diff --git a/finetune/make_captions_by_git.py b/finetune/make_captions_by_git.py
index ce6e66955..b3c5cc423 100644
--- a/finetune/make_captions_by_git.py
+++ b/finetune/make_captions_by_git.py
@@ -52,6 +52,9 @@ def collate_fn_remove_corrupted(batch):
 
 
 def main(args):
+    r"""
+    transformers 4.30.2で、バッチサイズ>1でも動くようになったので、以下コメントアウト
+
     # GITにバッチサイズが1より大きくても動くようにパッチを当てる: transformers 4.26.0用
     org_prepare_input_ids_for_generation = GenerationMixin._prepare_input_ids_for_generation
     curr_batch_size = [args.batch_size]  # ループの最後で件数がbatch_size未満になるので入れ替えられるように
@@ -65,6 +68,7 @@ def _prepare_input_ids_for_generation_patch(self, bos_token_id, encoder_outputs)
         return input_ids
 
     GenerationMixin._prepare_input_ids_for_generation = _prepare_input_ids_for_generation_patch
+    """
 
     print(f"load images from {args.train_data_dir}")
     train_data_dir_path = Path(args.train_data_dir)
@@ -81,7 +85,7 @@ def _prepare_input_ids_for_generation_patch(self, bos_token_id, encoder_outputs)
     def run_batch(path_imgs):
         imgs = [im for _, im in path_imgs]
 
-        curr_batch_size[0] = len(path_imgs)
+        # curr_batch_size[0] = len(path_imgs)
         inputs = git_processor(images=imgs, return_tensors="pt").to(DEVICE)  # 画像はpil形式
         generated_ids = git_model.generate(pixel_values=inputs.pixel_values, max_length=args.max_length)
         captions = git_processor.batch_decode(generated_ids, skip_special_tokens=True)