diff --git a/python/sglang/srt/model_parallel.py b/python/sglang/srt/model_parallel.py
index 6817bce0235..778347b8ef3 100644
--- a/python/sglang/srt/model_parallel.py
+++ b/python/sglang/srt/model_parallel.py
@@ -54,11 +54,7 @@ def _prepare_output_fn(output_layouts, use_local_output, mod, outputs, device_me
         )._prepare_output_fn(
             output_layouts, use_local_output, mod, outputs, device_mesh
         )
-        # wait for the output to be ready
-        if isinstance(outputs, AsyncCollectiveTensor):
-            return outputs.wait()
-        else:
-            return outputs
+        return torch.distributed._functional_collectives.wait_tensor(outputs)
 
 
 def tensor_parallel(