pass the matching trainer log level to deepspeed (#12401)

src/transformers/deepspeed.py

@@ -295,11 +295,13 @@ def deepspeed_init(trainer, num_training_steps, resume_from_checkpoint=None):
 
     """
     import deepspeed
+    from deepspeed.utils import logger as ds_logger
 
     model = trainer.model
+    args = trainer.args
 
-    hf_deepspeed_config = trainer.args.hf_deepspeed_config
-    hf_deepspeed_config.trainer_config_finalize(trainer.args, model, num_training_steps)
+    hf_deepspeed_config = args.hf_deepspeed_config
+    hf_deepspeed_config.trainer_config_finalize(args, model, num_training_steps)
 
     # resume config update - some bits like `model` and `num_training_steps` only become available during train
     config = hf_deepspeed_config.config
@@ -319,7 +321,7 @@ def deepspeed_init(trainer, num_training_steps, resume_from_checkpoint=None):
 
     optimizer = None
     if "optimizer" in config:
-        if trainer.args.adafactor:
+        if args.adafactor:
             raise ValueError(
                 "--adafactor was passed, but also found `optimizer` configured in the DeepSpeed config. "
                 "Only one optimizer can be configured."
@@ -356,6 +358,9 @@ def deepspeed_init(trainer, num_training_steps, resume_from_checkpoint=None):
     # keep for quick debug:
     # from pprint import pprint; pprint(config)
 
+    # set the Deepspeed log level consistent with the trainer
+    ds_logger.setLevel(args.get_process_log_level())
+
     model_parameters = filter(lambda p: p.requires_grad, model.parameters())
 
     model, optimizer, _, lr_scheduler = deepspeed.initialize(