plumed_pytorch_gnn/PytorchGNN.cpp

/* +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Copyright (c) 2022-2024 of Luigi Bonati and Enrico Trizio.

The pytorch module is free software: you can redistribute it and/or modify
it under the terms of the GNU Lesser General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.

The pytorch module is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU Lesser General Public License for more details.

You should have received a copy of the GNU Lesser General Public License
along with plumed.  If not, see <http://www.gnu.org/licenses/>.
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ */

#ifdef __PLUMED_HAS_LIBTORCH

#include <cmath>
#include <memory>
#include <fstream>
#include <torch/torch.h>
#include <torch/script.h>
#include <torch/csrc/jit/jit_log.h>

#include "core/PlumedMain.h"
#include "colvar/Colvar.h"
#include "colvar/ActionRegister.h"
#include "tools/NeighborList.h"
#include "tools/Communicator.h"
#include "tools/OpenMP.h"
#include "tools/File.h"
#include "tools/PDB.h"

// NOTE: Freezing a ScriptModule (torch::jit::freeze) works only in >=1.11
// For 1.8 <= versions <=1.10 we need a hack
// (see https://discuss.pytorch.org/t/how-to-check-libtorch-version/77709/4 and also
// https://github.com/pytorch/pytorch/blob/dfbd030854359207cb3040b864614affeace11ce/torch/csrc/jit/api/module.cpp#L479)
// adapted from NequIP https://github.com/mir-group/nequip
#if (TORCH_VERSION_MAJOR == 1 && TORCH_VERSION_MINOR <= 10)
#define DO_TORCH_FREEZE_HACK
// For the hack, need more headers:
#include <torch/csrc/jit/passes/freeze_module.h>
#include <torch/csrc/jit/passes/frozen_graph_optimizations.h>
#endif

using namespace std;

namespace PLMD {

class NeighborList;

namespace colvar {

namespace pytorch_gnn {

//+PLUMEDOC PYTORCH_GNN PYTORCH_GNN
/*
Load a Graph Neural Network (GNN) model compiled with TorchScript.

This module uses a fixed length unit of _Angstrom_. Thus, the GNN model read by
this module should be trained under the same unit convention. Besides, the
module constructs node attributes w.r.t to the atomic types. As a result, this
module require a PDB file which records names of _ALL_ atoms in the system
(the STRUCTURE keyword). Note that the atom names in this PDB file could
_ONLY_ be element symbols, e.g.:
\auxfile{plumed_topo.pdb}
ATOM      1  H   ACE A   1      15.100  12.940  29.390  1.00  0.00           H
ATOM      2  C   ACE A   1      14.970  13.860  29.960  1.00  0.00           C
ATOM      3  H   ACE A   1      15.720  13.820  30.760  1.00  0.00           H
ATOM      4  H   ACE A   1      13.980  13.920  30.410  1.00  0.00           H
ATOM      5  C   ACE A   1      15.300  15.070  29.100  1.00  0.00           C
\endauxfile

The module constructs graph edges between neighbors inside the selected atom
group, using the cutoff value recorded in the model file. By default, such an
atom group is defined by the single GROUPA keyword. Under this case, the node
number of the input graph in each MD step is fixed, and the number of edges
will change according to the relative postions of the atoms.
However, if the GROUPB parameter is given, the atom group mentioned above will
contain all atoms in GROUPA, _AND_ atoms in GROUPB which are within a radius of
_ANY_ atom in GROUPA. Such a radius of selecting atoms from GROUPB equals to the
cutoff radius recorded in the model file _plus_ the buffer size controlled by
the BUFFER keyword. Thus, when GROUPB is given, the node number of the input
graph could fluctuate in different MD steps. Besides, when SUBGROUPA is defined
the module will add long edges bewteen such a group. Cutoff radius of these
long edges will equal to the cutoff_l attribute recorded in the model file.

This module also support committor calculations. When the input PyTorch model
is a committor model, the outputs will assign the zeta value to the first
output (node-0) and the q value to the second output (node-1). If the KBIAS
keyword is also given, the module will also calculate the committor bias and
assign it with a label of kbias. These information will be shown in the log as
well.

Note that this function requires \ref installation-libtorch LibTorch C++ library.
Check the instructions in the \ref PYTORCH page to enable the module.
Specifically, we encourage the user to install the GPU-enabled version of
LibTorch, when dealing with large input graphs.

\par Examples
The following example instructs plumed to evaluate the GNN model using the atoms 1-10. The neighbor list for determining the edges will be updated every 100 steps.
\plumedfile
PYTORCH_GNN ...
  GROUPA=1-10
  MODEL=model.ptc
  STRUCTURE=plumed_topo.pdb
  NL_STRIDE=100
  LABEL=gnn
... PYTORCH_GNN
\endplumedfile

The following example instructs plumed to do the same calculation as the above example, but will evaluate the model on CUDA using double precision, and add an OPES bias potential on the CV.
\plumedfile
PYTORCH_GNN ...
  GROUPA=1-10
  MODEL=model.ptc
  STRUCTURE=plumed_topo.pdb
  NL_STRIDE=100
  CUDA
  FLOAT64
  LABEL=gnn
... PYTORCH_GNN

OPES_METAD ...
  LABEL=opes
  ARG=gnn.node-0
  FILE=KERNELS
  PACE=500
  TEMP=300
  BARRIER=35
... OPES_METAD
\endplumedfile

The following example instructs plumed to do the same calculation as the above example, but will sample Kang's (or Kolmogolov's) transition state ensemble.
\plumedfile
PYTORCH_GNN ...
  GROUPA=1-10
  MODEL=model.ptc
  STRUCTURE=plumed_topo.pdb
  NL_STRIDE=100
  FLOAT64
  KBIAS
  KBLAMBDA=-2.0
  LABEL=gnn
... PYTORCH_GNN

BIASVALUE ARG=gnn.kbias LABEL=vk
\endplumedfile

The following example instructs plumed to evaluate the GNN model using the atoms 1-10 as center atoms, and atoms 11-100 as the environment atoms. The buffer size used for selecting active atoms from the environment atoms is 2 PLUMED unit. The neighbor list for determining the edges will be updated every 2 steps.
\plumedfile
PYTORCH_GNN ...
  GROUPA=1-10
  GROUPB=11-100
  MODEL=model.ptc
  STRUCTURE=plumed_topo.pdb
  NL_STRIDE=2
  BUFFER=2.0
  LABEL=gnn
... PYTORCH_GNN
\endplumedfile

