Respect documentation on passive log level (#21700)

* Respect documentation on passive log level

* Fix test and set log level in examples

* Add doc

examples/pytorch/audio-classification/run_audio_classification.py

@@ -209,6 +209,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     transformers.utils.logging.set_verbosity(log_level)

examples/pytorch/contrastive-image-text/run_clip.py

@@ -246,6 +246,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     transformers.utils.logging.set_verbosity(log_level)

examples/pytorch/image-classification/run_image_classification.py

@@ -187,6 +187,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     transformers.utils.logging.set_verbosity(log_level)

examples/pytorch/image-pretraining/run_mae.py

@@ -186,6 +186,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     transformers.utils.logging.set_verbosity(log_level)

examples/pytorch/image-pretraining/run_mim.py

@@ -250,6 +250,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     transformers.utils.logging.set_verbosity(log_level)

examples/pytorch/language-modeling/run_clm.py

@@ -240,6 +240,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     datasets.utils.logging.set_verbosity(log_level)

examples/pytorch/language-modeling/run_mlm.py

@@ -240,6 +240,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     datasets.utils.logging.set_verbosity(log_level)

examples/pytorch/language-modeling/run_plm.py

@@ -231,6 +231,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     datasets.utils.logging.set_verbosity(log_level)

examples/pytorch/multiple-choice/run_swag.py

@@ -235,6 +235,11 @@ def main():
         datefmt="%m/%d/%Y %H:%M:%S",
         handlers=[logging.StreamHandler(sys.stdout)],
     )
+
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     datasets.utils.logging.set_verbosity(log_level)

examples/pytorch/question-answering/run_qa.py

@@ -238,6 +238,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     datasets.utils.logging.set_verbosity(log_level)

examples/pytorch/question-answering/run_qa_beam_search.py

@@ -236,6 +236,11 @@ def main():
         datefmt="%m/%d/%Y %H:%M:%S",
         handlers=[logging.StreamHandler(sys.stdout)],
     )
+
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     datasets.utils.logging.set_verbosity(log_level)

examples/pytorch/question-answering/run_seq2seq_qa.py

@@ -283,6 +283,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     datasets.utils.logging.set_verbosity(log_level)

examples/pytorch/semantic-segmentation/run_semantic_segmentation.py

@@ -276,6 +276,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     transformers.utils.logging.set_verbosity(log_level)

examples/pytorch/summarization/run_summarization.py

@@ -314,6 +314,11 @@ def main():
         datefmt="%m/%d/%Y %H:%M:%S",
         handlers=[logging.StreamHandler(sys.stdout)],
     )
+
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     datasets.utils.logging.set_verbosity(log_level)

examples/pytorch/text-classification/run_glue.py

@@ -227,6 +227,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     datasets.utils.logging.set_verbosity(log_level)

examples/pytorch/text-classification/run_xnli.py

@@ -186,6 +186,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     datasets.utils.logging.set_verbosity(log_level)

examples/pytorch/token-classification/run_ner.py

@@ -228,6 +228,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     datasets.utils.logging.set_verbosity(log_level)

examples/pytorch/translation/run_translation.py

@@ -272,6 +272,10 @@ def main():
         handlers=[logging.StreamHandler(sys.stdout)],
     )
 
+    if training_args.should_log:
+        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
+        transformers.utils.logging.set_verbosity_info()
+
     log_level = training_args.get_process_log_level()
     logger.setLevel(log_level)
     datasets.utils.logging.set_verbosity(log_level)

src/transformers/training_args.py

@@ -231,9 +231,9 @@ class TrainingArguments:
             Number of steps used for a linear warmup from 0 to `learning_rate`. Overrides any effect of `warmup_ratio`.
         log_level (`str`, *optional*, defaults to `passive`):
             Logger log level to use on the main process. Possible choices are the log levels as strings: 'debug',
-            'info', 'warning', 'error' and 'critical', plus a 'passive' level which doesn't set anything and lets the
-            application set the level.
-        log_level_replica (`str`, *optional*, defaults to `passive`):
+            'info', 'warning', 'error' and 'critical', plus a 'passive' level which doesn't set anything and keeps the
+            current log level for the Transformers library (which will be `"warning"` by default).
+        log_level_replica (`str`, *optional*, defaults to `"warning"`):
             Logger log level to use on replicas. Same choices as `log_level`"
         log_on_each_node (`bool`, *optional*, defaults to `True`):
             In multinode distributed training, whether to log using `log_level` once per node, or only on the main
@@ -690,7 +690,7 @@ class TrainingArguments:
         },
     )
     log_level_replica: Optional[str] = field(
-        default="passive",
+        default="warning",
         metadata={
             "help": "Logger log level to use on replica nodes. Same choices and defaults as ``log_level``",
             "choices": trainer_log_levels.keys(),
@@ -1774,7 +1774,8 @@ class TrainingArguments:
         Returns the log level to be used depending on whether this process is the main process of node 0, main process
         of node non-0, or a non-main process.
 
-        For the main process the log level defaults to `logging.INFO` unless overridden by `log_level` argument.
+        For the main process the log level defaults to the logging level set (`logging.WARNING` if you didn't do
+        anything) unless overridden by `log_level` argument.
 
         For the replica processes the log level defaults to `logging.WARNING` unless overridden by `log_level_replica`
         argument.
@@ -1786,8 +1787,8 @@ class TrainingArguments:
         log_level = trainer_log_levels[self.log_level]
         log_level_replica = trainer_log_levels[self.log_level_replica]
 
-        log_level_main_node = logging.INFO if log_level == -1 else log_level
-        log_level_replica_node = logging.WARNING if log_level_replica == -1 else log_level_replica
+        log_level_main_node = logging.get_verbosity() if log_level == -1 else log_level
+        log_level_replica_node = logging.get_verbosity() if log_level_replica == -1 else log_level_replica
         return log_level_main_node if self.should_log else log_level_replica_node
 
     @property

tests/trainer/test_trainer.py

@@ -1098,11 +1098,11 @@ class TrainerIntegrationTest(TestCasePlus, TrainerIntegrationCommon):
         logger = logging.get_logger()
         log_info_string = "Running training"
 
-        # test with the default log_level - should be info and thus log on the main process
+        # test with the default log_level - should be warning and thus not log on the main process
         with CaptureLogger(logger) as cl:
             trainer = get_regression_trainer()
             trainer.train()
-        self.assertIn(log_info_string, cl.out)
+        self.assertNotIn(log_info_string, cl.out)
 
         # test with low log_level - lower than info
         with CaptureLogger(logger) as cl: