pytorch hook 钩子函数的用法

# 设置hook func
def hook_func(name, module):
    def hook_function(module, inputs, outputs):
        # 请依据使用场景自定义函数
        print(name+' inputs', inputs)
        print(name+' outputs', outputs)
    return hook_function

# 注册正反向hook
for name, module in model.named_modules():
    module.register_forward_hook(hook_func('[forward]: '+name, module))
    module.register_backward_hook(hook_func('[backward]: '+name, module))

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 32, 3, 1)
        self.conv2 = nn.Conv2d(32, 64, 3, 1)
        self.dropout1 = nn.Dropout(0.25)
        self.dropout2 = nn.Dropout(0.5)
        self.fc1 = nn.Linear(9216, 128)
        self.fc2 = nn.Linear(128, 10)

    def forward(self, x):
        x = self.conv1(x)
        x = F.relu(x)
        x = self.conv2(x)
        x = F.relu(x)
        x = F.max_pool2d(x, 2)
        x = self.dropout1(x)
        x = torch.flatten(x, 1)
        x = self.fc1(x)
        x = F.relu(x)
        x = self.dropout2(x)
        x = self.fc2(x)
        output = F.log_softmax(x, dim=1)
        return output

Net(
  (conv1): Conv2d(1, 32, kernel_size=(3, 3), stride=(1, 1))
  (conv2): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1))
  (dropout1): Dropout(p=0.25, inplace=False)
  (dropout2): Dropout(p=0.5, inplace=False)
  (fc1): Linear(in_features=9216, out_features=128, bias=True)
  (fc2): Linear(in_features=128, out_features=10, bias=True)
)

# 设置hook func
def hook_func(name, module):
    def hook_function(module, inputs, outputs):
        with open("log_model.txt", 'a+') as f:
            # 请依据使用场景自定义函数
            f.write(name + '   len(inputs): ' + str(len(inputs)) + '\n')
            f.write(name + '   len(outputs):  ' + str(len(outputs)) + '\n')
    return hook_function

# 注册正反向hook
for name, module in model.named_modules():
    module.register_forward_hook(hook_func('[forward]: '+name, module))
    module.register_backward_hook(hook_func('[backward]: '+name, module))