Speed up test_crossencoder (#3587)

tests/nightly/gpu/test_bert.py

@@ -17,47 +17,36 @@ class TestBertModel(unittest.TestCase):
     samples on convai2
     """
 
-    @testing_utils.retry(ntries=3, log_retry=True)
     def test_biencoder(self):
         valid, test = testing_utils.train_model(
             dict(
-                task='convai2',
+                task='integration_tests:overfit',
                 model='bert_ranker/bi_encoder_ranker',
-                num_epochs=0.1,
-                batchsize=8,
-                learningrate=3e-4,
-                text_truncate=32,
-                validation_max_exs=20,
-                short_final_eval=True,
+                max_train_steps=500,
+                batchsize=2,
+                candidates="inline",
+                gradient_clip=1.0,
+                learningrate=1e-3,
+                text_truncate=8,
             )
         )
-        # can't conclude much from the biencoder after that little iterations.
-        # this test will just make sure it hasn't crashed and the accuracy isn't
-        # too high
-        self.assertLessEqual(test['accuracy'], 0.5)
+        self.assertGreaterEqual(test['accuracy'], 0.5)
 
-    @testing_utils.retry(ntries=3, log_retry=True)
     def test_crossencoder(self):
         valid, test = testing_utils.train_model(
             dict(
-                task='convai2',
+                task='integration_tests:overfit',
                 model='bert_ranker/cross_encoder_ranker',
-                num_epochs=0.002,
-                batchsize=1,
-                candidates="inline",
+                max_train_steps=500,
+                batchsize=2,
+                learningrate=1e-3,
+                gradient_clip=1.0,
                 type_optimization="all_encoder_layers",
-                warmup_updates=100,
-                text_truncate=32,
-                label_truncate=32,
-                validation_max_exs=20,
-                short_final_eval=True,
+                text_truncate=8,
+                label_truncate=8,
             )
         )
-        # The cross encoder reaches an interesting state MUCH faster
-        # accuracy should be present and somewhere between 0.2 and 0.8
-        # (large interval so that it doesn't flake.)
-        self.assertGreaterEqual(test['accuracy'], 0.03)
-        self.assertLessEqual(test['accuracy'], 0.8)
+        self.assertGreaterEqual(test['accuracy'], 0.8)
 
 
 if __name__ == '__main__':