*/
//+ENDPLUMEDOC


class PytorchGNN: public Colvar
{
  int n_out = 0;
  int summary_level = 3;
  bool pbc = true;
  bool serial = false;
  bool firsttime = true;
  bool invalidate_list = true;
  bool is_committor = false;
  bool k_bias = false;
  bool kb_weighted = false;
  bool kb_truncated = false;
  bool bailout_fusion = false;
  double r_max = 0.0; // In PLUMED length unit
  double buffer = 0.0; // In PLUMED length unit
  double r_max_l = -1.0; // In PLUMED length unit
  double kb_lambda = -1.0;
  double kb_epsilon = -1.0;
  double kb_sigmoid_p = -1.0;
  std::string model_file_name;
  std::string model_training_time;
  std::string structure_file_name;
  std::vector<int> system_node_types;
  std::vector<int> model_atomic_numbers;
  std::vector<double> model_atomic_masses;
  std::vector<AtomNumber> atom_list_a;
  std::vector<AtomNumber> atom_list_b;
  std::vector<AtomNumber> atom_list_sub_a;
  std::vector<int> atom_list_active; // local_ids
  std::vector<int> atom_list_active_subgroup; // local_ids
  std::unique_ptr<NeighborList> neighbor_list;
  torch::jit::script::Module model;
  torch::ScalarType torch_float_dtype = torch::kFloat32;
  torch::Device device = c10::Device(torch::kCPU);
  const std::array<std::string, 118> periodic_table = {
     "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"
  }; // TODO: add ghost atoms
  std::string model_summary(
    std::string model_name, torch::jit::Module module, int level_max, int level
  );
  int get_number_of_parameters(void);
  int atomic_number_from_name(std::string name);
  bool groups_have_intersection(void);
  bool subgroup_is_in_group_a(void);
  void find_active_atoms(int n_threads);
  void find_active_subgroup_atoms(void);

public:
  explicit PytorchGNN(const ActionOptions&);
  ~PytorchGNN();
  static void registerKeywords(Keywords& keys);
  void calculate() override;
  void prepare() override;
}; // class PytorchGNN

PLUMED_REGISTER_ACTION(PytorchGNN, "PYTORCH_GNN")

void PytorchGNN::registerKeywords(Keywords& keys)
{
  Colvar::registerKeywords(keys);

  keys.add(
    "atoms",
    "GROUPA",
    "First list of atoms (corresponding to the `system_selection in mlcolvar`)"
  );

  keys.add(
    "atoms",
    "GROUPB",
    "Second list of atoms (corresponding to the `environment_selection in mlcolvar`)"
  );

  keys.add(
    "atoms",
    "SUBGROUPA",
    "List of subsystem atoms (corresponding to the `subsystem_selection in mlcolvar`)"
  );

  keys.add(
    "compulsory",
    "MODEL",
    "Filename of the PyTorch compiled model"
  );

  keys.add(
    "compulsory",
    "STRUCTURE",
    "PDB file name that contains the whole simulated system, with currect atom names and orders"
  );

  keys.add(
    "optional",
    "NL_STRIDE",
    "The frequency with which we are updating the atoms in the neighbor list"
  );

  keys.add(
    "optional",
    "BUFFER",
    "Buffer size used in finding active environment atoms"
  );

  keys.add(
    "optional",
    "KBLAMBDA",
    "The LAMBDA value for calculating $V_K$. Only vaild for GNN committor models"
  );

  keys.add(
    "optional",
    "KBEPSILON",
    "The EPSILON value for calculating $V_K$. Only vaild for GNN committor models, the default value depends on the model precision"
  );

  keys.add(
    "optional",
    "_SUMMARYLEVEL",
    "Maximum verbosity level for the model structure printing"
  );

  keys.addFlag(
    "CUDA",
    false,
    "Perform the calculation on CUDA"
  );

  keys.addFlag(
    "SERIAL",
    false,
    "Perform the calculation in serial - for debug purpose"
  );

  keys.addFlag(
    "FLOAT64",
    false,
    "Evaluate the model in double precise"
  );

  keys.addFlag(
    "_BAILOUTFUSION",
    false,
    "Use a faster LibTorch fusion strategy (experimental)"
  );

  keys.addFlag(
    "KBIAS",
    false,
    "Calculate Kang's bias potential $V_K$ (a.k.a. Kolmogolov's bias). Only vaild for GNN committor models"
  );

  keys.addFlag(
    "KBWEIGHTED",
    false,
    "Calculate mass-weighted (exact) $V_K$. Only vaild for GNN committor models"
  );

  keys.addFlag(
    "KBTRUNCATED",
    false,
    "Calculate truncated (twisted) $V_K$. Only vaild for GNN committor models"
  );

  keys.addOutputComponent(
    "node",
    "default",
    "Model outputs"
  );

  keys.addOutputComponent(
    "kbias",
    "KBIAS",
    "Kang's bias potential $V_K$"
  );
}

PytorchGNN::PytorchGNN(const ActionOptions& ao):
  PLUMED_COLVAR_INIT(ao)
{
  // print libtorch version
  std::stringstream ss;
  ss << TORCH_VERSION_MAJOR << "." \
     << TORCH_VERSION_MINOR << "." \
     << TORCH_VERSION_PATCH;
  std::string version;
  ss >> version; // extract into the string.
  std::string version_info = "  LibTorch version: " + version + "\n";
  log.printf(version_info.data());

  // parse input
  parseAtomList("GROUPA", atom_list_a);
  parseAtomList("GROUPB", atom_list_b);
  parseAtomList("SUBGROUPA", atom_list_sub_a);

  parse("MODEL", model_file_name);

  parse("STRUCTURE", structure_file_name);

  int neighbor_list_stride = 1;
  parse("NL_STRIDE", neighbor_list_stride);
  if (neighbor_list_stride <= 0)
    plumed_merror("NL_STRIDE should be positive!");

  parse("BUFFER", buffer);
  if (buffer > 0 && atom_list_b.size() == 0)
    plumed_merror("Not GROUPB given! Cannot define the BUFFER key!");

  parse("KBLAMBDA", kb_lambda);
  parse("KBEPSILON", kb_epsilon);

  parse("_SUMMARYLEVEL", summary_level);

  bool use_cuda = false;
  bool required_cuda = false;
  parseFlag("CUDA", required_cuda);

  parseFlag("SERIAL", serial);
  if (required_cuda and serial)
    plumed_merror("Can not enable CUDA with SERIAL at the same time!");

  bool use_float64 = false;
  parseFlag("FLOAT64", use_float64);
  if (kb_epsilon < 0) {
    if (use_float64)
      kb_epsilon = 1E-14;
    else
      kb_epsilon = 1E-7;
  }

  bool nopbc = !pbc;
  parseFlag("NOPBC", nopbc);
  pbc = !nopbc;

  parseFlag("KBIAS", k_bias);
  parseFlag("KBWEIGHTED", kb_weighted);
  parseFlag("KBTRUNCATED", kb_truncated);

  parseFlag("_BAILOUTFUSION", bailout_fusion);

  checkRead();

