[Quickfix] update CachedRequestState as NewRequestData changed

Signed-off-by: MengqingCao <[email protected]>

Author: MengqingCao

vllm_ascend/worker/model_runner_v1.py

@@ -479,7 +479,7 @@ def _update_states(self, scheduler_output: "SchedulerOutput") -> None:
             self.requests[req_id] = CachedRequestState(
                 req_id=req_id,
                 prompt_token_ids=new_req_data.prompt_token_ids,
-                mm_inputs=new_req_data.mm_inputs,
+                mm_kwargs=new_req_data.mm_kwargs,
                 mm_positions=new_req_data.mm_positions,
                 sampling_params=sampling_params,
                 pooling_params=new_req_data.pooling_params,

vllm_ascend/worker/npu_input_batch.py

@@ -43,7 +43,7 @@ class CachedRequestState:
 
     req_id: str
     prompt_token_ids: list[int]
-    mm_inputs: list[MultiModalKwargs]
+    mm_kwargs: list[MultiModalKwargs]
     mm_positions: list[PlaceholderRange]
     sampling_params: Optional[SamplingParams]
     pooling_params: Optional[PoolingParams]