = _get_model_suffix(jdata)
    models = sorted(glob.glob(os.path.join(train_path, f"graph*{suffix}")))
    work_path = os.path.join(iter_name, model_devi_name)
    create_path(work_path)
    if model_devi_engine == "calypso":
        _calypso_run_opt_path = os.path.join(work_path, calypso_run_opt_name)
        calypso_model_devi_path = os.path.join(work_path, calypso_model_devi_name)
        create_path(calypso_model_devi_path)
        # run model devi script
        calypso_run_model_devi_script = os.path.join(
            calypso_model_devi_path, "calypso_run_model_devi.py"
        )
        shutil.copyfile(calypso_run_model_devi_file, calypso_run_model_devi_script)
        # Create work path list
        calypso_run_opt_path = []

        # mode 1: generate structures according to the user-provided input.dat file,
        # so calypso_input_path and model_devi_max_iter are needed
        if run_mode == 1:
            if jdata.get("vsc", False) and len(jdata.get("type_map")) > 1:
                # [input.dat.Li.250, input.dat.Li.300]
                one_ele_inputdat_list = glob.glob(
                    f"{jdata.get('calypso_input_path')}/input.dat.{jdata.get('type_map')[0]}.*"
                )
                if len(one_ele_inputdat_list) == 0:
                    number_of_pressure = 1
                else:
                    number_of_pressure = len(list(set(one_ele_inputdat_list)))

                # calypso_run_opt_path = ['gen_struc_analy.000','gen_struc_analy.001']
                for temp_idx in range(number_of_pressure):
                    calypso_run_opt_path.append(
                        "%s.%03d" % (_calypso_run_opt_path, temp_idx)  # noqa: UP031
                    )
            elif not jdata.get("vsc", False):
                calypso_run_opt_path.append("%s.%03d" % (_calypso_run_opt_path, 0))  # noqa: UP031

        # mode 2: control each iteration to generate structures in specific way
        # by providing model_devi_jobs key
        elif run_mode == 2:
            for iiidx, jobbs in enumerate(model_devi_jobs):
                if iter_index in jobbs.get("times"):
                    cur_job = model_devi_jobs[iiidx]

            pressures_list = cur_job.get("PSTRESS", [0.0001])
            for temp_idx in range(len(pressures_list)):
                calypso_run_opt_path.append(
                    "%s.%03d" % (_calypso_run_opt_path, temp_idx)  # noqa: UP031
                )
        # to different directory
        # calypso_run_opt_path = ['gen_struc_analy.000','gen_struc_analy.001','gen_struc_analy.002',]
        for temp_calypso_run_opt_path in calypso_run_opt_path:
            create_path(temp_calypso_run_opt_path)
            # run confs opt script
            run_opt_script = os.path.join(
                temp_calypso_run_opt_path, "calypso_run_opt.py"
            )
            shutil.copyfile(run_opt_file, run_opt_script)
            # check outcar script
            check_outcar_script = os.path.join(
                temp_calypso_run_opt_path, "check_outcar.py"
            )
            shutil.copyfile(check_outcar_file, check_outcar_script)

    for mm in models:
        model_name = os.path.basename(mm)
        if model_devi_engine != "calypso":
            os.symlink(mm, os.path.join(work_path, model_name))
        else:
            for temp_calypso_run_opt_path in calypso_run_opt_path:
                models_path = os.path.join(temp_calypso_run_opt_path, model_name)
                if not os.path.exists(models_path):
                    os.symlink(mm, models_path)

    with open(os.path.join(work_path, "cur_job.json"), "w") as outfile:
        json.dump(cur_job, outfile, indent=4)

    conf_path = os.path.join(work_path, "confs")
    create_path(conf_path)
    sys_counter = 0
    rng = np.random.default_rng()
    for ss in conf_systems:
        conf_counter = 0
        for cc in ss:
            if model_devi_engine == "lammps":
                conf_name = make_model_devi_conf_name(
                    sys_idx[sys_counter], conf_counter
                )
                poscar_name = conf_name + ".poscar"
                lmp_name = conf_name + ".lmp"
                os.symlink(cc, os.path.join(conf_path, poscar_name))
                if "sys_format" in jdata:
                    fmt = jdata["sys_format"]
                else:
                    fmt = "vasp/poscar"
                system = dpdata.System(
                    os.path.join(conf_path, poscar_name),
                    fmt=fmt,
                    type_map=jdata["type_map"],
                )
                if shuffle_poscar:
                    rng.shuffle(system.data["coords"], axis=1)
                if jdata.get("model_devi_nopbc", False):
                    system.remove_pbc()
                system.to_lammps_lmp(os.path.join(conf_path, lmp_name))
            elif model_devi_engine == "gromacs":
                pass
            elif model_devi_engine == "amber":
                # Jinzhe's specific Amber version
                conf_name = make_model_devi_conf_name(
                    sys_idx[sys_counter], conf_counter
                )
                rst7_name = conf_name + ".rst7"
                # link restart file
                os.symlink(cc, os.path.join(conf_path, rst7_name))
            conf_counter += 1
        sys_counter += 1

    input_mode = "native"
    if "calypso_input_path" in jdata:
        input_mode = "buffet"
    if "template" in cur_job:
        input_mode = "revise_template"
    use_plm = jdata.get("model_devi_plumed", False)
    use_plm_path = jdata.get("model_devi_plumed_path", False)
    if input_mode == "native":
        if model_devi_engine == "lammps":
            _make_model_devi_native(iter_index, jdata, mdata, conf_systems)
        elif model_devi_engine == "gromacs":
            _make_model_devi_native_gromacs(iter_index, jdata, mdata, conf_systems)
        elif model_devi_engine == "amber":
            _make_model_devi_amber(iter_index, jdata, mdata, conf_systems)
        elif model_devi_engine == "calypso":
            _make_model_devi_native_calypso(
                iter_index, model_devi_jobs, calypso_run_opt_path
            )  # generate input.dat automatic in each iter
        else:
            raise RuntimeError("unknown model_devi engine", model_devi_engine)
    elif input_mode == "revise_template":
        _make_model_devi_revmat(iter_index, jdata, mdata, conf_systems)
    elif input_mode == "buffet":
        _make_model_devi_buffet(
            jdata, calypso_run_opt_path
        )  # generate confs according to the input.dat provided
    else:
        raise RuntimeError("unknown model_devi input mode", input_mode)
    # Copy user defined forward_files
    symlink_user_forward_files(mdata=mdata, task_type="model_devi", work_path=work_path)
    return True


def _make_model_devi_revmat(iter_index, jdata, mdata, conf_systems):
    model_devi_jobs = jdata["model_devi_jobs"]
    if iter_index >= len(model_devi_jobs):
        return False
    cur_job = model_devi_jobs[iter_index]
    sys_idx = expand_idx(cur_job["sys_idx"])
    if len(sys_idx) != len(list(set(sys_idx))):
        raise RuntimeError("system index should be uniq")

    use_ele_temp = jdata.get("use_ele_temp", 0)
    mass_map = jdata["mass_map"]
    use_plm = jdata.get("model_devi_plumed", False)
    use_plm_path = jdata.get("model_devi_plumed_path", False)
    trj_freq = _get_param_alias(cur_job, ["t_freq", "trj_freq", "traj_freq"])

    rev_keys, rev_mat, num_lmp = parse_cur_job_revmat(cur_job, use_plm=use_plm)
    lmp_templ = cur_job["template"]["lmp"]
    lmp_templ = os.path.abspath(lmp_templ)
    if use_plm:
        plm_templ = cur_job["template"]["plm"]
        plm_templ = os.path.abspath(plm_templ)
        if use_plm_path:
            plm_path_templ = cur_job["template"]["plm_path"]
            plm_path_templ = os.path.abspath(plm_path_templ)

    iter_name = make_iter_name(iter_index)
    train_path = os.path.join(iter_name, train_name)
    train_path = os.path.abspath(train_path)
    suffix = _get_model_suffix(jdata)
    models = sorted(glob.glob(os.path.join(train_path, f"graph*{suffix}")))
    task_model_list = []
    for ii in models:
        task_model_list.append(os.path.join("..", os.path.basename(ii)))
    work_path = os.path.join(iter_name, model_devi_name)
    try:
        mdata["deepmd_version"]
    except KeyError:
        mdata = set_version(mdata)
    deepmd_version = mdata["deepmd_version"]

    sys_counter = 0
    for ss in conf_systems:
        conf_counter = 0
        task_counter = 0
        for cc in ss:
            sys_rev = cur_job.get("sys_rev_mat", None)
            total_rev_keys = rev_keys
            total_rev_mat = rev_mat
            total_num_lmp = num_lmp
            if sys_rev is not None:
                total_rev_mat = []
                sys_rev_keys, sys_rev_mat, sys_num_lmp = parse_cur_job_sys_revmat(
                    cur_job, sys_idx=sys_idx[sys_counter], use_plm=use_plm
                )
                _lmp_keys = rev_keys[:num_lmp] + sys_rev_keys[:sys_num_lmp]
                if use_plm:
                    _plm_keys = rev_keys[num_lmp:] + sys_rev_keys[sys_num_lmp:]
                    _lmp_keys += _plm_keys
                total_rev_keys = _lmp_keys
                total_num_lmp = num_lmp + sys_num_lmp
                for pub in rev_mat:
                    for pri in sys_rev_mat:
                        _lmp_mat = pub[:num_lmp] + pri[:sys_num_lmp]
                        if use_plm:
                            _plm_mat = pub[num_lmp:] + pri[sys_num_lmp:]
                            _lmp_mat += _plm_mat
                        total_rev_mat.append(_lmp_mat)
            for ii in range(len(total_rev_mat)):
                total_rev_item = total_rev_mat[ii]
                task_name = make_model_devi_task_name(
                    sys_idx[sys_counter], task_counter
                )
                conf_name = (
                    make_model_devi_conf_name(sys_idx[sys_counter], conf_counter)
                    + ".lmp"
                )
                task_path = os.path.join(work_path, task_name)
                # create task path
                create_path(task_path)
                model_devi_merge_traj = jdata.get("model_devi_merge_traj", False)
                if not model_devi_merge_traj:
                    create_path(os.path.join(task_path, "traj"))
                # link conf
                loc_conf_name = "conf.lmp"
                os.symlink(
                    os.path.join(os.path.join("..", "confs"), conf_name),
                    os.path.join(task_path, loc_conf_name),
                )
                cwd_ = os.getcwd()
                # chdir to task path
                os.chdir(task_path)
                shutil.copyfile(lmp_templ, "input.lammps")
                # revise input of lammps
                with open("input.lammps") as fp:
                    lmp_lines = fp.readlines()
                # only revise the line "pair_style deepmd" if the user has not written the full line (checked by then length of the line)
                template_has_pair_deepmd = 1
                for line_idx, line_context in enumerate(lmp_lines):
                    if (
                        (line_context[0] != "#")
                        and ("pair_style" in line_context)
                        and ("deepmd" in line_context)
                    ):
                        template_has_pair_deepmd = 0
                        template_pair_deepmd_idx = line_idx
                if template_has_pair_deepmd == 0:
                    if Version(deepmd_version) < Version("1"):
                        if len(lmp_lines[template_pair_deepmd_idx].split()) != (
                            len(models)
                            + len(["pair_style", "deepmd", "10", "model_devi.out"])
                        ):
                            lmp_lines = revise_lmp_input_model(
                                lmp_lines,
                                task_model_list,
                                trj_freq,
                                deepmd_version=deepmd_version,
                                use_ele_temp=use_ele_temp,
                                jdata=jdata,
                            )
                            # Add D3 pair_coeff and neigh_modify support for templates
                            lmp_lines = revise_lmp_input_pair_coeff(lmp_lines, jdata)
                            lmp_lines = revise_lmp_input_neigh_modify(lmp_lines, jdata)
                    else:
                        if len(lmp_lines[template_pair_deepmd_idx].split()) != (
                            len(models)
                            + len(
                                [
                                    "pair_style",
                                    "deepmd",
                                    "out_freq",
                                    "10",
                                    "out_file",
                                    "model_devi.out",
                                ]
                            )
                        ):
                            lmp_lines = revise_lmp_input_model(
                                lmp_lines,
                                task_model_list,
                                trj_freq,
                                deepmd_version=deepmd_version,
                                use_ele_temp=use_ele_temp,
                                jdata=jdata,
                            )
                            # Add D3 pair_coeff and neigh_modify support for templates
                            lmp_lines = revise_lmp_input_pair_coeff(lmp_lines, jdata)
                            lmp_lines = revise_lmp_input_neigh_modify(lmp_lines, jdata)
                # use revise_lmp_input_model to raise error message if "part_style" or "deepmd" not found
                else:
                    lmp_lines = revise_lmp_input_model(
                        lmp_lines,
                        task_model_list,
                        trj_freq,
                        deepmd_version=deepmd_version,
                        use_ele_temp=use_ele_temp,
                        jdata=jdata,
                    )

                # Add D3 pair_coeff and neigh_modify support for templates
                lmp_lines = revise_lmp_input_pair_coeff(lmp_lines, jdata)
                lmp_lines = revise_lmp_input_neigh_modify(lmp_lines, jdata)

                lmp_lines = revise_lmp_input_dump(
                    lmp_lines, trj_freq, model_devi_merge_traj
                )
                lmp_lines = revise_by_keys(
                    lmp_lines,
                    total_rev_keys[:total_num_lmp],
                    total_rev_item[:total_num_lmp],
                )
                # revise input of plumed
                if use_plm:
                    lmp_lines = revise_lmp_input_plm(lmp_lines, "input.plumed")
                    shutil.copyfile(plm_templ, "input.plumed")
                    with open("input.plumed") as fp:
                        plm_lines = fp.readlines()
                    # allow using the same list as lmp
                    # user should not use the same key name for plm
                    plm_lines = revise_by_keys(
                        plm_lines, total_rev_keys, total_rev_item
                    )
                    with open("input.plumed", "w") as fp:
                        fp.write("".join(plm_lines))
                    if use_plm_path:
                        shutil.copyfile(plm_path_templ, "plmpath.pdb")
                # dump input of lammps
                with open("input.lammps", "w") as fp:
                    fp.write("".join(lmp_lines))
                with open("job.json", "w") as fp:
                    job = {}
                    for ii, jj in zip(total_rev_keys, total_rev_item):
                        job[ii] = jj
                    json.dump(job, fp, indent=4)
                os.chdir(cwd_)
                task_counter += 1
            conf_counter += 1
        sys_counter += 1


def _make_model_devi_native(iter_index, jdata, mdata, conf_systems):
    model_devi_jobs = jdata["model_devi_jobs"]
    if iter_index >= len(model_devi_jobs):
        return False
    cur_job = model_devi_jobs[iter_index]
    ensemble, nsteps, trj_freq, temps, press, pka_e, dt, nbeads = parse_cur_job(cur_job)
    model_devi_f_avg_relative = jdata.get("model_devi_f_avg_relative", False)
    model_devi_merge_traj = jdata.get("model_devi_merge_traj", False)
    if (nbeads is not None) and model_devi_f_avg_relative:
        raise RuntimeError(
            "model_devi_f_avg_relative has not been supported for pimd. Set model_devi_f_avg_relative to False."
        )
    if (nbeads is not None) and (model_devi_merge_traj):
        raise RuntimeError(
            "model_devi_merge_traj has not been supported for pimd. Set model_devi_merge_traj to False."
        )
    if (nbeads is not None) and (not (nsteps % trj_freq == 0)):
        raise RuntimeError(
            "trj_freq should be a factor of nsteps for pimd. Please check your input."
        )
    if dt is not None:
        model_devi_dt = dt
    sys_idx = expand_idx(cur_job["sys_idx"])
    if len(sys_idx) != len(list(set(sys_idx))):
        raise RuntimeError("system index should be uniq")

    use_ele_temp = jdata.get("use_ele_temp", 0)
    model_devi_dt = jdata["model_devi_dt"]
    model_devi_neidelay = None
    if "model_devi_neidelay" in jdata:
        model_devi_neidelay = jdata["model_devi_neidelay"]
    model_devi_taut = 0.1
    if "model_devi_taut" in jdata:
        model_devi_taut = jdata["model_devi_taut"]
    model_devi_taup = 0.5
    if "model_devi_taup" in jdata:
        model_devi_taup = jdata["model_devi_taup"]
    mass_map = jdata["mass_map"]
    nopbc = jdata.get("model_devi_nopbc", False)

    iter_name = make_iter_name(iter_index)
    train_path = os.path.join(iter_name, train_name)
    train_path = os.path.abspath(train_path)
    suffix = _get_model_suffix(jdata)
    models = sorted(glob.glob(os.path.join(train_path, f"graph*{suffix}")))
    task_model_list = []
    for ii in models:
        task_model_list.append(Path(f"../{Path(ii).name}").as_posix())
    work_path = os.path.join(iter_name, model_devi_name)

    sys_counter = 0
    for ss in conf_systems:
        conf_counter = 0
        task_counter = 0
        for cc in ss:
            for tt_ in temps:
                if use_ele_temp:
                    if isinstance(tt_, list):
                        tt = tt_[0]
                        if use_ele_temp == 1:
                            te_f = tt_[1]
                            te_a = None
                        else:
                            te_f = None
                            te_a = tt_[1]
                    else:
                        assert isinstance(tt_, (float, int))
                        tt = float(tt_)
                        if use_ele_temp == 1:
                            te_f = tt
                            te_a = None
                        else:
                            te_f = None
                            te_a = tt
                else:
                    tt = tt_
                    te_f = None
                    te_a = None
                for pp in press:
                    task_name = make_model_devi_task_name(
                        sys_idx[sys_counter], task_counter
                    )
                    conf_name = (
                        make_model_devi_conf_name(sys_idx[sys_counter], conf_counter)
                        + ".lmp"
                    )
                    task_path = os.path.join(work_path, task_name)
                    # dlog.info(task_path)
                    create_path(task_path)
                    model_devi_merge_traj = jdata.get("model_devi_merge_traj", False)
                    if not model_devi_merge_traj:
                        create_path(os.path.join(task_path, "traj"))
                    loc_conf_name = "conf.lmp"
                    os.symlink(
                        os.path.join(os.path.join("..", "confs"), conf_name),
                        os.path.join(task_path, loc_conf_name),
                    )
                    cwd_ = os.getcwd()
                    os.chdir(task_path)
                    try:
                        mdata["deepmd_version"]
                    except KeyError:
                        mdata = set_version(mdata)
                    deepmd_version = mdata["deepmd_version"]
                    file_c = make_lammps_input(
                        ensemble,
                        loc_conf_name,
                        task_model_list,
                        nsteps,
                        model_devi_dt,
                        model_devi_neidelay,
                        trj_freq,
                        mass_map,
                        tt,
                        jdata=jdata,
                        tau_t=model_devi_taut,
                        pres=pp,
                        tau_p=model_devi_taup,
                        pka_e=pka_e,
                        ele_temp_f=te_f,
                        ele_temp_a=te_a,
                        nopbc=nopbc,
                        deepmd_version=deepmd_version,
                        nbeads=nbeads,
                    )
                    job = {}
                    job["ensemble"] = ensemble
                    job["press"] = pp
                    job["temps"] = tt
                    if te_f is not None:
                        job["ele_temp"] = te_f
                    if te_a is not None:
                        job["ele_temp"] = te_a
                    job["model_devi_dt"] = model_devi_dt
                    with open("job.json", "w") as _outfile:
                        json.dump(job, _outfile, indent=4)
                    os.chdir(cwd_)
                    with open(os.path.join(task_path, "input.lammps"), "w") as fp:
                        fp.write(file_c)
                    task_counter += 1
            conf_counter += 1
        sys_counter += 1


def _make_model_devi_native_gromacs(iter_index, jdata, mdata, conf_systems):
    try:
        from gromacs.fileformats.mdp import MDP
    except ImportError as e:
        raise RuntimeError(
            "GromacsWrapper>=0.8.0 is needed for DP-GEN + Gromacs."
        ) from e
    # only support for deepmd v2.0
    if Version(mdata["deepmd_version"]) < Version("2.0"):
        raise RuntimeError(
            "Only support deepmd-kit v2 or above for model_devi_engine='gromacs'"
        )
    model_devi_jobs = jdata["model_devi_jobs"]
    if iter_index >= len(model_devi_jobs):
        return False
    cur_job = model_devi_jobs[iter_index]
    dt = cur_job.get("dt", None)
    if dt is not None:
        model_devi_dt = dt
    else:
        model_devi_dt = jdata["model_devi_dt"]
    nsteps = cur_job.get("nsteps", None)
    lambdas = cur_job.get("lambdas", [1.0])
    temps = cur_job.get("temps", [298.0])

    for ll in lambdas:
        assert ll >= 0.0 and ll <= 1.0, "Lambda should be in [0,1]"

    if nsteps is None:
        raise RuntimeError("nsteps is None, you should set nsteps in model_devi_jobs!")
    # Currently Gromacs engine is not supported for different temperatures!
    # If you want to change temperatures, you should change it in mdp files.

