diff --git a/PaddleNLP/benchmark/bert/run_pretrain.py b/PaddleNLP/benchmark/bert/run_pretrain.py
index 4ecea128bb9eec6ab6b80f9136c9c6125e576090..ce5e2d1eb86cd48f025d7780d1c7ef3b32cef50d 100644
--- a/PaddleNLP/benchmark/bert/run_pretrain.py
+++ b/PaddleNLP/benchmark/bert/run_pretrain.py
@@ -326,7 +326,7 @@ def do_train(args):
                 if global_step % args.logging_steps == 0:
                     time_cost = time.time() - tic_train
                     print(
-                        "global step %d, epoch: %d, batch: %d, loss: %f, speed: %.2f step/s, ips :%.2f sequences/s"
+                        "global step %d, epoch: %d, batch: %d, loss: %f, speed: %.2f step/s, ips: %.2f sequences/s"
                         % (global_step, epoch, step, loss_return[0],
                            args.logging_steps / time_cost,
                            args.logging_steps * args.batch_size / time_cost))
diff --git a/PaddleNLP/benchmark/bert/run_pretrain_single.py b/PaddleNLP/benchmark/bert/run_pretrain_single.py
index 48b111623c11d98ebc49f9f6b160cfed102d576e..1bd2104907657e963503b72e609469e10c37c014 100644
--- a/PaddleNLP/benchmark/bert/run_pretrain_single.py
+++ b/PaddleNLP/benchmark/bert/run_pretrain_single.py
@@ -276,7 +276,7 @@ def do_train(args):
                 if global_step % args.logging_steps == 0:
                     time_cost = time.time() - tic_train
                     print(
-                        "global step %d, epoch: %d, batch: %d, loss: %f, speed: %.2f step/s, ips :%.2f sequences/s"
+                        "global step %d, epoch: %d, batch: %d, loss: %f, speed: %.2f step/s, ips: %.2f sequences/s"
                         % (global_step, epoch, step, loss_return[0],
                            args.logging_steps / time_cost,
                            args.logging_steps * args.batch_size / time_cost))