gistfile1.txt

import torch

x = torch.randn(3, 3, requires_grad=True)
print(x)

min = float('NaN')
max = 0.0
y = torch.clamp(x, min, max)
print('y', y)
y.sum().backward()
print('x.grad', x.grad)

y = torch.clamp(x, max=max)
y.sum().backward()
print('y', y)
print('x.grad', x.grad)

=====

tensor([[ 1.8375,  1.0008, -0.6007],
        [-0.3602,  0.4946, -1.0250],
        [-0.6579,  0.8329,  0.5419]], requires_grad=True)
y tensor([[ 0.0000,  0.0000, -0.6007],
        [-0.3602,  0.0000, -1.0250],
        [-0.6579,  0.0000,  0.0000]], grad_fn=<ClampBackward>)
x.grad tensor([[0., 0., 0.],
        [0., 0., 0.],
        [0., 0., 0.]])
y tensor([[ 0.0000,  0.0000, -0.6007],
        [-0.3602,  0.0000, -1.0250],
        [-0.6579,  0.0000,  0.0000]], grad_fn=<ClampMaxBackward>)
x.grad tensor([[0., 0., 1.],
        [1., 0., 1.],
        [1., 0., 0.]])