[Linux] conv2d 시행 시 Runtime error: GET was unable to find an engine to execute 에러 해결

Video editing model을 돌리다가 아래와 같은 에러가 conv2d에서 발생함을 확인했다 ㅋㅋ 

  File "/mnt/petrelfs/zhaozhiyuan/anaconda3/envs/minigpt4-nightly/lib/python3.9/site-packages/torch/nn/modules/conv.py", line 459, in _conv_forward
    return F.conv2d(input, weight, bias, self.stride,
RuntimeError: GET was unable to find an engine to execute this computation

여러가지 서치를 해본결과 뭐 cuda version이 안맞아서 그런거라는데.. 됐고 

그냥 아래와 같이 추가해주면 된다. 11.8 부분에는 자신의 cuda version을 넣어주면 된다. 

$ export PATH=/usr/local/cuda-11.8/bin${PATH:+:${PATH}}
$ export LD_LIBRARY_PATH=/usr/local/cuda-11.8/lib64

 

출처는 이 곳이다! 

https://github.com/pytorch/pytorch/issues/102535#issuecomment-1666535370

https://forums.developer.nvidia.com/t/path-ld-library-path/48080

PATH & LD_LIBRARY_PATH