为什么这个函数参数在每次调用中都是相同的,尽管传递的值不同(在循环中创建闭包)

我正在使用PyTorch并尝试注册模型参数上的钩子。下面的代码创建lambda函数来添加到每个模型参数中,因此我可以在hook中看到梯度属于哪个张量

import torch
import torchvision

# define model and random train batch
model = torchvision.models.alexnet()
input = torch.rand(10, 3, 224, 224)   # batch of 10 images
targets = torch.zeros(10).long()

def grad_hook_template(param, name, grad):
    print(f'Receive grad for {name} w whape {grad.shape}')

# add one lambda hook to each parameter
for name, param in model.named_parameters():
    print(f'Register hook for {name}')

    # use a lambda so we can pass additional information to the hook, which should only take one parameter
    param.register_hook(lambda grad: grad_hook_template(param, name, grad))

loss_fn = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.parameters(), lr=0.1)
optimizer.zero_grad()

prediction = model(input)
loss = loss_fn(prediction, targets)
loss.backward()
optimizer.step()

结果是 name 和 param 论据 grad_hook_template id ),但 grad

我读书。 here 循环不创建新的作用域和闭包在Python中是词法性的,即 名称 和 param copy.copy() 变量?