Feature/sg 644 upload big files properly on end

src/super_gradients/common/sg_loggers/base_sg_logger.py

@@ -169,7 +169,7 @@ def add_scalars(self, tag_scalar_dict: dict, global_step: int = None):
         self.tensorboard_writer.flush()
 
         # WRITE THE EPOCH RESULTS TO LOG FILE
-        log_line = f"\nEpoch ({global_step}/{self.max_global_steps})  - "
+        log_line = f"\nEpoch {global_step} ({global_step+1}/{self.max_global_steps})  - "
         for tag, value in tag_scalar_dict.items():
             if isinstance(value, torch.Tensor):
                 value = value.item()
@@ -249,6 +249,9 @@ def add_file(self, file_name: str = None):
 
     @multi_process_safe
     def upload(self):
+        """Upload the local tensorboard and log files to remote system."""
+        self.flush()
+
         if self.save_tensorboard_remote:
             self.model_checkpoints_data_interface.save_remote_tensorboard_event_files(self.experiment_name, self._local_dir)
 
@@ -259,12 +262,16 @@ def upload(self):
     @multi_process_safe
     def flush(self):
         self.tensorboard_writer.flush()
+        ConsoleSink.flush()
 
     @multi_process_safe
     def close(self):
+        self.upload()
+
         if self.system_monitor is not None:
             self.system_monitor.close()
             logger.info("[CLEANUP] - Successfully stopped system monitoring process")
+
         self.tensorboard_writer.close()
         if self.tensor_board_process is not None:
             try:

src/super_gradients/training/sg_trainer/sg_trainer.py

@@ -468,9 +468,6 @@ def _train_epoch(self, epoch: int, silent_mode: bool = False) -> tuple:
             ):
                 break
 
-        if not self.ddp_silent_mode:
-            self.sg_logger.upload()
-
         self.train_monitored_values = sg_trainer_utils.update_monitored_values_dict(
             monitored_values_dict=self.train_monitored_values, new_values_dict=pbar_message_dict
         )
@@ -1315,6 +1312,7 @@ def forward(self, inputs, targets):
                 if not self.ddp_silent_mode:
                     # SAVING AND LOGGING OCCURS ONLY IN THE MAIN PROCESS (IN CASES THERE ARE SEVERAL PROCESSES - DDP)
                     self._write_to_disk_operations(train_metrics_tuple, validation_results_tuple, inf_time, epoch, context)
+                    self.sg_logger.upload()
 
             # Evaluating the average model and removing snapshot averaging file if training is completed
             if self.training_params.average_best_models: