html css3 伸縮しない 画像表示効果

import torch
x = torch.randn((1,4),dtype=torch.float32,requires_grad=True)
y = x ** 2
z = y * 4
print(x)
print(y)
print(z)
loss1 = z.mean()
loss2 = z.sum()
print(loss1,loss2)
loss1.backward() # This code executes normally, but the middle variables are free after execution, so the next one has a problem
print(loss1,loss2)
loss2.backward() # This raises an error

1 # If you need to execute the backward twice, execute the first backward first, then the second backward
2 loss1.backward(retain_graph=True) # The argument here indicates that the intermediate arguments after the backward are retained.
3 loss2.backward() # After executing this, all intermediate variables will be released for the next loop
4 # If training network optimizer.step() # Update parameters

The create_graph parameter is relatively simple, refer to the official definition at

True