  // check groups
  if (atom_list_b.size() > 0) {
    if (groups_have_intersection())
      plumed_merror("GROUPA can not intersect with GROUPB!");
    atom_list_active.resize(atom_list_a.size() + atom_list_b.size());
    atom_list_active.clear();
  } else {
    atom_list_active.resize(atom_list_a.size());
    atom_list_active.clear();
    find_active_atoms(1);
  }
  if (atom_list_sub_a.size() > 0)
    find_active_subgroup_atoms();

  if (atom_list_sub_a.size() > 0)
    if (!subgroup_is_in_group_a())
      plumed_merror("Not all atoms in SUBGROUPA present in GROUPA!");

  // check precise
  if (use_float64)
    torch_float_dtype = torch::kFloat64;

  // check CUDA
  if (required_cuda && torch::cuda::is_available()) {
    device = c10::Device(torch::kCUDA);
    use_cuda = true;
  } else if (required_cuda) {
    use_cuda = false;
  }

  // check structure file
  PDB pdb;
  FILE *fp = fopen(structure_file_name.c_str(), "r");
  if (fp != NULL) {
    pdb.readFromFilepointer(
      fp,
      atoms.usingNaturalUnits(),         // TODO: remove the `atoms.` prefix when release
      0.1 / atoms.getUnits().getLength() // TODO: remove the `atoms.` prefix when release
    );
    fclose(fp);
  } else {
    plumed_merror("Can not open PDB file: '" + structure_file_name + "'");
  }

  // deserialize the model from file
  try {
    model = torch::jit::load(model_file_name, device);
  } catch (const c10::Error& e) {
    plumed_merror(
      "Cannot load model file: '" + model_file_name + "'. Reason: " + e.what()
    );
  }

  // disable parameter grads
  for (auto p: model.parameters())
    p.requires_grad_(false);

  // set up model precise
  model.to(torch_float_dtype);

  // summary
  int model_parameters = get_number_of_parameters();
  std::string model_architecture = model_summary("CV", model, summary_level, 0);

  // get CV length
  if (!model.hasattr("n_out") && !model.hasattr("n_cvs"))
    plumed_merror(
      "Can not find model attribute: 'n_out' or 'n_cvs'! One of these attributes has to be set during the compilation of the model!"
    );
  else if (model.hasattr("n_out") && model.hasattr("n_cvs"))
    plumed_merror(
      "Both model attribute: 'n_out' and 'n_cvs' are defined!"
    );
  if (model.hasattr("n_out"))
    n_out = model.attr("n_out").toTensor().item<int>();
  else
    n_out = model.attr("n_cvs").toTensor().item<int>();

  // get cutoff radius
  if (!model.hasattr("r_max") && !model.hasattr("cutoff") )
    plumed_merror(
      "Can not find model attribute: 'r_max' or 'cutoff'! One of these attributes has to be set during the compilation of the model!"
    );
  else if (model.hasattr("r_max") && model.hasattr("cutoff") )
    plumed_merror(
      "Both model attribute: 'r_max' and 'cutoff' are defined!"
    );

  // TODO: now, the `r_max` parameter in the model file is defined in unit of Angstrom.
  // We should warn the users about this default
  if (model.hasattr("cutoff"))
    r_max = model.attr("cutoff").toTensor().item<double>();
  else
    r_max = model.attr("r_max").toTensor().item<double>();
  r_max = r_max / atoms.getUnits().getLength() * 0.1; // TODO: remove the `atoms.` prefix when release

  // get long cutoff radius
  if (atom_list_sub_a.size() > 0) {
    if (!model.hasattr("cutoff_l")) {
      plumed_merror(
        "Can not find model attribute: 'cutoff_l'! Such an attributes is required for defining the subsystem group (SUBGROUPA)!"
      );
    } else if (model.attr("cutoff_l").toTensor().item<double>() < 0) {
      plumed_merror(
        "Model attribute: 'cutoff_l' is negative! A positive long cutoff radius is required for defining the subsystem group (SUBGROUPA)!"
      );
    } else {
      r_max_l = model.attr("cutoff_l").toTensor().item<double>();
      r_max_l = r_max_l / atoms.getUnits().getLength() * 0.1; // TODO: remove the `atoms.` prefix when release
    }
  } else if (
    model.hasattr("cutoff_l")
    && model.attr("cutoff_l").toTensor().item<double>() > 0
  ) {
    plumed_merror(
      "Found model attribute: 'cutoff_l'! Such an attributes requires defining the subsystem group (SUBGROUPA)!"
    );
  }

  // get atomic numbers
  if (!model.hasattr("atomic_numbers"))
    plumed_merror(
      "Can not find model attribute: 'atomic_numbers'! This attribute has to be set during the compilation of the model!"
    );
  auto atomic_numbers = model.attr("atomic_numbers").toTensor();
  for (int64_t i = 0; i < atomic_numbers.size(0); i++)
    model_atomic_numbers.push_back(atomic_numbers[i].item<int64_t>());

  // check model type
  if (model.hasattr("is_committor"))
    is_committor = model.attr("is_committor").toTensor().item<int>() != 0;
  if (!is_committor && k_bias)
    plumed_merror(
      "Can not calculate Kang's bias potential for a non-committor model!"
    );
  if (is_committor) {
    if (n_out != 2)
      plumed_merror(
        "The committor model should output two values!"
      );
    for (auto p: model.named_attributes())
      if (p.name == "sigmoid.p")
        kb_sigmoid_p = p.value.toDouble();
    if (kb_sigmoid_p < 0)
      plumed_merror(
        "Can not load the sigmoid_p value from the model!"
      );
    if (k_bias && kb_weighted) {
      auto atomic_masses = model.attr("atomic_masses").toTensor();
      for (int64_t i = 0; i < atomic_masses.size(0); i++)
        model_atomic_masses.push_back(atomic_masses[i].item<double>());
      if (model_atomic_masses.size() != model_atomic_numbers.size())
        plumed_merror(
          "Mismatch between model attributes: 'atomic_numbers' and 'atomic_masses'!"
        );
    }
  }

