feat(pt/tf/dp): support econf type embedding

deepmd/dpmodel/descriptor/dpa1.py

@@ -193,6 +193,12 @@ class DescrptDPA1(NativeOP, BaseDescriptor):
             Setting this parameter to `True` is equivalent to setting `tebd_input_mode` to 'strip'.
             Setting it to `False` is equivalent to setting `tebd_input_mode` to 'concat'.
             The default value is `None`, which means the `tebd_input_mode` setting will be used instead.
+    use_econf_tebd: bool, Optional
+            Whether to use electronic configuration type embedding.
+    type_map: List[str], Optional
+            A list of strings. Give the name to each type of atoms.
+            Only used if `use_econf_tebd` is `True` in type embedding net.
+
     spin
             (Only support None to keep consistent with other backend references.)
             (Not used in this version. Not-none option is not implemented.)
@@ -242,6 +248,8 @@ def __init__(
         concat_output_tebd: bool = True,
         spin: Optional[Any] = None,
         stripped_type_embedding: Optional[bool] = None,
+        use_econf_tebd: bool = False,
+        type_map: Optional[List[str]] = None,
         # consistent with argcheck, not used though
         seed: Optional[int] = None,
     ) -> None:
@@ -287,12 +295,16 @@ def __init__(
             trainable_ln=trainable_ln,
             ln_eps=ln_eps,
         )
+        self.use_econf_tebd = use_econf_tebd
+        self.type_map = type_map
         self.type_embedding = TypeEmbedNet(
             ntypes=ntypes,
             neuron=[tebd_dim],
             padding=True,
             activation_function="Linear",
             precision=precision,
+            use_econf_tebd=use_econf_tebd,
+            type_map=type_map,
         )
         self.tebd_dim = tebd_dim
         self.concat_output_tebd = concat_output_tebd
@@ -457,6 +469,8 @@ def serialize(self) -> dict:
             "smooth_type_embedding": obj.smooth,
             "type_one_side": obj.type_one_side,
             "concat_output_tebd": self.concat_output_tebd,
+            "use_econf_tebd": self.use_econf_tebd,
+            "type_map": self.type_map,
             # make deterministic
             "precision": np.dtype(PRECISION_DICT[obj.precision]).name,
             "embeddings": obj.embeddings.serialize(),

deepmd/dpmodel/descriptor/dpa2.py

@@ -323,6 +323,8 @@ def __init__(
         trainable: bool = True,
         seed: Optional[int] = None,
         add_tebd_to_repinit_out: bool = False,
+        use_econf_tebd: bool = False,
+        type_map: Optional[List[str]] = None,
     ):
         r"""The DPA-2 descriptor. see https://arxiv.org/abs/2312.15492.
 
@@ -350,6 +352,11 @@ def __init__(
             (Unused yet) Random seed for parameter initialization.
         add_tebd_to_repinit_out : bool, optional
             Whether to add type embedding to the output representation from repinit before inputting it into repformer.
+        use_econf_tebd : bool, Optional
+            Whether to use electronic configuration type embedding.
+        type_map : List[str], Optional
+            A list of strings. Give the name to each type of atoms.
+            Only used if `use_econf_tebd` is `True` in type embedding net.
 
         Returns
         -------
@@ -433,12 +440,16 @@ def init_subclass_params(sub_data, sub_class):
             trainable_ln=self.repformer_args.trainable_ln,
             ln_eps=self.repformer_args.ln_eps,
         )
+        self.use_econf_tebd = use_econf_tebd
+        self.type_map = type_map
         self.type_embedding = TypeEmbedNet(
             ntypes=ntypes,
             neuron=[self.repinit_args.tebd_dim],
             padding=True,
             activation_function="Linear",
             precision=precision,
+            use_econf_tebd=use_econf_tebd,
+            type_map=type_map,
         )
         self.concat_output_tebd = concat_output_tebd
         self.precision = precision
@@ -641,6 +652,8 @@ def serialize(self) -> dict:
             "env_protection": self.env_protection,
             "trainable": self.trainable,
             "add_tebd_to_repinit_out": self.add_tebd_to_repinit_out,
+            "use_econf_tebd": self.use_econf_tebd,
+            "type_map": self.type_map,
             "type_embedding": self.type_embedding.serialize(),
             "g1_shape_tranform": self.g1_shape_tranform.serialize(),
         }

