simplify more

dalle2_pytorch/train.py

@@ -265,9 +265,8 @@ class DiffusionPriorTrainer(nn.Module):
             with autocast(enabled = self.amp):
                 loss = self.diffusion_prior(*chunked_args, **chunked_kwargs)
 
-            loss = loss * chunk_size_frac
+                total_loss += loss.item() * chunk_size_frac
-            total_loss += loss.item()
+                self.scaler.scale(loss * chunk_size_frac).backward()
-            self.scaler.scale(loss).backward()
 
         return total_loss
 
@@ -389,8 +388,8 @@ class DecoderTrainer(nn.Module):
             with autocast(enabled = self.amp):
                 loss = self.decoder(*chunked_args, unet_number = unet_number, **chunked_kwargs)
 
-            loss = loss * chunk_size_frac
+                total_loss += loss.item() * chunk_size_frac
-            total_loss += loss.item()
+
-            self.scale(loss, unet_number = unet_number).backward()
+                self.scale(loss * chunk_size_frac, unet_number = unet_number).backward()
 
         return total_loss

setup.py

@@ -10,7 +10,7 @@ setup(
       'dream = dalle2_pytorch.cli:dream'
     ],
   },
-  version = '0.2.28',
+  version = '0.2.27',
   license='MIT',
   description = 'DALL-E 2',
   author = 'Phil Wang',