    sys_idx = expand_idx(cur_job["sys_idx"])
    if len(sys_idx) != len(list(set(sys_idx))):
        raise RuntimeError("system index should be uniq")

    mass_map = jdata["mass_map"]

    iter_name = make_iter_name(iter_index)
    train_path = os.path.join(iter_name, train_name)
    train_path = os.path.abspath(train_path)
    suffix = _get_model_suffix(jdata)
    models = sorted(glob.glob(os.path.join(train_path, f"graph*{suffix}")))
    task_model_list = []
    for ii in models:
        task_model_list.append(os.path.join("..", os.path.basename(ii)))
    work_path = os.path.join(iter_name, model_devi_name)

    sys_counter = 0
    for ss in conf_systems:
        conf_counter = 0
        task_counter = 0
        for cc in ss:
            for ll in lambdas:
                for tt in temps:
                    task_name = make_model_devi_task_name(
                        sys_idx[sys_counter], task_counter
                    )
                    task_path = os.path.join(work_path, task_name)
                    create_path(task_path)
                    gromacs_settings = jdata.get("gromacs_settings", "")
                    for key, file in gromacs_settings.items():
                        if (
                            key != "traj_filename"
                            and key != "mdp_filename"
                            and key != "group_name"
                            and key != "maxwarn"
                        ):
                            os.symlink(
                                os.path.join(cc, file), os.path.join(task_path, file)
                            )
                    # input.json for DP-Gromacs
                    with open(os.path.join(cc, "input.json")) as f:
                        input_json = json.load(f)
                    input_json["graph_file"] = models[0]
                    input_json["lambda"] = ll
                    with open(os.path.join(task_path, "input.json"), "w") as _outfile:
                        json.dump(input_json, _outfile, indent=4)

                    # trj_freq
                    trj_freq = cur_job.get("trj_freq", 10)
                    mdp = MDP()
                    mdp.read(os.path.join(cc, gromacs_settings["mdp_filename"]))
                    mdp["nstcomm"] = trj_freq
                    mdp["nstxout"] = trj_freq
                    mdp["nstlog"] = trj_freq
                    mdp["nstenergy"] = trj_freq
                    # dt
                    mdp["dt"] = model_devi_dt
                    # nsteps
                    mdp["nsteps"] = nsteps
                    # temps
                    if "ref_t" in list(mdp.keys()):
                        mdp["ref_t"] = tt
                    else:
                        mdp["ref-t"] = tt
                    mdp.write(os.path.join(task_path, gromacs_settings["mdp_filename"]))

                    cwd_ = os.getcwd()
                    os.chdir(task_path)
                    job = {}
                    job["trj_freq"] = cur_job["trj_freq"]
                    job["model_devi_dt"] = model_devi_dt
                    job["nsteps"] = nsteps
                    with open("job.json", "w") as _outfile:
                        json.dump(job, _outfile, indent=4)
                    os.chdir(cwd_)
                    task_counter += 1
            conf_counter += 1
        sys_counter += 1


def _make_model_devi_amber(
    iter_index: int, jdata: dict, mdata: dict, conf_systems: list
):
    """Make amber's MD inputs.

    Parameters
    ----------
    iter_index : int
        iter index
    jdata : dict
        run parameters. The following parameters will be used in this method:
            model_devi_jobs : list[dict]
                The list including the dict for information of each cycle:
                    sys_idx : list[int]
                        list of systems to run
                    trj_freq : int
                        freq to dump trajectory
            low_level : str
                low level method
            cutoff : float
                cutoff radius of the DPRc model
            parm7_prefix : str
                The path prefix to AMBER PARM7 files
            parm7 : list[str]
                List of paths to AMBER PARM7 files. Each file maps to a system.
            mdin_prefix : str
                The path prefix to AMBER mdin files
            mdin : list[str]
                List of paths to AMBER mdin files. Each files maps to a system.
                The following keywords will be replaced by the actual value:
                    @freq@ : freq to dump trajectory
                    @nstlim@ : total time step to run
                    @qm_region@ : AMBER mask of the QM region
                    @qm_theory@ : The QM theory, such as DFTB2
                    @qm_charge@ : The total charge of the QM theory, such as -2
                    @rcut@ : cutoff radius of the DPRc model
                    @GRAPH_FILE0@, @GRAPH_FILE1@, ... : graph files
            qm_region : list[str]
                AMBER mask of the QM region. Each mask maps to a system.
            qm_charge : list[int]
                Charge of the QM region. Each charge maps to a system.
            nsteps : list[int]
                The number of steps to run. Each number maps to a system.
            r : list[list[float]] or list[list[list[float]]]
                Constrict values for the enhanced sampling. The first dimension maps to systems.
                The second dimension maps to confs in each system. The third dimension is the
                constrict value. It can be a single float for 1D or list of floats for nD.
            disang_prefix : str
                The path prefix to disang prefix.
            disang : list[str]
                List of paths to AMBER disang files. Each file maps to a sytem.
                The keyword RVAL will be replaced by the constrict values, or RVAL1, RVAL2, ...
                for an nD system.
    mdata : dict
        machine parameters. Nothing will be used in this method.
    conf_systems : list
        conf systems

    References
    ----------
    .. [1] Development of Range-Corrected Deep Learning Potentials for Fast, Accurate Quantum
       Mechanical/Molecular Mechanical Simulations of Chemical Reactions in Solution,
       Jinzhe Zeng, Timothy J. Giese, Şölen Ekesan, and Darrin M. York, Journal of Chemical
       Theory and Computation 2021 17 (11), 6993-7009

    inputs: restart (coords), param, mdin, graph, disang (optional)

    """
    model_devi_jobs = jdata["model_devi_jobs"]
    if iter_index >= len(model_devi_jobs):
        return False
    cur_job = model_devi_jobs[iter_index]
    sys_idx = expand_idx(cur_job["sys_idx"])
    if len(sys_idx) != len(list(set(sys_idx))):
        raise RuntimeError("system index should be uniq")

    iter_name = make_iter_name(iter_index)
    train_path = os.path.join(iter_name, train_name)
    train_path = os.path.abspath(train_path)
    work_path = os.path.join(iter_name, model_devi_name)
    # parm7 - list
    parm7 = jdata["parm7"]
    parm7_prefix = jdata.get("parm7_prefix", "")
    parm7 = [os.path.join(parm7_prefix, pp) for pp in parm7]

    # link parm file
    for ii, pp in enumerate(parm7):
        os.symlink(pp, os.path.join(work_path, "qmmm%d.parm7" % ii))  # noqa: UP031
    # TODO: consider writing input in json instead of a given file
    # mdin
    mdin = jdata["mdin"]
    mdin_prefix = jdata.get("mdin_prefix", "")
    mdin = [os.path.join(mdin_prefix, pp) for pp in mdin]

    qm_region = jdata["qm_region"]
    qm_charge = jdata["qm_charge"]
    nsteps = jdata["nsteps"]

    for ii, pp in enumerate(mdin):
        with (
            open(pp) as f,
            open(os.path.join(work_path, "init%d.mdin" % ii), "w") as fw,  # noqa: UP031
        ):
            mdin_str = f.read()
            # freq, nstlim, qm_region, qm_theory, qm_charge, rcut, graph
            mdin_str = (
                mdin_str.replace("@freq@", str(cur_job.get("trj_freq", 50)))
                .replace("@nstlim@", str(nsteps[ii]))
                .replace("@qm_region@", qm_region[ii])
                .replace("@qm_charge@", str(qm_charge[ii]))
                .replace("@qm_theory@", jdata["low_level"])
                .replace("@rcut@", str(jdata["cutoff"]))
            )
            suffix = _get_model_suffix(jdata)
            models = sorted(glob.glob(os.path.join(train_path, f"graph.*{suffix}")))
            task_model_list = []
            for ii in models:
                task_model_list.append(os.path.join("..", os.path.basename(ii)))
            # graph
            for jj, mm in enumerate(task_model_list):
                # replace graph
                mdin_str = mdin_str.replace("@GRAPH_FILE%d@" % jj, mm)  # noqa: UP031
            fw.write(mdin_str)
    # disang - list
    disang = jdata["disang"]
    disang_prefix = jdata.get("disang_prefix", "")
    disang = [os.path.join(disang_prefix, pp) for pp in disang]

    for sys_counter, ss in enumerate(conf_systems):
        for idx_cc, cc in enumerate(ss):
            task_counter = idx_cc
            conf_counter = idx_cc

            task_name = make_model_devi_task_name(sys_idx[sys_counter], task_counter)
            conf_name = make_model_devi_conf_name(sys_idx[sys_counter], conf_counter)
            task_path = os.path.join(work_path, task_name)
            # create task path
            create_path(task_path)
            # link restart file
            loc_conf_name = "init.rst7"
            if cur_job.get("restart_from_iter") is None:
                os.symlink(
                    os.path.join(os.path.join("..", "confs"), conf_name + ".rst7"),
                    os.path.join(task_path, loc_conf_name),
                )
            else:
                restart_from_iter = cur_job["restart_from_iter"]
                restart_iter_name = make_iter_name(restart_from_iter)
                os.symlink(
                    os.path.relpath(
                        os.path.join(
                            restart_iter_name, model_devi_name, task_name, "rc.rst7"
                        ),
                        task_path,
                    ),
                    os.path.join(task_path, loc_conf_name),
                )
            cwd_ = os.getcwd()
            # chdir to task path
            os.chdir(task_path)

            # reaction coordinates of umbrella sampling
            # TODO: maybe consider a better name instead of `r`?
            if "r" in jdata:
                r = jdata["r"][sys_idx[sys_counter]][conf_counter]
                # r can either be a float or a list of float (for 2D coordinates)
                if not isinstance(r, Iterable) or isinstance(r, str):
                    r = [r]
                # disang file should include RVAL, RVAL2, ...
                with (
                    open(disang[sys_idx[sys_counter]]) as f,
                    open("TEMPLATE.disang", "w") as fw,
                ):
                    tl = f.read()
                    for ii, rr in enumerate(r):
                        if isinstance(rr, Iterable) and not isinstance(rr, str):
                            raise RuntimeError(
                                "rr should not be iterable! sys: %d rr: %s r: %s"  # noqa: UP031
                                % (sys_idx[sys_counter], str(rr), str(r))
                            )
                        tl = tl.replace("RVAL" + str(ii + 1), str(rr))
                    if len(r) == 1:
                        tl = tl.replace("RVAL", str(r[0]))
                    fw.write(tl)

            with open("job.json", "w") as fp:
                json.dump(cur_job, fp, indent=4)
            os.chdir(cwd_)


def run_md_model_devi(iter_index, jdata, mdata):
    # rmdlog.info("This module has been run !")
    model_devi_exec = mdata["model_devi_command"]

    model_devi_group_size = mdata["model_devi_group_size"]
    model_devi_resources = mdata["model_devi_resources"]
    use_plm = jdata.get("model_devi_plumed", False)
    use_plm_path = jdata.get("model_devi_plumed_path", False)
    model_devi_merge_traj = jdata.get("model_devi_merge_traj", False)

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, model_devi_name)
    assert os.path.isdir(work_path)

    all_task = glob.glob(os.path.join(work_path, "task.*"))
    all_task.sort()
    fp = open(os.path.join(work_path, "cur_job.json"))
    cur_job = json.load(fp)

    run_tasks_ = all_task
    # for ii in all_task:
    #     fres = os.path.join(ii, 'model_devi.out')
    #     if os.path.isfile(fres) :
    #         nlines = np.loadtxt(fres).shape[0]
    #         if nframes != nlines :
    #             run_tasks_.append(ii)
    #     else :
    #         run_tasks_.append(ii)

    run_tasks = [os.path.basename(ii) for ii in run_tasks_]
    # dlog.info("all_task is ", all_task)
    # dlog.info("run_tasks in run_model_deviation",run_tasks_)

    suffix = _get_model_suffix(jdata)
    all_models = glob.glob(os.path.join(work_path, f"graph*{suffix}"))
    model_names = [os.path.basename(ii) for ii in all_models]

    model_devi_engine = jdata.get("model_devi_engine", "lammps")
    if model_devi_engine == "lammps":
        nbeads = jdata["model_devi_jobs"][iter_index].get("nbeads")
        if nbeads is None:
            command = f"{{ if [ ! -f dpgen.restart.10000 ]; then {model_devi_exec} -i input.lammps -v restart 0; else {model_devi_exec} -i input.lammps -v restart 1; fi }}"
        else:
            command = f"{{ all_exist=true; for i in $(seq -w 1 {nbeads}); do [[ ! -f dpgen.restart${{i}}.10000 ]] && {{ all_exist=false; break; }}; done; $all_exist && {{ {model_devi_exec} -p {nbeads}x1 -i input.lammps -v restart 1; }} || {{ {model_devi_exec} -p {nbeads}x1 -i input.lammps -v restart 0; }} }}"
        command = f"/bin/bash -c {shlex.quote(command)}"
        commands = [command]

        forward_files = ["conf.lmp", "input.lammps"]
        backward_files = ["model_devi.log"]
        if nbeads is None:
            backward_files += ["model_devi.out"]
        else:
            num_digits = np.ceil(np.log10(nbeads + 1)).astype(int)
            backward_files += [
                f"model_devi{i + 1:0{num_digits}d}.out" for i in range(nbeads)
            ]
            backward_files += [f"log.lammps.{i:d}" for i in range(nbeads)]
        if model_devi_merge_traj:
            backward_files += ["all.lammpstrj"]
        else:
            forward_files += ["traj"]
            backward_files += ["traj"]

        if use_plm:
            forward_files += ["input.plumed"]
            # backward_files += ['output.plumed']
            backward_files += ["output.plumed", "COLVAR"]
            if use_plm_path:
                forward_files += ["plmpath.pdb"]
    elif model_devi_engine == "gromacs":
        gromacs_settings = jdata.get("gromacs_settings", {})
        mdp_filename = gromacs_settings.get("mdp_filename", "md.mdp")
        topol_filename = gromacs_settings.get("topol_filename", "processed.top")
        conf_filename = gromacs_settings.get("conf_filename", "conf.gro")
        index_filename = gromacs_settings.get("index_filename", "index.raw")
        type_filename = gromacs_settings.get("type_filename", "type.raw")
        ndx_filename = gromacs_settings.get("ndx_filename", "")
        # Initial reference to process pbc condition.
        # Default is em.tpr
        ref_filename = gromacs_settings.get("ref_filename", "em.tpr")
        deffnm = gromacs_settings.get("deffnm", "deepmd")
        maxwarn = gromacs_settings.get("maxwarn", 1)
        traj_filename = gromacs_settings.get("traj_filename", "deepmd_traj.gro")
        grp_name = gromacs_settings.get("group_name", "Other")
        trj_freq = cur_job.get("trj_freq", 10)

        command = "%s grompp -f %s -p %s -c %s -o %s -maxwarn %d" % (  # noqa: UP031
            model_devi_exec,
            mdp_filename,
            topol_filename,
            conf_filename,
            deffnm,
            maxwarn,
        )
        command += f"&& {model_devi_exec} mdrun -deffnm {deffnm} -cpi"
        if ndx_filename:
            command += f'&& echo -e "{grp_name}\\n{grp_name}\\n" | {model_devi_exec} trjconv -s {ref_filename} -f {deffnm}.trr -n {ndx_filename} -o {traj_filename} -pbc mol -ur compact -center'
        else:
            command += f'&& echo -e "{grp_name}\\n{grp_name}\\n" | {model_devi_exec} trjconv -s {ref_filename} -f {deffnm}.trr -o {traj_filename} -pbc mol -ur compact -center'
        command += "&& if [ ! -d traj ]; then \n mkdir traj; fi\n"
        command += f"python -c \"import dpdata;system = dpdata.System('{traj_filename}', fmt='gromacs/gro'); [system.to_gromacs_gro('traj/%d.gromacstrj' % (i * {trj_freq}), frame_idx=i) for i in range(system.get_nframes())]; system.to_deepmd_npy('traj_deepmd')\""
        _rel_model_names = " ".join([str(os.path.join("..", ii)) for ii in model_names])
        command += f"&& dp model-devi -m {_rel_model_names} -s traj_deepmd -o model_devi.out -f {trj_freq}"
        del _rel_model_names
        commands = [command]

        forward_files = [
            mdp_filename,
            topol_filename,
            conf_filename,
            index_filename,
            ref_filename,
            type_filename,
            "input.json",
            "job.json",
        ]
        if ndx_filename:
            forward_files.append(ndx_filename)
        backward_files = [
            f"{deffnm}.tpr",
            f"{deffnm}.log",
            traj_filename,
            "model_devi.out",
            "traj",
            "traj_deepmd",
        ]
    elif model_devi_engine == "amber":
        commands = [
            ("TASK=$(basename $(pwd)) && SYS1=${TASK:5:3} && SYS=$((10#$SYS1)) && ")
            + model_devi_exec
            + (
                " -O -p ../qmmm$SYS.parm7 -c init.rst7 -i ../init$SYS.mdin -o rc.mdout -r rc.rst7 -x rc.nc -inf rc.mdinfo -ref init.rst7"
            )
        ]
        forward_files = ["init.rst7", "TEMPLATE.disang"]
        backward_files = ["rc.mdout", "rc.nc", "rc.rst7", "TEMPLATE.dumpave"]
        model_names.extend(["qmmm*.parm7", "init*.mdin"])

    cwd = os.getcwd()

    user_forward_files = mdata.get("model_devi" + "_user_forward_files", [])
    forward_files += [os.path.basename(file) for file in user_forward_files]
    backward_files += mdata.get("model_devi" + "_user_backward_files", [])
    if len(run_tasks) == 0:
        raise RuntimeError(
            "run_tasks for model_devi should not be empty! Please check your files."
        )