  // training time
  if (model.hasattr("training_time")) {
    auto ts = model.attr("training_time").toTensor();
    std::stringstream ss_time;
    ss_time
      << "UTC"
      << (ts[0].item<int64_t>() >= 0 ? "+" : "")
      << std::to_string(ts[0].item<int64_t>())
      << " "
      << std::to_string(ts[1].item<int64_t>())
      << "-"
      << std::setw(2) << std::setfill('0') << std::to_string(ts[2].item<int64_t>())
      << "-"
      << std::setw(2) << std::setfill('0') << std::to_string(ts[3].item<int64_t>())
      << " "
      << std::setw(2) << std::setfill('0') << std::to_string(ts[4].item<int64_t>())
      << ":"
      << std::setw(2) << std::setfill('0') << std::to_string(ts[5].item<int64_t>())
      << ":"
      << std::setw(2) << std::setfill('0') << std::to_string(ts[6].item<int64_t>());
    model_training_time = ss_time.str();
  } else {
    model_training_time = "unknown";
  }

  // idk ... stolen from:
  // https://stackoverflow.com/questions/77102532/libtorch-performance-issue-when-using-multiple-gpus-in-multiple-threads
  if (bailout_fusion) {
    torch::jit::FusionStrategy bailout = {
      {torch::jit::FusionBehavior::STATIC,  0},
      {torch::jit::FusionBehavior::DYNAMIC, 0},
    };
    torch::jit::setFusionStrategy(bailout);
  }

  // optimize model
  model.eval();
#ifdef DO_TORCH_FREEZE_HACK
  // NOTE: do the hack
  // copied from the implementation of torch::jit::freeze,
  // except without the broken check
  // see https://github.com/pytorch/pytorch/blob/dfbd030854359207cb3040b864614affeace11ce/torch/csrc/jit/api/module.cpp
  bool optimize_numerics = true;  // the default
  // the {} is preserved_attrs
  auto out_mod = torch::jit::freeze_module(model, {});
  // see 1.11 bugfix in https://github.com/pytorch/pytorch/pull/71436
  auto graph = out_mod.get_method("forward").graph();
  OptimizeFrozenGraph(graph, optimize_numerics);
  model = out_mod;
#else
  // do it normally
  model = torch::jit::freeze(model);
#endif

  // optimize model for inference
#ifndef DO_TORCH_FREEZE_HACK
  model = torch::jit::optimize_for_inference(model);
#endif

  // send the model to device
  model.to(device);

  // create system atomic numbers
  std::vector<int> atom_is_required(pdb.getAtomNumbers().size());
  for (size_t i = 0; i < atom_list_a.size(); i++) {
    int index = atom_list_a[i].index();
    atom_is_required[index] = 1;
  }
  for (size_t i = 0; i < atom_list_b.size(); i++) {
    int index = atom_list_b[i].index();
    atom_is_required[index] = 1;
  }

  for (size_t i = 0; i < pdb.getAtomNumbers().size(); i++) {
    AtomNumber index = pdb.getAtomNumbers()[i];
    std::string name = pdb.getAtomName(index);
    int number = atomic_number_from_name(name);
    auto iter = std::find(
      model_atomic_numbers.begin(),
      model_atomic_numbers.end(),
      number
    );
    if (iter == model_atomic_numbers.end()) {
      if (atom_is_required[i])
        plumed_merror(
          "Element '" + name + "' does not present in model " + model_file_name
        );
      else
        system_node_types.push_back(-1);
    } else {
      int node_type = std::distance(model_atomic_numbers.begin(), iter);
      system_node_types.push_back(node_type);
    }
  }

  // create components
  if (!is_committor) {
    for (int i = 0; i < n_out; i++) {
      string name_comp = "node-" + std::to_string(i);
      addComponentWithDerivatives(name_comp);
      componentIsNotPeriodic(name_comp);
    }
  } else {
    string name_comp_z = "node-0";
    addComponentWithDerivatives(name_comp_z);
    componentIsNotPeriodic(name_comp_z);
    string name_comp_q = "node-1";
    addComponent(name_comp_q);
    componentIsNotPeriodic(name_comp_q);
    if (k_bias) {
      string name_comp_b = "kbias";
      addComponentWithDerivatives(name_comp_b);
      componentIsNotPeriodic(name_comp_b);
    }
  }

  // initialize the neighbor list
  if (atom_list_b.size() > 0)
    neighbor_list = Tools::make_unique<NeighborList>(
      atom_list_a,
      atom_list_b,
      serial,
      false,
      pbc,
      getPbc(),
      comm,
      r_max + buffer,
      neighbor_list_stride
    );
  else
    neighbor_list = Tools::make_unique<NeighborList>(
      atom_list_a,
      serial,
      pbc,
      getPbc(),
      comm,
      r_max + buffer,
      neighbor_list_stride
    );
  requestAtoms(neighbor_list->getFullAtomList());

