diff --git a/vllm/model_executor/layers/rotary_embedding.py b/vllm/model_executor/layers/rotary_embedding.py
index 71af9b26e2e93..d80e73bbe39e9 100644
--- a/vllm/model_executor/layers/rotary_embedding.py
+++ b/vllm/model_executor/layers/rotary_embedding.py
@@ -108,7 +108,7 @@ def _forward(
             query_pass = query[..., self.rotary_dim:]
             key_pass = key[..., self.rotary_dim:]
 
-        self.cos_sin_cache = self.cos_sin_cache.to(positions.get_device())
+        self.cos_sin_cache = self.cos_sin_cache.to(positions.device)
         cos_sin = self.cos_sin_cache[torch.add(positions, offsets)
                                      if offsets is not None else positions]
         cos, sin = cos_sin.chunk(2, dim=-1)
@@ -142,7 +142,7 @@ def forward(
         key: torch.Tensor,
         offsets: Optional[torch.Tensor] = None,
     ) -> Tuple[torch.Tensor, torch.Tensor]:
-        self.cos_sin_cache = self.cos_sin_cache.to(positions.get_device())
+        self.cos_sin_cache = self.cos_sin_cache.to(positions.device)
         # ops.rotary_embedding()/batched_rotary_embedding()
         # are in-place operations that update the query and key tensors.
         if offsets is not None: