Installation Guide

July 2, 2024 ยท View on GitHub

Ensure you are in a Python environment with the following version:

python==3.11.0

Execute the installation script:

bash scripts/install.sh

Possible Issues

When running scripts/install.sh, you may encounter an error during the installation of Mamba's selective_scan package. This can happen due to a version mismatch between nvcc and the CUDA version displayed by nvidia-smi. Below are steps to ensure that the CUDA versions match:

Ensuring CUDA Version Consistency

Before installing the selective_scan package, it is crucial to confirm that your CUDA environment is correctly configured. Make sure that the version of nvcc (CUDA compiler) matches the CUDA driver version shown by nvidia-smi. Version inconsistencies can lead to compilation errors or runtime failures. Follow these steps to check and configure your CUDA environment:

  1. Check CUDA Driver Version: Open a terminal and run the following command to view the CUDA driver version:

    nvidia-smi

    Note the CUDA version displayed on the screen.

  2. Check NVCC Version: In the terminal, run the following command to check the version of nvcc:

    nvcc --version

    Ensure that this version matches the version shown by nvidia-smi.

  3. Configure Environment Variables: If you find a version mismatch, you may need to adjust CUDA_HOME and other related environment variables. This usually means specifying the correct path to the version of CUDA you wish to use. For example, if you want to use CUDA 11.7, set:

    export CUDA_HOME=/usr/local/cuda-11.7
export PATH=$CUDA_HOME/bin:$PATH
export LD_LIBRARY_PATH=$CUDA_HOME/lib64:$LD_LIBRARY_PATH

    Replace /usr/local/cuda-11.7 with your actual CUDA installation path.

  4. Re-validate Environment Settings: After setting the environment variables, reopen a new terminal window and re-run the above nvidia-smi and nvcc --version commands to confirm the settings are correct and the versions match.

By ensuring the consistency of CUDA driver and toolkit versions, you can significantly reduce the risk of runtime issues and installation failures, particularly for libraries that rely on specific CUDA features.