  // print log
  std::string thename = getLabel();
  if(atom_list_b.size() > 0) {
    log.printf(
      "  Will build graphs using %u system and %u environment atoms\n",
      static_cast<unsigned>(atom_list_a.size()),
      static_cast<unsigned>(atom_list_b.size())
    );
    log.printf("  System atom list (GROUPA):\n");
    for (unsigned int i = 0; i < atom_list_a.size(); i++) {
      if (((i + 1) % 10) == 0)
        log.printf("\n");
      log.printf("  %d", atom_list_a[i].serial());
    }
    log.printf("\n");
    log.printf("  Environment atom list (GROUPB):\n");
    for (unsigned int i = 0; i < atom_list_b.size(); i++) {
      if (((i + 1) % 10) == 0)
        log.printf("\n");
      log.printf("  %d", atom_list_b[i].serial());
    }
    log.printf("\n");
  } else {
    log.printf(
      "  Will build graphs using %u atoms\n",
      static_cast<unsigned>(atom_list_a.size())
    );
    log.printf("  Atom list:\n");
    for (unsigned int i = 0; i < atom_list_a.size(); i++) {
      if (((i + 1) % 10) == 0)
        log.printf("\n");
      log.printf("  %d", atom_list_a[i].serial());
    }
    log.printf("\n");
  }
  if (atom_list_sub_a.size() > 0) {
    log.printf(
      "  Will add long edges between %u atoms\n",
      static_cast<unsigned>(atom_list_sub_a.size())
    );
    log.printf("  Subsystem atom list:\n");
    for (unsigned int i = 0; i < atom_list_sub_a.size(); i++) {
      if (((i + 1) % 10) == 0)
        log.printf("\n");
      log.printf("  %d", atom_list_sub_a[i].serial());
    }
    log.printf("\n");
  }
  log << "  Model atomic numbers: " << model_atomic_numbers;
  log.printf("\n");
  log.printf("  Boundary conditions: ");
  if (pbc)
    log.printf("periodic\n");
  else
    log.printf("non-periodic\n");
  log.printf("  Neighbor List update stride: %d\n", neighbor_list_stride);
  log.printf("  Graph cutoff radius: %f (PLUMED length unit)\n", r_max);
  if (atom_list_b.size() > 0)
    log.printf("  Environment buffer size: %f (PLUMED length unit)\n", buffer);
  if (atom_list_sub_a.size() > 0)
    log.printf("  Subsystem cutoff radius: %f (PLUMED length unit)\n", r_max_l);
  log.printf("  Number of outputs: %d \n", n_out);
  log.printf("  Is this a committor model: ");
  if (is_committor)
    log.printf("yes\n");
  else
    log.printf("no\n");
  if (is_committor) {
    log.printf("  If sample Kang's (or Kolmogolov's) ensemble: ");
    if (k_bias)
      log.printf("yes\n");
    else
      log.printf("no\n");
    if (k_bias) {
      log.printf("  If calculate truncated V_K: ");
        if (kb_truncated)
          log.printf("yes\n");
        else
          log.printf("no\n");
      log.printf("  If calculate mass-weighted V_K: ");
      if (kb_weighted) {
        log.printf("yes\n");
        log << "  Model atomic masses: " << model_atomic_masses << "\n";
      } else {
        log.printf("no\n");
      }
      log.printf("  LAMBDA    value for calculating V_K: %f\n", kb_lambda);
      log.printf("  EPSILON   value for calculating V_K: %e\n", kb_epsilon);
      log.printf("  SIGMOID_P value for calculating V_K: %e\n", kb_sigmoid_p);
    }
    if (k_bias) {
      log << "  Output alignment: " + thename + ".kbias  -> V_K\n";
      log << "  Output alignment: " + thename + ".node-0 -> zeta\n";
    } else {
      log << "  Output alignment: " + thename + ".node-0 -> zeta\n";
    }
    log << "  Output alignment: " + thename + ".node-1 -> q (no grad)\n";
  }
  log.printf("  Will run on device: ");
  if (use_cuda)
    log.printf("CUDA\n");
  else if (required_cuda)
    log.printf("CPU (CUDA device not found/LibTorch does not support CUDA)\n");
  else
    log.printf("CPU (as required)\n");
  log << "  Model file name: " + model_file_name + "\n";
  log << "  Model training time: " + model_training_time + "\n";
  log << "  Model parameters: " + std::to_string(model_parameters) + "\n";
  log << "  Model architecture: \n";
  log << model_architecture;
  log << "  Bibliography: ";
  if (is_committor || r_max_l > 0 || atom_list_b.size() > 0)
    log << plumed.cite("Kang et al. arXiv preprint arXiv:2510.18018 (2025)");
  log << plumed.cite("Zhang et al., J. Chem. Theory Comput. 20, 24, 10787–10797 (2024)");
  log << plumed.cite("Bonati, Trizio, Rizzi and Parrinello, J. Chem. Phys. 159, 014801 (2023)");
  log << plumed.cite("Bonati, Rizzi and Parrinello, J. Phys. Chem. Lett. 11, 2998-3004 (2020)");
  log.printf("\n");
}

PytorchGNN::~PytorchGNN()
{
  return;
}

void PytorchGNN::prepare()
{
  if (neighbor_list->getStride() > 0) {
    if (firsttime || ((getStep() % neighbor_list->getStride()) == 0)) {
      requestAtoms(neighbor_list->getFullAtomList());
      invalidate_list = true;
      firsttime = false;
    } else {
      requestAtoms(neighbor_list->getReducedAtomList());
      invalidate_list = false;
      if (getExchangeStep())
        plumed_merror(
          "Neighbor lists should be updated on exchange steps - choose a NL_STRIDE which divides the exchange stride!"
        );
    }
    if (getExchangeStep())
      firsttime = true;
  }
}


void PytorchGNN::calculate()
{
  // get some common data
  auto pbc_tools = getPbc();
  int n_atoms = getNumberOfAtoms();
  std::vector<PLMD::Vector> x_local = getPositions();

  // threads
  int n_threads = OpenMP::getNumThreads();
  if (!serial)
    n_threads = std::min(n_threads, n_atoms);
  else
    n_threads = 1;

  // perform the size check
  if (system_node_types.size() != (size_t)atoms.getNatoms())
    plumed_merror(
      "Structure file '" +
      structure_file_name +
      "' has different number of atoms with the simulated system!"
    );

  // update the neighbor list and number of atoms
  if (neighbor_list->getStride() > 0 && invalidate_list)
    neighbor_list->update(x_local);
  if (atom_list_b.size() > 0)
    find_active_atoms(n_threads);
  n_atoms = (int)atom_list_active.size();
  n_threads = std::min(n_threads, n_atoms);

  // get the unit
  double to_ang = 10 * atoms.getUnits().getLength(); // TODO: remove the `atoms.` prefix when release

  // get the positions
  // TODO: now, the positions used by the model file is in unit of Angstrom.
  // We should warn the users about this default
  std::vector<float> positions_vector(n_atoms * 3);
  #pragma omp parallel for num_threads(n_threads)
  for (int i = 0; i < n_atoms; i++) {
    int index = atom_list_active[i];
    positions_vector[i * 3 + 0] = x_local[index][0] * to_ang;
    positions_vector[i * 3 + 1] = x_local[index][1] * to_ang;
    positions_vector[i * 3 + 2] = x_local[index][2] * to_ang;
  }
  torch::Tensor positions = torch::from_blob(
    positions_vector.data(),
    n_atoms * 3,
    torch::TensorOptions().dtype(torch::kFloat32)
  );
  positions = positions.to(device).to(torch_float_dtype);
  positions = positions.reshape({n_atoms, 3});

  // cell
  // TODO: now, the box data used by the model file is in unit of Angstrom.
  // We should warn the users about this default
  PLMD::Tensor box = getBox();
  std::vector<float> cell_vector(9);
  for (int i = 0; i < 3; i++) {
    for (int j = 0; j < 3; j++)
      cell_vector[i * 3 + j] = box[i][j] * to_ang;
  }
  torch::Tensor cell = torch::from_blob(
    cell_vector.data(),
    9,
    torch::TensorOptions().dtype(torch::kFloat32)
  );
  cell = cell.to(device).to(torch_float_dtype);
  cell = cell.reshape({3, 3});

  // build node attributes
  // TODO: now, the node attributes are in MACE's format.
  // We should try to give more options, or warn the users about this default
  int n_node_feats = (int)model_atomic_numbers.size();
  std::vector<float> node_attrs_vector(n_node_feats * n_atoms);
  #pragma omp parallel for num_threads(n_threads)
  for (int i = 0; i < n_atoms; i++) {
    int index = atom_list_active[i];
    int node_type = system_node_types[getAbsoluteIndex(index).index()];
    node_attrs_vector[i * n_node_feats + node_type] = 1.0;
  }
  torch::Tensor node_attrs = torch::from_blob(
    node_attrs_vector.data(),
    n_node_feats * n_atoms,
    torch::TensorOptions().dtype(torch::kFloat32)
  );
  node_attrs = node_attrs.to(device).to(torch_float_dtype);
  node_attrs = node_attrs.reshape({n_atoms, n_node_feats});

