Fixed base model class name extraction from PeftModels (#27162)

* Fixed base model class name extraction from PeftModels

* Changes to first unwrap the model then extract the base model name

* Changed base_model to base_model.model to stay consistent with peft model abstractions

src/transformers/trainer.py

@@ -646,7 +646,7 @@ class Trainer:
         unwrapped_model = unwrap_model(model)
 
         if is_peft_available() and isinstance(unwrapped_model, PeftModel):
-            embeddings = unwrapped_model.base_model.get_input_embeddings()
+            embeddings = unwrapped_model.base_model.model.get_input_embeddings()
         else:
             embeddings = unwrapped_model.get_input_embeddings()
 
@@ -667,7 +667,7 @@ class Trainer:
         unwrapped_model = unwrap_model(model)
 
         if is_peft_available() and isinstance(unwrapped_model, PeftModel):
-            embeddings = unwrapped_model.base_model.get_input_embeddings()
+            embeddings = unwrapped_model.base_model.model.get_input_embeddings()
         else:
             embeddings = unwrapped_model.get_input_embeddings()
 
@@ -2752,10 +2752,11 @@ class Trainer:
             self._past = outputs[self.args.past_index]
 
         if labels is not None:
-            if is_peft_available() and isinstance(model, PeftModel):
-                model_name = unwrap_model(model.base_model)._get_name()
+            unwrapped_model = unwrap_model(model)
+            if is_peft_available() and isinstance(unwrapped_model, PeftModel):
+                model_name = unwrapped_model.base_model.model._get_name()
             else:
-                model_name = unwrap_model(model)._get_name()
+                model_name = unwrapped_model._get_name()
             if model_name in MODEL_FOR_CAUSAL_LM_MAPPING_NAMES.values():
                 loss = self.label_smoother(outputs, labels, shift_labels=True)
             else: