diff --git a/maml/describers/_m3gnet.py b/maml/describers/_m3gnet.py
index e16bb74e..46d41d30 100644
--- a/maml/describers/_m3gnet.py
+++ b/maml/describers/_m3gnet.py
@@ -6,14 +6,17 @@
 import numpy as np
 import pandas as pd
 
-from maml.base import BaseDescriber
+from maml.base import BaseDescriber, describer_type
 
 if TYPE_CHECKING:
     from pymatgen.core import Molecule, Structure
 
-DEFAULT_MODEL = Path(__file__).parent / "data/m3gnet_models/matbench_mp_e_form/0/m3gnet/"
+DEFAULT_MODEL = (
+    Path(__file__).parent / "data/m3gnet_models/matbench_mp_e_form/0/m3gnet/"
+)
 
 
+@describer_type("structure")
 class M3GNetStructure(BaseDescriber):
     """Use M3GNet pre-trained models as featurizer to get Structural features."""
 
@@ -57,7 +60,9 @@ def transform_one(self, structure: Structure | Molecule):
         graph = self.describer_model.graph_converter.convert(structure).as_list()
         graph = tf_compute_distance_angle(graph)
         three_basis = self.describer_model.basis_expansion(graph)
-        three_cutoff = polynomial(graph[Index.BONDS], self.describer_model.threebody_cutoff)
+        three_cutoff = polynomial(
+            graph[Index.BONDS], self.describer_model.threebody_cutoff
+        )
         g = self.describer_model.featurizer(graph)
         g = self.describer_model.feature_adjust(g)
         for i in range(self.describer_model.n_blocks):
@@ -67,6 +72,7 @@ def transform_one(self, structure: Structure | Molecule):
         return np.array(layer_before_readout(g))[0]
 
 
+@describer_type("site")
 class M3GNetSite(BaseDescriber):
     """Use M3GNet pre-trained models as featurizer to get atomic features."""
 
@@ -100,11 +106,17 @@ def __init__(
         else:
             self.describer_model = M3GNet.from_dir(DEFAULT_MODEL)
             self.model_path = str(DEFAULT_MODEL)
-        allowed_output_layers = ["embedding"] + [f"gc_{i + 1}" for i in range(self.describer_model.n_blocks)]
+        allowed_output_layers = ["embedding"] + [
+            f"gc_{i + 1}" for i in range(self.describer_model.n_blocks)
+        ]
         if output_layers is None:
             output_layers = ["gc_1"]
-        elif not isinstance(output_layers, list) or set(output_layers).difference(allowed_output_layers):
-            raise ValueError(f"Invalid output_layers, it must be a sublist of {allowed_output_layers}.")
+        elif not isinstance(output_layers, list) or set(output_layers).difference(
+            allowed_output_layers
+        ):
+            raise ValueError(
+                f"Invalid output_layers, it must be a sublist of {allowed_output_layers}."
+            )
         self.output_layers = output_layers
         self.return_type = return_type
         super().__init__(**kwargs)
@@ -123,14 +135,16 @@ def transform_one(self, structure: Structure | Molecule):
         graph = self.describer_model.graph_converter.convert(structure).as_list()
         graph = tf_compute_distance_angle(graph)
         three_basis = self.describer_model.basis_expansion(graph)
-        three_cutoff = polynomial(graph[Index.BONDS], self.describer_model.threebody_cutoff)
+        three_cutoff = polynomial(
+            graph[Index.BONDS], self.describer_model.threebody_cutoff
+        )
         g = self.describer_model.featurizer(graph)
         atom_fea = {"embedding": g[Index.ATOMS]}
         g = self.describer_model.feature_adjust(g)
         for i in range(self.describer_model.n_blocks):
             g = self.describer_model.three_interactions[i](g, three_basis, three_cutoff)
             g = self.describer_model.graph_layers[i](g)
-            atom_fea[f"gc_{i+1}"] = g[Index.ATOMS]
+            atom_fea[f"gc_{i + 1}"] = g[Index.ATOMS]
         atom_fea_dict = {k: v for k, v in atom_fea.items() if k in self.output_layers}
         if self.return_type == dict:
             return atom_fea_dict