diff --git a/rl/llm/engines/local.py b/rl/llm/engines/local.py
index 04b9136..f18dc69 100644
--- a/rl/llm/engines/local.py
+++ b/rl/llm/engines/local.py
@@ -392,7 +392,6 @@ def __exit__(self, exc_type, exc_value, traceback):
         del self.vllm
         gc.collect()
         torch.cuda.empty_cache()
-        torch.distributed.destroy_process_group()
         LOGGER.info("VLLM model unloaded.")
 
     def generate(self, prompt: InferenceInput) -> InferenceOutput: