turn excessive noise off (#1293)

Author: stas00

deepspeed/runtime/engine.py

@@ -2051,7 +2051,7 @@ def _copy_recovery_script(self, save_path):
         script = "zero_to_fp32.py"
         src = os.path.join(base_dir, "utils", script)
         dst = os.path.join(save_path, script)
-        logger.info(f"creating recovery script {dst}")
+        #logger.info(f"creating recovery script {dst}")
         copyfile(src, dst)
         # make executable
         os.chmod(dst, os.stat(dst).st_mode | stat.S_IEXEC)
@@ -2064,7 +2064,7 @@ def _save_zero_checkpoint(self, save_path, tag):
                        ds_version=version)
         torch.save(zero_sd, zero_checkpoint_name)
         self._copy_recovery_script(save_path)
-        logger.info('zero checkpoint saved {}'.format(zero_checkpoint_name))
+        #logger.info('zero checkpoint saved {}'.format(zero_checkpoint_name))
 
     def _zero3_consolidated_fp16_state_dict(self):
         """

deepspeed/runtime/pipe/module.py

@@ -577,10 +577,10 @@ def load_state_dir(self, load_dir, strict=True):
 
             layer.load_state_dict(checkpoint)
 
-            if self._grid.data_parallel_id == 0:
-                logger.info(
-                    f'RANK={self.global_rank} Loaded layer={idx+self._local_start} file={load_path}'
-                )
+            # if self._grid.data_parallel_id == 0:
+            #     logger.info(
+            #         f'RANK={self.global_rank} Loaded layer={idx+self._local_start} file={load_path}'
+            #     )
 
         self._synchronize_tied_weights()
 

deepspeed/runtime/state_dict_factory.py

@@ -78,7 +78,7 @@ def load(self,
         merge_count = 1
         if num_ckpt == mp_world_size:
             assert os.path.exists(load_path)
-            logger.info(f'rank: {mp_rank} loading checkpoint: {load_path}')
+            #logger.info(f'rank: {mp_rank} loading checkpoint: {load_path}')
             sd = torch.load(load_path, map_location=lambda storage, loc: storage)
 
             if quantize:
@@ -158,7 +158,7 @@ def set_module(self, sd, module):
         return sd
 
     def check_ckpt_list(self):
-        logger.info(f'checkpoint file list: {self.ckpt_list}')
+        #logger.info(f'checkpoint file list: {self.ckpt_list}')
         assert len(self.ckpt_list) > 0
 
         sd = torch.load(self.ckpt_list[0], map_location=lambda storage, loc: storage)