    ### Submit jobs
    check_api_version(mdata)

    submission = make_submission(
        mdata["model_devi_machine"],
        mdata["model_devi_resources"],
        commands=commands,
        work_path=work_path,
        run_tasks=run_tasks,
        group_size=model_devi_group_size,
        forward_common_files=model_names,
        forward_files=forward_files,
        backward_files=backward_files,
        outlog="model_devi.log",
        errlog="model_devi.log",
    )
    submission.run_submission()


def run_model_devi(iter_index, jdata, mdata):
    model_devi_engine = jdata.get("model_devi_engine", "lammps")
    if model_devi_engine != "calypso":
        run_md_model_devi(iter_index, jdata, mdata)
    else:
        run_calypso_model_devi(iter_index, jdata, mdata)


def post_model_devi(iter_index, jdata, mdata):
    pass


def _to_face_dist(box_):
    box = np.reshape(box_, [3, 3])
    vol = np.abs(np.linalg.det(box))
    dists = []
    for [ii, jj] in [[0, 1], [1, 2], [2, 0]]:
        vv = np.cross(box[ii], box[jj])
        dists.append(vol / np.linalg.norm(vv))
    return np.array(dists)


def check_cluster(conf_name, fp_cluster_vacuum, fmt="lammps/dump"):
    sys = dpdata.System(conf_name, fmt)
    assert sys.get_nframes() == 1
    cell = sys.data["cells"][0]
    coord = sys.data["coords"][0]
    xlim = max(coord[:, 0]) - min(coord[:, 0])
    ylim = max(coord[:, 1]) - min(coord[:, 1])
    zlim = max(coord[:, 2]) - min(coord[:, 2])
    a, b, c = map(norm, [cell[0, :], cell[1, :], cell[2, :]])
    min_vac = min([a - xlim, b - ylim, c - zlim])
    # print([a-xlim,b-ylim,c-zlim])
    # _,r3d=miniball.get_bounding_ball(coord)

    if min_vac < fp_cluster_vacuum:
        is_bad = True
    else:
        is_bad = False
    return is_bad


def check_bad_box(conf_name, criteria, fmt="lammps/dump"):
    all_c = criteria.split(";")
    sys = dpdata.System(conf_name, fmt)
    assert sys.get_nframes() == 1
    is_bad = False
    for ii in all_c:
        [key, value] = ii.split(":")
        if key == "length_ratio":
            lengths = np.linalg.norm(sys["cells"][0], axis=1)
            ratio = np.max(lengths) / np.min(lengths)
            if ratio > float(value):
                is_bad = True
        elif key == "height_ratio":
            lengths = np.linalg.norm(sys["cells"][0], axis=1)
            dists = _to_face_dist(sys["cells"][0])
            ratio = np.max(lengths) / np.min(dists)
            if ratio > float(value):
                is_bad = True
        #
        elif key == "wrap_ratio":
            ratio = [
                sys["cells"][0][1][0] / sys["cells"][0][0][0],
                sys["cells"][0][2][1] / sys["cells"][0][1][1],
                sys["cells"][0][2][0] / sys["cells"][0][0][0],
            ]
            if np.max(np.abs(ratio)) > float(value):
                is_bad = True
        elif key == "tilt_ratio":
            ratio = [
                sys["cells"][0][1][0] / sys["cells"][0][1][1],
                sys["cells"][0][2][1] / sys["cells"][0][2][2],
                sys["cells"][0][2][0] / sys["cells"][0][2][2],
            ]
            if np.max(np.abs(ratio)) > float(value):
                is_bad = True
        else:
            raise RuntimeError("unknow key", key)
    return is_bad


def _read_model_devi_file(
    task_path: str,
    model_devi_f_avg_relative: bool = False,
    model_devi_merge_traj: bool = False,
):
    model_devi_files = glob.glob(os.path.join(task_path, "model_devi*.out"))
    if len(model_devi_files) > 1:
        model_devi_files_sorted = sorted(
            model_devi_files,
            key=lambda x: int(re.search(r"model_devi(\d+)\.out", x).group(1)),
        )
        with open(model_devi_files_sorted[0]) as f:
            first_line = f.readline()
        if not (first_line.startswith("#")):
            first_line = "#"
        num_beads = len(model_devi_files_sorted)
        model_devi_contents = []
        for file in model_devi_files_sorted:
            model_devi_contents.append(np.loadtxt(file))
        assert all(
            model_devi_content.shape[0] == model_devi_contents[0].shape[0]
            for model_devi_content in model_devi_contents
        ), (
            r"Not all beads generated the same number of lines in the model_devi${ibead}.out file. Check your pimd task carefully."
        )
        last_step = model_devi_contents[0][-1, 0]
        for ibead in range(1, num_beads):
            model_devi_contents[ibead][:, 0] = model_devi_contents[ibead][
                :, 0
            ] + ibead * (last_step + 1)
        model_devi = np.concatenate(model_devi_contents, axis=0)
        num_columns = model_devi.shape[1]
        formats = ["%12d"] + ["%22.6e"] * (num_columns - 1)
        np.savetxt(
            os.path.join(task_path, "model_devi.out"),
            model_devi,
            fmt=formats,
            header=first_line.rstrip(),
            comments="",
        )

        if not model_devi_merge_traj:
            num_digits = np.ceil(np.log10(num_beads + 1)).astype(int)
            traj_files_sorted = []
            for ibead in range(num_beads):
                traj_files = glob.glob(
                    os.path.join(
                        task_path, "traj", f"*lammpstrj{ibead + 1:0{num_digits}d}"
                    )
                )
                traj_files_sorted.append(
                    sorted(
                        traj_files,
                        key=lambda x: int(
                            re.search(r"^(\d+)\.lammpstrj", os.path.basename(x)).group(
                                1
                            )
                        ),
                    )
                )
            assert all(
                len(traj_list) == len(traj_files_sorted[0])
                for traj_list in traj_files_sorted
            ), (
                "Not all beads generated the same number of frames. Check your pimd task carefully."
            )
            for ibead in range(num_beads):
                for itraj in range(len(traj_files_sorted[0])):
                    base_path, original_filename = os.path.split(
                        traj_files_sorted[ibead][itraj]
                    )
                    frame_number = int(original_filename.split(".")[0])
                    new_filename = os.path.join(
                        base_path,
                        f"{frame_number + ibead * (int(last_step) + 1):d}.lammpstrj",
                    )
                    os.rename(traj_files_sorted[ibead][itraj], new_filename)
    model_devi = np.loadtxt(os.path.join(task_path, "model_devi.out"))
    if model_devi_f_avg_relative:
        if model_devi_merge_traj is True:
            all_traj = os.path.join(task_path, "all.lammpstrj")
            all_f = get_all_dumped_forces(all_traj)
        else:
            trajs = glob.glob(os.path.join(task_path, "traj", "*.lammpstrj"))
            all_f = []
            for ii in trajs:
                all_f.append(get_dumped_forces(ii))

        all_f = np.array(all_f)
        all_f = all_f.reshape([-1, 3])
        avg_f = np.sqrt(np.average(np.sum(np.square(all_f), axis=1)))
        model_devi[:, 4:7] = model_devi[:, 4:7] / avg_f
        np.savetxt(
            os.path.join(task_path, "model_devi_avgf.out"), model_devi, fmt="%16.6e"
        )
    return model_devi


def _select_by_model_devi_standard(
    modd_system_task: list[str],
    f_trust_lo: float,
    f_trust_hi: float,
    v_trust_lo: float,
    v_trust_hi: float,
    cluster_cutoff: float,
    model_devi_engine: str,
    model_devi_skip: int = 0,
    model_devi_f_avg_relative: bool = False,
    model_devi_merge_traj: bool = False,
    detailed_report_make_fp: bool = True,
):
    if model_devi_engine == "calypso":
        iter_name = modd_system_task[0].split("/")[0]
        _work_path = os.path.join(iter_name, model_devi_name)
        # calypso_run_opt_path = os.path.join(_work_path,calypso_run_opt_name)
        calypso_run_opt_path = glob.glob(f"{_work_path}/{calypso_run_opt_name}.*")[0]
        numofspecies = _parse_calypso_input("NumberOfSpecies", calypso_run_opt_path)
        min_dis = _parse_calypso_dis_mtx(numofspecies, calypso_run_opt_path)
    fp_candidate = []
    fp_rest_accurate = []
    fp_rest_failed = []
    cc = 0
    counter = Counter()
    counter["candidate"] = 0
    counter["failed"] = 0
    counter["accurate"] = 0
    for tt in modd_system_task:
        with warnings.catch_warnings():
            warnings.simplefilter("ignore")
            all_conf = _read_model_devi_file(
                tt, model_devi_f_avg_relative, model_devi_merge_traj
            )

            if all_conf.shape == (7,):
                all_conf = all_conf.reshape(1, all_conf.shape[0])
            elif model_devi_engine == "calypso" and all_conf.shape == (8,):
                all_conf = all_conf.reshape(1, all_conf.shape[0])
            for ii in range(all_conf.shape[0]):
                if all_conf[ii][0] < model_devi_skip:
                    continue
                cc = int(all_conf[ii][0])
                if cluster_cutoff is None:
                    if model_devi_engine == "calypso":
                        if float(all_conf[ii][-1]) <= float(min_dis):
                            if detailed_report_make_fp:
                                fp_rest_failed.append([tt, cc])
                            counter["failed"] += 1
                            continue
                    if (
                        all_conf[ii][1] < v_trust_hi and all_conf[ii][1] >= v_trust_lo
                    ) or (
                        all_conf[ii][4] < f_trust_hi and all_conf[ii][4] >= f_trust_lo
                    ):
                        fp_candidate.append([tt, cc])
                        counter["candidate"] += 1
                    elif (all_conf[ii][1] >= v_trust_hi) or (
                        all_conf[ii][4] >= f_trust_hi
                    ):
                        if detailed_report_make_fp:
                            fp_rest_failed.append([tt, cc])
                        counter["failed"] += 1
                    elif all_conf[ii][1] < v_trust_lo and all_conf[ii][4] < f_trust_lo:
                        if detailed_report_make_fp:
                            fp_rest_accurate.append([tt, cc])
                        counter["accurate"] += 1
                    else:
                        if model_devi_engine == "calypso":
                            dlog.info(
                                "ase opt traj %s frame %d with f devi %f does not belong to either accurate, candidiate and failed "  # noqa: UP031
                                % (tt, ii, all_conf[ii][4])
                            )
                        else:
                            raise RuntimeError(
                                "md traj %s frame %d with f devi %f does not belong to either accurate, candidiate and failed, it should not happen"  # noqa: UP031
                                % (tt, ii, all_conf[ii][4])
                            )
                else:
                    idx_candidate = np.where(
                        np.logical_and(
                            all_conf[ii][7:] < f_trust_hi,
                            all_conf[ii][7:] >= f_trust_lo,
                        )
                    )[0]
                    for jj in idx_candidate:
                        fp_candidate.append([tt, cc, jj])
                    counter["candidate"] += len(idx_candidate)
                    idx_rest_accurate = np.where(all_conf[ii][7:] < f_trust_lo)[0]
                    if detailed_report_make_fp:
                        for jj in idx_rest_accurate:
                            fp_rest_accurate.append([tt, cc, jj])
                    counter["accurate"] += len(idx_rest_accurate)
                    idx_rest_failed = np.where(all_conf[ii][7:] >= f_trust_hi)[0]
                    if detailed_report_make_fp:
                        for jj in idx_rest_failed:
                            fp_rest_failed.append([tt, cc, jj])
                    counter["failed"] += len(idx_rest_failed)

    return fp_rest_accurate, fp_candidate, fp_rest_failed, counter


def _select_by_model_devi_adaptive_trust_low(
    modd_system_task: list[str],
    f_trust_hi: float,
    numb_candi_f: int,
    perc_candi_f: float,
    v_trust_hi: float,
    numb_candi_v: int,
    perc_candi_v: float,
    model_devi_skip: int = 0,
    model_devi_f_avg_relative: bool = False,
    model_devi_merge_traj: bool = False,
):
    """modd_system_task    model deviation tasks belonging to one system
    f_trust_hi
    numb_candi_f        number of candidate due to the f model deviation
    perc_candi_f        percentage of candidate due to the f model deviation
    v_trust_hi
    numb_candi_v        number of candidate due to the v model deviation
    perc_candi_v        percentage of candidate due to the v model deviation
    model_devi_skip.

    Returns
    -------
    accur               the accurate set
    candi               the candidate set
    failed              the failed set
    counter             counters, number of elements in the sets
    f_trust_lo          adapted trust level of f
    v_trust_lo          adapted trust level of v
    """
    idx_v = 1
    idx_f = 4
    accur = set()
    candi = set()
    failed = []
    coll_v = []
    coll_f = []
    for tt in modd_system_task:
        with warnings.catch_warnings():
            warnings.simplefilter("ignore")
            model_devi = _read_model_devi_file(
                tt, model_devi_f_avg_relative, model_devi_merge_traj
            )
            for ii in range(model_devi.shape[0]):
                if model_devi[ii][0] < model_devi_skip:
                    continue
                cc = int(model_devi[ii][0])
                # tt: name of task folder
                # cc: time step of the frame
                md_v = model_devi[ii][idx_v]
                md_f = model_devi[ii][idx_f]
                if md_f > f_trust_hi or md_v > v_trust_hi:
                    failed.append([tt, cc])
                else:
                    coll_v.append([model_devi[ii][idx_v], tt, cc])
                    coll_f.append([model_devi[ii][idx_f], tt, cc])
                    # now accur takes all non-failed frames,
                    # will be substracted by candidate lat er
                    accur.add((tt, cc))
    # sort
    coll_v.sort()
    coll_f.sort()
    assert len(coll_v) == len(coll_f)
    # calcuate numbers
    numb_candi_v = max(numb_candi_v, int(perc_candi_v * 0.01 * len(coll_v)))
    numb_candi_f = max(numb_candi_f, int(perc_candi_f * 0.01 * len(coll_f)))
    # adjust number of candidate
    if len(coll_v) < numb_candi_v:
        numb_candi_v = len(coll_v)
    if len(coll_f) < numb_candi_f:
        numb_candi_f = len(coll_f)
    # compute trust lo
    if numb_candi_v == 0:
        v_trust_lo = v_trust_hi
    else:
        v_trust_lo = coll_v[-numb_candi_v][0]
    if numb_candi_f == 0:
        f_trust_lo = f_trust_hi
    else:
        f_trust_lo = coll_f[-numb_candi_f][0]
    # add to candidate set
    for ii in range(len(coll_v) - numb_candi_v, len(coll_v)):
        candi.add(tuple(coll_v[ii][1:]))
    for ii in range(len(coll_f) - numb_candi_f, len(coll_f)):
        candi.add(tuple(coll_f[ii][1:]))
    # accurate set is substracted by the candidate set
    accur = accur - candi
    # convert to list
    candi = [list(ii) for ii in candi]
    accur = [list(ii) for ii in accur]
    # counters
    counter = Counter()
    counter["candidate"] = len(candi)
    counter["failed"] = len(failed)
    counter["accurate"] = len(accur)

    return accur, candi, failed, counter, f_trust_lo, v_trust_lo


def _make_fp_vasp_inner(
    iter_index,
    modd_path,
    work_path,
    model_devi_skip,
    v_trust_lo,
    v_trust_hi,
    f_trust_lo,
    f_trust_hi,
    fp_task_min,
    fp_task_max,
    fp_link_files,
    type_map,
    jdata,
):
    """iter_index          int             iter index
    modd_path           string          path of model devi
    work_path           string          path of fp
    fp_task_max         int             max number of tasks
    fp_link_files       [string]        linked files for fp, POTCAR for example
    fp_params           map             parameters for fp.
    """
    # --------------------------------------------------------------------------------------------------------------------------------------
    model_devi_engine = jdata.get("model_devi_engine", "lammps")
    if model_devi_engine == "calypso":
        iter_name = work_path.split("/")[0]
        _work_path = os.path.join(iter_name, model_devi_name)
        # calypso_run_opt_path = os.path.join(_work_path,calypso_run_opt_name)
        calypso_run_opt_path = glob.glob(f"{_work_path}/{calypso_run_opt_name}.*")[0]
        numofspecies = _parse_calypso_input("NumberOfSpecies", calypso_run_opt_path)
        min_dis = _parse_calypso_dis_mtx(numofspecies, calypso_run_opt_path)

