We will leverage the NVIDIA specific conda setup provided by ALCF.
module load conda; conda activate
conda create --prefix=/path/MISTRAL_CL_CONDA --clone base
To activate this environment, use
$ conda activate /path/MISTRAL_CL_CONDA
To deactivate an active environment, use
$ conda deactivate
Update and install additional packages.
conda install conda-forge::rdkit
pip install equinox ruff sentencepiece openpyxl quax
All the production runs were performed on NVIDIA GH200. We used a Miniconda for 'Linux-aarch64'. Next steps are easy.
conda env create -f env/env_gh200.yml
To activate this environment, use
$ conda activate py_gh200
To deactivate an active environment, use
$ conda deactivate