Merge pull request kohya-ss#79 from mgz-dev/tensorboard-improvements

expand details in tensorboard logs
Fannovel16 · Jan 21, 2023 · 7ee808d · 7ee808d
2 parents 7dbcef7 e5d9f48
commit 7ee808d
Show file tree

Hide file tree

Showing 2 changed files with 18 additions and 5 deletions.
diff --git a/library/train_util.py b/library/train_util.py
@@ -1408,5  1408,17 @@ def save_state_on_train_end(args: argparse.Namespace, accelerator):
   model_name = DEFAULT_LAST_OUTPUT_NAME if args.output_name is None else args.output_name
   accelerator.save_state(os.path.join(args.output_dir, LAST_STATE_NAME.format(model_name)))
 
 def generate_step_logs(args: argparse.Namespace, current_loss, avr_loss, lr_scheduler):
     logs = {"loss/current": current_loss, "loss/average": avr_loss}
 
     if args.network_train_unet_only:
       logs["lr/unet"] = lr_scheduler.get_last_lr()[0]
     elif args.network_train_text_encoder_only:
       logs["lr/textencoder"] = lr_scheduler.get_last_lr()[0]
     else:
       logs["lr/textencoder"] = lr_scheduler.get_last_lr()[0]
       logs["lr/unet"] = lr_scheduler.get_last_lr()[-1]
 
     return logs
 
 # endregion
diff --git a/train_network.py b/train_network.py
@@ -346,20  346,21 @@ def train(args):
         global_step  = 1
 
       current_loss = loss.detach().item()
-      if args.logging_dir is not None:
-        logs = {"loss": current_loss, "lr": lr_scheduler.get_last_lr()[0]}
-        accelerator.log(logs, step=global_step)
-
       loss_total  = current_loss
       avr_loss = loss_total / (step 1)
       logs = {"loss": avr_loss}  # , "lr": lr_scheduler.get_last_lr()[0]}
       progress_bar.set_postfix(**logs)
 
       if args.logging_dir is not None:
         logs = train_util.generate_step_logs(args, current_loss, avr_loss, lr_scheduler)
 
         accelerator.log(logs, step=global_step)
 
       if global_step >= args.max_train_steps:
         break
 
     if args.logging_dir is not None:
-      logs = {"epoch_loss": loss_total / len(train_dataloader)}
       logs = {"loss/epoch": loss_total / len(train_dataloader)}
       accelerator.log(logs, step=epoch 1)
 
     accelerator.wait_for_everyone()