empty cuda cache

flux/sampling_lore.py

@@ -348,7 +348,9 @@ def denoise_with_noise_optim(
                 trainable_noise.grad *= source_mask[0]
             optimizer.step()
             print(f"Time {t_curr:.4f} Step {j+1}/{training_epochs}, Loss: {total_loss.item():.6f}")
-            
+            torch.cuda.empty_cache()
+            torch.cuda.synchronize()
+       
         step_list.append(t_curr)
         trainable_noise = trainable_noise.detach()
         trainable_noise_list.append(trainable_noise.clone())