diff --git a/vllm/model_executor/models/qwen2_5_omni_thinker.py b/vllm/model_executor/models/qwen2_5_omni_thinker.py
index b09cc504bd..26b69bd6dd 100644
--- a/vllm/model_executor/models/qwen2_5_omni_thinker.py
+++ b/vllm/model_executor/models/qwen2_5_omni_thinker.py
@@ -1473,14 +1473,12 @@ class Qwen2_5OmniThinkerForConditionalGeneration(
         if multimodal_embeddings is None or is_multimodal is None:
             return super().embed_input_ids(input_ids)
 
-        inputs_embeds = self._embed_text_input_ids(
-            input_ids,
-            self.get_language_model().embed_input_ids,
-            is_multimodal=is_multimodal,
-        )
-
         if len(multimodal_embeddings) == 0:
-            return inputs_embeds
+            return super().embed_input_ids(
+                input_ids,
+                multimodal_embeddings=multimodal_embeddings,
+                is_multimodal=is_multimodal,
+            )
 
         # Check for audio-in-video: interleaved video and audio tokens
         # in the multimodal region. Only use the interleaved path when