Added integration tests for sequence builders.

pytorch_transformers/tests/tokenization_xlnet_test.py

@@ -89,6 +89,18 @@ class XLNetTokenizationTest(CommonTestCases.CommonTokenizerTester):
                                       u'9', u'2', u'0', u'0', u'0', u',', SPIECE_UNDERLINE + u'and', SPIECE_UNDERLINE + u'this',
                                       SPIECE_UNDERLINE + u'is', SPIECE_UNDERLINE + u'f', u'al', u'se', u'.'])
 
+    def test_sequence_builders(self):
+        tokenizer = XLNetTokenizer.from_pretrained("xlnet-base-cased")
+
+        text = tokenizer.encode("sequence builders")
+        text_2 = tokenizer.encode("multi-sequence build")
+
+        encoded_sentence = tokenizer.add_special_tokens_single_sentence(text)
+        encoded_pair = tokenizer.add_special_tokens_sentences_pair(text, text_2)
+
+        assert encoded_sentence == text + [4, 3]
+        assert encoded_pair == text + [4] + text_2 + [4, 3]
+
 
 if __name__ == '__main__':
     unittest.main()