diff --git a/toolkit/data_loader.py b/toolkit/data_loader.py
index 830c8e80..1be02c67 100644
--- a/toolkit/data_loader.py
+++ b/toolkit/data_loader.py
@@ -1,6 +1,7 @@
 import json
 import os
 import random
+from functools import lru_cache
 from typing import List, TYPE_CHECKING
 
 import cv2
@@ -417,6 +418,7 @@ class AiToolkitDataset(LatentCachingMixin, BucketsMixin, CaptionMixin, Dataset):
         file_item.load_caption(self.caption_dict)
         return file_item
 
+    @lru_cache(maxsize=300)
     def __getitem__(self, item):
         if self.dataset_config.buckets:
             # for buckets we collate ourselves for now