From ffafaeb1333c03f08f18ed4503040d485d6cd733 Mon Sep 17 00:00:00 2001
From: jinghuan-Chen <275923410@qq.com>
Date: Wed, 18 Jun 2025 16:17:31 +0800
Subject: [PATCH] Release CUDA Graphs resource before exit.

---
 nanovllm/engine/model_runner.py | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/nanovllm/engine/model_runner.py b/nanovllm/engine/model_runner.py
index 0c7b2ef..6f838a6 100644
--- a/nanovllm/engine/model_runner.py
+++ b/nanovllm/engine/model_runner.py
@@ -53,6 +53,9 @@ class ModelRunner:
             dist.barrier()
             if self.rank == 0:
                 self.shm.unlink()
+        if not self.enforce_eager:
+            del self.graphs, self.graph_pool
+        torch.cuda.synchronize()
         dist.destroy_process_group()
 
     def loop(self):