  // build edges
  int n_edges = 0;
  int n_edges_l = 0;
  torch::Tensor edge_index;

  if (atom_list_b.size() > 0) {
    n_edges = n_atoms * (n_atoms - 1);
    std::vector<float> distance_vector(n_edges);
    std::vector<std::vector<int64_t>> edge_index_vector;
    edge_index_vector.resize(2, std::vector<int64_t>(n_edges));

    #pragma omp parallel for num_threads(n_threads)
    for (int i = 0; i < n_atoms; i++) {
      int count = 0;
      for (int j = 0; j < n_atoms; j++) {
        if (i != j) {
          // NOTE: build edge index using the full local index list
          edge_index_vector[0][i * (n_atoms - 1) + count] = i;
          edge_index_vector[1][i * (n_atoms - 1) + count] = j;
          count++;
        }
      }
    }

    #pragma omp parallel for num_threads(n_threads)
    for (int i = 0; i < n_edges; i++) {
      distance_vector[i] = pbc_tools.distance(
        pbc,
        x_local[atom_list_active[edge_index_vector[0][i]]],
        x_local[atom_list_active[edge_index_vector[1][i]]]
      );
    }

    torch::Tensor distances = torch::from_blob(
      distance_vector.data(),
      n_edges,
      torch::TensorOptions().dtype(torch::kFloat32)
    );
    torch::Tensor senders = torch::from_blob(
      edge_index_vector[0].data(),
      n_edges,
      torch::TensorOptions().dtype(torch::kInt64)
    );
    torch::Tensor receivers = torch::from_blob(
      edge_index_vector[1].data(),
      n_edges,
      torch::TensorOptions().dtype(torch::kInt64)
    );

    const torch::Tensor mask = distances <= r_max;
    senders = senders.index({mask});
    receivers = receivers.index({mask});
    edge_index = torch::vstack({senders, receivers});
    n_edges = (int)edge_index.size(1);
  } else {
    n_edges = (int)neighbor_list->size() * 2;
    int n_pairs = (int)neighbor_list->size();
    std::vector<std::vector<int64_t>> edge_index_vector;
    edge_index_vector.resize(2, std::vector<int64_t>(n_edges));

    #pragma omp parallel for num_threads(n_threads)
    for (int i = 0; i < n_pairs; i++) {
      auto pair = neighbor_list->getClosePair(i);
      edge_index_vector[0][i] = pair.first;
      edge_index_vector[1][i] = pair.second;
      edge_index_vector[0][n_pairs + i] = pair.second;
      edge_index_vector[1][n_pairs + i] = pair.first;
    }

    torch::Tensor senders = torch::from_blob(
      edge_index_vector[0].data(),
      n_edges,
      torch::TensorOptions().dtype(torch::kInt64)
    );
    torch::Tensor receivers = torch::from_blob(
      edge_index_vector[1].data(),
      n_edges,
      torch::TensorOptions().dtype(torch::kInt64)
    );
    edge_index = torch::vstack({senders, receivers});
  }

  if (atom_list_sub_a.size() > 0) {
    torch::Tensor edge_index_l;
    int n_atoms_l = atom_list_sub_a.size();
    n_edges_l = n_atoms_l * (n_atoms_l - 1);
    std::vector<float> distance_vector_l(n_edges_l);
    std::vector<std::vector<int64_t>> edge_index_vector_l;
    edge_index_vector_l.resize(2, std::vector<int64_t>(n_edges_l));

    #pragma omp parallel for num_threads(n_threads)
    for (int i = 0; i < n_atoms_l; i++) {
      int count = 0;
      for (int j = 0; j < n_atoms_l; j++) {
        if (i != j) {
          // NOTE: build edge index using the partial local index list
          edge_index_vector_l[0][
            i * (n_atoms_l - 1) + count
          ] = atom_list_active_subgroup[i];
          edge_index_vector_l[1][
            i * (n_atoms_l - 1) + count
          ] = atom_list_active_subgroup[j];
          count++;
        }
      }
    }

    #pragma omp parallel for num_threads(n_threads)
    for (int i = 0; i < n_edges_l; i++) {
      distance_vector_l[i] = pbc_tools.distance(
        pbc,
        x_local[edge_index_vector_l[0][i]],
        x_local[edge_index_vector_l[1][i]]
      );
    }

    torch::Tensor distances_l = torch::from_blob(
      distance_vector_l.data(),
      n_edges_l,
      torch::TensorOptions().dtype(torch::kFloat32)
    );
    const torch::Tensor mask_l = distances_l <= r_max_l;

    if (mask_l.size(0) > 0) {
      torch::Tensor senders_l = torch::from_blob(
        edge_index_vector_l[0].data(),
        n_edges_l,
        torch::TensorOptions().dtype(torch::kInt64)
      );
      torch::Tensor receivers_l = torch::from_blob(
        edge_index_vector_l[1].data(),
        n_edges_l,
        torch::TensorOptions().dtype(torch::kInt64)
      );
      senders_l = senders_l.index({mask_l});
      receivers_l = receivers_l.index({mask_l});
      edge_index_l = torch::vstack({senders_l, receivers_l});
      n_edges_l = (int)edge_index_l.size(1);

      edge_index = torch::hstack({edge_index, edge_index_l});
      n_edges = n_edges + n_edges_l;
    }
  }

  edge_index = edge_index.to(device);

  // edge shifts
  torch::Tensor shifts;
  torch::Tensor unit_shifts;

  if (pbc) {
    if (pbc_tools.isOrthorombic()) {
      auto deltas = (
        torch::index_select(positions, 0, edge_index[0])
        - torch::index_select(positions, 0, edge_index[1])
      );
      unit_shifts = torch::round(deltas / torch::diagonal(cell, 0));
      shifts = unit_shifts * torch::diagonal(cell, 0);
    } else {
      auto cell_inv = torch::linalg_pinv(cell.transpose(1, 0));
      auto positions_s = torch::matmul(
        cell_inv, positions.transpose(1, 0)
      );
      auto deltas = (
        torch::index_select(positions_s, 1, edge_index[0])
        - torch::index_select(positions_s, 1, edge_index[1])
      );
      unit_shifts = torch::round(deltas).transpose(1, 0);
      shifts = torch::matmul(unit_shifts, cell);
    }
  } else {
    shifts = torch::zeros({n_edges, 3}, torch_float_dtype);
    unit_shifts = torch::zeros({n_edges, 3}, torch_float_dtype);
  }

  shifts = shifts.to(device);
  unit_shifts = unit_shifts.to(device);