        calypso_total_fp_num = 300
        modd_path = os.path.join(modd_path, calypso_model_devi_name)
        model_devi_skip = -1
        with open(os.path.join(modd_path, "Model_Devi.out")) as summfile:
            summary = np.loadtxt(summfile)
        summaryfmax = summary[:, -4]
        dis = summary[:, -1]
        acc = np.where((summaryfmax <= f_trust_lo) & (dis > float(min_dis)))
        fail = np.where((summaryfmax > f_trust_hi) | (dis <= float(min_dis)))
        nnan = np.where(np.isnan(summaryfmax))

        acc_num = len(acc[0])
        fail_num = len(fail[0])
        nan_num = len(nnan[0])
        tot = len(summaryfmax) - nan_num
        candi_num = tot - acc_num - fail_num
        dlog.info(
            f"summary  accurate_ratio: {acc_num * 100 / tot:8.4f}%  candidata_ratio: {candi_num * 100 / tot:8.4f}%  failed_ratio: {fail_num * 100 / tot:8.4f}%  in {tot:d} structures"
        )
    # --------------------------------------------------------------------------------------------------------------------------------------

    modd_task = glob.glob(os.path.join(modd_path, "task.*"))
    modd_task.sort()
    system_index = []
    for ii in modd_task:
        system_index.append(os.path.basename(ii).split(".")[1])

    set_tmp = set(system_index)
    system_index = list(set_tmp)
    system_index.sort()

    fp_tasks = []

    charges_recorder = []  # record charges for each fp_task
    charges_map = jdata.get("sys_charges", [])

    cluster_cutoff = jdata.get("cluster_cutoff", None)
    model_devi_adapt_trust_lo = jdata.get("model_devi_adapt_trust_lo", False)
    model_devi_f_avg_relative = jdata.get("model_devi_f_avg_relative", False)
    model_devi_merge_traj = jdata.get("model_devi_merge_traj", False)
    # skip save *.out if detailed_report_make_fp is False, default is True
    detailed_report_make_fp = jdata.get("detailed_report_make_fp", True)
    # skip bad box criteria
    skip_bad_box = jdata.get("fp_skip_bad_box")
    # skip discrete structure in cluster
    fp_cluster_vacuum = jdata.get("fp_cluster_vacuum", None)

    def _trust_limitation_check(sys_idx, lim):
        if isinstance(lim, list):
            sys_lim = lim[sys_idx]
        elif isinstance(lim, dict):
            sys_lim = lim[str(sys_idx)]
        else:
            sys_lim = lim
        return sys_lim

    for ss in system_index:
        modd_system_glob = os.path.join(modd_path, "task." + ss + ".*")
        modd_system_task = glob.glob(modd_system_glob)
        modd_system_task.sort()
        if model_devi_engine in ("lammps", "gromacs", "calypso"):
            # convert global trust limitations to local ones
            f_trust_lo_sys = _trust_limitation_check(int(ss), f_trust_lo)
            f_trust_hi_sys = _trust_limitation_check(int(ss), f_trust_hi)
            v_trust_lo_sys = _trust_limitation_check(int(ss), v_trust_lo)
            v_trust_hi_sys = _trust_limitation_check(int(ss), v_trust_hi)

            # assumed e -> v
            if not model_devi_adapt_trust_lo:
                (
                    fp_rest_accurate,
                    fp_candidate,
                    fp_rest_failed,
                    counter,
                ) = _select_by_model_devi_standard(
                    modd_system_task,
                    f_trust_lo_sys,
                    f_trust_hi_sys,
                    v_trust_lo_sys,
                    v_trust_hi_sys,
                    cluster_cutoff,
                    model_devi_engine,
                    model_devi_skip,
                    model_devi_f_avg_relative=model_devi_f_avg_relative,
                    model_devi_merge_traj=model_devi_merge_traj,
                    detailed_report_make_fp=detailed_report_make_fp,
                )
            else:
                numb_candi_f = jdata.get("model_devi_numb_candi_f", 10)
                numb_candi_v = jdata.get("model_devi_numb_candi_v", 0)
                perc_candi_f = jdata.get("model_devi_perc_candi_f", 0.0)
                perc_candi_v = jdata.get("model_devi_perc_candi_v", 0.0)
                (
                    fp_rest_accurate,
                    fp_candidate,
                    fp_rest_failed,
                    counter,
                    f_trust_lo_ad,
                    v_trust_lo_ad,
                ) = _select_by_model_devi_adaptive_trust_low(
                    modd_system_task,
                    f_trust_hi_sys,
                    numb_candi_f,
                    perc_candi_f,
                    v_trust_hi_sys,
                    numb_candi_v,
                    perc_candi_v,
                    model_devi_skip=model_devi_skip,
                    model_devi_f_avg_relative=model_devi_f_avg_relative,
                    model_devi_merge_traj=model_devi_merge_traj,
                )
                dlog.info(
                    "system {:s} {:9s} : f_trust_lo {:6.3f}   v_trust_lo {:6.3f}".format(
                        ss, "adapted", f_trust_lo_ad, v_trust_lo_ad
                    )
                )
        elif model_devi_engine == "amber":
            counter = Counter()
            counter["candidate"] = 0
            counter["failed"] = 0
            counter["accurate"] = 0
            fp_rest_accurate = []
            fp_candidate = []
            fp_rest_failed = []
            for tt in modd_system_task:
                cc = 0
                with open(os.path.join(tt, "rc.mdout")) as f:
                    skip_first = False
                    first_active = True
                    for line in f:
                        if line.startswith("     ntx     =       1"):
                            skip_first = True
                        if line.startswith(
                            "Active learning frame written with max. frc. std.:"
                        ):
                            if skip_first and first_active:
                                first_active = False
                                continue
                            model_devi = (
                                float(line.split()[-2])
                                * dpdata.unit.EnergyConversion("kcal_mol", "eV").value()
                            )
                            if model_devi < f_trust_lo:
                                # accurate
                                if detailed_report_make_fp:
                                    fp_rest_accurate.append([tt, cc])
                                counter["accurate"] += 1
                            elif model_devi > f_trust_hi:
                                # failed
                                if detailed_report_make_fp:
                                    fp_rest_failed.append([tt, cc])
                                counter["failed"] += 1
                            else:
                                # candidate
                                fp_candidate.append([tt, cc])
                                counter["candidate"] += 1
                            cc += 1

        else:
            raise RuntimeError("unknown model_devi_engine", model_devi_engine)

        # print a report
        fp_sum = sum(counter.values())

        if fp_sum == 0:
            dlog.info(f"system {ss:s} has no fp task, maybe the model devi is nan %")
            continue
        for cc_key, cc_value in counter.items():
            dlog.info(
                f"system {ss:s} {cc_key:9s} : {cc_value:6d} in {fp_sum:6d} {cc_value / fp_sum * 100:6.2f} %"
            )
        random.shuffle(fp_candidate)
        if detailed_report_make_fp:
            random.shuffle(fp_rest_failed)
            random.shuffle(fp_rest_accurate)
            with open(
                os.path.join(work_path, f"candidate.shuffled.{ss}.out"), "w"
            ) as fp:
                for ii in fp_candidate:
                    fp.write(" ".join([str(nn) for nn in ii]) + "\n")
            with open(
                os.path.join(work_path, f"rest_accurate.shuffled.{ss}.out"), "w"
            ) as fp:
                for ii in fp_rest_accurate:
                    fp.write(" ".join([str(nn) for nn in ii]) + "\n")
            with open(
                os.path.join(work_path, f"rest_failed.shuffled.{ss}.out"), "w"
            ) as fp:
                for ii in fp_rest_failed:
                    fp.write(" ".join([str(nn) for nn in ii]) + "\n")

        # set number of tasks
        accurate_ratio = float(counter["accurate"]) / float(fp_sum)
        fp_accurate_threshold = jdata.get("fp_accurate_threshold", 1)
        fp_accurate_soft_threshold = jdata.get(
            "fp_accurate_soft_threshold", fp_accurate_threshold
        )
        if accurate_ratio < fp_accurate_soft_threshold:
            this_fp_task_max = fp_task_max
        elif (
            accurate_ratio >= fp_accurate_soft_threshold
            and accurate_ratio < fp_accurate_threshold
        ):
            this_fp_task_max = int(
                fp_task_max
                * (accurate_ratio - fp_accurate_threshold)
                / (fp_accurate_soft_threshold - fp_accurate_threshold)
            )
        else:
            this_fp_task_max = 0
        # ----------------------------------------------------------------------------
        if model_devi_engine == "calypso":
            calypso_intend_fp_num_temp = (
                len(fp_candidate) / candi_num
            ) * calypso_total_fp_num
            if calypso_intend_fp_num_temp < 1:
                calypso_intend_fp_num = 1
            else:
                calypso_intend_fp_num = int(calypso_intend_fp_num_temp)
        # ----------------------------------------------------------------------------
        numb_task = min(this_fp_task_max, len(fp_candidate))
        if numb_task < fp_task_min:
            numb_task = 0

        # ----------------------------------------------------------------------------
        if (model_devi_engine == "calypso" and len(jdata.get("type_map")) == 1) or (
            model_devi_engine == "calypso"
            and len(jdata.get("type_map")) > 1
            and candi_num <= calypso_total_fp_num
        ):
            numb_task = min(this_fp_task_max, len(fp_candidate))
            if numb_task < fp_task_min:
                numb_task = 0
        elif (
            model_devi_engine == "calypso"
            and len(jdata.get("type_map")) > 1
            and candi_num > calypso_total_fp_num
        ):
            numb_task = calypso_intend_fp_num
            if len(fp_candidate) < numb_task:
                numb_task = 0
        # ----------------------------------------------------------------------------
        dlog.info(
            f"system {ss:s} accurate_ratio: {accurate_ratio:8.4f}    thresholds: {fp_accurate_soft_threshold:6.4f} and {fp_accurate_threshold:6.4f}   eff. task min and max {fp_task_min:4d} {this_fp_task_max:4d}   number of fp tasks: {numb_task:6d}"
        )
        # make fp tasks

        # read all.lammpstrj, save in all_sys for each system_index
        all_sys = []
        trj_freq = None
        netcdftraj = None
        if model_devi_merge_traj:
            for ii in modd_system_task:
                all_traj = os.path.join(ii, "all.lammpstrj")
                all_sys_per_task = dpdata.System(
                    all_traj, fmt="lammps/dump", type_map=type_map
                )
                all_sys.append(all_sys_per_task)
            model_devi_jobs = jdata["model_devi_jobs"]
            cur_job = model_devi_jobs[iter_index]
            trj_freq = int(
                _get_param_alias(cur_job, ["t_freq", "trj_freq", "traj_freq"])
            )

        count_bad_box = 0
        count_bad_cluster = 0
        fp_candidate = sorted(fp_candidate[:numb_task])

        for cc in range(numb_task):
            tt = fp_candidate[cc][0]
            ii = fp_candidate[cc][1]
            ss = os.path.basename(tt).split(".")[1]
            conf_name = os.path.join(tt, "traj")
            conf_sys = None
            if model_devi_engine == "lammps":
                if model_devi_merge_traj:
                    conf_sys = all_sys[int(os.path.basename(tt).split(".")[-1])][
                        int(int(ii) / trj_freq)
                    ]
                else:
                    conf_name = os.path.join(conf_name, str(ii) + ".lammpstrj")
                ffmt = "lammps/dump"
            elif model_devi_engine == "gromacs":
                conf_name = os.path.join(conf_name, str(ii) + ".gromacstrj")
                ffmt = "lammps/dump"
            elif model_devi_engine == "amber":
                conf_name = os.path.join(tt, "rc.nc")
                rst_name = os.path.abspath(os.path.join(tt, "init.rst7"))
            elif model_devi_engine == "calypso":
                conf_name = os.path.join(conf_name, str(ii) + ".poscar")
                ffmt = "vasp/poscar"
            else:
                raise RuntimeError("unknown model_devi engine", model_devi_engine)
            conf_name = os.path.abspath(conf_name)
            if skip_bad_box is not None:
                skip = check_bad_box(conf_name, skip_bad_box, fmt=ffmt)
                if skip:
                    count_bad_box += 1
                    continue

            if fp_cluster_vacuum is not None:
                assert fp_cluster_vacuum > 0
                skip_cluster = check_cluster(conf_name, fp_cluster_vacuum)
                if skip_cluster:
                    count_bad_cluster += 1
                    continue

            if model_devi_engine != "calypso":
                # link job.json
                job_name = os.path.join(tt, "job.json")
                job_name = os.path.abspath(job_name)

            if cluster_cutoff is not None:
                # take clusters
                jj = fp_candidate[cc][2]
                poscar_name = f"{conf_name}.cluster.{jj}.POSCAR"
                new_system = take_cluster(conf_name, type_map, jj, jdata)
                new_system.to_vasp_poscar(poscar_name)
            fp_task_name = make_fp_task_name(int(ss), cc)
            fp_task_path = os.path.join(work_path, fp_task_name)
            create_path(fp_task_path)
            fp_tasks.append(fp_task_path)
            if charges_map:
                charges_recorder.append(charges_map[int(ss)])
            cwd = os.getcwd()
            os.chdir(fp_task_path)
            if cluster_cutoff is None:
                if model_devi_engine == "lammps":
                    if model_devi_merge_traj:
                        conf_sys.to("lammps/lmp", "conf.dump")
                    else:
                        os.symlink(os.path.relpath(conf_name), "conf.dump")
                    os.symlink(os.path.relpath(job_name), "job.json")
                elif model_devi_engine == "gromacs":
                    os.symlink(os.path.relpath(conf_name), "conf.dump")
                    os.symlink(os.path.relpath(job_name), "job.json")
                elif model_devi_engine == "amber":
                    # read and write with ase
                    from ase.io.netcdftrajectory import (
                        NetCDFTrajectory,
                        write_netcdftrajectory,
                    )

                    if cc > 0 and tt == fp_candidate[cc - 1][0]:
                        # same MD task, use the same file
                        pass
                    else:
                        # not the same file
                        if cc > 0:
                            # close the old file
                            netcdftraj.close()
                        netcdftraj = NetCDFTrajectory(conf_name)
                    # write nc file
                    write_netcdftrajectory("rc.nc", netcdftraj[ii])
                    if cc >= numb_task - 1:
                        netcdftraj.close()
                    # link restart since it's necessary to start Amber
                    os.symlink(os.path.relpath(rst_name), "init.rst7")
                    os.symlink(os.path.relpath(job_name), "job.json")
                elif model_devi_engine == "calypso":
                    os.symlink(os.path.relpath(conf_name), "POSCAR")
                    fjob = open("job.json", "w+")
                    fjob.write('{"model_devi_engine":"calypso"}')
                    fjob.close()
                    # os.system('touch job.json')
                else:
                    raise RuntimeError("unknown model_devi_engine", model_devi_engine)
            else:
                os.symlink(os.path.relpath(poscar_name), "POSCAR")
                np.save("atom_pref", new_system.data["atom_pref"])
            for pair in fp_link_files:
                os.symlink(pair[0], pair[1])
            os.chdir(cwd)
        if count_bad_box > 0:
            dlog.info(
                f"system {ss:s} skipped {count_bad_box:6d} confs with bad box, {numb_task - count_bad_box:6d} remains"
            )
        if count_bad_cluster > 0:
            dlog.info(
                f"system {ss:s} skipped {count_bad_cluster:6d} confs with bad cluster, {numb_task - count_bad_cluster:6d} remains"
            )
    if model_devi_engine == "calypso":
        dlog.info(
            f"summary  accurate_ratio: {acc_num * 100 / tot:8.4f}%  candidata_ratio: {candi_num * 100 / tot:8.4f}%  failed_ratio: {fail_num * 100 / tot:8.4f}%  in {tot:d} structures"
        )
    if cluster_cutoff is None:
        cwd = os.getcwd()
        for idx, task in enumerate(fp_tasks):
            os.chdir(task)
            if model_devi_engine == "lammps":
                sys = None
                if model_devi_merge_traj:
                    sys = dpdata.System(
                        "conf.dump", fmt="lammps/lmp", type_map=type_map
                    )
                else:
                    sys = dpdata.System(
                        "conf.dump", fmt="lammps/dump", type_map=type_map
                    )
                sys.to_vasp_poscar("POSCAR")
                # dump to poscar

                if charges_map:
                    warnings.warn(
                        '"sys_charges" keyword only support for gromacs engine now.'
                    )
            elif model_devi_engine == "gromacs":
                # dump_to_poscar('conf.dump', 'POSCAR', type_map, fmt = "gromacs/gro")
                if charges_map:
                    dump_to_deepmd_raw(
                        "conf.dump",
                        "deepmd.raw",
                        type_map,
                        fmt="gromacs/gro",
                        charge=charges_recorder[idx],
                    )
                else:
                    dump_to_deepmd_raw(
                        "conf.dump",
                        "deepmd.raw",
                        type_map,
                        fmt="gromacs/gro",
                        charge=None,
                    )
            elif model_devi_engine in ("amber", "calypso"):
                pass
            else:
                raise RuntimeError("unknown model_devi engine", model_devi_engine)
            os.chdir(cwd)
    return fp_tasks


def make_vasp_incar(jdata, filename):
    if "fp_incar" in jdata.keys():
        fp_incar_path = jdata["fp_incar"]
        assert os.path.exists(fp_incar_path)
        fp_incar_path = os.path.abspath(fp_incar_path)
        fr = open(fp_incar_path)
        incar = fr.read()
        fr.close()
    elif "user_fp_params" in jdata.keys():
        incar = write_incar_dict(jdata["user_fp_params"])
    else:
        incar = make_vasp_incar_user_dict(jdata["fp_params"])
    with open(filename, "w") as fp:
        fp.write(incar)
    return incar


def make_pwmat_input(jdata, filename):
    if "fp_incar" in jdata.keys():
        fp_incar_path = jdata["fp_incar"]
        assert os.path.exists(fp_incar_path)
        fp_incar_path = os.path.abspath(fp_incar_path)
        fr = open(fp_incar_path)
        input = fr.read()
        fr.close()
    elif "user_fp_params" in jdata.keys():
        fp_params = jdata["user_fp_params"]
        node1 = fp_params["node1"]
        node2 = fp_params["node2"]
        atom_config = fp_params["in.atom"]
        ecut = fp_params["ecut"]
        e_error = fp_params["e_error"]
        rho_error = fp_params["rho_error"]
        kspacing = fp_params["kspacing"]
        flag_symm = fp_params["flag_symm"]
        os.system("command -v poscar2config.x | wc -l > 1.txt")
        fc = open("1.txt")
        flag_command = fc.read()
        fc.close()
        if int(flag_command) == 1:
            os.system("poscar2config.x < POSCAR > tmp.config")
        else:
            os.system(
                "cp ../../../out_data_post_fp_pwmat/02.fp/task.000.000000/poscar2config.x ./"
            )
            os.system("./poscar2config.x < POSCAR > tmp.config")
        os.system("rm -rf tmp.config")
        input_dict = make_pwmat_input_dict(
            node1,
            node2,
            atom_config,
            ecut,
            e_error,
            rho_error,
            icmix=None,
            smearing=None,
            sigma=None,
            kspacing=kspacing,
            flag_symm=flag_symm,
        )

