diff --git a/swift/llm/infer/infer_engine/vllm_engine.py b/swift/llm/infer/infer_engine/vllm_engine.py
index 07850e6783..f9633a1aa2 100644
--- a/swift/llm/infer/infer_engine/vllm_engine.py
+++ b/swift/llm/infer/infer_engine/vllm_engine.py
@@ -307,6 +307,12 @@ def _prepare_generation_config(self, request_config: RequestConfig) -> SamplingP
         if kwargs.get('seed') is None:
             kwargs['seed'] = get_seed()
         res = SamplingParams(**kwargs)
+
+        if hasattr(res, 'output_kind'):
+            # fix n > 1 in V1 Engine
+            from vllm.sampling_params import RequestOutputKind
+            res.output_kind = RequestOutputKind.FINAL_ONLY
+
         res.top_logprobs = request_config.top_logprobs
         return res