fix custom diffusion training with concept list (huggingface#6710)

Co-authored-by: Sayak Paul <[email protected]>

Author: 2 people

examples/custom_diffusion/train_custom_diffusion.py

@@ -753,7 +753,7 @@ def main(args):
                     num_new_images = args.num_class_images - cur_class_images
                     logger.info(f"Number of class images to sample: {num_new_images}.")
 
-                    sample_dataset = PromptDataset(args.class_prompt, num_new_images)
+                    sample_dataset = PromptDataset(concept["class_prompt"], num_new_images)
                     sample_dataloader = torch.utils.data.DataLoader(sample_dataset, batch_size=args.sample_batch_size)
 
                     sample_dataloader = accelerator.prepare(sample_dataloader)