Pytorch 2.0.1+cu118 不兼容 Cuda 12.1。

huangapple go评论79阅读模式
英文:

Pytorch 2.0.1+cu118 is not working with Cuda 12.1

问题

我在我的计算机上安装了CUDA 12.1,以及PyTorch 2.0.1+cu118。

我正尝试安装CLIP-FIELDS。当我运行gridencoder的"python setup.py install"时,出现以下错误:

检测到的CUDA版本(12.1)与用于编译PyTorch的版本(11.8)不匹配。请确保使用相同的CUDA版本。

我在官方PyTorch论坛上看到了关于此问题的讨论:https://discuss.pytorch.org/t/install-pytorch-with-cuda-12-1/174294,其中提到cuda 12.1与Torch 2.0.1+cu118兼容。

有人知道导致崩溃的原因吗?

英文:

I have CUDA 12.1 installed on my computer, and Pytorch 2.0.1+cu118 installed.

I am trying to the installation of CLIP-FIELDS. When I install gridencoder "python setup.py install", it appears the following error:

The detected CUDA version (12.1) mismatches the version that was used to compile
PyTorch (11.8). Please make sure to use the same CUDA versions.

I saw on official pytorch forums <https://discuss.pytorch.org/t/install-pytorch-with-cuda-12-1/174294> that cuda 12.1 is compatible with Torch 2.0.1+cu118.

Does anyone know the reason of crashing then?

答案1

得分: 2

匹配CUDA和Pytorch的版本通常很麻烦。
我看到2个解决方案:

  • 本地安装CUDA 11.8。这可能会很麻烦,而且可能会破坏其他Python安装,最坏的情况下还可能破坏计算机的图形可视化。
  • 创建一个具有正确版本的Pytorch和CUDA的Docker容器。第一次学习可能会有点麻烦,但之后可以用于任何后续的Git存储库安装。并且没有危害主要计算机的主要CUDA的风险。
英文:

To match the version of CUDA and Pytorch is usually a pain in the mass.
I see 2 solutions :

  • Install CUDA 11.8 natively. This can be painful and break other python installs, and in the worst case also the graphical visualization in the computer
  • Create a Docker Container with the proper version of pytorch and CUDA. This can be painful to learn the first time, but then is a skill that can be used for any later install of a git repository. And there is no risk to compromise the main CUDA of the computer.

huangapple
  • 本文由 发表于 2023年6月1日 18:50:26
  • 转载请务必保留本文链接:https://go.coder-hub.com/76381130.html
匿名

发表评论

匿名网友

:?: :razz: :sad: :evil: :!: :smile: :oops: :grin: :eek: :shock: :???: :cool: :lol: :mad: :twisted: :roll: :wink: :idea: :arrow: :neutral: :cry: :mrgreen:

确定