Pytorch one of the variables needed for gradient computation has been modified by an inplace

报错如下所示：

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: 
[torch.FloatTensor [4, 1, 1, 155]], which is output 0 of UnsqueezeBackward0, is at version 1066; expected version 1065 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

造成这个原因主要是代码中有以下几种情况：

1. a = a
2. a += b
3. 循环中 a[i,:,:]=...a[i,:,:]

我报这个错就是因为在代码中写了：

attention_sum += attention

贴一个总结比较好的知乎
关于 pytorch inplace operation, 需要知道的几件事