Centralize logging (#6434)

* Logging

* Style

* hf_logging > utils.logging

* Address @thomwolf's comments

* Update test

* Update src/transformers/benchmark/benchmark_utils.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Revert bad change
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

src/transformers/convert_bert_original_tf2_checkpoint_to_pytorch.py

@@ -8,7 +8,6 @@ The script re-maps the TF2.x Bert weight names to the original names, so the mod
 You may adapt this script to include classification/MLM/NSP/etc. heads.
 """
 import argparse
-import logging
 import os
 import re
 
@@ -17,9 +16,11 @@ import torch
 
 from transformers import BertConfig, BertModel
 
+from .utils import logging
 
-logging.basicConfig(level=logging.INFO)
-logger = logging.getLogger(__name__)
+
+logging.set_verbosity_info()
+logger = logging.get_logger(__name__)
 
 
 def load_tf2_weights_in_bert(model, tf_checkpoint_path, config):

src/transformers/convert_bert_original_tf_checkpoint_to_pytorch.py

@@ -16,14 +16,15 @@
 
 
 import argparse
-import logging
 
 import torch
 
 from transformers import BertConfig, BertForPreTraining, load_tf_weights_in_bert
 
+from .utils import logging
 
-logging.basicConfig(level=logging.INFO)
+
+logging.set_verbosity_info()
 
 
 def convert_tf_checkpoint_to_pytorch(tf_checkpoint_path, bert_config_file, pytorch_dump_path):

src/transformers/convert_electra_original_tf_checkpoint_to_pytorch.py

@@ -16,14 +16,15 @@
 
 
 import argparse
-import logging
 
 import torch
 
 from transformers import ElectraConfig, ElectraForMaskedLM, ElectraForPreTraining, load_tf_weights_in_electra
 
+from .utils import logging
 
-logging.basicConfig(level=logging.INFO)
+
+logging.set_verbosity_info()
 
 
 def convert_tf_checkpoint_to_pytorch(tf_checkpoint_path, config_file, pytorch_dump_path, discriminator_or_generator):

src/transformers/convert_gpt2_original_tf_checkpoint_to_pytorch.py

@@ -16,14 +16,15 @@
 
 
 import argparse
-import logging
 
 import torch
 
 from transformers import CONFIG_NAME, WEIGHTS_NAME, GPT2Config, GPT2Model, load_tf_weights_in_gpt2
 
+from .utils import logging
 
-logging.basicConfig(level=logging.INFO)
+
+logging.set_verbosity_info()
 
 
 def convert_gpt2_checkpoint_to_pytorch(gpt2_checkpoint_path, gpt2_config_file, pytorch_dump_folder_path):

src/transformers/convert_mobilebert_original_tf_checkpoint_to_pytorch.py

 import argparse
-import logging
 
 import torch
 
 from transformers import MobileBertConfig, MobileBertForPreTraining, load_tf_weights_in_mobilebert
 
+from .utils import logging
 
-logging.basicConfig(level=logging.INFO)
+
+logging.set_verbosity_info()
 
 
 def convert_tf_checkpoint_to_pytorch(tf_checkpoint_path, mobilebert_config_file, pytorch_dump_path):

src/transformers/convert_openai_original_tf_checkpoint_to_pytorch.py

@@ -16,14 +16,15 @@
 
 
 import argparse
-import logging
 
 import torch
 
 from transformers import CONFIG_NAME, WEIGHTS_NAME, OpenAIGPTConfig, OpenAIGPTModel, load_tf_weights_in_openai_gpt
 
+from .utils import logging
 
-logging.basicConfig(level=logging.INFO)
+
+logging.set_verbosity_info()
 
 
 def convert_openai_checkpoint_to_pytorch(openai_checkpoint_folder_path, openai_config_file, pytorch_dump_folder_path):

src/transformers/convert_pytorch_checkpoint_to_tf2.py

@@ -16,7 +16,6 @@
 
 
 import argparse
-import logging
 import os
 
 from transformers import (
@@ -76,6 +75,8 @@ from transformers import (
 )
 from transformers.file_utils import hf_bucket_url
 
+from .utils import logging
+
 
 if is_torch_available():
     import numpy as np
@@ -104,7 +105,7 @@ if is_torch_available():
     )
 
