Pytorch register_ Hook (operate on gradient grad)

stay CAM Meet in  register_hook

import torch
 
v = torch.ones((1, 3), dtype=torch.float32, requires_grad=True)
y = v**2
z = y.sum()
y.register_hook(lambda grad: print(grad))
z.backward()
print(v.grad)

register_hook It's for the right tensor Of grad operational