        input = write_input_dict(input_dict)
    else:
        input = make_pwmat_input_user_dict(jdata["fp_params"])
    if "IN.PSP" in input or "in.psp" in input:
        with open(filename, "w") as fp:
            fp.write(input)
            fp.write("job=scf\n")
            if "OUT.MLMD" in input or "out.mlmd" in input:
                return input
            else:
                fp.write("OUT.MLMD = T")
                return input
    else:
        with open(filename, "w") as fp:
            fp.write(input)
            fp.write("job=scf\n")
            fp_pp_files = jdata["fp_pp_files"]
            for idx, ii in enumerate(fp_pp_files):
                fp.write("IN.PSP%d = %s\n" % (idx + 1, ii))  # noqa: UP031
            if "OUT.MLMD" in input or "out.mlmd" in input:
                return input
            else:
                fp.write("OUT.MLMD = T")
                return input


def make_vasp_incar_ele_temp(jdata, filename, ele_temp, nbands_esti=None):
    from pymatgen.io.vasp import Incar

    with open(filename) as fp:
        incar = fp.read()
    try:
        incar = incar_upper(Incar.from_string(incar))
    except AttributeError:
        incar = incar_upper(Incar.from_str(incar))
    incar["ISMEAR"] = -1
    incar["SIGMA"] = ele_temp * pc.Boltzmann / pc.electron_volt
    incar.write_file("INCAR")
    if nbands_esti is not None:
        nbands = nbands_esti.predict(".")
        with open(filename) as fp:
            try:
                incar = Incar.from_string(fp.read())
            except AttributeError:
                incar = Incar.from_str(fp.read())
        incar["NBANDS"] = nbands
        incar.write_file("INCAR")


def make_fp_vasp_incar(iter_index, jdata, nbands_esti=None):
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return
    cwd = os.getcwd()
    for ii in fp_tasks:
        os.chdir(ii)
        make_vasp_incar(jdata, "INCAR")
        if os.path.exists("job.json"):
            with open("job.json") as fp:
                job_data = json.load(fp)
            if "ele_temp" in job_data:
                make_vasp_incar_ele_temp(
                    jdata, "INCAR", job_data["ele_temp"], nbands_esti=nbands_esti
                )
        os.chdir(cwd)


def _make_fp_pwmat_input(iter_index, jdata):
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return
    cwd = os.getcwd()
    for ii in fp_tasks:
        os.chdir(ii)
        make_pwmat_input(jdata, "etot.input")
        os.system("sed -i '1,2c 4 1' etot.input")
        os.chdir(cwd)


def make_fp_vasp_cp_cvasp(iter_index, jdata):
    # Move cvasp interface to jdata
    if ("cvasp" in jdata) and (jdata["cvasp"] is True):
        pass
    else:
        return
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return
    cwd = os.getcwd()
    for ii in fp_tasks:
        os.chdir(ii)
        # copy cvasp.py
        shutil.copyfile(cvasp_file, "cvasp.py")
        os.chdir(cwd)


def make_fp_vasp_kp(iter_index, jdata):
    from pymatgen.io.vasp import Incar, Kpoints

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_aniso_kspacing = jdata.get("fp_aniso_kspacing")

    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return
    cwd = os.getcwd()
    for ii in fp_tasks:
        os.chdir(ii)
        # get kspacing and kgamma from incar
        assert os.path.exists("INCAR")
        with open("INCAR") as fp:
            incar = fp.read()
        try:
            standard_incar = incar_upper(Incar.from_string(incar))
        except AttributeError:
            standard_incar = incar_upper(Incar.from_str(incar))
        if fp_aniso_kspacing is None:
            try:
                kspacing = standard_incar["KSPACING"]
            except KeyError:
                raise RuntimeError("KSPACING must be given in INCAR")
        else:
            kspacing = fp_aniso_kspacing
        try:
            gamma = standard_incar["KGAMMA"]
            if isinstance(gamma, bool):
                pass
            else:
                if gamma[0].upper() == "T":
                    gamma = True
                else:
                    gamma = False
        except KeyError:
            raise RuntimeError("KGAMMA must be given in INCAR")
        # check poscar
        assert os.path.exists("POSCAR")
        # make kpoints
        ret = make_kspacing_kpoints("POSCAR", kspacing, gamma)
        try:
            kp = Kpoints.from_string(ret)
        except AttributeError:
            kp = Kpoints.from_str(ret)
        kp.write_file("KPOINTS")
        os.chdir(cwd)


def _link_fp_vasp_pp(iter_index, jdata):
    fp_pp_path = jdata["fp_pp_path"]
    fp_pp_files = jdata["fp_pp_files"]
    assert os.path.exists(fp_pp_path)
    fp_pp_path = os.path.abspath(fp_pp_path)

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)

    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return
    cwd = os.getcwd()
    for ii in fp_tasks:
        os.chdir(ii)
        for jj in fp_pp_files:
            pp_file = os.path.join(fp_pp_path, jj)
            os.symlink(pp_file, jj)
        os.chdir(cwd)


def sys_link_fp_vasp_pp(iter_index, jdata):
    fp_pp_path = jdata["fp_pp_path"]
    fp_pp_files = jdata["fp_pp_files"]
    fp_pp_path = os.path.abspath(fp_pp_path)
    type_map = jdata["type_map"]
    assert os.path.exists(fp_pp_path)
    assert len(fp_pp_files) == len(type_map), (
        "size of fp_pp_files should be the same as the size of type_map"
    )

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)

    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return

    system_idx_str = [os.path.basename(ii).split(".")[1] for ii in fp_tasks]
    system_idx_str = list(set(system_idx_str))
    system_idx_str.sort()
    for ii in system_idx_str:
        potcars = []
        sys_tasks = glob.glob(os.path.join(work_path, f"task.{ii}.*"))
        assert len(sys_tasks) != 0
        sys_poscar = os.path.join(sys_tasks[0], "POSCAR")
        sys = dpdata.System(sys_poscar, fmt="vasp/poscar")
        for ele_name in sys["atom_names"]:
            ele_idx = jdata["type_map"].index(ele_name)
            potcars.append(fp_pp_files[ele_idx])
        with open(os.path.join(work_path, f"POTCAR.{ii}"), "w") as fp_pot:
            for jj in potcars:
                with open(os.path.join(fp_pp_path, jj)) as fp:
                    fp_pot.write(fp.read())
        sys_tasks = glob.glob(os.path.join(work_path, f"task.{ii}.*"))
        cwd = os.getcwd()
        for jj in sys_tasks:
            os.chdir(jj)
            os.symlink(os.path.join("..", f"POTCAR.{ii}"), "POTCAR")
            os.chdir(cwd)


def _link_fp_abacus_pporb_descript(iter_index, jdata):
    # assume pp orbital files, numerical descrptors and model for dpks are all in fp_pp_path.
    fp_pp_path = os.path.abspath(jdata["fp_pp_path"])
    type_map = jdata["type_map"]

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return

    cwd = os.getcwd()
    for ii in fp_tasks:
        os.chdir(ii)

        # get value of 'deepks_model' from INPUT
        input_param = get_abacus_input_parameters("INPUT")
        fp_dpks_model = input_param.get("deepks_model", None)
        if fp_dpks_model is not None:
            model_file = os.path.join(
                fp_pp_path, os.path.split(fp_dpks_model)[1]
            )  # only the filename
            assert os.path.isfile(model_file), (
                f"Can not find the deepks model file {model_file}, which is defined in {ii}/INPUT"
            )
            os.symlink(model_file, fp_dpks_model)  # link to the model file

        # get pp, orb, descriptor filenames from STRU
        stru_param = get_abacus_STRU("STRU")
        atom_names = stru_param["atom_names"]
        pp_files_stru = stru_param.get("pp_files", None)
        orb_files_stru = stru_param.get("orb_files", None)
        descriptor_file_stru = stru_param.get("dpks_descriptor", None)

        if pp_files_stru:
            assert "fp_pp_files" in jdata, "need to define fp_pp_files in jdata"
        if orb_files_stru:
            assert "fp_orb_files" in jdata, "need to define fp_orb_files in jdata"
        if descriptor_file_stru:
            assert "fp_dpks_descriptor" in jdata, (
                "need to define fp_dpks_descriptor in jdata"
            )

        for idx, iatom in enumerate(atom_names):
            type_map_idx = type_map.index(iatom)
            if iatom not in type_map:
                raise RuntimeError(
                    f"atom name {iatom} in STRU is not defined in type_map"
                )
            if pp_files_stru:
                src_file = os.path.join(fp_pp_path, jdata["fp_pp_files"][type_map_idx])
                assert os.path.isfile(src_file), (
                    f"Can not find the pseudopotential file {src_file}"
                )
                os.symlink(src_file, pp_files_stru[idx])
            if orb_files_stru:
                src_file = os.path.join(fp_pp_path, jdata["fp_orb_files"][type_map_idx])
                assert os.path.isfile(src_file), (
                    f"Can not find the orbital file {src_file}"
                )
                os.symlink(src_file, orb_files_stru[idx])
        if descriptor_file_stru:
            src_file = os.path.join(fp_pp_path, jdata["fp_dpks_descriptor"])
            assert os.path.isfile(src_file), (
                f"Can not find the descriptor file {src_file}"
            )
            os.symlink(src_file, descriptor_file_stru)

        os.chdir(cwd)


def _make_fp_vasp_configs(iter_index: int, jdata: dict):
    """Read the model deviation from model_devi step, and then generate the candidated structures
    in 02.fp directory.

    Currently, the formats of generated structures are decided by model_devi_eigne.

    Parameters
    ----------
    iter_index : int
        The index of iteration.
    jdata : dict
        The json data.

    Returns
    -------
    int
        The number of the candidated structures.
    """
    # TODO: we need to unify different data formats
    fp_task_max = jdata["fp_task_max"]
    model_devi_skip = jdata["model_devi_skip"]
    type_map = jdata["type_map"]
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    create_path(work_path)

    modd_path = os.path.join(iter_name, model_devi_name)
    task_min = -1
    if os.path.isfile(os.path.join(modd_path, "cur_job.json")):
        cur_job = json.load(open(os.path.join(modd_path, "cur_job.json")))
        if "task_min" in cur_job:
            task_min = cur_job["task_min"]
    else:
        cur_job = {}
    # support iteration dependent trust levels
    v_trust_lo = cur_job.get(
        "model_devi_v_trust_lo", jdata.get("model_devi_v_trust_lo", 1e10)
    )
    v_trust_hi = cur_job.get(
        "model_devi_v_trust_hi", jdata.get("model_devi_v_trust_hi", 1e10)
    )
    if cur_job.get("model_devi_f_trust_lo") is not None:
        f_trust_lo = cur_job.get("model_devi_f_trust_lo")
    else:
        f_trust_lo = jdata["model_devi_f_trust_lo"]
    if cur_job.get("model_devi_f_trust_hi") is not None:
        f_trust_hi = cur_job.get("model_devi_f_trust_hi")
    else:
        f_trust_hi = jdata["model_devi_f_trust_hi"]

    # make configs
    fp_tasks = _make_fp_vasp_inner(
        iter_index,
        modd_path,
        work_path,
        model_devi_skip,
        v_trust_lo,
        v_trust_hi,
        f_trust_lo,
        f_trust_hi,
        task_min,
        fp_task_max,
        [],
        type_map,
        jdata,
    )
    return fp_tasks


def make_fp_vasp(iter_index, jdata):
    # abs path for fp_incar if it exists
    if "fp_incar" in jdata:
        jdata["fp_incar"] = os.path.abspath(jdata["fp_incar"])
    # get nbands esti if it exists
    if "fp_nbands_esti_data" in jdata:
        nbe = NBandsEsti(jdata["fp_nbands_esti_data"])
    else:
        nbe = None
    # order is critical!
    # 1, create potcar
    sys_link_fp_vasp_pp(iter_index, jdata)
    # 2, create incar
    make_fp_vasp_incar(iter_index, jdata, nbands_esti=nbe)
    # 3, create kpoints
    make_fp_vasp_kp(iter_index, jdata)
    # 4, copy cvasp
    make_fp_vasp_cp_cvasp(iter_index, jdata)


def make_fp_pwscf(iter_index, jdata):
    work_path = os.path.join(make_iter_name(iter_index), fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    # make pwscf input
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_pp_files = jdata["fp_pp_files"]
    if "user_fp_params" in jdata.keys():
        fp_params = jdata["user_fp_params"]
        user_input = True
    else:
        fp_params = jdata["fp_params"]
        user_input = False
    cwd = os.getcwd()
    for ii in fp_tasks:
        os.chdir(ii)
        sys_data = dpdata.System("POSCAR").data
        sys_data["atom_masses"] = []
        pps = []
        for iii in sys_data["atom_names"]:
            sys_data["atom_masses"].append(
                jdata["mass_map"][jdata["type_map"].index(iii)]
            )
            pps.append(fp_pp_files[jdata["type_map"].index(iii)])
        ret = make_pwscf_input(sys_data, pps, fp_params, user_input=user_input)
        with open("input", "w") as fp:
            fp.write(ret)
        os.chdir(cwd)
    # link pp files
    _link_fp_vasp_pp(iter_index, jdata)


def make_fp_abacus_scf(iter_index, jdata):
    work_path = os.path.join(make_iter_name(iter_index), fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    pporb_path = "pporb"
    # make abacus/pw/scf input
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_pp_files = jdata["fp_pp_files"]
    fp_orb_files = None
    fp_dpks_descriptor = None
    # get paramters for writting INPUT file
    fp_params = {}
    if "user_fp_params" in jdata.keys():
        fp_params = jdata["user_fp_params"]
    elif "fp_incar" in jdata.keys():
        fp_input_path = jdata["fp_incar"]
        assert os.path.exists(fp_input_path)
        fp_input_path = os.path.abspath(fp_input_path)
        fp_params = get_abacus_input_parameters(fp_input_path)
    else:
        raise RuntimeError(
            "Set 'user_fp_params' or 'fp_incar' in json file to make INPUT of ABACUS"
        )
    ret_input = make_abacus_scf_input(fp_params, extra_file_path=pporb_path)

    # Get orbital and deepks setting
    if "basis_type" in fp_params:
        if fp_params["basis_type"] == "lcao":
            assert (
                "fp_orb_files" in jdata
                and isinstance(jdata["fp_orb_files"], list)
                and len(jdata["fp_orb_files"]) == len(fp_pp_files)
            )
            fp_orb_files = jdata["fp_orb_files"]
    dpks_out_labels = fp_params.get("deepks_out_labels", 0)
    dpks_scf = fp_params.get("deepks_scf", 0)
    if dpks_out_labels or dpks_scf:
        assert "fp_dpks_descriptor" in jdata and isinstance(
            jdata["fp_dpks_descriptor"], str
        )
        fp_dpks_descriptor = jdata["fp_dpks_descriptor"]

    # get paramters for writting KPT file
    if "kspacing" not in fp_params.keys():
        if "gamma_only" in fp_params.keys():
            if fp_params["gamma_only"] == 1:
                gamma_param = {"k_points": [1, 1, 1, 0, 0, 0]}
                ret_kpt = make_abacus_scf_kpt(gamma_param)
            else:
                if "k_points" in jdata.keys():
                    ret_kpt = make_abacus_scf_kpt(jdata)
                elif "fp_kpt_file" in jdata.keys():
                    fp_kpt_path = jdata["fp_kpt_file"]
                    assert os.path.exists(fp_kpt_path)
                    fp_kpt_path = os.path.abspath(fp_kpt_path)
                    fk = open(fp_kpt_path)
                    ret_kpt = fk.read()
                    fk.close()
                else:
                    raise RuntimeError("Cannot find any k-points information")
        else:
            if "k_points" in jdata.keys():
                ret_kpt = make_abacus_scf_kpt(jdata)
            elif "fp_kpt_file" in jdata.keys():
                fp_kpt_path = jdata["fp_kpt_file"]
                assert os.path.exists(fp_kpt_path)
                fp_kpt_path = os.path.abspath(fp_kpt_path)
                fk = open(fp_kpt_path)
                ret_kpt = fk.read()
                fk.close()
            else:
                gamma_param = {"k_points": [1, 1, 1, 0, 0, 0]}
                ret_kpt = make_abacus_scf_kpt(gamma_param)
                warnings.warn(
                    "Cannot find k-points information, gamma_only will be generated."
                )

    cwd = os.getcwd()
    for ii in fp_tasks:
        os.chdir(ii)
        sys_data = dpdata.System("POSCAR").data
        if "mass_map" in jdata:
            sys_data["atom_masses"] = jdata["mass_map"]
        with open("INPUT", "w") as fp:
            fp.write(ret_input)
        if "kspacing" not in fp_params.keys():
            with open("KPT", "w") as fp:
                fp.write(ret_kpt)
        ret_stru = make_abacus_scf_stru(
            sys_data,
            fp_pp_files,
            fp_orb_files,
            fp_dpks_descriptor,
            type_map=jdata["type_map"],
            pporb=pporb_path,
        )
        with open("STRU", "w") as fp:
            fp.write(ret_stru)

        if not os.path.isdir(pporb_path):
            os.makedirs(pporb_path)

        os.chdir(cwd)
    # link pp and orbital files
    _link_fp_abacus_pporb_descript(iter_index, jdata)


def make_fp_siesta(iter_index, jdata):
    work_path = os.path.join(make_iter_name(iter_index), fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    # make siesta input
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_pp_files = jdata["fp_pp_files"]
    if "user_fp_params" in jdata.keys():
        fp_params = jdata["user_fp_params"]
        user_input = True
    else:
        fp_params = jdata["fp_params"]
        user_input = False
    cwd = os.getcwd()
    for ii in fp_tasks:
        os.chdir(ii)
        sys_data = dpdata.System("POSCAR").data
        ret = make_siesta_input(sys_data, fp_pp_files, fp_params)
        with open("input", "w") as fp:
            fp.write(ret)
        os.chdir(cwd)
    # link pp files
    _link_fp_vasp_pp(iter_index, jdata)


def make_fp_gaussian(iter_index, jdata):
    work_path = os.path.join(make_iter_name(iter_index), fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    # make gaussian gjf file
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    if "user_fp_params" in jdata.keys():
        fp_params = jdata["user_fp_params"]
    else:
        fp_params = jdata["fp_params"]
    cwd = os.getcwd()

    model_devi_engine = jdata.get("model_devi_engine", "lammps")
    for ii in fp_tasks:
        os.chdir(ii)
        if model_devi_engine == "lammps":
            sys_data = dpdata.System("POSCAR").data
        elif model_devi_engine == "gromacs":
            sys_data = dpdata.System("deepmd.raw", fmt="deepmd/raw").data
            if os.path.isfile("deepmd.raw/charge"):
                sys_data["charge"] = int(np.loadtxt("deepmd.raw/charge", dtype=int))
        ret = make_gaussian_input(sys_data, fp_params)
        with open("input", "w") as fp:
            fp.write(ret)
        os.chdir(cwd)


def make_fp_cp2k(iter_index, jdata):
    work_path = os.path.join(make_iter_name(iter_index), fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    # make cp2k input
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    if "user_fp_params" in jdata.keys():
        fp_params = jdata["user_fp_params"]
    # some users might use own inputs
    # specify the input path string
    elif "external_input_path" in jdata.keys():
        fp_params = None
        exinput_path = os.path.abspath(jdata["external_input_path"])
    else:
        fp_params = jdata["fp_params"]
    cwd = os.getcwd()