 
-logging.basicConfig(level=logging.INFO)
+logging.set_verbosity_info()
 
 MODEL_CLASSES = {
     "bert": (BertConfig, TFBertForPreTraining, BertForPreTraining, BERT_PRETRAINED_CONFIG_ARCHIVE_MAP,),

src/transformers/convert_reformer_trax_checkpoint_to_pytorch.py

@@ -16,7 +16,6 @@
 
 
 import argparse
-import logging
 import pickle
 
 import numpy as np
@@ -24,8 +23,10 @@ import torch
 
 from transformers import ReformerConfig, ReformerModelWithLMHead
 
+from .utils import logging
 
-logging.basicConfig(level=logging.INFO)
+
+logging.set_verbosity_info()
 
 
 def set_param(torch_layer, weight, bias=None):

src/transformers/convert_roberta_original_pytorch_checkpoint_to_pytorch.py

@@ -16,7 +16,6 @@
 
 
 import argparse
-import logging
 import pathlib
 
 import fairseq
@@ -28,13 +27,15 @@ 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
+
 
 if version.parse(fairseq.__version__) < version.parse("0.9.0"):
     raise Exception("requires fairseq >= 0.9.0")
 
 
-logging.basicConfig(level=logging.INFO)
-logger = logging.getLogger(__name__)
+logging.set_verbosity_info()
+logger = logging.get_logger(__name__)
 
 SAMPLE_TEXT = "Hello world! cécé herlolip"
 

src/transformers/convert_t5_original_tf_checkpoint_to_pytorch.py

@@ -16,14 +16,15 @@
 
 
 import argparse
-import logging
 
 import torch
 
 from transformers import T5Config, T5Model, load_tf_weights_in_t5
 
+from .utils import logging
 
-logging.basicConfig(level=logging.INFO)
+
+logging.set_verbosity_info()
 
 
 def convert_tf_checkpoint_to_pytorch(tf_checkpoint_path, config_file, pytorch_dump_path):

src/transformers/convert_transfo_xl_original_tf_checkpoint_to_pytorch.py

@@ -16,7 +16,6 @@
 
 
 import argparse
-import logging
 import os
 import pickle
 import sys
@@ -33,8 +32,10 @@ from transformers import (
 )
 from transformers.tokenization_transfo_xl import CORPUS_NAME, VOCAB_FILES_NAMES
 
+from .utils import logging
 
-logging.basicConfig(level=logging.INFO)
+
+logging.set_verbosity_info()
 
 # We do this to be able to load python 2 datasets pickles
 # See e.g. https://stackoverflow.com/questions/2121874/python-pickling-after-changing-a-modules-directory/2121918#2121918

src/transformers/convert_xlm_original_pytorch_checkpoint_to_pytorch.py

@@ -17,7 +17,6 @@
 
 import argparse
 import json
-import logging
 
 import numpy
 import torch
@@ -25,8 +24,10 @@ import torch
 from transformers import CONFIG_NAME, WEIGHTS_NAME
 from transformers.tokenization_xlm import VOCAB_FILES_NAMES
 
+from .utils import logging
 
-logging.basicConfig(level=logging.INFO)
+
+logging.set_verbosity_info()
 
 
 def convert_xlm_checkpoint_to_pytorch(xlm_checkpoint_path, pytorch_dump_folder_path):

src/transformers/convert_xlnet_original_tf_checkpoint_to_pytorch.py

@@ -16,7 +16,6 @@
 
 
 import argparse
-import logging
 import os
 
 import torch
@@ -31,6 +30,8 @@ from transformers import (
     load_tf_weights_in_xlnet,
 )
 
+from .utils import logging
+
 
 GLUE_TASKS_NUM_LABELS = {
     "cola": 2,
@@ -45,7 +46,7 @@ GLUE_TASKS_NUM_LABELS = {
 }
 
 
-logging.basicConfig(level=logging.INFO)
+logging.set_verbosity_info()
 
 
 def convert_xlnet_checkpoint_to_pytorch(

src/transformers/data/datasets/glue.py

-import logging
 import os
 import time
 from dataclasses import dataclass, field
@@ -14,11 +13,12 @@ from ...tokenization_bart import BartTokenizer, BartTokenizerFast
 from ...tokenization_roberta import RobertaTokenizer, RobertaTokenizerFast
 from ...tokenization_utils import PreTrainedTokenizer
 from ...tokenization_xlm_roberta import XLMRobertaTokenizer
+from ...utils import logging
 from ..processors.glue import glue_convert_examples_to_features, glue_output_modes, glue_processors
 from ..processors.utils import InputFeatures
 
