用pytorch检测cuda环境
#判断是否安装了cuda
import torch
print(torch.cuda.is_available()) #返回True则说明已经安装了cuda
#判断是否安装了cuDNN
from torch.backends import cudnn
print(cudnn.is_available()) #返回True则说明已经安装了cuDNN
CUDA capability sm_86 is not compatible with the current PyTorch installation
警告描述:
NVIDIA GeForce RTX 3050 Ti Laptop GPU with CUDA capability sm_86 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_61 sm_70 sm_75 compute_37.
sm那一串可以用如下代码查看
>>> import torch
>>> torch.__version__
'1.10.0+cu113'
>>> torch.cuda.get_arch_list()
['sm_37', 'sm_50', 'sm_60', 'sm_70', 'sm_75', 'sm_80', 'sm_86']
出现这个问题,就是cuda版本低了。我当时是cuda10.2,于是往上升为cuda11.3(就像上面那种组合就没问题)
(之后pytorch和mmcv-full可能也要对应安装)
多个libiomp5md.dll
问题描述:
OMP: Error #15: Initializing libiomp5md.dll, but found libiomp5md.dll already initialized.
问题原因:
为了装对应pytorch,没删干净,装重复了东西。(正常安装应该不会有问题)
解决办法是删了重装或者删掉多余的dll
详见:https://zhuanlan.zhihu.com/p/371649016
(我是卸载了pytorch再重装的)
标签:总结,遇到,torch,环境,dll,current,pytorch,cuda,sm From: https://www.cnblogs.com/faf4r/p/17031710.html