diff --git a/lerobot/common/policies/pi0/paligemma_with_expert.py b/lerobot/common/policies/pi0/paligemma_with_expert.py
index 49c844c7bf..fb5077fb2c 100644
--- a/lerobot/common/policies/pi0/paligemma_with_expert.py
+++ b/lerobot/common/policies/pi0/paligemma_with_expert.py
@@ -223,7 +223,7 @@ def embed_image(self, image: torch.Tensor):
             return self.paligemma.model.get_image_features(image)
 
     def embed_language_tokens(self, tokens: torch.Tensor):
-        return self.paligemma.language_model.model.embed_tokens(tokens)
+        return self.paligemma.language_model.embed_tokens(tokens)
 
     # TODO: break down this huge forward into modules or functions
     def forward(
@@ -235,7 +235,7 @@ def forward(
         use_cache: Optional[bool] = None,
         fill_kv_cache: Optional[bool] = None,
     ):
-        models = [self.paligemma.language_model.model, self.gemma_expert.model]
+        models = [self.paligemma.language_model, self.gemma_expert.model]
 
         for hidden_states in inputs_embeds:
             # TODO this is very inefficient