Fixing Backward compatiblity for zero-shot (#13855)

Fixes #13846

tests/test_pipelines_zero_shot.py

@@ -61,6 +61,24 @@ class ZeroShotClassificationPipelineTests(unittest.TestCase, metaclass=PipelineT
         )
         self.assertEqual(outputs, {"sequence": ANY(str), "labels": [ANY(str)], "scores": [ANY(float)]})
 
+        # https://github.com/huggingface/transformers/issues/13846
+        outputs = classifier(["I am happy"], ["positive", "negative"])
+        self.assertEqual(
+            outputs,
+            [
+                {"sequence": ANY(str), "labels": [ANY(str), ANY(str)], "scores": [ANY(float), ANY(float)]}
+                for i in range(1)
+            ],
+        )
+        outputs = classifier(["I am happy", "I am sad"], ["positive", "negative"])
+        self.assertEqual(
+            outputs,
+            [
+                {"sequence": ANY(str), "labels": [ANY(str), ANY(str)], "scores": [ANY(float), ANY(float)]}
+                for i in range(2)
+            ],
+        )
+
         with self.assertRaises(ValueError):
             classifier("", candidate_labels="politics")