文章目录
一. requires_grad 属性：查看是否记录梯度
x = torch.rand(3, 3) # 直接创建的tensor变量默认是没有梯度的x.requires_grad结果如下：
False二. requires_grad_ ()函数：调用函数设置记录梯度与否
函数：requires_grad_(requires_grad=True)
x = torch.tensor([1., 2., 3.])print('x_base: ', x.requires_grad) # 正常状态：Falsex.requires_grad_(False) #Falseprint('x.requires_grad_(False): ', x.requires_grad) #Falsex.requires_grad_(True) # 等价：requires_grad_() #Trueprint('x.requires_grad_(True): ', x.requires_grad) #True结果如下：
x_base: Falsex.requires_grad_(False): Falsex.requires_grad_(True): True三. requires_grad属性参数，创建tensor时设置是否记录梯度
x = torch.tensor([1., 2., 3.], requires_grad=True) #Trueprint('x_base: ', x.requires_grad) # 正常状态结果如下：
x_base: True四. 查看模型的权重名称和参数值
[name for name, param in model.named_parameters()] #查看模型的权重名称[param for name, param in model.named_parameters()] #查看模型的权重值五. 查看模型权重梯度值
网络结构如下：
class Simple(nn.Module): def __init__(self): super().__init__() self.conv1 = nn.Conv2d(3, 16, 3, 1, padding=1, bias=False) self.conv2 = nn.Conv2d(16, 32, 3, 1, padding=1, bias=False) self.linear = nn.Linear(32*10*10, 20, bias=False) def forward(self, x): x = self.conv1(x) x = self.conv2(x) x = self.linear(x.view(x.size(0), -1)) return x model = Simple()查看模型权重梯度值：
print(model.conv1.weight.grad)来源地址：https://blog.csdn.net/flyingluohaipeng/article/details/129192987
