[update] bugfix in dist_trainer

fastNLP/core/_logger.py

@@ -176,4 +176,4 @@ logger = _init_logger(path=None, level='INFO')
 def init_logger_dist():
     global logger
     rank = dist.get_rank()
-    logger.setLevel(logging.INFO if rank else logging.WARNING)
+    logger.setLevel(logging.INFO if rank == 0 else logging.WARNING)

fastNLP/core/dist_trainer.py

@@ -117,6 +117,8 @@ class DistTrainer():
         else:
             self.device = torch.device(device)
 
+        init_logger_dist()
+
         self.world_size = dist.get_world_size()
         self.rank = dist.get_rank() # unique id for each process
 
@@ -180,7 +182,6 @@ class DistTrainer():
         else:
             self.cp_save_path = None
         # use INFO in the master, WARN for others
-        init_logger_dist()
         self.logger = logger
         self.logger.info("Setup Distributed Trainer")
         self.logger.warning("Process pid: {}, rank: {}, local rank: {}, device: {}, fp16: {}".format(
@@ -307,6 +308,7 @@ class DistTrainer():
         return results
 
     def _train(self):
+        dist.barrier()
         if not self.use_tqdm:
             from .utils import _pseudo_tqdm as inner_tqdm
         else: