I tried installing the conda environment coming with this project on my fairly standard machine:
Distributor ID: Ubuntu
Description: Ubuntu 22.04.1 LTS
Release: 22.04
Codename: jammy
Kernel: 5.15.0-52-generic
Nvidia Driver Version: 515.65.01
CUDA Version: 11.7
RTX 2080 Ti
However, I ran into two cascading issues:
The repository seems to assume that you have the complete CUDA dev kit installed on a system level. Although this is a fair assumption, I think it decreases the portability of the project, especially in the current world of researchers having dozens of different conda envs with different CUDA versions. I addressed this issue by adding the cuda-toolkit=11.3 package from the official nvidia conda channel
After installing the correct toolkit for this project, I found out that it was incompatible with my system-installed gcc v11.3.0 (standard ubuntu distribution). The nvcc version coming with CUDA 11.3 is only compatible with gcc/g++ <10.0. Thus I added the cxx-compiler dependency from the conda-forge channel.
(Please correct me if my understanding of these issues is wrong)
I tested this on the machine listed above as well as an Ubuntu 20.04 system. Seems to be working as expected using test_env.py and train_shac.py
To summarise, this PR only increases the portability of the project with respect with respect to its dependencies and doesn't alter the functionality.
I tried installing the conda environment coming with this project on my fairly standard machine:
However, I ran into two cascading issues:
cuda-toolkit=11.3package from the official nvidia conda channelcxx-compilerdependency from the conda-forge channel.(Please correct me if my understanding of these issues is wrong)
I tested this on the machine listed above as well as an Ubuntu 20.04 system. Seems to be working as expected using
test_env.pyandtrain_shac.pyTo summarise, this PR only increases the portability of the project with respect with respect to its dependencies and doesn't alter the functionality.