Feat(pt): Support Density fitting

deepmd/infer/deep_pot.py

@@ -212,6 +212,11 @@ def eval(
             aparam=aparam,
             **kwargs,
         )
+        # TODO: if the grid is requested, we can directly return it without reshaping to energy, force and virial. We can also consider to return the grid in a separate key in the results dict, instead of reshaping it to energy, force and virial.
+        if "grid" in kwargs:
+            result = results["density"].reshape(nframes, -1)
+            return result
+
         energy = results["energy_redu"].reshape(nframes, 1)
         force = results["energy_derv_r"].reshape(nframes, natoms, 3)
         virial = results["energy_derv_c_redu"].reshape(nframes, 9)

deepmd/pt/entrypoints/main.py

@@ -421,7 +421,12 @@ def train(
 
     # Initialize DDP
     if os.environ.get("LOCAL_RANK") is not None:
-        dist.init_process_group(backend="cuda:nccl,cpu:gloo")
+        import datetime
+
+        timeout = datetime.timedelta(
+            seconds=18000
+        )  # set a longer timeout for for large datasets or slow file systems
+        dist.init_process_group(backend="cuda:nccl,cpu:gloo", timeout=timeout)
 
     trainer = get_trainer(
         config,

deepmd/pt/infer/deep_eval.py

@@ -406,11 +406,7 @@ def eval(
             coords, atom_types, len(atom_types.shape) > 1
         )
         request_defs = self._get_request_defs(atomic)
-        if "spin" not in kwargs or kwargs["spin"] is None:
-            out = self._eval_func(self._eval_model, numb_test, natoms)(
-                coords, cells, atom_types, fparam, aparam, request_defs, charge_spin
-            )
-        else:
+        if "spin" in kwargs and kwargs["spin"] is not None:
             out = self._eval_func(self._eval_model_spin, numb_test, natoms)(
                 coords,
                 cells,
@@ -421,6 +417,21 @@ def eval(
                 request_defs,
                 charge_spin,
             )
+        elif "grid" in kwargs and kwargs["grid"] is not None:
+            out = self._eval_func(self._eval_model_density, numb_test, natoms)(
+                coords,
+                cells,
+                atom_types,
+                np.array(kwargs["grid"]),
+                fparam,
+                aparam,
+                request_defs,
+            )
+            return {"density": out}
+        else:
+            out = self._eval_func(self._eval_model, numb_test, natoms)(
+                coords, cells, atom_types, fparam, aparam, request_defs
+            )
         return dict(
             zip(
                 [x.name for x in request_defs],
@@ -688,6 +699,80 @@ def _eval_model_spin(
                 )  # this is kinda hacky
         return tuple(results)
 
+    def _eval_model_density(
+        self,
+        coords: np.ndarray,
+        cells: np.ndarray | None,
+        atom_types: np.ndarray,
+        grid: np.ndarray,
+        fparam: np.ndarray | None,
+        aparam: np.ndarray | None,
+        request_defs: list[OutputVariableDef],
+    ):
+        model = self.dp.to(DEVICE)
+
+        nframes = coords.shape[0]
+        if len(atom_types.shape) == 1:
+            natoms = len(atom_types)
+            atom_types = np.tile(atom_types, nframes).reshape(nframes, -1)
+        else:
+            natoms = len(atom_types[0])
+
+        coord_input = torch.tensor(
+            coords.reshape([nframes, natoms, 3]),
+            dtype=GLOBAL_PT_FLOAT_PRECISION,
+            device=DEVICE,
+        )
+        type_input = torch.tensor(atom_types, dtype=torch.long, device=DEVICE)
+        grid_input = torch.tensor(
+            grid.reshape([nframes, -1, 3]),
+            dtype=GLOBAL_PT_FLOAT_PRECISION,
+            device=DEVICE,
+        )
+        ngrid = grid_input.shape[1]
+        if cells is not None:
+            box_input = torch.tensor(
+                cells.reshape([nframes, 3, 3]),
+                dtype=GLOBAL_PT_FLOAT_PRECISION,
+                device=DEVICE,
+            )
+        else:
+            box_input = None
+        if fparam is not None:
+            fparam_input = to_torch_tensor(
+                fparam.reshape(nframes, self.get_dim_fparam())
+            )
+        else:
+            fparam_input = None
+        if aparam is not None:
+            aparam_input = to_torch_tensor(
+                aparam.reshape(nframes, natoms, self.get_dim_aparam())
+            )
+        else:
+            aparam_input = None
+
+        do_atomic_virial = any(
+            x.category == OutputVariableCategory.DERV_C_REDU for x in request_defs
+        )
+        batch_output = model(
+            coord_input,
+            type_input,
+            grid=grid_input,
+            box=box_input,
+            do_atomic_virial=do_atomic_virial,
+            fparam=fparam_input,
+            aparam=aparam_input,
+        )
+        if isinstance(batch_output, tuple):
+            batch_output = batch_output[0]
+
+        results = []
+        pt_name = "density"
+        density_shape = [nframes, ngrid]
+        out = batch_output[pt_name].reshape(density_shape).detach().cpu().numpy()
+        results.append(out)
+        return tuple(results)
+
     def _get_output_shape(
         self, odef: OutputVariableDef, nframes: int, natoms: int
     ) -> list[int]:

