diff --git a/library/model_util.py b/library/model_util.py
index e227ced8..f3f236af 100644
--- a/library/model_util.py
+++ b/library/model_util.py
@@ -831,7 +831,7 @@ def is_safetensors(path):
   return os.path.splitext(path)[1].lower() == '.safetensors'
 
 
-def load_checkpoint_with_text_encoder_conversion(ckpt_path, device):
+def load_checkpoint_with_text_encoder_conversion(ckpt_path, device="cpu"):
   # text encoderの格納形式が違うモデルに対応する ('text_model'がない)
   TEXT_ENCODER_KEY_REPLACEMENTS = [
       ('cond_stage_model.transformer.embeddings.', 'cond_stage_model.transformer.text_model.embeddings.'),
@@ -866,7 +866,7 @@ def load_checkpoint_with_text_encoder_conversion(ckpt_path, device):
 
 # TODO dtype指定の動作が怪しいので確認する text_encoderを指定形式で作れるか未確認
 def load_models_from_stable_diffusion_checkpoint(v2, ckpt_path, device='cpu', dtype=None):
-  _, state_dict = load_checkpoint_with_text_encoder_conversion(ckpt_path, device)
+  _, state_dict = load_checkpoint_with_text_encoder_conversion(ckpt_path) # no need to specify device in loading state_dict
 
   # Convert the UNet2DConditionModel model.
   unet_config = create_unet_diffusers_config(v2)