diff --git a/ci/L0_multi_gpu_vllm/vllm_backend/vllm_multi_gpu_test.py b/ci/L0_multi_gpu_vllm/vllm_backend/vllm_multi_gpu_test.py
index 21591845..f809d9a3 100644
--- a/ci/L0_multi_gpu_vllm/vllm_backend/vllm_multi_gpu_test.py
+++ b/ci/L0_multi_gpu_vllm/vllm_backend/vllm_multi_gpu_test.py
@@ -29,13 +29,15 @@
 import unittest
 from functools import partial
 
-import pynvml
 import tritonclient.grpc as grpcclient
 from tritonclient.utils import *
+from vllm.utils import import_pynvml
 
 sys.path.append("../../common")
 from test_util import TestResultCollector, UserData, callback, create_vllm_request
 
+pynvml = import_pynvml()
+
 
 class VLLMMultiGPUTest(TestResultCollector):
     def setUp(self):