Personal code repo for bioinformatics analysis

On Linux systems, install additional CUDA dependencies:

# Install cusparselt (for Linux) # Download from: https://developer.nvidia.com/cusparselt-downloads # Install cuDNN # Download from https://developer.nvidia.com/cudnn-downloads # Install nccl # Download from https://developer.nvidia.com/nccl/nccl-download # on hpc, if cant install those manually, use conda to install them and set env variable to the conda lib folder conda install nvidia::libcusparse conda-force::nccl nvidia::cuda-toolkit # then export LD_LIBRARY_PATH=$CONDA_PREFIX/lib:$LD_LIBRARY_PATH

For sksparse, it depends on openblas, so need to make sure both libblas.so.3 and liblapack.so.3 point to the right version of openblas (if system openblas failed, try build from source and configure alternative) Also, for sksparse, suite-sparse is needed.

For juila, use 1.12.0-beta1+0.x64.linux.gnu if 1.11 failed due to curl issue.

For rpy2, it needs libtirpc-dev

If libcublas cant be found, check /usr/local/cuda-13/lib64/ If torch still complain about libcublas, try update torch cuda index