[Misc] Clean up type annotation for SupportsMultiModal (#14794)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

vllm/model_executor/models/blip2.py

@@ -15,8 +15,7 @@ from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.layers.sampler import SamplerOutput, get_sampler
 from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
-from vllm.multimodal.inputs import (MultiModalFieldConfig, MultiModalKwargs,
-                                    NestedTensors)
+from vllm.multimodal.inputs import MultiModalFieldConfig, MultiModalKwargs
 from vllm.multimodal.parse import MultiModalDataItems
 from vllm.multimodal.processing import (BaseMultiModalProcessor,
                                         BaseProcessingInfo, PromptIndexTargets,
@@ -25,7 +24,7 @@ from vllm.multimodal.profiling import BaseDummyInputsBuilder, ProcessorInputs
 from vllm.sequence import IntermediateTensors
 
 from .blip import BlipVisionModel
-from .interfaces import SupportsMultiModal, SupportsPP
+from .interfaces import MultiModalEmbeddings, SupportsMultiModal, SupportsPP
 from .utils import (AutoWeightsLoader, init_vllm_registered_model,
                     maybe_prefix, merge_multimodal_embeddings)
 
@@ -629,8 +628,7 @@ class Blip2ForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP):
         return self.language_projection(query_output)
 
     def get_multimodal_embeddings(
-        self, **kwargs
-    ) -> Union[list[torch.Tensor], torch.Tensor, tuple[torch.Tensor, ...]]:
+            self, **kwargs: object) -> Optional[MultiModalEmbeddings]:
         image_input = self._parse_and_validate_image_input(**kwargs)
         if image_input is None:
             return None
@@ -640,7 +638,7 @@ class Blip2ForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP):
     def get_input_embeddings(
         self,
         input_ids: torch.Tensor,
-        multimodal_embeddings: Optional[NestedTensors] = None,
+        multimodal_embeddings: Optional[MultiModalEmbeddings] = None,
     ) -> torch.Tensor:
         inputs_embeds = self.language_model.get_input_embeddings(input_ids)
         if multimodal_embeddings is not None: