diff --git a/modules/trainer/GenericTrainer.py b/modules/trainer/GenericTrainer.py
index dd17ad76e..a0b069267 100644
--- a/modules/trainer/GenericTrainer.py
+++ b/modules/trainer/GenericTrainer.py
@@ -193,7 +193,7 @@ def __prune_backups(self, backups_to_keep: int):
                 try:
                     shutil.rmtree(dirpath)
                 except Exception:
-                    print(f"Could not delete old rolling backup {dirpath}")
+                    tqdm.write(f"Could not delete old rolling backup {dirpath}")
 
         return
 
@@ -270,7 +270,7 @@ def on_sample_custom(sampler_output: ModelSamplerOutput):
                 )
             except Exception:
                 traceback.print_exc()
-                print("Error during sampling, proceeding without sampling")
+                tqdm.write("Error during sampling, proceeding without sampling")
 
             torch_gc()
 
@@ -303,7 +303,7 @@ def __sample_during_training(
             # We absolutely do not want to fail training just because the sample definition file becomes missing or broken right before sampling.
             except Exception:
                 traceback.print_exc()
-                print("Error during loading the sample definition file, proceeding without sampling")
+                tqdm.write("Error during loading the sample definition file, proceeding without sampling")
                 sample_params_list = []
 
         if self.model.ema:
@@ -428,7 +428,7 @@ def __save_backup_config(self, backup_path):
         if os.path.isfile(self.config.sample_definition_file_name):
             shutil.copy2(self.config.sample_definition_file_name, samples_path)
 
-    def __backup(self, train_progress: TrainProgress, print_msg: bool = True, print_cb: Callable[[str], None] = print):
+    def __backup(self, train_progress: TrainProgress, print_msg: bool = True):
         torch_gc()
 
         self.callbacks.on_update_status("Creating backup")
@@ -443,7 +443,7 @@ def __backup(self, train_progress: TrainProgress, print_msg: bool = True, print_
 
         try:
             if print_msg:
-                print_cb("Creating Backup " + backup_path)
+                tqdm.write("Creating Backup " + backup_path)
 
             self.model_saver.save(
                 self.model,
@@ -456,13 +456,13 @@ def __backup(self, train_progress: TrainProgress, print_msg: bool = True, print_
             self.__save_backup_config(backup_path)
         except Exception:
             traceback.print_exc()
-            print("Could not save backup. Check your disk space!")
+            tqdm.write("Could not save backup. Check your disk space!")
             try:
                 if os.path.isdir(backup_path):
                     shutil.rmtree(backup_path)
             except Exception:
                 traceback.print_exc()
-                print("Could not delete partial backup")
+                tqdm.write("Could not delete partial backup")
         finally:
             if self.config.rolling_backup:
                 self.__prune_backups(self.config.rolling_backup_count)
@@ -475,7 +475,7 @@ def __backup(self, train_progress: TrainProgress, print_msg: bool = True, print_
 
         torch_gc()
 
-    def __save(self, train_progress: TrainProgress, print_msg: bool = True, print_cb: Callable[[str], None] = print):
+    def __save(self, train_progress: TrainProgress, print_msg: bool = True):
         torch_gc()
 
         self.callbacks.on_update_status("Saving")
@@ -486,7 +486,7 @@ def __save(self, train_progress: TrainProgress, print_msg: bool = True, print_cb
             f"{self.config.save_filename_prefix}{get_string_timestamp()}-save-{train_progress.filename_string()}{self.config.output_model_format.file_extension()}"
         )
         if print_msg:
-            print_cb("Saving " + save_path)
+            tqdm.write("Saving " + save_path)
 
         try:
             if self.model.ema:
@@ -508,13 +508,13 @@ def __save(self, train_progress: TrainProgress, print_msg: bool = True, print_cb
                 self.model.optimizer.train()
         except Exception:
             traceback.print_exc()
-            print("Could not save model. Check your disk space!")
+            tqdm.write("Could not save model. Check your disk space!")
             try:
                 if os.path.isfile(save_path):
                     shutil.rmtree(save_path)
             except Exception:
                 traceback.print_exc()
-                print("Could not delete partial save")
+                tqdm.write("Could not delete partial save")
         finally:
             if self.model.ema:
                 self.model.ema.copy_temp_to(self.parameters)
@@ -719,9 +719,9 @@ def sample_commands_fun():
                     if multi.is_master() and (backup or save):
                         self.model.to(self.temp_device)
                         if backup:
-                            self.__backup(train_progress, True, step_tqdm.write)
+                            self.__backup(train_progress, True)
                         if save:
-                            self.__save(train_progress, True, step_tqdm.write)
+                            self.__save(train_progress, True)
                         self.model_setup.setup_train_device(self.model, self.config)
 
                 self.callbacks.on_update_status("Training ...")
diff --git a/modules/util/multi_gpu_util.py b/modules/util/multi_gpu_util.py
index d983a2e8c..962434b65 100644
--- a/modules/util/multi_gpu_util.py
+++ b/modules/util/multi_gpu_util.py
@@ -6,6 +6,8 @@
 
 import torch
 
+from tqdm import tqdm
+
 
 def is_enabled() -> bool:
     return torch.distributed.is_available() and torch.distributed.is_initialized()
@@ -149,7 +151,7 @@ def parameter_divergence(params: list[torch.Tensor], train_device: torch.device)
 def warn_parameter_divergence(params: list[torch.Tensor], train_device: torch.device):
     divergence = parameter_divergence(params, train_device)
     if divergence is not None and divergence > 0:
-        print(f"\n\nWARNING: Parameter divergence between GPUs of {divergence}\n\n")
+        tqdm.write(f"WARNING: Parameter divergence between GPUs of {divergence}")
 
 
 @torch.no_grad()