diff --git a/src/transformers/convert_albert_original_tf_checkpoint_to_pytorch.py b/src/transformers/convert_albert_original_tf_checkpoint_to_pytorch.py
index 8fabca0fbdd..10c018170fc 100644
--- a/src/transformers/convert_albert_original_tf_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_albert_original_tf_checkpoint_to_pytorch.py
@@ -20,8 +20,7 @@ import argparse
 import torch
 
 from transformers import AlbertConfig, AlbertForPreTraining, load_tf_weights_in_albert
-
-from .utils import logging
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()
diff --git a/src/transformers/convert_bart_original_pytorch_checkpoint_to_pytorch.py b/src/transformers/convert_bart_original_pytorch_checkpoint_to_pytorch.py
index 8f460a5914e..c07b50fef0c 100644
--- a/src/transformers/convert_bart_original_pytorch_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_bart_original_pytorch_checkpoint_to_pytorch.py
@@ -31,8 +31,7 @@ from transformers import (
     BartTokenizer,
 )
 from transformers.modeling_bart import _make_linear_from_emb
-
-from .utils import logging
+from transformers.utils import logging
 
 
 FAIRSEQ_MODELS = ["bart.large", "bart.large.mnli", "bart.large.cnn", "bart_xsum/model.pt"]
diff --git a/src/transformers/convert_bert_original_tf2_checkpoint_to_pytorch.py b/src/transformers/convert_bert_original_tf2_checkpoint_to_pytorch.py
index a68bf25487f..e4e0e3f55ae 100644
--- a/src/transformers/convert_bert_original_tf2_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_bert_original_tf2_checkpoint_to_pytorch.py
@@ -15,8 +15,7 @@ import tensorflow as tf
 import torch
 
 from transformers import BertConfig, BertModel
-
-from .utils import logging
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()
diff --git a/src/transformers/convert_bert_original_tf_checkpoint_to_pytorch.py b/src/transformers/convert_bert_original_tf_checkpoint_to_pytorch.py
index bc1c8d128f9..d1cb69a2eb4 100755
--- a/src/transformers/convert_bert_original_tf_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_bert_original_tf_checkpoint_to_pytorch.py
@@ -20,8 +20,7 @@ import argparse
 import torch
 
 from transformers import BertConfig, BertForPreTraining, load_tf_weights_in_bert
-
-from .utils import logging
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()
diff --git a/src/transformers/convert_electra_original_tf_checkpoint_to_pytorch.py b/src/transformers/convert_electra_original_tf_checkpoint_to_pytorch.py
index b5f1278ddb2..9cbfcf665dc 100644
--- a/src/transformers/convert_electra_original_tf_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_electra_original_tf_checkpoint_to_pytorch.py
@@ -20,8 +20,7 @@ import argparse
 import torch
 
 from transformers import ElectraConfig, ElectraForMaskedLM, ElectraForPreTraining, load_tf_weights_in_electra
-
-from .utils import logging
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()
diff --git a/src/transformers/convert_gpt2_original_tf_checkpoint_to_pytorch.py b/src/transformers/convert_gpt2_original_tf_checkpoint_to_pytorch.py
index 4324bc5a8dd..e42ebd888d1 100755
--- a/src/transformers/convert_gpt2_original_tf_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_gpt2_original_tf_checkpoint_to_pytorch.py
@@ -20,8 +20,7 @@ import argparse
 import torch
 
 from transformers import CONFIG_NAME, WEIGHTS_NAME, GPT2Config, GPT2Model, load_tf_weights_in_gpt2
-
-from .utils import logging
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()
diff --git a/src/transformers/convert_mobilebert_original_tf_checkpoint_to_pytorch.py b/src/transformers/convert_mobilebert_original_tf_checkpoint_to_pytorch.py
index 468c503fd72..767b36a5703 100644
--- a/src/transformers/convert_mobilebert_original_tf_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_mobilebert_original_tf_checkpoint_to_pytorch.py
@@ -3,8 +3,7 @@ import argparse
 import torch
 
 from transformers import MobileBertConfig, MobileBertForPreTraining, load_tf_weights_in_mobilebert
-
-from .utils import logging
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()
diff --git a/src/transformers/convert_openai_original_tf_checkpoint_to_pytorch.py b/src/transformers/convert_openai_original_tf_checkpoint_to_pytorch.py
index 83760e00d6b..397884e32c0 100755
--- a/src/transformers/convert_openai_original_tf_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_openai_original_tf_checkpoint_to_pytorch.py
@@ -20,8 +20,7 @@ import argparse
 import torch
 
