diff --git a/basicsr/data/ffhq_blind_dataset.py b/basicsr/data/ffhq_blind_dataset.py
index 9f900606..f0564375 100755
--- a/basicsr/data/ffhq_blind_dataset.py
+++ b/basicsr/data/ffhq_blind_dataset.py
@@ -47,10 +47,11 @@ def __init__(self, opt):
             self.crop_components = False
 
         if self.latent_gt_path is not None:
-            self.load_latent_gt = True            
-            self.latent_gt_dict = torch.load(self.latent_gt_path)
+            self.load_latent_gt = True
+            self.latent_gt_dict = None  # Lazy load to avoid memory issues with multi-process DataLoader
         else:
-            self.load_latent_gt = False  
+            self.load_latent_gt = False
+            self.latent_gt_dict = None  
 
         if self.io_backend_opt['type'] == 'lmdb':
             self.io_backend_opt['db_paths'] = self.gt_folder
@@ -190,6 +191,9 @@ def __getitem__(self, index):
         img_gt, status = augment(img_gt, hflip=self.opt['use_hflip'], rotation=False, return_status=True)
 
         if self.load_latent_gt:
+            # Lazy load latent_gt_dict to avoid memory issues with multi-process DataLoader
+            if self.latent_gt_dict is None:
+                self.latent_gt_dict = torch.load(self.latent_gt_path)
             if status[0]:
                 latent_gt = self.latent_gt_dict['hflip'][name]
             else: