diff --git a/modelscope/pipelines/builder.py b/modelscope/pipelines/builder.py
index 6495a5db..3c97c2be 100644
--- a/modelscope/pipelines/builder.py
+++ b/modelscope/pipelines/builder.py
@@ -18,6 +18,7 @@ PIPELINES = Registry('pipelines')
 DEFAULT_MODEL_FOR_PIPELINE = {
     # TaskName: (pipeline_module_name, model_repo)
     Tasks.image_matting: ('image-matting', 'damo/image-matting-person'),
+    Tasks.nli: ('nli', 'damo/nlp_structbert_nli_chinese-base'),
     Tasks.text_classification:
     ('bert-sentiment-analysis', 'damo/bert-base-sst2'),
     Tasks.text_generation: ('palm', 'damo/nlp_palm_text-generation_chinese'),
diff --git a/tests/pipelines/test_nli.py b/tests/pipelines/test_nli.py
new file mode 100644
index 00000000..9167b897
--- /dev/null
+++ b/tests/pipelines/test_nli.py
@@ -0,0 +1,48 @@
+# Copyright (c) Alibaba, Inc. and its affiliates.
+import unittest
+
+from maas_hub.snapshot_download import snapshot_download
+
+from modelscope.models import Model
+from modelscope.models.nlp import SbertForNLI
+from modelscope.pipelines import NLIPipeline, pipeline
+from modelscope.preprocessors import NLIPreprocessor
+from modelscope.utils.constant import Tasks
+
+
+class NLITest(unittest.TestCase):
+    model_id = 'damo/nlp_structbert_nli_chinese-base'
+    sentence1 = '四川商务职业学院和四川财经职业学院哪个好？'
+    sentence2 = '四川商务职业学院商务管理在哪个校区？'
+
+    def test_run_from_local(self):
+        cache_path = snapshot_download(self.model_id)
+        tokenizer = NLIPreprocessor(cache_path)
+        model = SbertForNLI(cache_path, tokenizer=tokenizer)
+        pipeline1 = NLIPipeline(model, preprocessor=tokenizer)
+        pipeline2 = pipeline(Tasks.nli, model=model, preprocessor=tokenizer)
+        print(f'sentence1: {self.sentence1}\nsentence2: {self.sentence2}\n'
+              f'pipeline1:{pipeline1(input=(self.sentence1, self.sentence2))}')
+        print()
+        print(
+            f'sentence1: {self.sentence1}\nsentence2: {self.sentence2}\n'
+            f'pipeline1: {pipeline2(input=(self.sentence1, self.sentence2))}')
+
+    def test_run_with_model_from_modelhub(self):
+        model = Model.from_pretrained(self.model_id)
+        tokenizer = NLIPreprocessor(model.model_dir)
+        pipeline_ins = pipeline(
+            task=Tasks.nli, model=model, preprocessor=tokenizer)
+        print(pipeline_ins(input=(self.sentence1, self.sentence2)))
+
+    def test_run_with_model_name(self):
+        pipeline_ins = pipeline(task=Tasks.nli, model=self.model_id)
+        print(pipeline_ins(input=(self.sentence1, self.sentence2)))
+
+    def test_run_with_default_model(self):
+        pipeline_ins = pipeline(task=Tasks.nli)
+        print(pipeline_ins(input=(self.sentence1, self.sentence2)))
+
+
+if __name__ == '__main__':
+    unittest.main()