 from transformers import CONFIG_NAME, WEIGHTS_NAME, OpenAIGPTConfig, OpenAIGPTModel, load_tf_weights_in_openai_gpt
-
-from .utils import logging
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()
diff --git a/src/transformers/convert_pytorch_checkpoint_to_tf2.py b/src/transformers/convert_pytorch_checkpoint_to_tf2.py
index cb0ef848485..43779c964dd 100755
--- a/src/transformers/convert_pytorch_checkpoint_to_tf2.py
+++ b/src/transformers/convert_pytorch_checkpoint_to_tf2.py
@@ -78,8 +78,7 @@ from transformers import (
     load_pytorch_checkpoint_in_tf2_model,
 )
 from transformers.file_utils import hf_bucket_url
-
-from .utils import logging
+from transformers.utils import logging
 
 
 if is_torch_available():
diff --git a/src/transformers/convert_reformer_trax_checkpoint_to_pytorch.py b/src/transformers/convert_reformer_trax_checkpoint_to_pytorch.py
index 97d1a63f51f..ec58e2f9132 100755
--- a/src/transformers/convert_reformer_trax_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_reformer_trax_checkpoint_to_pytorch.py
@@ -22,8 +22,7 @@ import numpy as np
 import torch
 
 from transformers import ReformerConfig, ReformerModelWithLMHead
-
-from .utils import logging
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()
diff --git a/src/transformers/convert_roberta_original_pytorch_checkpoint_to_pytorch.py b/src/transformers/convert_roberta_original_pytorch_checkpoint_to_pytorch.py
index 5b4857cb76f..dbb73db2820 100644
--- a/src/transformers/convert_roberta_original_pytorch_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_roberta_original_pytorch_checkpoint_to_pytorch.py
@@ -26,8 +26,7 @@ from packaging import version
 
 from transformers.modeling_bert import BertIntermediate, BertLayer, BertOutput, BertSelfAttention, BertSelfOutput
 from transformers.modeling_roberta import RobertaConfig, RobertaForMaskedLM, RobertaForSequenceClassification
-
-from .utils import logging
+from transformers.utils import logging
 
 
 if version.parse(fairseq.__version__) < version.parse("0.9.0"):
diff --git a/src/transformers/convert_t5_original_tf_checkpoint_to_pytorch.py b/src/transformers/convert_t5_original_tf_checkpoint_to_pytorch.py
index 2e1b5c35e2a..67b65faba97 100755
--- a/src/transformers/convert_t5_original_tf_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_t5_original_tf_checkpoint_to_pytorch.py
@@ -20,8 +20,7 @@ import argparse
 import torch
 
 from transformers import T5Config, T5Model, load_tf_weights_in_t5
-
-from .utils import logging
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()
diff --git a/src/transformers/convert_transfo_xl_original_tf_checkpoint_to_pytorch.py b/src/transformers/convert_transfo_xl_original_tf_checkpoint_to_pytorch.py
index d07ae69f4f2..9135b0d088e 100755
--- a/src/transformers/convert_transfo_xl_original_tf_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_transfo_xl_original_tf_checkpoint_to_pytorch.py
@@ -31,8 +31,7 @@ from transformers import (
     load_tf_weights_in_transfo_xl,
 )
 from transformers.tokenization_transfo_xl import CORPUS_NAME, VOCAB_FILES_NAMES
-
-from .utils import logging
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()
diff --git a/src/transformers/convert_xlm_original_pytorch_checkpoint_to_pytorch.py b/src/transformers/convert_xlm_original_pytorch_checkpoint_to_pytorch.py
index 7a72d3daa49..9baf1159125 100755
--- a/src/transformers/convert_xlm_original_pytorch_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_xlm_original_pytorch_checkpoint_to_pytorch.py
@@ -23,8 +23,7 @@ import torch
 
 from transformers import CONFIG_NAME, WEIGHTS_NAME
 from transformers.tokenization_xlm import VOCAB_FILES_NAMES
-
-from .utils import logging
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()
diff --git a/src/transformers/convert_xlnet_original_tf_checkpoint_to_pytorch.py b/src/transformers/convert_xlnet_original_tf_checkpoint_to_pytorch.py
index 633fd01ca68..5ad56c73b51 100755
--- a/src/transformers/convert_xlnet_original_tf_checkpoint_to_pytorch.py
+++ b/src/transformers/convert_xlnet_original_tf_checkpoint_to_pytorch.py
@@ -29,8 +29,7 @@ from transformers import (
     XLNetLMHeadModel,
     load_tf_weights_in_xlnet,
 )
-
-from .utils import logging
+from transformers.utils import logging
 
 
 GLUE_TASKS_NUM_LABELS = {