[Fix] Set right modality string when adding multi-modal item to the prompt.

Signed-off-by: Michele Gazzetti <michele.gazzetti1@ibm.com>

Author: mgazz

vllm/entrypoints/chat_utils.py

@@ -772,7 +772,7 @@ def parse_tensors(self,
 
         if isinstance(tensor_encodings, str):
             tensor= self._connector.fetch_tensor_encoding(tensor_encodings)
-            placeholder = self._tracker.add("image_embeds", tensor)
+            placeholder = self._tracker.add(modality_str, tensor)
 
         self._add_placeholder(placeholder)