    # skip bad box criteria
    skip_bad_box = jdata.get("fp_skip_bad_box")
    count_bad_box = 0

    for ii in fp_tasks:
        os.chdir(ii)
        sys_data = dpdata.System("POSCAR").data
        # make input for every task
        # if fp_params exits, make keys
        if skip_bad_box is not None:
            # Check the box directly from POSCAR
            skip = check_bad_box("POSCAR", skip_bad_box, fmt="vasp/poscar")
            if skip:
                count_bad_box += 1
                os.chdir(cwd)
                shutil.rmtree(ii)  # Clean up inconsistent task directory
                continue

        if fp_params:
            cp2k_input = make_cp2k_input(sys_data, fp_params)
        else:
            # else read from user input
            cp2k_input = make_cp2k_input_from_external(sys_data, exinput_path)
        with open("input.inp", "w") as fp:
            fp.write(cp2k_input)
        # make coord.xyz used by cp2k for every task
        cp2k_coord = make_cp2k_xyz(sys_data)
        with open("coord.xyz", "w") as fp:
            fp.write(cp2k_coord)
        os.chdir(cwd)

    if count_bad_box > 0:
        dlog.info(
            f"skipped {count_bad_box:6d} confs with bad box, {len(fp_tasks) - count_bad_box:6d} remains"
        )

    # link pp files
    _link_fp_vasp_pp(iter_index, jdata)


def make_fp_pwmat(iter_index, jdata):
    # abs path for fp_incar if it exists
    if "fp_incar" in jdata:
        jdata["fp_incar"] = os.path.abspath(jdata["fp_incar"])
    # order is critical!
    # 1, link pp files
    _link_fp_vasp_pp(iter_index, jdata)
    # 2, create pwmat input
    _make_fp_pwmat_input(iter_index, jdata)


def make_fp_amber_diff(iter_index: int, jdata: dict):
    """Run amber twice to calculate high-level and low-level potential,
    and then generate difference between them.

    Besides AMBER, one needs to install `dpamber` package, which is avaiable at
    https://github.com/njzjz/dpamber

    Currently, it should be used with the AMBER model_devi driver.

    Parameters
    ----------
    iter_index : int
        iter index
    jdata : dict
        Run parameters. The following parameters are used in this method:
            mdin_prefix : str
                The path prefix to AMBER mdin files
            qm_region : list[str]
                AMBER mask of the QM region. Each mask maps to a system.
            qm_charge : list[int]
                Charge of the QM region. Each charge maps to a system.
            high_level : str
                high level method
            low_level : str
                low level method
            fp_params : dict
                This parameters includes:
                    high_level_mdin : str
                        High-level AMBER mdin file. %qm_theory%, %qm_region%,
                        and %qm_charge% will be replace.
                    low_level_mdin : str
                        Low-level AMBER mdin file. %qm_theory%, %qm_region%,
                        and %qm_charge% will be replace.
            parm7_prefix : str
                The path prefix to AMBER PARM7 files
            parm7 : list[str]
                List of paths to AMBER PARM7 files. Each file maps to a system.

    References
    ----------
    .. [1] Development of Range-Corrected Deep Learning Potentials for Fast, Accurate Quantum
       Mechanical/Molecular Mechanical Simulations of Chemical Reactions in Solution,
       Jinzhe Zeng, Timothy J. Giese, Şölen Ekesan, and Darrin M. York, Journal of Chemical
       Theory and Computation 2021 17 (11), 6993-7009
    """
    assert jdata["model_devi_engine"] == "amber"
    work_path = os.path.join(make_iter_name(iter_index), fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    # make amber input
    cwd = os.getcwd()
    # link two mdin files and param7
    os.chdir(os.path.join(fp_tasks[0], ".."))
    mdin_prefix = jdata.get("mdin_prefix", "")
    low_level_mdin = jdata["fp_params"]["low_level_mdin"]
    low_level_mdin = os.path.join(mdin_prefix, low_level_mdin)
    high_level_mdin = jdata["fp_params"]["high_level_mdin"]
    high_level_mdin = os.path.join(mdin_prefix, high_level_mdin)
    with open(low_level_mdin) as f:
        low_level_mdin_str = f.read()
    with open(high_level_mdin) as f:
        high_level_mdin_str = f.read()

    qm_region = jdata["qm_region"]
    high_level = jdata["high_level"]
    low_level = jdata["low_level"]
    qm_charge = jdata["qm_charge"]
    # qm_theory qm_region qm_charge
    for ii, _ in enumerate(qm_region):
        mdin_new_str = (
            low_level_mdin_str.replace("%qm_theory%", low_level)
            .replace("%qm_region%", qm_region[ii])
            .replace("%qm_charge%", str(qm_charge[ii]))
        )
        with open("low_level%d.mdin" % ii, "w") as f:  # noqa: UP031
            f.write(mdin_new_str)

        mdin_new_str = (
            high_level_mdin_str.replace("%qm_theory%", high_level)
            .replace("%qm_region%", qm_region[ii])
            .replace("%qm_charge%", str(qm_charge[ii]))
        )
        with open("high_level%d.mdin" % ii, "w") as f:  # noqa: UP031
            f.write(mdin_new_str)

    parm7 = jdata["parm7"]
    parm7_prefix = jdata.get("parm7_prefix", "")
    parm7 = [os.path.join(parm7_prefix, pp) for pp in parm7]
    for ii, pp in enumerate(parm7):
        os.symlink(pp, "qmmm%d.parm7" % ii)  # noqa: UP031

    rst7_prefix = jdata.get("sys_configs_prefix", "")
    for ii, ss in enumerate(jdata["sys_configs"]):
        os.symlink(os.path.join(rst7_prefix, ss[0]), "init%d.rst7" % ii)  # noqa: UP031

    with open("qm_region", "w") as f:
        f.write("\n".join(qm_region))
    os.chdir(cwd)


def make_fp_cpx(iter_index, jdata):
    """Make input file for Quantum Espresso Car-Parrinello (cp.x) run.

    Convert the POSCAR file to cp.x input file using prepared template.

    Parameters
    ----------
    iter_index : int
        iter index
    jdata : dict
        Run parameters.
    """
    work_path = os.path.join(make_iter_name(iter_index), fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_params = jdata["fp_params"]
    input_fn = fp_params["input_fn"]
    template_fn = fp_params["template_fn"]

    with open(template_fn) as tn:
        template = tn.read()

    for ii in fp_tasks:
        itemp = copy.deepcopy(template)
        with set_directory(Path(ii)):
            system = dpdata.System("POSCAR", fmt="vasp/poscar")
            # convert POSCAR to cp.in
            cell_param = ""
            for a in system["cells"][0]:
                for v in a:
                    cell_param += f"{v:.16e} "
                cell_param += "\n"
            itemp = itemp.replace("%CELL%", cell_param)

            pos = ""
            ntypes = system.get_ntypes()
            numbs = system.get_atom_numbs()
            names = system.get_atom_names()
            coords = system["coords"][0]
            for t in range(ntypes):
                ts = sum(numbs[:t])
                for a in range(numbs[t]):
                    pos += f"{names[t]:2} {coords[a + ts, 0]:20.16f} {coords[a + ts, 1]:20.16f} {coords[a + ts, 2]:20.16f}\n"
            itemp = itemp.replace("%POSITIONS%", pos)
            with open(input_fn + ".in", "w") as fp:
                fp.write(itemp)


def make_fp_custom(iter_index, jdata):
    """Make input file for customized FP style.

    Convert the POSCAR file to custom format.

    Parameters
    ----------
    iter_index : int
        iter index
    jdata : dict
        Run parameters.
    """
    work_path = os.path.join(make_iter_name(iter_index), fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_params = jdata["fp_params"]
    input_fn = fp_params["input_fn"]
    input_fmt = fp_params["input_fmt"]

    for ii in fp_tasks:
        with set_directory(Path(ii)):
            system = dpdata.System("POSCAR", fmt="vasp/poscar")
            system.to(input_fmt, input_fn)


def make_fp(iter_index, jdata, mdata):
    """Select the candidate strutures and make the input file of FP calculation.

    Parameters
    ----------
    iter_index : int
        iter index
    jdata : dict
        Run parameters.
    mdata : dict
        Machine parameters.
    """
    fp_tasks = _make_fp_vasp_configs(iter_index, jdata)
    if len(fp_tasks) == 0:
        return
    make_fp_calculation(iter_index, jdata, mdata)


def make_fp_calculation(iter_index, jdata, mdata):
    """Make the input file of FP calculation.

    Parameters
    ----------
    iter_index : int
        iter index
    jdata : dict
        Run parameters.
    mdata : dict
        Machine parameters.
    """
    fp_style = jdata["fp_style"]
    if fp_style == "vasp":
        make_fp_vasp(iter_index, jdata)
    elif fp_style == "pwscf":
        make_fp_pwscf(iter_index, jdata)
    elif fp_style == "abacus":
        make_fp_abacus_scf(iter_index, jdata)
    elif fp_style == "siesta":
        make_fp_siesta(iter_index, jdata)
    elif fp_style == "gaussian":
        make_fp_gaussian(iter_index, jdata)
    elif fp_style == "cp2k":
        make_fp_cp2k(iter_index, jdata)
    elif fp_style == "pwmat":
        make_fp_pwmat(iter_index, jdata)
    elif fp_style == "amber/diff":
        make_fp_amber_diff(iter_index, jdata)
    elif fp_style == "cpx":
        make_fp_cpx(iter_index, jdata)
    elif fp_style == "custom":
        make_fp_custom(iter_index, jdata)
    else:
        raise RuntimeError("unsupported fp style")
    # Copy user defined forward_files
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    symlink_user_forward_files(mdata=mdata, task_type="fp", work_path=work_path)


def _vasp_check_fin(ii):
    if os.path.isfile(os.path.join(ii, "OUTCAR")):
        with open(os.path.join(ii, "OUTCAR")) as fp:
            content = fp.read()
            count = content.count("Elapse")
            if count != 1:
                return False
    else:
        return False
    return True


def _qe_check_fin(ii):
    if os.path.isfile(os.path.join(ii, "output")):
        with open(os.path.join(ii, "output")) as fp:
            content = fp.read()
            count = content.count("JOB DONE")
            if count != 1:
                return False
    else:
        return False
    return True


def _abacus_scf_check_fin(ii):
    if os.path.isfile(os.path.join(ii, "OUT.ABACUS/running_scf.log")):
        with open(os.path.join(ii, "OUT.ABACUS/running_scf.log")) as fp:
            content = fp.read()
            count = content.count("!FINAL_ETOT_IS")
            if count != 1:
                return False
    else:
        return False
    return True


def _siesta_check_fin(ii):
    if os.path.isfile(os.path.join(ii, "output")):
        with open(os.path.join(ii, "output")) as fp:
            content = fp.read()
            count = content.count("End of run")
            if count != 1:
                return False
    else:
        return False
    return True


def _gaussian_check_fin(ii):
    if os.path.isfile(os.path.join(ii, "output")):
        with open(os.path.join(ii, "output")) as fp:
            content = fp.read()
            count = content.count("termination")
            if count == 0:
                return False
    else:
        return False
    return True


def _cp2k_check_fin(ii):
    if os.path.isfile(os.path.join(ii, "output")):
        with open(os.path.join(ii, "output")) as fp:
            content = fp.read()
            count = content.count("SCF run converged")
            if count == 0:
                return False
    else:
        return False
    return True


def _pwmat_check_fin(ii):
    if os.path.isfile(os.path.join(ii, "REPORT")):
        with open(os.path.join(ii, "REPORT")) as fp:
            content = fp.read()
            count = content.count("time")
            if count != 1:
                return False
    else:
        return False
    return True


def run_fp_inner(
    iter_index,
    jdata,
    mdata,
    forward_files,
    backward_files,
    check_fin,
    log_file="fp.log",
    forward_common_files=[],
):
    fp_command = mdata["fp_command"]
    fp_group_size = mdata["fp_group_size"]
    fp_resources = mdata["fp_resources"]
    mark_failure = fp_resources.get("mark_failure", False)

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)

    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return

    fp_style = jdata["fp_style"]
    if fp_style == "amber/diff":
        # firstly get sys_idx
        fp_command = (
            (
                "TASK=$(basename $(pwd)) && "
                "SYS1=${TASK:5:3} && "
                "SYS=$((10#$SYS1)) && "
                'QM_REGION=$(awk "NR==$SYS+1" ../qm_region) &&'
            )
            + fp_command
            + (
                " -O -p ../qmmm$SYS.parm7 -c ../init$SYS.rst7 -i ../low_level$SYS.mdin -o low_level.mdout -r low_level.rst7 "
                "-x low_level.nc -y rc.nc -frc low_level.mdfrc -inf low_level.mdinfo && "
            )
            + fp_command
            + (
                " -O -p ../qmmm$SYS.parm7 -c ../init$SYS.rst7 -i ../high_level$SYS.mdin -o high_level.mdout -r high_level.rst7 "
                "-x high_level.nc -y rc.nc -frc high_level.mdfrc -inf high_level.mdinfo && "
            )
            + (
                'dpamber corr --cutoff {:f} --parm7_file ../qmmm$SYS.parm7 --nc rc.nc --hl high_level --ll low_level --qm_region "$QM_REGION"'
            ).format(jdata["cutoff"])
        )

    fp_run_tasks = fp_tasks
    # for ii in fp_tasks :
    #     if not check_fin(ii) :
    #         fp_run_tasks.append(ii)
    run_tasks = [os.path.basename(ii) for ii in fp_run_tasks]

    user_forward_files = mdata.get("fp" + "_user_forward_files", [])
    forward_files += [os.path.basename(file) for file in user_forward_files]
    backward_files += mdata.get("fp" + "_user_backward_files", [])

    ### Submit jobs
    check_api_version(mdata)

    submission = make_submission(
        mdata["fp_machine"],
        mdata["fp_resources"],
        commands=[fp_command],
        work_path=work_path,
        run_tasks=run_tasks,
        group_size=fp_group_size,
        forward_common_files=forward_common_files,
        forward_files=forward_files,
        backward_files=backward_files,
        outlog=log_file,
        errlog=log_file,
    )
    submission.run_submission()


def run_fp(iter_index, jdata, mdata):
    fp_style = jdata["fp_style"]
    fp_pp_files = jdata.get("fp_pp_files", [])

    if fp_style == "vasp":
        forward_files = ["POSCAR", "INCAR", "POTCAR", "KPOINTS"]
        backward_files = ["fp.log", "OUTCAR", "vasprun.xml"]
        # Move cvasp interface to jdata
        if ("cvasp" in jdata) and (jdata["cvasp"] is True):
            mdata["fp_resources"]["cvasp"] = True
        if ("cvasp" in mdata["fp_resources"]) and (
            mdata["fp_resources"]["cvasp"] is True
        ):
            dlog.info("cvasp is on !")
            forward_files.append("cvasp.py")
            forward_common_files = []
        else:
            forward_common_files = []
        run_fp_inner(
            iter_index,
            jdata,
            mdata,
            forward_files,
            backward_files,
            _vasp_check_fin,
            forward_common_files=forward_common_files,
        )
    elif fp_style == "pwscf":
        forward_files = ["input"] + fp_pp_files
        backward_files = ["output"]
        run_fp_inner(
            iter_index,
            jdata,
            mdata,
            forward_files,
            backward_files,
            _qe_check_fin,
            log_file="output",
        )
    elif fp_style == "abacus":
        fp_params = {}
        if "user_fp_params" in jdata.keys():
            fp_params = jdata["user_fp_params"]
        elif "fp_incar" in jdata.keys():
            fp_input_path = jdata["fp_incar"]
            assert os.path.exists(fp_input_path)
            fp_input_path = os.path.abspath(fp_input_path)
            fp_params = get_abacus_input_parameters(fp_input_path)
        forward_files = ["INPUT", "STRU", "pporb"]
        if "kspacing" not in fp_params.keys():
            forward_files.append("KPT")
        backward_files = ["output", "OUT.ABACUS"]
        run_fp_inner(
            iter_index,
            jdata,
            mdata,
            forward_files,
            backward_files,
            _abacus_scf_check_fin,
            log_file="output",
        )
    elif fp_style == "siesta":
        forward_files = ["input"] + fp_pp_files
        backward_files = ["output"]
        run_fp_inner(
            iter_index,
            jdata,
            mdata,
            forward_files,
            backward_files,
            _siesta_check_fin,
            log_file="output",
        )
    elif fp_style == "gaussian":
        forward_files = ["input"]
        backward_files = ["output"]
        run_fp_inner(
            iter_index,
            jdata,
            mdata,
            forward_files,
            backward_files,
            _gaussian_check_fin,
            log_file="output",
        )
    elif fp_style == "cp2k":
        forward_files = ["input.inp", "coord.xyz"]
        backward_files = ["output"]
        run_fp_inner(
            iter_index,
            jdata,
            mdata,
            forward_files,
            backward_files,
            _cp2k_check_fin,
            log_file="output",
        )
    elif fp_style == "pwmat":
        forward_files = ["atom.config", "etot.input"] + fp_pp_files
        backward_files = ["REPORT", "OUT.MLMD", "output"]
        run_fp_inner(
            iter_index,
            jdata,
            mdata,
            forward_files,
            backward_files,
            _pwmat_check_fin,
            log_file="output",
        )
    elif fp_style == "amber/diff":
        forward_files = ["rc.nc"]
        backward_files = [
            "low_level.mdfrc",
            "low_level.mdout",
            "high_level.mdfrc",
            "high_level.mdout",
            "output",
            "dataset",
        ]
        forward_common_files = [
            "low_level*.mdin",
            "high_level*.mdin",
            "qmmm*.parm7",
            "qm_region",
            "init*.rst7",
        ]
        run_fp_inner(
            iter_index,
            jdata,
            mdata,
            forward_files,
            backward_files,
            None,
            log_file="output",
            forward_common_files=forward_common_files,
        )
    elif fp_style == "cpx":
        extensions = [".cel", ".evp", ".for", ".pos", ".str"]
        input_fn = jdata["fp_params"]["input_fn"]
        forward_files = [input_fn + ".in"]
        backward_files = [input_fn + ext for ext in extensions] + ["output"]
        run_fp_inner(
            iter_index,
            jdata,
            mdata,
            forward_files,
            backward_files,
            _qe_check_fin,
            log_file="output",
        )
    elif fp_style == "custom":
        fp_params = jdata["fp_params"]
        forward_files = [fp_params["input_fn"]]
        backward_files = [fp_params["output_fn"]]
        run_fp_inner(
            iter_index,
            jdata,
            mdata,
            forward_files,
            backward_files,
            None,
            log_file="output",
        )
    else:
        raise RuntimeError("unsupported fp style")


def post_fp_check_fail(iter_index, jdata, rfailed=None):
    ratio_failed = rfailed if rfailed else jdata.get("ratio_failed", 0.05)
    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return
    ntask = len(fp_tasks)
    nfail = 0

