Merge pull request #24 from jinghuan-Chen/fix/Release-CUDA-Graphs-resource-before-exit

2025-06-18 17:15:28 +08:00
parent 4fc764f175 ffafaeb133
commit fa0078174e
1 changed files with 3 additions and 0 deletions
--- 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):