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

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

vllm/model_executor/models/qwen2_5_vl.py

@@ -59,7 +59,8 @@ from vllm.platforms import _Backend
 from vllm.sequence import IntermediateTensors
 from vllm.transformers_utils.config import uses_mrope
 
-from .interfaces import SupportsLoRA, SupportsMultiModal, SupportsPP
+from .interfaces import (MultiModalEmbeddings, SupportsLoRA,
+                         SupportsMultiModal, SupportsPP)
 from .qwen2_vl import Qwen2VLDummyInputsBuilder as Qwen2_5_VLDummyInputsBuilder
 from .qwen2_vl import (Qwen2VLMultiModalProcessor, Qwen2VLProcessingInfo,
                        apply_rotary_pos_emb_vision)
@@ -952,7 +953,7 @@ class Qwen2_5_VLForConditionalGeneration(nn.Module, SupportsMultiModal,
         return modalities
 
     def get_multimodal_embeddings(
-            self, **kwargs) -> Optional[tuple[torch.Tensor, ...]]:
+            self, **kwargs: object) -> Optional[MultiModalEmbeddings]:
 
         modalities = self._parse_and_validate_multimodal_inputs(**kwargs)
         if not modalities:
@@ -978,7 +979,7 @@ class Qwen2_5_VLForConditionalGeneration(nn.Module, SupportsMultiModal,
     def get_input_embeddings(
         self,
         input_ids: torch.Tensor,
-        multimodal_embeddings: Optional[tuple[torch.Tensor, ...]] = None,
+        multimodal_embeddings: Optional[MultiModalEmbeddings] = None,
     ) -> torch.Tensor:
         inputs_embeds = self.language_model.get_input_embeddings(input_ids)
         if multimodal_embeddings is not None:
@@ -990,10 +991,9 @@ class Qwen2_5_VLForConditionalGeneration(nn.Module, SupportsMultiModal,
     def get_input_embeddings_v0(
         self,
         input_ids: torch.Tensor,
-        image_input: Optional[tuple[torch.Tensor, ...]] = None,
-        video_input: Optional[tuple[torch.Tensor, ...]] = None,
+        image_input: Optional[Qwen2_5_VLImageInputs] = None,
+        video_input: Optional[Qwen2_5_VLVideoInputs] = None,
     ) -> torch.Tensor:
-
         inputs_embeds = self.get_input_embeddings(input_ids)
         if image_input is not None:
             image_embeds = self._process_image_input(image_input)