avoid configuring model twice

energizer/estimator.py

@@ -499,8 +499,9 @@ def test_epoch_end(self, output: list[BATCH_OUTPUT], metrics: METRIC | None) ->
     """
 
     def configure_model(self) -> None:
-        with self.fabric.init_module():
-            self.model.configure_model()
+        if not self.model.is_configured:
+            with self.fabric.init_module():
+                self.model.configure_model()
 
     def configure_optimization_args(
         self,

energizer/models.py

@@ -18,6 +18,7 @@
 
 class Model(ABC):
     _model_instance: torch.nn.Module | None = None
+    _is_configured: bool = False
 
     @property
     def model_instance(self) -> torch.nn.Module:
@@ -32,14 +33,18 @@ def summary(self) -> str:
     def configure_model(self, *args, **kwargs) -> None:
         ...
 
+    @property
+    def is_configured(self) -> bool:
+        return self._is_configured
+
 
 class TorchModel(Model):
     def __init__(self, model: torch.nn.Module) -> None:
         super().__init__()
         self._model_instance = model
 
     def configure_model(self, *args, **kwargs) -> None:
-        pass
+        self._is_configured = True
 
 
 class HFModel(Model):
@@ -147,6 +152,8 @@ def configure_model(self) -> None:
             if self._convert_to_bettertransformer:
                 self.convert_to_bettertransformer()
 
+        self._is_configured = True
+
     def convert_to_bettertransformer(self) -> None:
         assert self._model_instance is not None
         self._model_instance = self._model_instance.to_bettertransformer()