fixed arguments for each aggregation test

tests.py

@@ -265,6 +265,7 @@ class TestSemF1(unittest.TestCase):
             aggregate=True,
         )
         self.assertIsInstance(scores, Scores)
+        print(f'Score: {scores}')
 
     def test_aggregate_untokenized_single_ref(self):
         scores = self.semf1_metric.compute(
@@ -278,12 +279,13 @@ class TestSemF1(unittest.TestCase):
             aggregate=True,
         )
         self.assertIsInstance(scores, Scores)
+        print(f'Score: {scores}')
 
     def test_aggregate_tokenized_single_ref(self):
         scores = self.semf1_metric.compute(
             predictions=self.tokenized_single_reference_predictions,
             references=self.tokenized_single_reference_references,
-            tokenize_sentences=True,
+            tokenize_sentences=False,
             multi_references=False,
             gpu=False,
             batch_size=32,
@@ -291,38 +293,41 @@ class TestSemF1(unittest.TestCase):
             aggregate=True,
         )
         self.assertIsInstance(scores, Scores)
+        print(f'Score: {scores}')
 
     def test_aggregate_untokenized_multi_ref(self):
         scores = self.semf1_metric.compute(
             predictions=self.untokenized_multi_reference_predictions,
             references=self.untokenized_multi_reference_references,
             tokenize_sentences=True,
-            multi_references=False,
+            multi_references=True,
             gpu=False,
             batch_size=32,
             verbose=False,
             aggregate=True,
         )
         self.assertIsInstance(scores, Scores)
+        print(f'Score: {scores}')
 
     def test_aggregate_tokenized_multi_ref(self):
         scores = self.semf1_metric.compute(
             predictions=self.tokenized_multi_reference_predictions,
             references=self.tokenized_multi_reference_references,
-            tokenize_sentences=True,
-            multi_references=False,
+            tokenize_sentences=False,
+            multi_references=True,
             gpu=False,
             batch_size=32,
             verbose=False,
             aggregate=True,
         )
         self.assertIsInstance(scores, Scores)
+        print(f'Score: {scores}')
 
     def test_aggregate_same_pred_and_ref(self):
         scores = self.semf1_metric.compute(
             predictions=self.tokenized_single_reference_predictions,
             references=self.tokenized_single_reference_predictions,
-            tokenize_sentences=True,
+            tokenize_sentences=False,
             multi_references=False,
             gpu=False,
             batch_size=32,
@@ -330,6 +335,7 @@ class TestSemF1(unittest.TestCase):
             aggregate=True,
         )
         self.assertIsInstance(scores, Scores)
+        print(f'Score: {scores}')
 
     def test_untokenized_single_reference(self):
         scores = self.semf1_metric.compute(