Log when removing bested checkpoint

dan/ocr/manager/training.py

 # -*- coding: utf-8 -*-
+import logging
 import os
 import random
 from copy import deepcopy
@@ -28,6 +29,7 @@ from dan.utils import fix_ddp_layers_names, ind_to_token
 if MLFLOW_AVAILABLE:
     import mlflow
 
+logger = logging.getLogger(__name__)
 MODEL_NAMES = ("encoder", "decoder")
 
 
@@ -430,6 +432,7 @@ class GenericTrainingManager:
 
         # Remove other checkpoints
         for path_to_del in self.paths["checkpoints"].glob(f"{name}_*.pt"):
+            logger.warning(f"Removing checkpoint `{path_to_del}`")
             path_to_del.unlink()
 
         torch.save(content, path)