4 files changed, 4 insertions, 6 deletions

diff --git a/training/functional.py b/training/functional.py
index e14aeea..46d25f6 100644
--- a/training/functional.py
+++ b/training/functional.py
@@ -644,11 +644,9 @@ def train(
     min_snr_gamma: int = 5,
     **kwargs,
 ):
-    text_encoder, unet, optimizer, train_dataloader, val_dataloader, lr_scheduler, extra = strategy.prepare(
+    text_encoder, unet, optimizer, train_dataloader, val_dataloader, lr_scheduler = strategy.prepare(
         accelerator, text_encoder, unet, optimizer, train_dataloader, val_dataloader, lr_scheduler, **kwargs)
 
-    kwargs.update(extra)
-
     vae.to(accelerator.device, dtype=dtype)
     vae.requires_grad_(False)
     vae.eval()
diff --git a/training/strategy/dreambooth.py b/training/strategy/dreambooth.py
index 695174a..42624cd 100644
--- a/training/strategy/dreambooth.py
+++ b/training/strategy/dreambooth.py
@@ -198,7 +198,7 @@ def dreambooth_prepare(
 
     text_encoder.text_model.embeddings.requires_grad_(False)
 
-    return text_encoder, unet, optimizer, train_dataloader, val_dataloader, lr_scheduler, {}
+    return text_encoder, unet, optimizer, train_dataloader, val_dataloader, lr_scheduler
 
 
 dreambooth_strategy = TrainingStrategy(
diff --git a/training/strategy/lora.py b/training/strategy/lora.py
index ae85401..73ec8f2 100644
--- a/training/strategy/lora.py
+++ b/training/strategy/lora.py
@@ -184,7 +184,7 @@ def lora_prepare(
 
     text_encoder.text_model.embeddings.token_override_embedding.params.requires_grad_(True)
 
-    return text_encoder, unet, optimizer, train_dataloader, val_dataloader, lr_scheduler, {}
+    return text_encoder, unet, optimizer, train_dataloader, val_dataloader, lr_scheduler
 
 
 lora_strategy = TrainingStrategy(
diff --git a/training/strategy/ti.py b/training/strategy/ti.py
index 9cdc1bb..363c3f9 100644
--- a/training/strategy/ti.py
+++ b/training/strategy/ti.py
@@ -207,7 +207,7 @@ def textual_inversion_prepare(
     text_encoder.text_model.embeddings.position_embedding.requires_grad_(False)
     text_encoder.text_model.embeddings.token_embedding.requires_grad_(False)
 
-    return text_encoder, unet, optimizer, train_dataloader, val_dataloader, lr_scheduler, {}
+    return text_encoder, unet, optimizer, train_dataloader, val_dataloader, lr_scheduler
 
 
 textual_inversion_strategy = TrainingStrategy(