diff --git a/tests/test_datasets.py b/tests/test_datasets.py
index 05d9b90..8792ae5 100644
--- a/tests/test_datasets.py
+++ b/tests/test_datasets.py
@@ -37,21 +37,21 @@ def test_load_snli():
     train_data, classes = mz.datasets.snli.load_data('train',
                                                      'classification',
                                                      return_classes=True)
-    num_samples = 550146
+    num_samples = 549361
     assert len(train_data) == num_samples
     x, y = train_data.unpack()
     assert len(x['text_left']) == num_samples
     assert len(x['text_right']) == num_samples
     assert y.shape == (num_samples, 1)
-    assert classes == ['entailment', 'contradiction', 'neutral', '-']
+    assert classes == ['entailment', 'contradiction', 'neutral']
     dev_data, classes = mz.datasets.snli.load_data('dev', 'classification',
                                                    return_classes=True)
-    assert len(dev_data) == 10000
-    assert classes == ['entailment', 'contradiction', 'neutral', '-']
+    assert len(dev_data) == 9842
+    assert classes == ['entailment', 'contradiction', 'neutral']
     test_data, classes = mz.datasets.snli.load_data('test', 'classification',
                                                     return_classes=True)
-    assert len(test_data) == 10000
-    assert classes == ['entailment', 'contradiction', 'neutral', '-']
+    assert len(test_data) == 9824
+    assert classes == ['entailment', 'contradiction', 'neutral']
 
     train_data = mz.datasets.snli.load_data('train', 'ranking')
     x, y = train_data.unpack()