    # check fail according to the number of collected data
    sys_data = glob.glob(os.path.join(work_path, "data.*"))
    sys_data.sort()
    nframe = 0
    for ii in sys_data:
        sys_paths = expand_sys_str(ii)
        for single_sys in sys_paths:
            sys = dpdata.LabeledSystem(os.path.join(single_sys), fmt="deepmd/npy")
            nframe += len(sys)
    nfail = ntask - nframe

    rfail = float(nfail) / float(ntask)
    dlog.info("failed tasks: %6d in %6d  %6.2f %% " % (nfail, ntask, rfail * 100.0))  # noqa: UP031
    if rfail > ratio_failed:
        raise RuntimeError("find too many unsuccessfully terminated jobs")


def post_fp_vasp(iter_index, jdata, rfailed=None):
    ratio_failed = rfailed if rfailed else jdata.get("ratio_failed", 0.05)
    model_devi_engine = jdata.get("model_devi_engine", "lammps")
    if model_devi_engine != "calypso":
        model_devi_jobs = jdata["model_devi_jobs"]
        assert iter_index < len(model_devi_jobs)
    use_ele_temp = jdata.get("use_ele_temp", 0)

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return

    system_index = []
    for ii in fp_tasks:
        system_index.append(os.path.basename(ii).split(".")[1])
    system_index.sort()
    set_tmp = set(system_index)
    system_index = list(set_tmp)
    system_index.sort()

    cwd = os.getcwd()

    tcount = 0
    icount = 0
    for ss in system_index:
        sys_outcars = glob.glob(os.path.join(work_path, f"task.{ss}.*/OUTCAR"))
        sys_outcars.sort()
        tcount += len(sys_outcars)
        all_sys = None
        all_te = []
        for oo in sys_outcars:
            try:
                _sys = dpdata.LabeledSystem(oo, type_map=jdata["type_map"])
            except Exception:
                dlog.info("Try to parse from vasprun.xml")
                try:
                    _sys = dpdata.LabeledSystem(
                        oo.replace("OUTCAR", "vasprun.xml"), type_map=jdata["type_map"]
                    )
                except Exception:
                    _sys = dpdata.LabeledSystem()
                    dlog.info("Failed fp path: {}".format(oo.replace("OUTCAR", "")))
            if len(_sys) == 1:
                # save ele_temp, if any
                if os.path.exists(oo.replace("OUTCAR", "job.json")):
                    with open(oo.replace("OUTCAR", "job.json")) as fp:
                        job_data = json.load(fp)
                    if "ele_temp" in job_data:
                        assert use_ele_temp
                        ele_temp = job_data["ele_temp"]
                        all_te.append(ele_temp)
                        if use_ele_temp == 0:
                            raise RuntimeError(
                                "should not get ele temp at setting: use_ele_temp == 0"
                            )
                        elif use_ele_temp == 1:
                            _sys.data["fparam"] = np.array(ele_temp).reshape(1, 1)
                        elif use_ele_temp == 2:
                            tile_te = np.tile(ele_temp, [_sys.get_natoms()])
                            _sys.data["aparam"] = tile_te.reshape(
                                1, _sys.get_natoms(), 1
                            )
                        else:
                            raise RuntimeError(
                                "invalid setting of use_ele_temp " + str(use_ele_temp)
                            )
                        # check if ele_temp shape is correct
                        _sys.check_data()
                if all_sys is None:
                    all_sys = dpdata.MultiSystems(_sys, type_map=jdata["type_map"])
                else:
                    all_sys.append(_sys)
            elif len(_sys) >= 2:
                raise RuntimeError("The vasp parameter NSW should be set as 1")
            else:
                icount += 1
        all_te = np.array(all_te)
        if all_sys is not None:
            sys_data_path = os.path.join(work_path, f"data.{ss}")
            all_sys.to_deepmd_raw(sys_data_path)
            all_sys.to_deepmd_npy(sys_data_path, set_size=len(sys_outcars))

    if tcount == 0:
        rfail = 0.0
        dlog.info("failed frame: %6d in %6d " % (icount, tcount))  # noqa: UP031
    else:
        rfail = float(icount) / float(tcount)
        dlog.info(
            "failed frame: %6d in %6d  %6.2f %% " % (icount, tcount, rfail * 100.0)  # noqa: UP031
        )

    if rfail > ratio_failed:
        raise RuntimeError(
            "find too many unsuccessfully terminated jobs. Too many FP tasks are not converged. Please check your input parameters (e.g. INCAR) or configuration (e.g. POSCAR) in directories 'iter.*.*/02.fp/task.*.*/.'"
        )


def post_fp_pwscf(iter_index, jdata):
    model_devi_jobs = jdata["model_devi_jobs"]
    assert iter_index < len(model_devi_jobs)

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return

    system_index = []
    for ii in fp_tasks:
        system_index.append(os.path.basename(ii).split(".")[1])
    system_index.sort()
    set_tmp = set(system_index)
    system_index = list(set_tmp)
    system_index.sort()

    cwd = os.getcwd()
    for ss in system_index:
        sys_output = glob.glob(os.path.join(work_path, f"task.{ss}.*/output"))
        sys_input = glob.glob(os.path.join(work_path, f"task.{ss}.*/input"))
        sys_output.sort()
        sys_input.sort()

        flag = True
        for ii, oo in zip(sys_input, sys_output):
            if flag:
                _sys = dpdata.LabeledSystem(
                    oo, fmt="qe/pw/scf", type_map=jdata["type_map"]
                )
                if len(_sys) > 0:
                    all_sys = _sys
                    flag = False
                else:
                    pass
            else:
                _sys = dpdata.LabeledSystem(
                    oo, fmt="qe/pw/scf", type_map=jdata["type_map"]
                )
                if len(_sys) > 0:
                    all_sys.append(_sys)

        sys_data_path = os.path.join(work_path, f"data.{ss}")
        all_sys.to_deepmd_raw(sys_data_path)
        all_sys.to_deepmd_npy(sys_data_path, set_size=len(sys_output))


def post_fp_abacus_scf(iter_index, jdata):
    model_devi_jobs = jdata["model_devi_jobs"]
    assert iter_index < len(model_devi_jobs)

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return

    system_index = []
    for ii in fp_tasks:
        system_index.append(os.path.basename(ii).split(".")[1])
    system_index.sort()
    set_tmp = set(system_index)
    system_index = list(set_tmp)
    system_index.sort()

    cwd = os.getcwd()
    for ss in system_index:
        sys_output = glob.glob(os.path.join(work_path, f"task.{ss}.*"))
        sys_input = glob.glob(os.path.join(work_path, f"task.{ss}.*/INPUT"))
        sys_output.sort()
        sys_input.sort()

        all_sys = None
        for ii, oo in zip(sys_input, sys_output):
            _sys = dpdata.LabeledSystem(
                oo, fmt="abacus/scf", type_map=jdata["type_map"]
            )
            if len(_sys) > 0:
                if all_sys is None:
                    all_sys = _sys
                else:
                    all_sys.append(_sys)

        if all_sys is not None:
            sys_data_path = os.path.join(work_path, f"data.{ss}")
            all_sys.to_deepmd_raw(sys_data_path)
            all_sys.to_deepmd_npy(sys_data_path, set_size=len(sys_output))


def post_fp_siesta(iter_index, jdata):
    model_devi_jobs = jdata["model_devi_jobs"]
    assert iter_index < len(model_devi_jobs)

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return

    system_index = []
    for ii in fp_tasks:
        system_index.append(os.path.basename(ii).split(".")[1])
    system_index.sort()
    set_tmp = set(system_index)
    system_index = list(set_tmp)
    system_index.sort()

    cwd = os.getcwd()
    for ss in system_index:
        sys_output = glob.glob(os.path.join(work_path, f"task.{ss}.*/output"))
        sys_input = glob.glob(os.path.join(work_path, f"task.{ss}.*/input"))
        sys_output.sort()
        sys_input.sort()
        for idx, oo in enumerate(sys_output):
            _sys = dpdata.LabeledSystem()
            (
                _sys.data["atom_names"],
                _sys.data["atom_numbs"],
                _sys.data["atom_types"],
                _sys.data["cells"],
                _sys.data["coords"],
                _sys.data["energies"],
                _sys.data["forces"],
                _sys.data["virials"],
            ) = dpdata.siesta.output.obtain_frame(oo)
            if idx == 0:
                all_sys = _sys
            else:
                all_sys.append(_sys)

        sys_data_path = os.path.join(work_path, f"data.{ss}")
        all_sys.to_deepmd_raw(sys_data_path)
        all_sys.to_deepmd_npy(sys_data_path, set_size=len(sys_output))


def post_fp_gaussian(iter_index, jdata):
    model_devi_jobs = jdata["model_devi_jobs"]
    assert iter_index < len(model_devi_jobs)

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return

    system_index = []
    for ii in fp_tasks:
        system_index.append(os.path.basename(ii).split(".")[1])
    system_index.sort()
    set_tmp = set(system_index)
    system_index = list(set_tmp)
    system_index.sort()

    cwd = os.getcwd()
    for ss in system_index:
        sys_output = glob.glob(os.path.join(work_path, f"task.{ss}.*/output"))
        sys_output.sort()
        for idx, oo in enumerate(sys_output):
            sys = dpdata.LabeledSystem(oo, fmt="gaussian/log")
            if len(sys) > 0:
                sys.check_type_map(type_map=jdata["type_map"])
            if jdata.get("use_atom_pref", False):
                sys.data["atom_pref"] = np.load(
                    os.path.join(os.path.dirname(oo), "atom_pref.npy")
                )
            if idx == 0:
                if jdata.get("use_clusters", False):
                    all_sys = dpdata.MultiSystems(sys, type_map=jdata["type_map"])
                else:
                    all_sys = sys
            else:
                all_sys.append(sys)
        sys_data_path = os.path.join(work_path, f"data.{ss}")
        all_sys.to_deepmd_raw(sys_data_path)
        all_sys.to_deepmd_npy(sys_data_path, set_size=len(sys_output))


def post_fp_cp2k(iter_index, jdata, rfailed=None):
    ratio_failed = rfailed if rfailed else jdata.get("ratio_failed", 0.10)
    model_devi_jobs = jdata["model_devi_jobs"]
    assert iter_index < len(model_devi_jobs)

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return

    system_index = []
    for ii in fp_tasks:
        system_index.append(os.path.basename(ii).split(".")[1])
    system_index.sort()
    set_tmp = set(system_index)
    system_index = list(set_tmp)
    system_index.sort()

    cwd = os.getcwd()
    # tcount: num of all fp tasks
    tcount = 0
    # icount: num of converged fp tasks
    icount = 0
    for ss in system_index:
        sys_output = glob.glob(os.path.join(work_path, f"task.{ss}.*/output"))
        sys_output.sort()
        tcount += len(sys_output)
        all_sys = dpdata.MultiSystems(type_map=jdata["type_map"])
        for oo in sys_output:
            _sys = dpdata.LabeledSystem(
                oo, fmt="cp2kdata/e_f", type_map=jdata["type_map"]
            )
            if len(_sys) > 0:
                all_sys.append(_sys)
                icount += 1

        if (all_sys is not None) and (len(all_sys) > 0):
            sys_data_path = os.path.join(work_path, f"data.{ss}")
            all_sys.to_deepmd_raw(sys_data_path)
            all_sys.to_deepmd_npy(sys_data_path)

    if tcount == 0:
        rfail = 0.0
        dlog.info("failed frame: %6d in %6d " % (tcount - icount, tcount))  # noqa: UP031
    else:
        rfail = float(tcount - icount) / float(tcount)
        dlog.info(
            "failed frame: %6d in %6d  %6.2f %% "  # noqa: UP031
            % (tcount - icount, tcount, rfail * 100.0)
        )

    if rfail > ratio_failed:
        raise RuntimeError(
            "find too many unsuccessfully terminated jobs. Too many FP tasks are not converged. Please check your files in directories 'iter.*.*/02.fp/task.*.*/.'"
        )


def post_fp_pwmat(iter_index, jdata, rfailed=None):
    ratio_failed = rfailed if rfailed else jdata.get("ratio_failed", 0.05)
    model_devi_jobs = jdata["model_devi_jobs"]
    assert iter_index < len(model_devi_jobs)

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return

    system_index = []
    for ii in fp_tasks:
        system_index.append(os.path.basename(ii).split(".")[1])
    system_index.sort()
    set_tmp = set(system_index)
    system_index = list(set_tmp)
    system_index.sort()

    cwd = os.getcwd()

    tcount = 0
    icount = 0
    for ss in system_index:
        sys_output = glob.glob(os.path.join(work_path, f"task.{ss}.*/OUT.MLMD"))
        sys_output.sort()
        tcount += len(sys_output)
        all_sys = None
        for oo in sys_output:
            _sys = dpdata.LabeledSystem(oo, type_map=jdata["type_map"])
            if len(_sys) == 1:
                if all_sys is None:
                    all_sys = _sys
                else:
                    all_sys.append(_sys)
            else:
                icount += 1
        if all_sys is not None:
            sys_data_path = os.path.join(work_path, f"data.{ss}")
            all_sys.to_deepmd_raw(sys_data_path)
            all_sys.to_deepmd_npy(sys_data_path, set_size=len(sys_output))
    dlog.info(f"failed frame number: {icount} ")
    dlog.info(f"total frame number: {tcount} ")
    reff = icount / tcount
    dlog.info(f"ratio of failed frame:  {reff:.2%}")

    if reff > ratio_failed:
        raise RuntimeError("find too many unsuccessfully terminated jobs")


def post_fp_amber_diff(iter_index, jdata):
    model_devi_jobs = jdata["model_devi_jobs"]
    assert iter_index < len(model_devi_jobs)

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return

    system_index = []
    for ii in fp_tasks:
        system_index.append(os.path.basename(ii).split(".")[1])
    system_index.sort()
    set_tmp = set(system_index)
    system_index = list(set_tmp)
    system_index.sort()

    for ss in system_index:
        sys_output = glob.glob(os.path.join(work_path, f"task.{ss}.*"))
        sys_output.sort()
        all_sys = dpdata.MultiSystems(type_map=jdata["type_map"])
        for oo in sys_output:
            sys = dpdata.MultiSystems(type_map=jdata["type_map"]).from_deepmd_npy(
                os.path.join(oo, "dataset")
            )
            all_sys.append(sys)
        sys_data_path = os.path.join(work_path, f"data.{ss}")
        all_sys.to_deepmd_raw(sys_data_path)
        all_sys.to_deepmd_npy(sys_data_path, set_size=len(sys_output), prec=np.float64)


def post_fp_cpx(iter_index, jdata):
    """Post fp for cp.x. Collect data from qe/cp/traj labeled system.

    Parameters
    ----------
    iter_index : int
        The index of the current iteration.
    jdata : dict
        The parameter data.
    """
    model_devi_jobs = jdata["model_devi_jobs"]
    assert iter_index < len(model_devi_jobs)

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return

    system_index = []
    for ii in fp_tasks:
        system_index.append(os.path.basename(ii).split(".")[1])
    system_index.sort()
    set_tmp = set(system_index)
    system_index = list(set_tmp)
    system_index.sort()

    fp_params = jdata["fp_params"]
    input_fn = fp_params["input_fn"]

    for ss in system_index:
        sys_output = glob.glob(os.path.join(work_path, f"task.{ss}.*"))
        sys_output.sort()
        all_sys = dpdata.MultiSystems(type_map=jdata["type_map"])
        for oo in sys_output:
            if os.path.exists(os.path.join(oo, "output")):
                sys = dpdata.LabeledSystem(os.path.join(oo, input_fn), fmt="qe/cp/traj")
                all_sys.append(sys)
        sys_data_path = os.path.join(work_path, f"data.{ss}")
        all_sys.to_deepmd_raw(sys_data_path)
        all_sys.to_deepmd_npy(sys_data_path, set_size=len(sys_output), prec=np.float64)


def post_fp_custom(iter_index, jdata):
    """Post fp for custom fp. Collect data from user-defined `output_fn`.

