with torch.autograd.set_detect_anomaly(True)

with torch.autograd.set_detect_anomaly(True)

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [16, 384, 4, 4]], which is output 0 of HardtanhBackward1, is at version 2; expected version 1 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

感谢网友在评论里补充了答案：

出现这种问题一般是tensor进行操作的时候没有进行clone()造成的。比如你带有梯度信息的tensor进了sin变换，你应该写img = torch.sin(img.clone())。如果直接写img = torch.sin(img)就会报错

x[i+1] = torch.sin(x[i]) * w[i]

改成下面这样就行了：

x[i+1] = torch.sin(x[i].clone()) * w[i]