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

根据项目要求，需要使用两个优化器对模型进行优化

运行中一直出现：
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [100, 5000]], which is output 0 of TBackward, is at version 3; expected version 2 instead. Hint: enable anomaly detection

经过网上查询发现，这个问题可能是由于出现了x+=3类似的代码，改成x = x + 1,更改过后发现仍旧不行，后来发现，在使用两个优化器进行优化是需要是如下格式：

	 optimizerG.zero_grad() 
	 optimizerD.zero_grad() 
	 loss1.backward(retain_graph=True)  
	 loss2.backward()  
	 optimizerD.step()  
	 optimizerG.step()  

必须将参数更新放在最后