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

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
2025-08-13 22:18:07 +08:00
parent 20d65aa755
commit 19b927e52d
24 changed files with 549 additions and 486 deletions
--- a/tests/v1/worker/test_gpu_model_runner.py
+++ b/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(),