Less eval steps during training

Author: anton-l

examples/train_unconditional.py

@@ -147,9 +147,9 @@ def transforms(examples):
 
         accelerator.wait_for_everyone()
 
-        # Generate a sample image for visual inspection
+        # Generate sample images for visual inspection
         if accelerator.is_main_process:
-            with torch.no_grad():
+            if epoch % args.save_images_epochs == 0 or epoch == args.num_epochs - 1:
                 pipeline = DDPMPipeline(
                     unet=accelerator.unwrap_model(ema_model.averaged_model if args.use_ema else model),
                     scheduler=noise_scheduler,
@@ -159,9 +159,11 @@ def transforms(examples):
                 # run pipeline in inference (sample random noise and denoise)
                 images = pipeline(generator=generator, batch_size=args.eval_batch_size, output_type="numpy")["sample"]
 
-            # denormalize the images and save to tensorboard
-            images_processed = (images * 255).round().astype("uint8")
-            accelerator.trackers[0].writer.add_images("test_samples", images_processed.transpose(0, 3, 1, 2), epoch)
+                # denormalize the images and save to tensorboard
+                images_processed = (images * 255).round().astype("uint8")
+                accelerator.trackers[0].writer.add_images(
+                    "test_samples", images_processed.transpose(0, 3, 1, 2), epoch
+                )
 
             if epoch % args.save_model_epochs == 0 or epoch == args.num_epochs - 1:
                 # save the model
@@ -184,7 +186,8 @@ def transforms(examples):
     parser.add_argument("--train_batch_size", type=int, default=16)
     parser.add_argument("--eval_batch_size", type=int, default=16)
     parser.add_argument("--num_epochs", type=int, default=100)
-    parser.add_argument("--save_model_epochs", type=int, default=5)
+    parser.add_argument("--save_images_epochs", type=int, default=10)
+    parser.add_argument("--save_model_epochs", type=int, default=10)
     parser.add_argument("--gradient_accumulation_steps", type=int, default=1)
     parser.add_argument("--learning_rate", type=float, default=1e-4)
     parser.add_argument("--lr_scheduler", type=str, default="cosine")