deepmd/pt/loss/__init__.py

@@ -1,4 +1,7 @@
 # SPDX-License-Identifier: LGPL-3.0-or-later
+from .charge import (
+    GridDensityLoss,
+)
 from .denoise import (
     DenoiseLoss,
 )
@@ -28,6 +31,7 @@
     "EnergyHessianStdLoss",
     "EnergySpinLoss",
     "EnergyStdLoss",
+    "GridDensityLoss",
     "PropertyLoss",
     "TaskLoss",
     "TensorLoss",

deepmd/pt/loss/charge.py

@@ -0,0 +1,125 @@
+# SPDX-License-Identifier: LGPL-3.0-or-later
+import torch
+
+from deepmd.pt.loss.loss import (
+    TaskLoss,
+)
+from deepmd.pt.utils import (
+    env,
+)
+from deepmd.pt.utils.env import (
+    GLOBAL_PT_FLOAT_PRECISION,
+)
+from deepmd.utils.data import (
+    DataRequirementItem,
+)
+
+
+class GridDensityLoss(TaskLoss):
+    def __init__(
+        self,
+        starter_learning_rate=1.0,
+        start_pref_d=0.0,
+        limit_pref_d=0.0,
+        inference=False,
+        **kwargs,
+    ):
+        r"""Construct a layer to compute loss on grid density.
+
+        Parameters
+        ----------
+        starter_learning_rate : float
+            The learning rate at the start of the training.
+        start_pref_d : float
+            The prefactor of charge density loss at the start of the training.
+        limit_pref_d : float
+            The prefactor of charge density loss at the end of the training.
+        inference : bool
+            If true, it will output all losses found in output, ignoring the pre-factors.
+        **kwargs
+            Other keyword arguments.
+        """
+        super().__init__()
+        self.starter_learning_rate = starter_learning_rate
+        self.has_d = (start_pref_d != 0.0 and limit_pref_d != 0.0) or inference
+
+        self.start_pref_d = start_pref_d
+        self.limit_pref_d = limit_pref_d
+        self.inference = inference
+
+    def forward(self, input_dict, model, label, natoms, learning_rate, mae=False):
+        """Return loss on energy and force.
+
+        Parameters
+        ----------
+        input_dict : dict[str, torch.Tensor]
+            Model inputs.
+        model : torch.nn.Module
+            Model to be used to output the predictions.
+        label : dict[str, torch.Tensor]
+            Labels.
+        natoms : int
+            The local atom number.
+
+        Returns
+        -------
+        model_pred: dict[str, torch.Tensor]
+            Model predictions.
+        loss: torch.Tensor
+            Loss for model to minimize.
+        more_loss: dict[str, torch.Tensor]
+            Other losses for display.
+        """
+        model_pred = model(**input_dict)
+        coef = learning_rate / self.starter_learning_rate
+        pref_d = self.limit_pref_d + (self.start_pref_d - self.limit_pref_d) * coef
+
+        loss = torch.zeros(1, dtype=env.GLOBAL_PT_FLOAT_PRECISION, device=env.DEVICE)[0]
+        more_loss = {}
+        # more_loss['log_keys'] = []  # showed when validation on the fly
+        # more_loss['test_keys'] = []  # showed when doing dp test
+        atom_norm = 1.0 / natoms
+        if self.has_d and "density" in model_pred and "density" in label:
+            density_pred = model_pred["density"]
+            density_label = label["density"]
+            find_density = label.get("find_density", 0.0)
+            pref_d = pref_d * find_density
+            density_pred_reshape = density_pred.reshape(-1)
+            density_label_reshape = density_label.reshape(-1)
+            l2_density_loss = torch.square(
+                density_label_reshape - density_pred_reshape
+            ).mean()
+            rmse_d = l2_density_loss.sqrt()
+            more_loss["rmse_d"] = self.display_if_exist(rmse_d.detach(), find_density)
+            l1_density_loss = torch.abs(
+                density_label_reshape - density_pred_reshape
+            ).mean()
+            loss += (pref_d * l1_density_loss).to(GLOBAL_PT_FLOAT_PRECISION)
+            mae_d = l1_density_loss
+            more_loss["mae_d"] = self.display_if_exist(mae_d.detach(), find_density)
+        return model_pred, loss, more_loss
+
+    @property
+    def label_requirement(self) -> list[DataRequirementItem]:
+        """Return data label requirements needed for this loss calculation."""
+        label_requirement = []
+        label_requirement.append(
+            DataRequirementItem(
+                "grid",
+                ndof=3,
+                atomic=True,  # the grid is defined for each atom, so it is atomic
+                must=True,
+                high_prec=True,
+            )
+        )
+        if self.has_d:
+            label_requirement.append(
+                DataRequirementItem(
+                    "density",
+                    ndof=1,
+                    atomic=True,
+                    must=False,
+                    high_prec=True,
+                )
+            )
+        return label_requirement

deepmd/pt/model/atomic_model/__init__.py

@@ -17,6 +17,9 @@
 from .base_atomic_model import (
     BaseAtomicModel,
 )
+from .density_atomic_model import (
+    DPDensityAtomicModel,
+)
 from .dipole_atomic_model import (
     DPDipoleAtomicModel,
 )
@@ -47,6 +50,7 @@
     "BaseAtomicModel",
     "DPAtomicModel",
     "DPDOSAtomicModel",
+    "DPDensityAtomicModel",
     "DPDipoleAtomicModel",
     "DPEnergyAtomicModel",
     "DPPolarAtomicModel",