fix version

Signed-off-by: wangli <wangli858794774@gmail.com>

Author: Potabk

vllm_ascend/worker/model_runner_v1.py

@@ -468,7 +468,7 @@ def _update_states(self, scheduler_output: "SchedulerOutput") -> None:
                 to_update.apply(pooling_params)
 
             backward_kwargs = {}
-            if vllm_version_is("0.10.2"):
+            if vllm_version_is("0.10.2rc2"):
                 backward_kwargs["mm_kwargs"] = new_req_data.mm_kwargs
                 backward_kwargs["mm_hashes"] = new_req_data.mm_hashes
                 backward_kwargs["mm_positions"] = new_req_data.mm_positions
@@ -490,7 +490,7 @@ def _update_states(self, scheduler_output: "SchedulerOutput") -> None:
 
             # Only relevant for models using M-RoPE (e.g, Qwen2-VL)
             if self.uses_mrope:
-                if vllm_version_is("0.10.2"):
+                if vllm_version_is("0.10.2rc2"):
                     self._init_mrope_positions_0102(self.requests[req_id])
                 else:
                     self._init_mrope_positions(self.requests[req_id])
@@ -827,7 +827,7 @@ def _execute_mm_encoder(self, scheduler_output: "SchedulerOutput"):
             return
 
         # Batch the multi-modal inputs.
-        if vllm_version_is("0.10.2"):
+        if vllm_version_is("0.10.2rc2"):
             mm_kwargs, mm_hashes_pos = self._batch_mm_kwargs_from_scheduler_0102(
                 scheduler_output)
         else:
@@ -928,7 +928,7 @@ def _gather_mm_embeddings(
     ) -> list[torch.Tensor]:
 
         def _iter_mm_features(req_state: CachedRequestState):
-            if vllm_version_is("0.10.2"):
+            if vllm_version_is("0.10.2rc2"):
                 # legacy path (to be removed later)
                 assert req_state.mm_hashes is not None
                 assert req_state.mm_positions is not None

vllm_ascend/worker/npu_input_batch.py

@@ -79,7 +79,7 @@ def num_tokens(self) -> int:
     @deprecated("`mm_inputs` is superseded by `mm_kwargs` and will be "
                 "removed in v0.13. Please use `mm_kwargs` instead.")
     def mm_inputs(self) -> list[MultiModalKwargsItems]:
-        if vllm_version_is("0.10.2"):
+        if vllm_version_is("0.10.2rc2"):
             assert self.mm_kwargs is not None
             return [
                 MultiModalKwargsItems.from_seq([item])