diff --git a/megatron/training/training.py b/megatron/training/training.py
index 4d00bd1c8a..d071673aae 100644
--- a/megatron/training/training.py
+++ b/megatron/training/training.py
@@ -76,7 +76,6 @@
     reduce_max_stat_across_model_parallel_group,
     is_last_rank,
     print_rank_0,
-    print_rank_last,
     report_memory,
     unwrap_model,
     update_use_dist_ckpt,
@@ -1073,7 +1072,7 @@ def training_log(loss_dict, total_loss_dict, learning_rate, decoupled_learning_r
         total_loss_dict[advanced_iters_key] = 0
         total_loss_dict[skipped_iters_key] = 0
         total_loss_dict[nan_iters_key] = 0
-        print_rank_last(log_string)
+        print_rank_0(log_string)
         if report_memory_flag:
             # Report memory after optimizer state has been initialized.
             if torch.distributed.get_rank() == 0:
@@ -1784,9 +1783,9 @@ def evaluate_and_print_results(prefix, forward_step_func,
         process_non_loss_data_func(collected_non_loss_data, iteration, writer)
 
     length = len(string) + 1
-    print_rank_last('-' * length)
-    print_rank_last(string)
-    print_rank_last('-' * length)
+    print_rank_0('-' * length)
+    print_rank_0(string)
+    print_rank_0('-' * length)
 
 
 def cyclic_iter(iter):