  // other things
  // TODO: some of these things are required by MACE. We should disable
  // some of them when not using MACE, maybe by distinguishing the MACE model.
  auto batch = torch::zeros({n_atoms}, torch::dtype(torch::kInt64));
  auto ptr = torch::empty({2}, torch::dtype(torch::kInt64));
  auto weight = torch::empty({1}, torch_float_dtype);
  ptr[0] = 0;
  ptr[1] = n_atoms;
  weight[0] = 1.0;

  // load data to device
  // TODO: some of these things are required by MACE. We should disable
  // some of them when not using MACE, maybe by distinguishing the MACE model.
  batch = batch.to(device);
  ptr = ptr.to(device);
  weight = weight.to(device);

  // pack the input, call the model
  // TODO: some of these things are required by MACE. We should disable
  // some of them when not using MACE, maybe by distinguishing the MACE model.
  c10::Dict<std::string, torch::Tensor> input;
  input.insert("batch", batch);
  input.insert("cell", cell);
  input.insert("edge_index", edge_index);
  input.insert("node_attrs", node_attrs);
  input.insert("positions", positions);
  input.insert("ptr", ptr);
  input.insert("weight", weight);
  input.insert("shifts", shifts);
  input.insert("unit_shifts", unit_shifts);

  // Optional fields.
  if (atom_list_b.size() > 0) {
    auto system_masks = torch::vstack({
      torch::ones(
        {(int64_t)atom_list_a.size(), 1}, torch::dtype(torch::kBool)
      ),
      torch::zeros(
        {n_atoms - (int64_t)atom_list_a.size(), 1}, torch::dtype(torch::kBool)
      ),
    });
    system_masks = system_masks.to(device);
    input.insert("system_masks", system_masks);

    auto n_system = torch::ones({1, 1}, torch::dtype(torch::kInt64));
    n_system = n_system.to(device);
    n_system[0][0] = (int64_t)atom_list_a.size();
    input.insert("n_system", n_system);
  }

  if (atom_list_sub_a.size() > 0){
    auto edge_masks_le = torch::vstack({
      torch::zeros({n_edges - n_edges_l, 1}, torch::dtype(torch::kBool)),
      torch::ones({n_edges_l, 1}, torch::dtype(torch::kBool)),
    });
    edge_masks_le = edge_masks_le.to(device);
    input.insert("edge_masks_le", edge_masks_le);
  }

  // TODO: figure out how to enable virials. Maybe we could port MACE's python
  // code to our python module.
  auto output = model.forward({input, false}).toTensor();

  // helper variables
  std::vector<PLMD::Vector> derivatives(n_atoms);
  auto grad_output = torch::ones({1}).expand({1, 1}).to(device);

  if (!is_committor) {
    for (int i = 0; i < n_out; i++) {
      // set CV values
      string name_comp = "node-" + std::to_string(i);
      getPntrToComponent(name_comp)->set(output[0][i].cpu().item<double>());
      // set derivatives
      auto gradients = torch::autograd::grad(
        {output.slice(1, i, (i + 1))},
        {positions},
        {grad_output}, // grad_outputs
        true,          // retain_graph
        false          // create_graph
      )[0].cpu();
      #pragma omp parallel for num_threads(n_threads)
      for (int j = 0; j < n_atoms; j++) {
        derivatives[j][0] = gradients[j][0].item<double>() * to_ang;
        derivatives[j][1] = gradients[j][1].item<double>() * to_ang;
        derivatives[j][2] = gradients[j][2].item<double>() * to_ang;
      }
      #pragma omp parallel for num_threads(n_threads)
      for (int j = 0; j < n_atoms; j++) {
        int index = atom_list_active[j];
        setAtomsDerivatives(
          getPntrToComponent(name_comp), index, derivatives[j]
        );
      }
    }
  } else if (!k_bias) {
    // set committor values
    string name_comp_z = "node-0";
    getPntrToComponent(name_comp_z)->set(output[0][0].cpu().item<double>());
    string name_comp_q = "node-1";
    getPntrToComponent(name_comp_q)->set(output[0][1].cpu().item<double>());
    // set derivatives of z
    auto gradients = torch::autograd::grad(
      {output.slice(1, 0, 1)},
      {positions},
      {grad_output}, // grad_outputs
      false,         // retain_graph
      false          // create_graph
    )[0].cpu();
    #pragma omp parallel for num_threads(n_threads)
    for (int j = 0; j < n_atoms; j++) {
      derivatives[j][0] = gradients[j][0].item<double>() * to_ang;
      derivatives[j][1] = gradients[j][1].item<double>() * to_ang;
      derivatives[j][2] = gradients[j][2].item<double>() * to_ang;
    }
    #pragma omp parallel for num_threads(n_threads)
    for (int j = 0; j < n_atoms; j++) {
      int index = atom_list_active[j];
      setAtomsDerivatives(
        getPntrToComponent(name_comp_z), index, derivatives[j]
      );
    }
  } else {
    auto z = output[0][0];
    auto q = output[0][1];
    auto epsilon = torch::tensor(kb_epsilon, torch_float_dtype).to(device);
    auto sigmoid_p = torch::tensor(kb_sigmoid_p, torch_float_dtype).to(device);
    // get bias value
    auto gradients_z = torch::autograd::grad(
      {z},
      {positions},
      {grad_output}, // grad_outputs
      true,          // retain_graph
      true           // create_graph
    )[0];
    // square
    auto gradients_z_2 = torch::pow(gradients_z, 2);
    // mass-weighting
    if (kb_weighted) {
      std::vector<float> node_masses_vector(n_atoms);
      #pragma omp parallel for num_threads(n_threads)
      for (int i = 0; i < n_atoms; i++) {
        int index = atom_list_active[i];
        int node_type = system_node_types[getAbsoluteIndex(index).index()];
        node_masses_vector[i] = model_atomic_masses[node_type];
      }
      torch::Tensor node_masses = torch::from_blob(
        node_masses_vector.data(),
        n_atoms,
        torch::TensorOptions().dtype(torch::kFloat32)
      );
      node_masses = node_masses.to(device).to(torch_float_dtype);
      node_masses = node_masses.reshape({n_atoms, 1});
      gradients_z_2 = gradients_z_2 / node_masses;
    }
    // sum over all dims
    auto gradients_z_sum = torch::sum(gradients_z_2);
    // chain rules
    auto k_bias_value = torch::ones(1, torch_float_dtype).to(device);
    if (!kb_truncated)
      k_bias_value = kb_lambda * (
        torch::log(gradients_z_sum + epsilon)
        - 4.0 * torch::log(1.0 + torch::exp(-sigmoid_p * z))
        - 2.0 * sigmoid_p * z
        - torch::log(epsilon)
      );
    else
      k_bias_value = kb_lambda * (
        torch::log(gradients_z_sum * torch::pow(q * (1 - q), 2) + epsilon)
        - torch::log(epsilon)
      );

