use hash to check valid optim

Author: aria1th

modules/hypernetworks/hypernetwork.py

@@ -177,12 +177,13 @@ def save(self, filename):
         state_dict['sd_checkpoint_name'] = self.sd_checkpoint_name
         if self.optimizer_name is not None:
             optimizer_saved_dict['optimizer_name'] = self.optimizer_name
+
+        torch.save(state_dict, filename)
         if self.optimizer_state_dict:
+            optimizer_saved_dict['hash'] = sd_models.model_hash(filename)
             optimizer_saved_dict['optimizer_state_dict'] = self.optimizer_state_dict
             torch.save(optimizer_saved_dict, filename + '.optim')
 
-        torch.save(state_dict, filename)
-
     def load(self, filename):
         self.filename = filename
         if self.name is None:
@@ -204,7 +205,10 @@ def load(self, filename):
         optimizer_saved_dict = torch.load(self.filename + '.optim', map_location = 'cpu') if os.path.exists(self.filename + '.optim') else {}
         self.optimizer_name = optimizer_saved_dict.get('optimizer_name', 'AdamW')
         print(f"Optimizer name is {self.optimizer_name}")
-        self.optimizer_state_dict = optimizer_saved_dict.get('optimizer_state_dict', None)
+        if sd_models.model_hash(filename) == optimizer_saved_dict.get('hash', None):
+            self.optimizer_state_dict = optimizer_saved_dict.get('optimizer_state_dict', None)
+        else:
+            self.optimizer_state_dict = None
         if self.optimizer_state_dict:
             print("Loaded existing optimizer from checkpoint")
         else:
@@ -229,7 +233,7 @@ def list_hypernetworks(path):
         name = os.path.splitext(os.path.basename(filename))[0]
         # Prevent a hypothetical "None.pt" from being listed.
         if name != "None":
-            res[name] = filename
+            res[name + f"({sd_models.model_hash(filename)})"] = filename
     return res
 
 
@@ -375,6 +379,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
     else:
         hypernetwork_dir = None
 
+    hypernetwork_name = hypernetwork_name.rsplit('(', 1)[0]
     if create_image_every > 0:
         images_dir = os.path.join(log_directory, "images")
         os.makedirs(images_dir, exist_ok=True)