PyTorch

PyTorch查看CUDA和CUDNN版本

>>> import torch
>>> print(torch)
<module 'torch' from '/home/torch/code_pytorch/pytorch.master/torch/__init__.py'>
>>> torch.__version__
'2.0.0a0+gitb8151d2'
>>> 
>>> torch.cuda.is_available()
True
>>> torch.version.cuda
'11.7'
>>> 
>>> torch.backends.cudnn.is_available()
False
>>> torch.backends.cudnn.version()
>>> 
>>>

查看PyTorch版本的对应的git commit id

python -c 'import torch; print(torch.version.git_version)'

求 Minimum 的梯度

# $ cat test_minimum_grad.py 
#!/usr/bin/env python

import torch

x = torch.tensor([1.0, 5.0, 3.0], dtype=torch.float, requires_grad=True)
y = torch.tensor([4.0, 2.0, 6.0], dtype=torch.float, requires_grad=True)
z = torch.minimum(x, y)
print(z.detach().numpy())

d = torch.tensor([1.0, 2.0, 3.0], dtype=torch.float, requires_grad=True)
z.backward(gradient=d)
print('grad of x is ', x.grad)

# $ ./test_minimum_grad.py 
# [1. 2. 3.]
# grad of x is  tensor([1., 0., 3.])

参考:

posted @ 2023-02-27 18:39  憶藝  阅读(49)  评论(0)    收藏  举报