赞
踩
Traceback (most recent call last):
File "/data/dev/projects/fsgan/download_fsgan_models.py", line 8, in <module>
from fsgan.utils.utils import download_from_url
File "/data/dev/projects/fsgan/fsgan/utils/utils.py", line 6, in <module>
import torch
File "/home/yukang/anaconda3/envs/fsgan/lib/python3.9/site-packages/torch/__init__.py", line 192, in <module>
from torch._C import * # noqa: F403
ImportError: /home/yukang/anaconda3/envs/fsgan/lib/python3.9/site-packages/torch/lib/libtorch_cuda_cpp.so: undefined symbol: cudaGraphRetainUserObject, version libcudart.so.11.0
问题是因为 anaconda 的人写了一些错误的包装。
conda uninstall pytorch
conda uninstall libtorch
pip uninstall torch
pip uninstall torch # yes twice
然后现在使用 pytorch.org 中的命令全新安装 pytorch
不是(conda install pytorch)而是下面的指令。
conda install pytorch cuda80 -c soumith
已解决。 {\Huge \color{red}已解决。} 已解决。
参考链接:https://github.com/pytorch/pytorch/issues/2091
♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠ ⊕ ♠
Copyright © 2003-2013 www.wpsshop.cn 版权所有,并保留所有权利。