Fix: prevent loading best model when PEFT adapters are active (#3470)

* Fix: prevent loading best model when PEFT adapters are active (#3056)

* Reuse _load_from_checkpoint in _load_best_model

---------

Co-authored-by: Tom Aarsen <[email protected]>

Author: sahibpreetsingh12

sentence_transformers/trainer.py

@@ -594,26 +594,19 @@ def evaluation_loop(
     def _load_best_model(self) -> None:
         # Attempt to load the model from self.state.best_model_checkpoint
         logger.info(f"Loading best model from {self.state.best_model_checkpoint} (score: {self.state.best_metric}).")
-        try:
-            dummy_model = self.model.__class__(
-                self.state.best_model_checkpoint,
-                trust_remote_code=self.model.trust_remote_code,
-            )
-        except Exception as exc:
-            logger.error(f"Could not load the best model from {self.state.best_model_checkpoint}. Error: {str(exc)}")
-            return
 
-        # Store the best model checkpoint in the model card
         try:
             if checkpoint := self.state.best_model_checkpoint:
                 step = checkpoint.rsplit("-", 1)[-1]
                 self.model.model_card_data.set_best_model_step(int(step))
         except Exception:
             pass
 
-        # Ideally, the only changes between self.model and the dummy model are the weights
-        # so we should be able to just copy the state dict
-        self.model.load_state_dict(dummy_model.state_dict())
+        try:
+            self._load_from_checkpoint(self.state.best_model_checkpoint)
+        except Exception as exc:
+            logger.error(f"Could not load the best model from {self.state.best_model_checkpoint}. Error: {str(exc)}")
+            return
 
     def validate_column_names(self, dataset: Dataset, dataset_name: str | None = None) -> None:
         if isinstance(dataset, dict):