diff --git a/vllm/model_executor/layers/rotary_embedding/base.py b/vllm/model_executor/layers/rotary_embedding/base.py
index 17cd39bb8cd6..2c0f4dd13080 100644
--- a/vllm/model_executor/layers/rotary_embedding/base.py
+++ b/vllm/model_executor/layers/rotary_embedding/base.py
@@ -168,7 +168,8 @@ def forward_hip(
         else:
             # ops.rotary_embedding() is an in-place operation
             # that updates the query and key tensors.
-            self.forward_cuda(positions, query, key)
+            # FIXME: self.forward_cuda is not a in-place operation in eager mode.
+            return self.forward_cuda(positions, query, key)
         return query, key
 
     def forward_xpu(