    // set committor values
    string name_comp_z = "node-0";
    getPntrToComponent(name_comp_z)->set(output[0][0].cpu().item<double>());
    string name_comp_q = "node-1";
    getPntrToComponent(name_comp_q)->set(output[0][1].cpu().item<double>());
    string name_comp_b = "kbias";
    getPntrToComponent(name_comp_b)->set(k_bias_value.cpu().item<double>());
    // set derivatives of z
    #pragma omp parallel for num_threads(n_threads)
    for (int j = 0; j < n_atoms; j++) {
      derivatives[j][0] = gradients_z[j][0].item<double>() * to_ang;
      derivatives[j][1] = gradients_z[j][1].item<double>() * to_ang;
      derivatives[j][2] = gradients_z[j][2].item<double>() * to_ang;
    }
    #pragma omp parallel for num_threads(n_threads)
    for (int j = 0; j < n_atoms; j++) {
      int index = atom_list_active[j];
      setAtomsDerivatives(
        getPntrToComponent(name_comp_z), index, derivatives[j]
      );
    }
    // set derivatives of bias
    auto gradients_b = torch::autograd::grad(
      {k_bias_value},
      {positions},
      {grad_output}, // grad_outputs
      false,         // retain_graph
      false          // create_graph
    )[0].cpu();
    #pragma omp parallel for num_threads(n_threads)
    for (int j = 0; j < n_atoms; j++) {
      derivatives[j][0] = gradients_b[j][0].item<double>() * to_ang;
      derivatives[j][1] = gradients_b[j][1].item<double>() * to_ang;
      derivatives[j][2] = gradients_b[j][2].item<double>() * to_ang;
    }
    #pragma omp parallel for num_threads(n_threads)
    for (int j = 0; j < n_atoms; j++) {
      int index = atom_list_active[j];
      setAtomsDerivatives(
        getPntrToComponent(name_comp_b), index, derivatives[j]
      );
    }
  }
}

int PytorchGNN::atomic_number_from_name(std::string name)
{
  std::transform(
    name.begin(),
    name.end(),
    name.begin(),
    [](unsigned char c){return std::tolower(c);}
  );
  auto iter = std::find(periodic_table.begin(), periodic_table.end(), name);
  if (iter == periodic_table.end())
    plumed_merror(
      "Can not find element name '" + name + "' from the periodic table!"
    );
  return std::distance(periodic_table.begin(), iter) + 1;
}

std::string PytorchGNN::model_summary(
    std::string model_name, torch::jit::Module module, int level_max, int level
) {
  std::stringstream ss;

  std::string model_type = module.type()->name()->name();
  ss << "  (" << model_name << "): " << model_type;

  if (module.named_children().size() != 0) {
    if (level <= level_max) {
      ss << " {\n";
      for (const torch::jit::NameModule& s : module.named_children())
        ss << torch::jit::jit_log_prefix(
          "  ",
          model_summary(s.name, s.value, level_max, level + 1)
        );
      ss << "  }\n";
    } else {
      ss << " { ... }";
    }
  } else {
    ss << "\n";
  }

  return ss.str();
}

bool PytorchGNN::groups_have_intersection(void) {
  std::vector<AtomNumber> intersections;
  std::vector<AtomNumber> atom_list_a_copy(atom_list_a);
  std::vector<AtomNumber> atom_list_b_copy(atom_list_b);

  std::sort(atom_list_a_copy.begin(), atom_list_a_copy.end());
  std::sort(atom_list_b_copy.begin(), atom_list_b_copy.end());

  std::set_intersection(
    atom_list_a_copy.begin(),
    atom_list_a_copy.end(),
    atom_list_b_copy.begin(),
    atom_list_b_copy.end(),
    back_inserter(intersections)
  );

  return intersections.size() > 0;
}

bool PytorchGNN::subgroup_is_in_group_a(void) {
  std::vector<AtomNumber> atom_list_a_copy(atom_list_a);
  std::vector<AtomNumber> atom_list_sub_a_copy(atom_list_sub_a);
  for (auto atom_elt: atom_list_sub_a_copy)
    if (
      std::find(atom_list_a_copy.begin(),
      atom_list_a_copy.end(), atom_elt) == atom_list_a_copy.end()
    )
      return false;
  return true;
}

void PytorchGNN::find_active_atoms(int n_threads) {
  if (atom_list_b.size() > 0) {
    atom_list_active.clear();
    std::vector<int> neighbors(neighbor_list->size());

    #pragma omp parallel for num_threads(n_threads)
    for (size_t i = 0; i < neighbor_list->size(); i++)
      neighbors[i] = neighbor_list->getClosePair(i).second;

    // TODO: make this faster
    std::unordered_set<int> neighbors_set;
    for (int i : neighbors)
      neighbors_set.insert(i);
    neighbors.assign(neighbors_set.begin(), neighbors_set.end());

    // NOTE: the system atoms (atom_list_a) should always appear at the head of
    // this list. Do NOT change the order!
    for (size_t i = 0; i < atom_list_a.size(); i++)
      atom_list_active.push_back(i);
    // NOTE: the neighbors should be appended to the tail of this list.
    for (size_t i = 0; i < neighbors.size(); i++)
      atom_list_active.push_back(neighbors[i]);
  } else if (atom_list_active.size() == 0) {
    atom_list_active.clear();

    for (size_t i = 0; i < atom_list_a.size(); i++)
      atom_list_active.push_back(i);
  }
}

void PytorchGNN::find_active_subgroup_atoms(void) {
  atom_list_active_subgroup.clear();
  // NOTE: since system atoms (atom_list_a) always appear at the head of the
  // local indices, we simply find subsystem indices in atom_list_a.
  for (auto atom_sub: atom_list_sub_a) {
    int index = (int)std::distance(
      atom_list_a.begin(),
      find(atom_list_a.begin(), atom_list_a.end(), atom_sub)
    );
    atom_list_active_subgroup.push_back(index);
  }
}

int PytorchGNN::get_number_of_parameters(void)
{
  int n_parameters = 0;

  for (auto p: model.parameters())
    n_parameters += p.numel();

  return n_parameters;
}

} // pytorch_gnn

} // colvar

} // PLMD

#endif // PLUMED_HAS_LIBTORCH