    Parameters
    ----------
    iter_index : int
        The index of the current iteration.
    jdata : dict
        The parameter data.
    """
    model_devi_jobs = jdata["model_devi_jobs"]
    assert iter_index < len(model_devi_jobs)

    iter_name = make_iter_name(iter_index)
    work_path = os.path.join(iter_name, fp_name)
    fp_tasks = glob.glob(os.path.join(work_path, "task.*"))
    fp_tasks.sort()
    if len(fp_tasks) == 0:
        return

    system_index = []
    for ii in fp_tasks:
        system_index.append(os.path.basename(ii).split(".")[1])
    system_index.sort()
    set_tmp = set(system_index)
    system_index = list(set_tmp)
    system_index.sort()

    fp_params = jdata["fp_params"]
    output_fn = fp_params["output_fn"]
    output_fmt = fp_params["output_fmt"]

    for ss in system_index:
        sys_output = glob.glob(os.path.join(work_path, f"task.{ss}.*"))
        sys_output.sort()
        all_sys = dpdata.MultiSystems(type_map=jdata["type_map"])
        for oo in sys_output:
            if os.path.exists(os.path.join(oo, output_fn)):
                sys = dpdata.LabeledSystem(os.path.join(oo, output_fn), fmt=output_fmt)
                all_sys.append(sys)
        sys_data_path = os.path.join(work_path, f"data.{ss}")
        all_sys.to_deepmd_raw(sys_data_path)
        all_sys.to_deepmd_npy(sys_data_path, set_size=len(sys_output), prec=np.float64)


def post_fp(iter_index, jdata):
    fp_style = jdata["fp_style"]
    if fp_style == "vasp":
        post_fp_vasp(iter_index, jdata)
    elif fp_style == "pwscf":
        post_fp_pwscf(iter_index, jdata)
    elif fp_style == "abacus":
        post_fp_abacus_scf(iter_index, jdata)
    elif fp_style == "siesta":
        post_fp_siesta(iter_index, jdata)
    elif fp_style == "gaussian":
        post_fp_gaussian(iter_index, jdata)
    elif fp_style == "cp2k":
        post_fp_cp2k(iter_index, jdata)
    elif fp_style == "pwmat":
        post_fp_pwmat(iter_index, jdata)
    elif fp_style == "amber/diff":
        post_fp_amber_diff(iter_index, jdata)
    elif fp_style == "cpx":
        post_fp_cpx(iter_index, jdata)
    elif fp_style == "custom":
        post_fp_custom(iter_index, jdata)
    else:
        raise RuntimeError("unsupported fp style")
    post_fp_check_fail(iter_index, jdata)
    # clean traj
    clean_traj = True
    if "model_devi_clean_traj" in jdata:
        clean_traj = jdata["model_devi_clean_traj"]
    modd_path = None
    if isinstance(clean_traj, bool):
        iter_name = make_iter_name(iter_index)
        if clean_traj:
            modd_path = os.path.join(iter_name, model_devi_name)
    elif isinstance(clean_traj, int):
        clean_index = iter_index - clean_traj
        if clean_index >= 0:
            modd_path = os.path.join(make_iter_name(clean_index), model_devi_name)
    if modd_path is not None:
        md_trajs = glob.glob(os.path.join(modd_path, "task*/traj"))
        for ii in md_trajs:
            shutil.rmtree(ii)


def set_version(mdata):
    deepmd_version = "2"
    mdata["deepmd_version"] = deepmd_version
    return mdata


def run_iter(param_file, machine_file, allow_ref: bool = False):
    jdata = load_file(param_file)
    mdata = load_file(machine_file)

    jdata_arginfo = run_jdata_arginfo()
    jdata = normalize(jdata_arginfo, jdata, strict_check=False, allow_ref=allow_ref)

    update_mass_map(jdata)

    # set up electron temperature
    use_ele_temp = jdata.get("use_ele_temp", 0)
    if use_ele_temp == 1:
        setup_ele_temp(False)
    elif use_ele_temp == 2:
        setup_ele_temp(True)

    if jdata.get("pretty_print", False):
        from monty.serialization import dumpfn

        # assert(jdata["pretty_format"] in ['json','yaml'])
        fparam = (
            SHORT_CMD
            + "_"
            + param_file.split(".")[0]
            + "."
            + jdata.get("pretty_format", "json")
        )
        dumpfn(jdata, fparam, indent=4)
        fmachine = (
            SHORT_CMD
            + "_"
            + machine_file.split(".")[0]
            + "."
            + jdata.get("pretty_format", "json")
        )
        dumpfn(mdata, fmachine, indent=4)

    if mdata.get("handlers", None):
        if mdata["handlers"].get("smtp", None):
            que = queue.Queue(-1)
            queue_handler = logging.handlers.QueueHandler(que)
            smtp_handler = logging.handlers.SMTPHandler(**mdata["handlers"]["smtp"])
            listener = logging.handlers.QueueListener(que, smtp_handler)
            dlog.addHandler(queue_handler)
            listener.start()
    # Convert mdata
    mdata = convert_mdata(mdata)
    max_tasks = 10000
    numb_task = 9
    record = "record.dpgen"
    iter_rec = [0, -1]
    if os.path.isfile(record):
        with open(record) as frec:
            for line in frec:
                iter_rec = [int(x) for x in line.split()]
        if len(iter_rec) == 0:
            raise ValueError("There should not be blank lines in record.dpgen.")
        dlog.info("continue from iter %03d task %02d" % (iter_rec[0], iter_rec[1]))  # noqa: UP031

    cont = True
    ii = -1
    while cont:
        ii += 1
        if ii < iter_rec[0]:
            continue
        iter_name = make_iter_name(ii)
        sepline(iter_name, "=")
        for jj in range(numb_task):
            if ii * max_tasks + jj <= iter_rec[0] * max_tasks + iter_rec[1]:
                continue
            task_name = "task %02d" % jj  # noqa: UP031
            sepline(f"{iter_name} {task_name}", "-")
            if jj == 0:
                log_iter("make_train", ii, jj)
                make_train(ii, jdata, mdata)
            elif jj == 1:
                log_iter("run_train", ii, jj)
                run_train(ii, jdata, mdata)
            elif jj == 2:
                log_iter("post_train", ii, jj)
                post_train(ii, jdata, mdata)
            elif jj == 3:
                log_iter("make_model_devi", ii, jj)
                cont = make_model_devi(ii, jdata, mdata)
                if not cont:
                    break
            elif jj == 4:
                log_iter("run_model_devi", ii, jj)
                run_model_devi(ii, jdata, mdata)

            elif jj == 5:
                log_iter("post_model_devi", ii, jj)
                post_model_devi(ii, jdata, mdata)
            elif jj == 6:
                log_iter("make_fp", ii, jj)
                make_fp(ii, jdata, mdata)
            elif jj == 7:
                log_iter("run_fp", ii, jj)
                run_fp(ii, jdata, mdata)
            elif jj == 8:
                log_iter("post_fp", ii, jj)
                post_fp(ii, jdata)
            else:
                raise RuntimeError("unknown task %d, something wrong" % jj)  # noqa: UP031
            record_iter(record, ii, jj)


def get_atomic_masses(atom):
    element_names = [
        "Hydrogen",
        "Helium",
        "Lithium",
        "Beryllium",
        "Boron",
        "Carbon",
        "Nitrogen",
        "Oxygen",
        "Fluorine",
        "Neon",
        "Sodium",
        "Magnesium",
        "Aluminium",
        "Silicon",
        "Phosphorus",
        "Sulfur",
        "Chlorine",
        "Argon",
        "Potassium",
        "Calcium",
        "Scandium",
        "Titanium",
        "Vanadium",
        "Chromium",
        "Manganese",
        "Iron",
        "Cobalt",
        "Nickel",
        "Copper",
        "Zinc",
        "Gallium",
        "Germanium",
        "Arsenic",
        "Selenium",
        "Bromine",
        "Krypton",
        "Rubidium",
        "Strontium",
        "Yttrium",
        "Zirconium",
        "Niobium",
        "Molybdenum",
        "Technetium",
        "Ruthenium",
        "Rhodium",
        "Palladium",
        "Silver",
        "Cadmium",
        "Indium",
        "Tin",
        "Antimony",
        "Tellurium",
        "Iodine",
        "Xenon",
        "Caesium",
        "Barium",
        "Lanthanum",
        "Cerium",
        "Praseodymium",
        "Neodymium",
        "Promethium",
        "Samarium",
        "Europium",
        "Gadolinium",
        "Terbium",
        "Dysprosium",
        "Holmium",
        "Erbium",
        "Thulium",
        "Ytterbium",
        "Lutetium",
        "Hafnium",
        "Tantalum",
        "Tungsten",
        "Rhenium",
        "Osmium",
        "Iridium",
        "Platinum",
        "Gold",
        "Mercury",
        "Thallium",
        "Lead",
        "Bismuth",
        "Polonium",
        "Astatine",
        "Radon",
        "Francium",
        "Radium",
        "Actinium",
        "Thorium",
        "Protactinium",
        "Uranium",
        "Neptunium",
        "Plutonium",
        "Americium",
        "Curium",
        "Berkelium",
        "Californium",
        "Einsteinium",
        "Fermium",
        "Mendelevium",
        "Nobelium",
        "Lawrencium",
        "Rutherfordium",
        "Dubnium",
        "Seaborgium",
        "Bohrium",
        "Hassium",
        "Meitnerium",
        "Darmastadtium",
        "Roentgenium",
        "Copernicium",
        "Nihonium",
        "Flerovium",
        "Moscovium",
        "Livermorium",
        "Tennessine",
        "Oganesson",
    ]
    chemical_symbols = [
        "H",
        "He",
        "Li",
        "Be",
        "B",
        "C",
        "N",
        "O",
        "F",
        "Ne",
        "Na",
        "Mg",
        "Al",
        "Si",
        "P",
        "S",
        "Cl",
        "Ar",
        "K",
        "Ca",
        "Sc",
        "Ti",
        "V",
        "Cr",
        "Mn",
        "Fe",
        "Co",
        "Ni",
        "Cu",
        "Zn",
        "Ga",
        "Ge",
        "As",
        "Se",
        "Br",
        "Kr",
        "Rb",
        "Sr",
        "Y",
        "Zr",
        "Nb",
        "Mo",
        "Tc",
        "Ru",
        "Rh",
        "Pd",
        "Ag",
        "Cd",
        "In",
        "Sn",
        "Sb",
        "Te",
        "I",
        "Xe",
        "Cs",
        "Ba",
        "La",
        "Ce",
        "Pr",
        "Nd",
        "Pm",
        "Sm",
        "Eu",
        "Gd",
        "Tb",
        "Dy",
        "Ho",
        "Er",
        "Tm",
        "Yb",
        "Lu",
        "Hf",
        "Ta",
        "W",
        "Re",
        "Os",
        "Ir",
        "Pt",
        "Au",
        "Hg",
        "Tl",
        "Pb",
        "Bi",
        "Po",
        "At",
        "Rn",
        "Fr",
        "Ra",
        "Ac",
        "Th",
        "Pa",
        "U",
        "Np",
        "Pu",
        "Am",
        "Cm",
        "Bk",
        "Cf",
        "Es",
        "Fm",
        "Md",
        "No",
        "Lr",
        "Rf",
        "Db",
        "Sg",
        "Bh",
        "Hs",
        "Mt",
        "Ds",
        "Rg",
        "Cn",
        "Nh",
        "Fl",
        "Mc",
        "Lv",
        "Ts",
        "Og",
    ]
    atomic_number = [i + 1 for i in range(len(chemical_symbols))]

    # NIST Standard Reference Database 144
    # URL: https://physics.nist.gov/cgi-bin/Compositions/stand_alone.pl?ele=&ascii=ascii&isotype=all
    atomic_masses_common = [
        1.00782503223,
        4.00260325413,
        7.0160034366,
        9.012183065,
        11.00930536,
        12.0,
        14.00307400443,
        15.99491461957,
        18.99840316273,
        19.9924401762,
        22.989769282,
        23.985041697,
        26.98153853,
        27.97692653465,
        30.97376199842,
        31.9720711744,
        34.968852682,
        39.9623831237,
        38.9637064864,
        39.962590863,
        44.95590828,
        47.94794198,
        50.94395704,
        51.94050623,
        54.93804391,
        55.93493633,
        58.93319429,
        57.93534241,
        62.92959772,
        63.92914201,
        68.9255735,
        73.921177761,
        74.92159457,
        79.9165218,
        78.9183376,
        83.9114977282,
        84.9117897379,
        87.9056125,
        88.9058403,
        89.9046977,
        92.906373,
        97.90540482,
        96.9063667,
        101.9043441,
        102.905498,
        105.9034804,
        106.9050916,
        113.90336509,
        114.903878776,
        119.90220163,
        120.903812,
        129.906222748,
        126.9044719,
        131.9041550856,
        132.905451961,
        137.905247,
        138.9063563,
        139.9054431,
        140.9076576,
        141.907729,
        144.9127559,
        151.9197397,
        152.921238,
        157.9241123,
        158.9253547,
        163.9291819,
        164.9303288,
        165.9302995,
        168.9342179,
        173.9388664,
        174.9407752,
        179.946557,
        180.9479958,
        183.95093092,
        186.9557501,
        191.961477,
        192.9629216,
        194.9647917,
        196.96656879,
        201.9706434,
        204.9744278,
        207.9766525,
        208.9803991,
        208.9824308,
        209.9871479,
        222.0175782,
        223.019736,
        226.0254103,
        227.0277523,
        232.0380558,
        231.0358842,
        238.0507884,
        237.0481736,
        244.0642053,
        243.0613813,
        247.0703541,
        247.0703073,
        251.0795886,
        252.08298,
        257.0951061,
        258.0984315,
        259.10103,
        262.10961,
        267.12179,
        268.12567,
        271.13393,
        272.13826,
        270.13429,
        276.15159,
        281.16451,
        280.16514,
        285.17712,
        284.17873,
        289.19042,
        288.19274,
        293.20449,
        292.20746,
        294.21392,
    ]
    # IUPAC Technical Report
    # doi:10.1515/pac-2015-0305
    atomic_masses_2013 = [
        1.00784,
        4.002602,
        6.938,
        9.0121831,
        10.806,
        12.0096,
        14.00643,
        15.99903,
        18.99840316,
        20.1797,
        22.98976928,
        24.304,
        26.9815385,
        28.084,
        30.973762,
        32.059,
        35.446,
        39.948,
        39.0983,
        40.078,
        44.955908,
        47.867,
        50.9415,
        51.9961,
        54.938044,
        55.845,
        58.933194,
        58.6934,
        63.546,
        65.38,
        69.723,
        72.63,
        74.921595,
        78.971,
        79.901,
        83.798,
        85.4678,
        87.62,
        88.90584,
        91.224,
        92.90637,
        95.95,
        None,
        101.07,
        102.9055,
        106.42,
        107.8682,
        112.414,
        114.818,
        118.71,
        121.76,
        127.6,
        126.90447,
        131.293,
        132.905452,
        137.327,
        138.90547,
        140.116,
        140.90766,
        144.242,
        None,
        150.36,
        151.964,
        157.25,
        158.92535,
        162.5,
        164.93033,
        167.259,
        168.93422,
        173.054,
        174.9668,
        178.49,
        180.94788,
        183.84,
        186.207,
        190.23,
        192.217,
        195.084,
        196.966569,
        200.592,
        204.382,
        207.2,
        208.9804,
        None,
        None,
        None,
        None,
        None,
        None,
        232.0377,
        231.03588,
        238.02891,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
    ]
    # IUPAC Technical Report
    # doi:10.1515/pac-2019-0603
    atomic_masses_2021 = [
        1.00784,
        4.002602,
        6.938,
        9.0121831,
        10.806,
        12.0096,
        14.00643,
        15.99903,
        18.99840316,
        20.1797,
        22.98976928,
        24.304,
        26.9815384,
        28.084,
        30.973762,
        32.059,
        35.446,
        39.792,
        39.0983,
        40.078,
        44.955907,
        47.867,
        50.9415,
        51.9961,
        54.938043,
        55.845,
        58.933194,
        58.6934,
        63.546,
        65.38,
        69.723,
        72.63,
        74.921595,
        78.971,
        79.901,
        83.798,
        85.4678,
        87.62,
        88.905838,
        91.224,
        92.90637,
        95.95,
        None,
        101.07,
        102.90549,
        106.42,
        107.8682,
        112.414,
        114.818,
        118.71,
        121.76,
        127.6,
        126.90447,
        131.293,
        132.905452,
        137.327,
        138.90547,
        140.116,
        140.90766,
        144.242,
        None,
        150.36,
        151.964,
        157.25,
        158.925354,
        162.5,
        164.930329,
        167.259,
        168.934219,
        173.045,
        174.9668,
        178.486,
        180.94788,
        183.84,
        186.207,
        190.23,
        192.217,
        195.084,
        196.96657,
        200.592,
        204.382,
        206.14,
        208.9804,
        None,
        None,
        None,
        None,
        None,
        None,
        232.0377,
        231.03588,
        238.02891,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
        None,
    ]

    atomic_masses = [
        atomic_masses_common[n] if i is None else i
        for n, i in enumerate(atomic_masses_2021)
    ]

    if atom in element_names:
        return atomic_masses[element_names.index(atom)]
    elif atom in chemical_symbols:
        return atomic_masses[chemical_symbols.index(atom)]
    elif atom in atomic_number:
        return atomic_masses[atomic_number.index(atom)]
    else:
        raise RuntimeError(
            "unknown atomic identifier",
            atom,
            'if one want to use isotopes, or non-standard element names, chemical symbols, or atomic number in the type_map list, please customize the mass_map list instead of using "auto".',
        )


def update_mass_map(jdata):
    if jdata["mass_map"] == "auto":
        jdata["mass_map"] = [get_atomic_masses(i) for i in jdata["type_map"]]


def gen_run(args):
    if args.PARAM and args.MACHINE:
        if args.debug:
            dlog.setLevel(logging.DEBUG)
        dlog.info("start running")
        run_iter(args.PARAM, args.MACHINE, allow_ref=args.allow_ref)
        dlog.info("finished")


def _main():
    parser = argparse.ArgumentParser()
    parser.add_argument("PARAM", type=str, help="The parameters of the generator")
    parser.add_argument(
        "MACHINE", type=str, help="The settings of the machine running the generator"
    )
    args = parser.parse_args()

    logging.basicConfig(level=logging.INFO, format="%(asctime)s %(message)s")
    # logging.basicConfig (filename="compute_string.log", filemode="a", level=logging.INFO, format='%(asctime)s %(message)s')
    logging.getLogger("paramiko").setLevel(logging.WARNING)

    logging.info("start running")
    run_iter(args.PARAM, args.MACHINE)
    logging.info("finished!")


if __name__ == "__main__":
    _main()