mskrajnowski/2023-machine-learning-talk

2023-machine-learning-talk

CPU-only setup

1. Install docker

2. Build container images

   docker compose build shell

3. Setup the virtualenv

   docker compose run --rm shell python -m venv .venv
   docker compose run --rm shell pip install --no-deps -r requirements.cpu.txt

4. Enable nbextensions

   jupyter contrib nbextension install --sys-prefix

5. Run jupyter notebook

   docker compose up notebook

6. Jupyter server should print out a link you can use to open the notebook, it should look like http://127.0.0.1:8888/?token=abcd1234...

GPU acceleration

Nvidia CUDA

1. Install and enable nvidia-container-toolkit

   https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html

2. Enable CUDA-specific compose overrides

   ln -s ./docker-compose.cuda.yml ./docker-compose.override.yml

3. Rebuild container images

   docker compose build shell

4. Recreate the virtualenv

   rm -rf .venv
   docker compose run --rm shell python -m venv .venv
   docker compose run --rm shell pip install --no-deps -r requirements.cuda.txt

5. Check if everything worked

   docker compose run --rm shell python -c \
       'import torch; print(f"CUDA available: {torch.cuda.is_available()}")'

   should print out

   CUDA available: True