[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_input_batch.py
+++ b/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,