diff --git a/byte_infer_perf/llm_perf/server/endpoint.py b/byte_infer_perf/llm_perf/server/endpoint.py index e9b5d476..e3176108 100644 --- a/byte_infer_perf/llm_perf/server/endpoint.py +++ b/byte_infer_perf/llm_perf/server/endpoint.py @@ -49,6 +49,9 @@ def __init__(self, xpu_cfg) -> None: def __del__(self): self.scheduler.stop() + def __del__(self): + self.scheduler.stop() + def warmup(self, max_batch_size): prompt = "中国的首都是哪里?"