I encountered a problem when running the model

GeForce RTX 3090 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_ 70 .

Check out the current cuda version:

cat /usr/local/cuda/version.txt
CUDA Version 11 . 0 . 182

Solution:

1. Change python to python3.8 environment (At the time, an answer to the online information said, I directly changed it to 3.8 at the time, and did not verify whether 3.7 and 3.6 are feasible)
2. Install nightly’s pytorch, corresponding to the cuda version of 11.0conda install pytorch torchvision cudatoolkit=11 -c pytorch-nightly

Successfully resolved

Similar Posts:

• Pytorch Install Error: OSError: [WinError 126] Could not Find the Module, Error loading “xx\lib\site-packages\torch\lib\asmjit.dl
• [Solved] pytorchImportError: numpy.core.multiarray failed to import
• Rust encountered error[E0554]: `#![feature]` may not be used on the stable release channel (switch nightly version)
• [Solved] DCNv2 Compile Error: error: identifier “THCudaBlas_SgemmBatched” is undefined
• Runtimeerror: CUDA error: out of memory [How to Solve]
• [Solved] PIP Install torch1.7 Error: ERROR: torch has an invalid wheel, .dist-info directory not found
• How to Solve Anaconda Error (Three Errors to Solve)
• Examples of torch.NN.Functional.Relu() and torch.NN.Relu()
• [Solved] Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR
• one of the variables needed for gradient computation has been modified by an inplace operation