diff --git a/tests/generation/test_synced_gpus.py b/tests/generation/test_synced_gpus.py
index f0a2ffc96185..ffdd1a1887c1 100644
--- a/tests/generation/test_synced_gpus.py
+++ b/tests/generation/test_synced_gpus.py
@@ -20,15 +20,18 @@
 from paddlenlp.generation import GenerationConfig
 from paddlenlp.trainer import PdArgumentParser, Trainer, TrainingArguments
 from paddlenlp.transformers import AutoModelForCausalLM, AutoTokenizer
-from tests.parallel_launch import TestMultipleGpus
-from tests.transformers.test_modeling_common import ids_tensor
 
+try:
+    from tests.parallel_launch import TestMultipleGpus
+    from tests.transformers.test_modeling_common import ids_tensor
 
-class ShardingStage3Tester(TestMultipleGpus):
-    def test_synced_gpus_greedy(self):
-        # test this file
-        self.run_2gpu(__file__)
+    class ShardingStage3Tester(TestMultipleGpus):
+        def test_synced_gpus_greedy(self):
+            # test this file
+            self.run_2gpu(__file__)
 
+except:
+    pass
 
 if __name__ == "__main__":
     tokenizer = AutoTokenizer.from_pretrained("__internal_testing__/tiny-random-llama")