Open varoudis opened 1 year ago
delete from the requirements.txt torch run: pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu116 then install requirements.txt and the rest
Installing g++ resolved this issue for me
thank you both, Ill check it
thank you both, Ill check it
Hi , Have you solve it , I have the same issue like yours.
On a fresh build of ubuntu 22.04 I am seeing this same issue. Everything looks good till installing the "bash scripts/ install" step. Turns out this was a mismatch between torch and nvcc because I was using the ubuntu repo install instead of the installer from nvidia. After switching to CUDA 11.7 it worked fine.
Same issue here. Tried rm -r venv and recreating it with torch removed from requirements.txt and installed with one of these instead: pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu116 pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu117 But neither worked, same error :(
Okay so I had
nvcc --version -> 11.5
but
import torch; torch.__version__ -> 11.7
So I had to manually install https://developer.nvidia.com/cuda-11-7-1-download-archive?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=22.04&target_type=runfile_local
And now I have Successfully built raymarching
Alright, it s been a while, but for those in the future. If you see problems like "[Ninja, -v]" or "nvcc error" (something like this), Just use gcc-10 version since CUDA11.6 is incompatiable when (gcc -v > 10)
So, just use gcc-10, you will go to good
apt install gcc-10 g++-10
rm /usr/bin/gcc
you can totaly delete /usr/bin/gcc since it just some gcc's link (wrong version tho
ln -s /usr/bin/gcc-10 /usr/bin/gcc
ln -s /usr/bin/g++-10 /usr/bin/g++
You may need sudo for above
Hope this helps
Anyone had a similar fail to build? Clean Ubuntu 2204 with nvidia-470 drivers and base 22-4 cuda toolkit.