Polaris Installation

1. Load Polaris modules.

   module load cray-python/3.9.12.1
   module load cudatoolkit-standalone/11.7.1
   

2. Clone and create your virtual environment.

   git clone git@github.com:gpauloski/llm-pytorch
   # git clone https://github.com/gpauloski/llm-pytorch
   cd llm-pytorch
   

3. Create your virtual environment.

   python -m venv venv
   . venv/bin/activate
   

   Note: anytime you want to use the virtual environment you will need to load the above module and activate the virtual environment. It may be helpful to add these to your ~/.bashrc or PBS job scripts.
4. Install PyTorch.

   pip install torch==1.13.1
   

5. Install the llm package.

   pip install -e .  # Use the [dev] extras install for developing
   

6. Install ColossalAI. This can be done directly from pip (pip install colossalai) which will JIT compile the CUDA extensions, or you can install with CUDA extensions pre-build. This must be done from a compute node. E.g., qsub -A [ALLOCATION] -l select=1:system=polaris -l walltime=1:00:00 -I -q debug -l filesystems=home:grand.

   # Activate your modules and virtual environment
   module load cray-python/3.9.12.1 cudatoolkit-standalone/11.7.1
   . /path/to/venv/bin/activate
   
   module load gcc
   
   CUDA_EXT=1 pip install colossalai
   

7. Done!