Fixing mishandling of ignore_labels. (#14274)

Fixes #14272

tests/test_pipelines_token_classification.py

@@ -627,6 +627,15 @@ class TokenClassificationPipelineTests(unittest.TestCase, metaclass=PipelineTest
             ],
         )
 
+        token_classifier = pipeline(
+            task="token-classification", model=model_name, framework="pt", ignore_labels=["O", "I-MISC"]
+        )
+        outputs = token_classifier("This is a test !")
+        self.assertEqual(
+            nested_simplify(outputs),
+            [],
+        )
+
     @require_torch
     def test_pt_ignore_subwords_slow_tokenizer_raises(self):
         model_name = "sshleifer/tiny-dbmdz-bert-large-cased-finetuned-conll03-english"