diff --git a/python/paddle/fluid/tests/unittests/test_imperative_transformer_sorted_gradient.py b/python/paddle/fluid/tests/unittests/test_imperative_transformer_sorted_gradient.py
index a4fabad60c88c38e83f7b4ab4af8b87bacfd7a0b..010c8aeccacd6550a5b8963c63f3d28af898550e 100644
--- a/python/paddle/fluid/tests/unittests/test_imperative_transformer_sorted_gradient.py
+++ b/python/paddle/fluid/tests/unittests/test_imperative_transformer_sorted_gradient.py
@@ -315,7 +315,7 @@ use_py_reader = False
 sync = False
 
 # how many batches we use
-batch_num = 2
+batch_num = 5
 
 np.random.seed(90)
 src_word_np = np.arange(1, TrainTaskConfig.batch_size * seq_len + 1).reshape(