 
-logger = logging.getLogger(__name__)
+logger = logging.get_logger(__name__)
 
 
 @dataclass

src/transformers/data/datasets/language_modeling.py

-import logging
 import os
 import pickle
 import time
@@ -9,9 +8,10 @@ from torch.utils.data.dataset import Dataset
 from filelock import FileLock
 
 from ...tokenization_utils import PreTrainedTokenizer
+from ...utils import logging
 
 
-logger = logging.getLogger(__name__)
+logger = logging.get_logger(__name__)
 
 
 class TextDataset(Dataset):

src/transformers/data/datasets/squad.py

-import logging
 import os
 import time
 from dataclasses import dataclass, field
@@ -12,10 +11,11 @@ from filelock import FileLock
 
 from ...modeling_auto import MODEL_FOR_QUESTION_ANSWERING_MAPPING
 from ...tokenization_utils import PreTrainedTokenizer
+from ...utils import logging
 from ..processors.squad import SquadFeatures, SquadV1Processor, SquadV2Processor, squad_convert_examples_to_features
 
 
-logger = logging.getLogger(__name__)
+logger = logging.get_logger(__name__)
 
 MODEL_CONFIG_CLASSES = list(MODEL_FOR_QUESTION_ANSWERING_MAPPING.keys())
 MODEL_TYPES = tuple(conf.model_type for conf in MODEL_CONFIG_CLASSES)

src/transformers/data/metrics/squad_metrics.py

@@ -10,15 +10,16 @@ that a question is unanswerable.
 
 import collections
 import json
-import logging
 import math
 import re
 import string
 
 from transformers.tokenization_bert import BasicTokenizer
 
+from ...utils import logging
 
-logger = logging.getLogger(__name__)
+
+logger = logging.get_logger(__name__)
 
 
 def normalize_answer(s):

src/transformers/data/processors/glue.py

@@ -15,7 +15,6 @@
 # limitations under the License.
 """ GLUE processors and helpers """
 
-import logging
 import os
 from dataclasses import asdict
 from enum import Enum
@@ -23,13 +22,14 @@ from typing import List, Optional, Union
 
 from ...file_utils import is_tf_available
 from ...tokenization_utils import PreTrainedTokenizer
+from ...utils import logging
 from .utils import DataProcessor, InputExample, InputFeatures
 
 
 if is_tf_available():
     import tensorflow as tf
 
-logger = logging.getLogger(__name__)
+logger = logging.get_logger(__name__)
 
 
 def glue_convert_examples_to_features(

src/transformers/data/processors/squad.py

 import json
-import logging
 import os
 from functools import partial
 from multiprocessing import Pool, cpu_count
@@ -10,6 +9,7 @@ from tqdm import tqdm
 from ...file_utils import is_tf_available, is_torch_available
 from ...tokenization_bert import whitespace_tokenize
 from ...tokenization_utils_base import TruncationStrategy
+from ...utils import logging
 from .utils import DataProcessor
 
 
@@ -24,7 +24,7 @@ if is_torch_available():
 if is_tf_available():
     import tensorflow as tf
 
-logger = logging.getLogger(__name__)
+logger = logging.get_logger(__name__)
 
 
 def _improve_answer_span(doc_tokens, input_start, input_end, tokenizer, orig_answer_text):

src/transformers/data/processors/utils.py

@@ -17,14 +17,14 @@
 import csv
 import dataclasses
 import json
-import logging
 from dataclasses import dataclass
 from typing import List, Optional, Union
 
 from ...file_utils import is_tf_available, is_torch_available
+from ...utils import logging
 
 
-logger = logging.getLogger(__name__)
+logger = logging.get_logger(__name__)
 
 
 @dataclass