diff --git a/official/modeling/model_training_utils.py b/official/nlp/bert/model_training_utils.py
similarity index 100%
rename from official/modeling/model_training_utils.py
rename to official/nlp/bert/model_training_utils.py
diff --git a/official/modeling/model_training_utils_test.py b/official/nlp/bert/model_training_utils_test.py
similarity index 99%
rename from official/modeling/model_training_utils_test.py
rename to official/nlp/bert/model_training_utils_test.py
index e4e787a6439..87ef7b6a03c 100644
--- a/official/modeling/model_training_utils_test.py
+++ b/official/nlp/bert/model_training_utils_test.py
@@ -28,7 +28,7 @@
 
 from tensorflow.python.distribute import combinations
 from tensorflow.python.distribute import strategy_combinations
-from official.modeling import model_training_utils
+from official.nlp.bert import model_training_utils
 
 
 def eager_strategy_combinations():
diff --git a/official/nlp/bert/run_classifier.py b/official/nlp/bert/run_classifier.py
index d45a78820b6..35c4388609e 100644
--- a/official/nlp/bert/run_classifier.py
+++ b/official/nlp/bert/run_classifier.py
@@ -25,8 +25,6 @@
 from absl import flags
 from absl import logging
 import tensorflow as tf
-
-from official.modeling import model_training_utils
 from official.modeling import performance
 from official.nlp import optimization
 from official.nlp.bert import bert_models
@@ -34,6 +32,7 @@
 from official.nlp.bert import configs as bert_configs
 from official.nlp.bert import input_pipeline
 from official.nlp.bert import model_saving_utils
+from official.nlp.bert import model_training_utils
 from official.utils.misc import distribution_utils
 from official.utils.misc import keras_utils
 
diff --git a/official/nlp/bert/run_pretraining.py b/official/nlp/bert/run_pretraining.py
index b7d28fb4354..822955d6006 100644
--- a/official/nlp/bert/run_pretraining.py
+++ b/official/nlp/bert/run_pretraining.py
@@ -22,14 +22,13 @@
 from absl import logging
 import gin
 import tensorflow as tf
-
-from official.modeling import model_training_utils
 from official.modeling import performance
 from official.nlp import optimization
 from official.nlp.bert import bert_models
 from official.nlp.bert import common_flags
 from official.nlp.bert import configs
 from official.nlp.bert import input_pipeline
+from official.nlp.bert import model_training_utils
 from official.utils.misc import distribution_utils
 
 
diff --git a/official/nlp/bert/run_squad_helper.py b/official/nlp/bert/run_squad_helper.py
index 51f3731ef46..293a42693b0 100644
--- a/official/nlp/bert/run_squad_helper.py
+++ b/official/nlp/bert/run_squad_helper.py
@@ -23,14 +23,13 @@
 from absl import flags
 from absl import logging
 import tensorflow as tf
-
-from official.modeling import model_training_utils
 from official.modeling import performance
 from official.nlp import optimization
 from official.nlp.bert import bert_models
 from official.nlp.bert import common_flags
 from official.nlp.bert import input_pipeline
 from official.nlp.bert import model_saving_utils
+from official.nlp.bert import model_training_utils
 from official.nlp.bert import squad_evaluate_v1_1
 from official.nlp.data import squad_lib_sp
 from official.utils.misc import keras_utils
diff --git a/official/nlp/xlnet/training_utils.py b/official/nlp/xlnet/training_utils.py
index 627fa975113..293e4633d8f 100644
--- a/official/nlp/xlnet/training_utils.py
+++ b/official/nlp/xlnet/training_utils.py
@@ -28,7 +28,7 @@
 import tensorflow as tf
 from typing import Any, Callable, Dict, Text, Optional
 
-from official.modeling import model_training_utils
+from official.nlp.bert import model_training_utils
 from official.nlp.xlnet import data_utils
 from official.nlp.xlnet import xlnet_modeling as modeling