First Time Steps

1. Log into Ondemand on Ubelix.
2. Create a new session with
   1. GPU-Type - RTX 4090
   2. Time - 2 hours
   3. Instance - small or medium.
   4. Cuda - 12.8
   5. Mode - Jupyter Lab.
   6. Account : teaching
   7. wckey: Leave it blank
   8. Partition: teaching (GPU)
   9. Select the Advanced mode checkbox
      1. Reservation = CAS_NLP_4
   10. Select the email on start and give your email.
   11. Submit.
   12. Once the session has started, open it.
3. On the terminal, fork the github repository with git clone https://github.com/dsl-unibe-ch/CAS_NLP_M4.git
4. Change into the directory with cd CAS_NLP_M4
5. Load Anaconda module load Anaconda3
6. Create a virtual environment with python -m venv .venv
7. Initialise conda eval "$(conda shell.bash hook)"
8. Add conda source
   1. conda config --add channels conda-forge
   2. conda config --set channel_priority strict
9. Create venv conda create --name myenv
10. Conda activate environment conda activate myenv
11. Install pip requirements with pip install -r requirements.txt
12. Create kernel python -m ipykernel install --user --name conda_kernel
13. Set the kernel to this new conda environment conda_kernel on the notebook.
14. Create an account on Huggingface and create an access token. Make a .env file by copying the contents of .env.example. Inside .env file, assign the access token to the variable HF_TOKEN.

Before each session

1. Perform the steps 1, 2, 4, 5, 10 from First Time Steps.
2. Pull latest github repo updates with git pull. In case of conflicts, resolve them.
3. Perform the steps 13 from First Time Steps.

Recreating a new env/ kernel

In case of conflicts, you may create a new conda env and kernel again.

1. conda create -n llm-cu128 python=3.11 -y
2. conda activate llm-cu128
3. python -m pip install -U pip
4. python -m pip install -r requirements.txt
5. python -m ipykernel install --user --name llm-cu128 --display-name "Python (llm-cu128)"

Useful Links

1. Ubelix HPC guide