No cache after all

1 files changed, 7 insertions, 17 deletions

diff --git a/data/csv.py b/data/csv.py
index 58c833e..2f0a392 100644
--- a/data/csv.py
+++ b/data/csv.py
@@ -300,22 +300,6 @@ class VlpnDataset(IterableDataset):
         self.cache = {}
         self.length_ = (self.bucket_assignments.bincount() / self.batch_size).ceil().long().sum().item()
 
-    def get_static_example(self, item: VlpnDataItem, item_index: int):
-        if item_index in self.cache:
-            return copy.copy(self.cache[item_index])
-
-        example = {}
-
-        example["prompt_ids"] = self.prompt_processor.get_input_ids(keywords_to_prompt(item.prompt))
-        example["nprompt_ids"] = self.prompt_processor.get_input_ids(item.nprompt)
-
-        if self.num_class_images != 0:
-            example["class_prompt_ids"] = self.prompt_processor.get_input_ids(item.cprompt)
-
-        self.cache[item_index] = example
-
-        return example
-
     def __len__(self):
         return self.length_
 
@@ -374,12 +358,18 @@ class VlpnDataset(IterableDataset):
                 item = self.items[item_index]
                 mask[self.bucket_item_range[bucket_mask][0]] = False
 
-                example = self.get_static_example(item, item_index)
+                example = {}
+
+                example["prompt_ids"] = self.prompt_processor.get_input_ids(keywords_to_prompt(item.prompt))
+                example["nprompt_ids"] = self.prompt_processor.get_input_ids(item.nprompt)
+
                 example["instance_prompt_ids"] = self.prompt_processor.get_input_ids(
                     keywords_to_prompt(item.prompt, self.dropout, True)
                 )
                 example["instance_images"] = image_transforms(get_image(item.instance_image_path))
+
                 if self.num_class_images != 0:
+                    example["class_prompt_ids"] = self.prompt_processor.get_input_ids(item.cprompt)
                     example["class_images"] = image_transforms(get_image(item.class_image_path))
 
                 batch.append(example)