[Bugfix] Fix MRoPE dispatch on CPU (vllm-project#24712)

Signed-off-by: jiang1.li <[email protected]>

Author: bigPYJ1151

vllm/model_executor/layers/rotary_embedding/mrope.py

@@ -300,6 +300,15 @@ def forward_cuda(
         key = torch.cat((key_rot, key_pass), dim=-1).reshape(key_shape)
         return query, key
 
+    def forward_cpu(
+        self,
+        positions: torch.Tensor,
+        query: torch.Tensor,
+        key: Optional[torch.Tensor] = None,
+        offsets: Optional[torch.Tensor] = None,
+    ) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
+        return self.forward_native(positions, query, key, offsets)
+
     @classmethod
     def get_input_positions(
         cls,