From 4dd79783744adbfdc86f9454bffb5a92715a7f61 Mon Sep 17 00:00:00 2001
From: Roger Wang <hey@rogerw.io>
Date: Tue, 2 Dec 2025 18:33:45 -0800
Subject: [PATCH] [Bugfix] Fix regression on pooling models from PR#29621
 (#29921)

Signed-off-by: Roger Wang <hey@rogerw.io>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
---
 vllm/multimodal/parse.py | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/vllm/multimodal/parse.py b/vllm/multimodal/parse.py
index 0d3b8289e4e1..650368dcb8fc 100644
--- a/vllm/multimodal/parse.py
+++ b/vllm/multimodal/parse.py
@@ -134,11 +134,17 @@ class EmbeddingItems(
     or a list of embedding tensors (one per item).
     """
 
+    def _unwrap(
+        self, item: torch.Tensor | MediaWithBytes[torch.Tensor]
+    ) -> torch.Tensor:
+        """Extract media from wrapper if present."""
+        return item.media if isinstance(item, MediaWithBytes) else item
+
     def get_count(self) -> int:
         return len(self.data)
 
     def get(self, index: int) -> torch.Tensor:
-        return self.data[index]
+        return self._unwrap(self.data[index])
 
     def get_processor_data(self) -> Mapping[str, object]:
         return {}