[Core] Use individual MM items in P0/P1 cache and model runner (#22570)

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

tests/v1/worker/test_gpu_input_batch.py

@@ -203,7 +203,7 @@ def _construct_cached_request_state(req_id_suffix: int):
         prompt_token_ids=prompt_token_ids,
         sampling_params=_create_sampling_params(),
         pooling_params=None,
-        mm_inputs=[],
+        mm_kwargs=[],
         mm_positions=[],
         block_ids=([], ),
         generator=None,

tests/v1/worker/test_gpu_model_runner.py

@@ -120,7 +120,7 @@ def _schedule_new_request(*req_ids: str) -> SchedulerOutput:
             NewRequestData(
                 req_id=req_id,
                 prompt_token_ids=[1, 2, 3],
-                mm_inputs=[],
+                mm_kwargs=[],
                 mm_hashes=[],
                 mm_positions=[],
                 sampling_params=SamplingParams(),