[V1][Molmo] Fix get_multimodal_embeddings() in molmo.py (#14161)

2025-03-04 07:43:59 -08:00
parent c8525f06fc
commit b3cf368d79
22 changed files with 249 additions and 150 deletions
--- a/vllm/model_executor/models/deepseek_vl2.py
+++ b/vllm/model_executor/models/deepseek_vl2.py
@@ -606,7 +606,9 @@ class DeepseekVLV2ForCausalLM(nn.Module, SupportsMultiModal, SupportsPP):
        return self._pixel_values_to_embedding(
            pixel_values=pixel_values, images_spatial_crop=images_spatial_crop)

-    def get_multimodal_embeddings(self, **kwargs: object) -> torch.Tensor:
+    def get_multimodal_embeddings(
+        self, **kwargs: object
+    ) -> Union[list[torch.Tensor], torch.Tensor, tuple[torch.Tensor, ...]]:
        image_input = self._parse_and_validate_image_input(**kwargs)
        if image_input is None:
            return None