[Model] Add SupportsMultiModal.get_language_model interface (#16007)

Signed-off-by: NickLucche <nlucches@redhat.com>

vllm/model_executor/models/mllama4.py

@@ -742,6 +742,9 @@ class Llama4ForConditionalGeneration(nn.Module, SupportsMultiModal,
             for img in vision_embeddings_flat.split(patches_per_image, dim=0)
         ]
 
+    def get_language_model(self) -> torch.nn.Module:
+        return self.language_model
+
     def get_multimodal_embeddings(self,
                                   **kwargs) -> Optional[MultiModalEmbeddings]:
         image_input = self._parse_and_validate_image_input(**kwargs)