Thank you for your research.
I am working on a Ubuntu 24.04 machine, which originally supports CUDA 12.8 and gcc-13. The python version in my environment is 3.10.9. For the purpose of installing this software, I installed CUDA 11.3, gcc-9 and g++-9, appropriately setting the CXX and CUDA_HOME variables before installation, and using the linux-5.4-headers in the setup of MinkowskiEngine.
I used a slightly different command for installing torch-scatter since the version given by the link was incorrect: pip3 install torch-scatter==2.1.0 -f https://data.pyg.org/whl/torch-1.12.1+cu113.html
I also explicitly installed numpy==1.26.4 to avoid errors in the installation of mmcv==2.0.0 and other packages.
After following the installation instructions up until Pointnet2, installing pytorch-lightning==1.7.2 makes pip remove the existing installation of torch==1.12.1 and replace it with torch==2.7.1. YOLO-World also explicitly requires torch>=2.1.0.
Finishing all the steps, I tried to run the single_scene_inference.py file without arguments but torch is explicitly deactivated in the beginning of the program:
Disabling PyTorch because PyTorch >= 2.1 is required but found 1.12.1+cu113
None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
As such, instantiating the YOLOWorld class (more precisely, initializing network_2d) leads to this error:
CLIPTextModelWithProjection requires the PyTorch library but it was not found in your environment.
Check out the instructions on the installation page: https://pytorch.org/get-started/locally/ and follow the ones that match your environment.
Please note that you may need to restart your runtime after installation.
How can I correct this error? Do you use a different python version? Is there any detail of the setup that I might have missed?
I appreciate your time and attention.
Thank you for your research.
I am working on a Ubuntu 24.04 machine, which originally supports CUDA 12.8 and gcc-13. The python version in my environment is 3.10.9. For the purpose of installing this software, I installed CUDA 11.3, gcc-9 and g++-9, appropriately setting the
CXXandCUDA_HOMEvariables before installation, and using thelinux-5.4-headersin the setup of MinkowskiEngine.I used a slightly different command for installing
torch-scattersince the version given by the link was incorrect:pip3 install torch-scatter==2.1.0 -f https://data.pyg.org/whl/torch-1.12.1+cu113.htmlI also explicitly installed
numpy==1.26.4to avoid errors in the installation ofmmcv==2.0.0and other packages.After following the installation instructions up until Pointnet2, installing
pytorch-lightning==1.7.2makes pip remove the existing installation oftorch==1.12.1and replace it withtorch==2.7.1. YOLO-World also explicitly requirestorch>=2.1.0.Finishing all the steps, I tried to run the
single_scene_inference.pyfile without arguments buttorchis explicitly deactivated in the beginning of the program:As such, instantiating the YOLOWorld class (more precisely, initializing network_2d) leads to this error:
How can I correct this error? Do you use a different python version? Is there any detail of the setup that I might have missed?
I appreciate your time and attention.