[Solved] help with detrex build needed

[DanielWit]

[Solved] help with detrex build needed

[rentainhe]

Hello, I was wondering what's your GCC version.

I can not actually tell what's the detailed error in your env, if your GCC version is too high (like 10.0+), it may not work with detectron2 I guess.

[DanielWit]

oh ok, that's most likely the issue, my gcc version is 13.1. Thank you for quick response. I will try to solve this somehow

[rentainhe]

oh ok, that's most likely the issue, my gcc version is 13.1. Thank you for quick response. I will try to solve this somehow

You can try GCC 9.4 or lower I guess.

[DanielWit]

oh ok, that's most likely the issue, my gcc version is 13.1. Thank you for quick response. I will try to solve this somehow

You can try GCC 9.4 or lower I guess.

That's what I am trying to do, although ton of issues given that I am just starting with Linux and stuff. Apparently, you can not just downgrade these files as nearly the whole system depends on them. But thanks for your help! I wish there was a docker file available for this awesome repo!

[rentainhe]

Thanks a lot for reporting this situation, we will try to update a Docker file for detrex for better usage

[DanielWit]

managed to run DINO. After all, I decided to install Ubuntu 18.04 instead of Debian 11. Given that Ubuntu is older distribution all gcc and g++ were maxed out at 9.3 therefore compatible with the Deformable Attention implementation. Together with Nvidia toolkit 11.6 and the appropriate torch version everything went without a single issue.

edit:
sudo apt install g++-9 gcc-9
for RTX 3090:
https://developer.nvidia.com/cuda-11-6-0-download-archive?target_os=Linux&target_arch=x86_64&Distribution=WSL-Ubuntu&target_version=2.0&target_type=deb_local
pip install torch==1.13.0+cu116 torchvision==0.14.0+cu116 torchaudio==0.13.0 --extra-index-url https://download.pytorch.org/whl/cu116

[rentainhe]

managed to run DINO. After all, I decided to install Ubuntu 18.04 instead of Debian 11. Given that Ubuntu is older distribution all gcc and g++ were maxed out at 9.3 therefore compatible with the Deformable Attention implementation. Together with Nvidia toolkit 11.6 and the appropriate torch version everything went without a single issue.

edit: sudo apt install g++-9 gcc-9 for RTX 3090: https://developer.nvidia.com/cuda-11-6-0-download-archive?target_os=Linux&target_arch=x86_64&Distribution=WSL-Ubuntu&target_version=2.0&target_type=deb_local pip install torch==1.13.0+cu116 torchvision==0.14.0+cu116 torchaudio==0.13.0 --extra-index-url https://download.pytorch.org/whl/cu116

Thanks for your nice advice, we will add it to docs to help others who have met the same issue.

[k-styles]

This issue still persists for newer GCC versions, please try to solve them asap.