[Bugfix] Avoid transferring cached multi-modal items from P0 to P1 (#16273)

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

vllm/v1/request.py

@@ -3,17 +3,16 @@
 import enum
 from typing import TYPE_CHECKING, Optional, Union
 
+from vllm.multimodal.inputs import MultiModalKwargs, PlaceholderRange
 from vllm.sampling_params import SamplingParams
+from vllm.utils import is_list_of
 from vllm.v1.engine import (EngineCoreEvent, EngineCoreEventType,
                             EngineCoreRequest, FinishReason)
 from vllm.v1.structured_output.request import StructuredOutputRequest
 from vllm.v1.utils import ConstantList
 
 if TYPE_CHECKING:
-
     from vllm.lora.request import LoRARequest
-    from vllm.multimodal import MultiModalKwargs
-    from vllm.multimodal.inputs import PlaceholderRange
 
 
 class Request:
@@ -23,9 +22,9 @@ class Request:
         request_id: str,
         prompt: Optional[str],
         prompt_token_ids: list[int],
-        multi_modal_inputs: Optional[list["MultiModalKwargs"]],
+        multi_modal_inputs: Optional[list[MultiModalKwargs]],
         multi_modal_hashes: Optional[list[str]],
-        multi_modal_placeholders: Optional[list["PlaceholderRange"]],
+        multi_modal_placeholders: Optional[list[PlaceholderRange]],
         sampling_params: SamplingParams,
         eos_token_id: Optional[int],
         arrival_time: float,
@@ -75,6 +74,11 @@ class Request:
 
     @classmethod
     def from_engine_core_request(cls, request: EngineCoreRequest) -> "Request":
+        if request.mm_inputs is not None:
+            assert isinstance(request.mm_inputs, list)
+            assert is_list_of(request.mm_inputs, MultiModalKwargs), (
+                "mm_inputs was not updated in EngineCore.add_request")
+
         return cls(
             request_id=request.request_id,
             prompt=request.prompt,