diff --git a/python/sglang/srt/model_executor/model_runner.py b/python/sglang/srt/model_executor/model_runner.py
index 923482d72..317734578 100644
--- a/python/sglang/srt/model_executor/model_runner.py
+++ b/python/sglang/srt/model_executor/model_runner.py
@@ -378,6 +378,12 @@ class ModelRunner:
             )
             server_args.attention_backend = "torch_native"
 
+        if server_args.prefill_attention_backend is not None and (
+            server_args.prefill_attention_backend
+            == server_args.decode_attention_backend
+        ):  # override the default attention backend
+            server_args.attention_backend = server_args.prefill_attention_backend
+
         if server_args.attention_backend is None:
             """
             Auto select the fastest attention backend.