pytorch retain_grad vs requires_grad
requires_grad大家都挺熟悉的,因此穿插在retain_grad的例子里进行捎带讲解就行。下面看一个代码片段:
import torch
# 创建一个标量 tensor,并开启梯度计算
x = torch.tensor(2.0, requires_grad=True)
# 中间计算:y 依赖于 x,是非叶子节点
y = x * 3
# 继续计算,得到 z
z = y * 4
# 反向传播
z.backward()
# 查看梯度
print("x.grad:", x.grad)
print("y.grad:", y.grad)
输出结果为:
x.grad: tensor(12.)
y.grad: None
/tmp/ipykernel_219007/1060175670.py:17: UserWarning: The .grad attribute of a Tensor that is not a leaf Tensor is being accessed. Its .grad attribute won't be populated during autograd.backward(). If you indeed want the .grad field to be populated for a non-leaf Tensor, use .retain_grad() on the non-leaf Tensor. If you access the non-leaf Tensor by mistake, make sure you access the leaf Tensor instead. See github.com/pytorch/pytorch/pull/30531 for more informations. (Triggered internally at aten/src/ATen/core/TensorBody.h:489.)
print("y.grad:", y.grad)
警告的大致意思是:访问了非叶子节点的.grad
属性,但非叶子节点的.grad
属性并不会在反向传播的过程中被自动保存下来(这是为了节省内存,毕竟我们只需要计算那些手动设置.requires_grad
为True
的张量的梯度,并进行梯度更新,对吧?)
因此,我们只需要添加一行代码y.retain_grad()
,修改后的代码如下:
import torch
# 创建一个标量 tensor,并开启梯度计算
x = torch.tensor(2.0, requires_grad=True)
# 中间计算:y 依赖于 x,是非叶子节点
y = x * 3
y.retain_grad()
# 继续计算,得到 z
z = y * 4
# 反向传播
z.backward()
# 查看梯度
print("x.grad:", x.grad)
print("y.grad:", y.grad)
输出结果为:
x.grad: tensor(12.)
y.grad: tensor(4.)
可以看到,现在非叶子节点y
的梯度也在反向传播以后被正确保存了!