diff --git a/vllm/model_executor/models/aya_vision.py b/vllm/model_executor/models/aya_vision.py
index 65f3063936..f4c9bbee9e 100644
--- a/vllm/model_executor/models/aya_vision.py
+++ b/vllm/model_executor/models/aya_vision.py
@@ -358,6 +358,10 @@ class AyaVisionForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsP
                 architectures=["Cohere2ForCausalLM"],
             )
 
+        self.make_empty_intermediate_tensors = (
+            self.language_model.make_empty_intermediate_tensors
+        )
+
     @property
     def dtype(self):
         return next(self.parameters()).dtype
diff --git a/vllm/model_executor/models/voxtral.py b/vllm/model_executor/models/voxtral.py
index 42e811edae..baf6d5c739 100644
--- a/vllm/model_executor/models/voxtral.py
+++ b/vllm/model_executor/models/voxtral.py
@@ -349,6 +349,10 @@ class VoxtralForConditionalGeneration(
                 dim=config.text_config.hidden_size,
             )
 
+        self.make_empty_intermediate_tensors = (
+            self.language_model.make_empty_intermediate_tensors
+        )
+
     def get_mm_mapping(self) -> MultiModelKeys:
         """Get module prefix for multimodal models to filter LoRA modules."""
         return MultiModelKeys.from_string_field(