使用PyTorch时CUDA内存不足

Question

试图使用Google Colab从此存储库中复制超级分辨率GAN-Super Resolution，但每次执行最后一块代码时，都会发生以下错误：

RuntimeError                              Traceback (most recent call last)
<ipython-input-23-b9349075c05d> in <module>()
     16 
     17         gen_out = gen(torch.from_numpy(lr_images).to(cuda).float())
---> 18         _,f_label = disc(gen_out)
     19         _,r_label = disc(torch.from_numpy(hr_images).to(cuda).float())
     20         d1_loss = (disc_loss(f_label,torch.zeros_like(f_label,dtype=torch.float)))

4 frames
/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py in batch_norm(input, running_mean, running_var, weight, bias, training, momentum, eps)
   2280 
   2281     return torch.batch_norm(
-> 2282         input, weight, bias, running_mean, running_var, training, momentum, eps, torch.backends.cudnn.enabled
   2283     )
   2284 

RuntimeError: CUDA out of memory. Tried to allocate 256.00 MiB (GPU 0; 11.17 GiB total capacity; 10.29 GiB already allocated; 63.81 MiB free; 10.65 GiB reserved in total by PyTorch)

我已经尝试过减少批量大小，但没有效果。如何解决这个问题

Answer 1

将批量大小从64更改为1最终解决了问题。荣誉归于@Farhood ET