deepmd/dpmodel/utils/type_embed.py

@@ -39,6 +39,11 @@ class TypeEmbedNet(NativeOP):
         Random seed for initializing the network parameters.
     padding
         Concat the zero padding to the output, as the default embedding of empty type.
+    use_econf_tebd: bool, Optional
+        Whether to use electronic configuration type embedding.
+    type_map: List[str], Optional
+        A list of strings. Give the name to each type of atoms.
+        Only used if `use_econf_tebd` is `True` in type embedding net.
     """
 
     def __init__(
@@ -52,6 +57,8 @@ def __init__(
         trainable: bool = True,
         seed: Optional[int] = None,
         padding: bool = False,
+        use_econf_tebd: bool = False,
+        type_map: Optional[List[str]] = None,
     ) -> None:
         self.ntypes = ntypes
         self.neuron = neuron
@@ -61,8 +68,33 @@ def __init__(
         self.activation_function = str(activation_function)
         self.trainable = trainable
         self.padding = padding
+        self.use_econf_tebd = use_econf_tebd
+        self.type_map = type_map
+        embed_input_dim = ntypes
+        if self.use_econf_tebd:
+            from deepmd.utils.econf_embd import (
+                ECONF_DIM,
+                electronic_configuration_embedding,
+            )
+            from deepmd.utils.econf_embd import type_map as periodic_table
+
+            assert (
+                self.type_map is not None
+            ), "When using electronic configuration type embedding, type_map must be provided!"
+
+            missing_types = [t for t in self.type_map if t not in periodic_table]
+            assert not missing_types, (
+                "When using electronic configuration type embedding, "
+                "all element in type_map should be in periodic table! "
+                f"Found these invalid elements: {missing_types}"
+            )
+            self.econf_tebd = np.array(
+                [electronic_configuration_embedding[kk] for kk in self.type_map],
+                dtype=PRECISION_DICT[self.precision],
+            )
+            embed_input_dim = ECONF_DIM
         self.embedding_net = EmbeddingNet(
-            ntypes,
+            embed_input_dim,
             self.neuron,
             self.activation_function,
             self.resnet_dt,
@@ -71,9 +103,12 @@ def __init__(
 
     def call(self) -> np.ndarray:
         """Compute the type embedding network."""
-        embed = self.embedding_net(
-            np.eye(self.ntypes, dtype=PRECISION_DICT[self.precision])
-        )
+        if not self.use_econf_tebd:
+            embed = self.embedding_net(
+                np.eye(self.ntypes, dtype=PRECISION_DICT[self.precision])
+            )
+        else:
+            embed = self.embedding_net(self.econf_tebd)
         if self.padding:
             embed = np.pad(embed, ((0, 1), (0, 0)), mode="constant")
         return embed
@@ -120,5 +155,7 @@ def serialize(self) -> dict:
             "activation_function": self.activation_function,
             "trainable": self.trainable,
             "padding": self.padding,
+            "use_econf_tebd": self.use_econf_tebd,
+            "type_map": self.type_map,
             "embedding": self.embedding_net.serialize(),
         }

