A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 12, 03:30 UTC
[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>
Roger Wang committed
Dec 2, 2025 at 18:33 UTC
4dd79783744adbfdc86f9454bffb5a92715a7f61
1 file changed
+7
-1
vllm/multimodal/parse.py
+7
-1
@@ -134,11 +134,17 @@ class EmbeddingItems(
134
or a list of embedding tensors (one per item).
135
"""
136
137
+ def _unwrap(
138
+ self, item: torch.Tensor | MediaWithBytes[torch.Tensor]
139
+ ) -> torch.Tensor:
140
+ """Extract media from wrapper if present."""
141
+ return item.media if isinstance(item, MediaWithBytes) else item
142
+
143
def get_count(self) -> int:
144
return len(self.data)
145
146
def get(self, index: int) -> torch.Tensor:
141
- return self.data[index]
147
+ return self._unwrap(self.data[index])
148
149
def get_processor_data(self) -> Mapping[str, object]:
150
return {}