make opt state/scheduler state/epoch non mandatory to be compatible w…

…ith current behavior
lucidrains · Jun 12, 2021 · dda155f · dda155f
1 parent d48ea80
commit dda155f
Showing 1 changed file with 6 additions and 4 deletions.
diff --git a/train_dalle.py b/train_dalle.py
@@ -178,7 +178,9 @@ def cp_path_to_dir(cp_path, tag):
         assert dalle_path.exists(), 'DALL-E model file does not exist'
     loaded_obj = torch.load(str(dalle_path), map_location='cpu')
 
-    dalle_params, vae_params, weights, opt_state, scheduler_state = loaded_obj['hparams'], loaded_obj['vae_params'], loaded_obj['weights'], loaded_obj['opt_state'], loaded_obj['scheduler_state']
+    dalle_params, vae_params, weights= loaded_obj['hparams'], loaded_obj['vae_params'], loaded_obj['weights']
+    opt_state = loaded_obj.get('opt_state')
+    scheduler_state = loaded_obj.get('scheduler_state')
 
     if vae_params is not None:
         vae = DiscreteVAE(**vae_params)
@@ -190,7 +192,7 @@ def cp_path_to_dir(cp_path, tag):
         **dalle_params
     )
     IMAGE_SIZE = vae.image_size
-    resume_epoch = loaded_obj['epoch']
+    resume_epoch = loaded_obj.get('epoch', 0)
 else:
     if exists(VAE_PATH):
         vae_path = Path(VAE_PATH)
@@ -296,7 +298,7 @@ def group_weight(model):
 # optimizer
 
 opt = Adam(get_trainable_params(dalle), lr=LEARNING_RATE)
-if RESUME:
+if RESUME and opt_state:
     opt.load_state_dict(opt_state)
 
 if LR_DECAY:
@@ -309,7 +311,7 @@ def group_weight(model):
         min_lr=1e-6,
         verbose=True,
     )
-    if RESUME:
+    if RESUME and scheduler_state:
         scheduler.load_state_dict(scheduler_state)
 else:
     scheduler = None