deepmd/pt/model/descriptor/dpa1.py

@@ -172,6 +172,11 @@ class DescrptDPA1(BaseDescriptor, torch.nn.Module):
             Setting this parameter to `True` is equivalent to setting `tebd_input_mode` to 'strip'.
             Setting it to `False` is equivalent to setting `tebd_input_mode` to 'concat'.
             The default value is `None`, which means the `tebd_input_mode` setting will be used instead.
+    use_econf_tebd: bool, Optional
+            Whether to use electronic configuration type embedding.
+    type_map: List[str], Optional
+            A list of strings. Give the name to each type of atoms.
+            Only used if `use_econf_tebd` is `True` in type embedding net.
     spin
             (Only support None to keep consistent with other backend references.)
             (Not used in this version. Not-none option is not implemented.)
@@ -220,6 +225,8 @@ def __init__(
         smooth_type_embedding: bool = True,
         type_one_side: bool = False,
         stripped_type_embedding: Optional[bool] = None,
+        use_econf_tebd: bool = False,
+        type_map: Optional[List[str]] = None,
         # not implemented
         spin=None,
         type: Optional[str] = None,
@@ -270,7 +277,15 @@ def __init__(
             ln_eps=ln_eps,
             old_impl=old_impl,
         )
-        self.type_embedding = TypeEmbedNet(ntypes, tebd_dim, precision=precision)
+        self.use_econf_tebd = use_econf_tebd
+        self.type_map = type_map
+        self.type_embedding = TypeEmbedNet(
+            ntypes,
+            tebd_dim,
+            precision=precision,
+            use_econf_tebd=use_econf_tebd,
+            type_map=type_map,
+        )
         self.tebd_dim = tebd_dim
         self.concat_output_tebd = concat_output_tebd
         self.trainable = trainable
@@ -415,6 +430,8 @@ def serialize(self) -> dict:
             "smooth_type_embedding": obj.smooth,
             "type_one_side": obj.type_one_side,
             "concat_output_tebd": self.concat_output_tebd,
+            "use_econf_tebd": self.use_econf_tebd,
+            "type_map": self.type_map,
             # make deterministic
             "precision": RESERVED_PRECISON_DICT[obj.prec],
             "embeddings": obj.filter_layers.serialize(),

deepmd/pt/model/descriptor/dpa2.py

@@ -76,6 +76,8 @@ def __init__(
         trainable: bool = True,
         seed: Optional[int] = None,
         add_tebd_to_repinit_out: bool = False,
+        use_econf_tebd: bool = False,
+        type_map: Optional[List[str]] = None,
         old_impl: bool = False,
     ):
         r"""The DPA-2 descriptor. see https://arxiv.org/abs/2312.15492.
@@ -104,6 +106,11 @@ def __init__(
             (Unused yet) Random seed for parameter initialization.
         add_tebd_to_repinit_out : bool, optional
             Whether to add type embedding to the output representation from repinit before inputting it into repformer.
+        use_econf_tebd : bool, Optional
+            Whether to use electronic configuration type embedding.
+        type_map : List[str], Optional
+            A list of strings. Give the name to each type of atoms.
+            Only used if `use_econf_tebd` is `True` in type embedding net.
 
         Returns
         -------
@@ -189,8 +196,14 @@ def init_subclass_params(sub_data, sub_class):
             ln_eps=self.repformer_args.ln_eps,
             old_impl=old_impl,
         )
+        self.use_econf_tebd = use_econf_tebd
+        self.type_map = type_map
         self.type_embedding = TypeEmbedNet(
-            ntypes, self.repinit_args.tebd_dim, precision=precision
+            ntypes,
+            self.repinit_args.tebd_dim,
+            precision=precision,
+            use_econf_tebd=self.use_econf_tebd,
+            type_map=type_map,
         )
         self.concat_output_tebd = concat_output_tebd
         self.precision = precision
@@ -368,6 +381,8 @@ def serialize(self) -> dict:
             "env_protection": self.env_protection,
             "trainable": self.trainable,
             "add_tebd_to_repinit_out": self.add_tebd_to_repinit_out,
+            "use_econf_tebd": self.use_econf_tebd,
+            "type_map": self.type_map,
             "type_embedding": self.type_embedding.embedding.serialize(),
             "g1_shape_tranform": self.g1_shape_tranform.serialize(),
         }

deepmd/pt/model/model/__init__.py

@@ -107,6 +107,8 @@
     ntypes = len(model_params["type_map"])
     # descriptor
     model_params["descriptor"]["ntypes"] = ntypes
+    if model_params["descriptor"].get("use_econf_tebd", False):
+        model_params["descriptor"]["type_map"] = copy.deepcopy(model_params["type_map"])
     descriptor = BaseDescriptor(**model_params["descriptor"])
     # fitting
     fitting_net = model_params.get("fitting_net", None)
@@ -152,6 +154,8 @@
     ntypes = len(model_params["type_map"])
     # descriptor
     model_params["descriptor"]["ntypes"] = ntypes
+    if model_params["descriptor"].get("use_econf_tebd", False):
+        model_params["descriptor"]["type_map"] = copy.deepcopy(model_params["type_map"])
     descriptor = BaseDescriptor(**model_params["descriptor"])
     # fitting
     fitting_net = model_params.get("fitting_net", None)