Author: Lev Rozanov

HKU Docker Environment

This Docker image sets up a shared dev environment for the HKU competition projects, cloning repos via HTTPS with a GitHub token.

What is this?

• Clones both hku-data and hku-comp-fix repos automatically on build using a GitHub token
• Python 3.11 with all required dependencies pre-installed in a uv venv
• GPU acceleration support with CUDA 12.1, PyTorch, TensorFlow, and CuPy
• Everyone works in the same environment
• Git LFS support for large data files

What's inside

• CUDA 12.1 runtime for GPU acceleration
• Python 3.11 and essential packages
• Git and Git LFS configured
• Pre-installed: pandas, numpy, scikit-learn, pyarrow, psutil, jupyter
• GPU frameworks: PyTorch (CUDA), TensorFlow (GPU), CuPy (CUDA)
• RAPIDS 25.8: cuDF, cuML, cuGraph, cuxfilter, cuCIM, cuVS
• Tools: vim, nano, curl, wget, tree, htop
• Both repos cloned into /workspace-hku/

Quick Start

Build the GPU-enabled image (using buildkit and secret token):

Windows PowerShell:
$env:DOCKER_BUILDKIT=1; docker build --secret id=GITHUB_TOKEN,src=./token.txt -t hku-docker-env-gpu  .

Linux:
DOCKER_BUILDKIT=1 docker build --secret id=GITHUB_TOKEN,src=./token.txt -t hku-docker-env-gpu  .

# Run the container with GPU access:
docker run --gpus all -it --rm hku-docker-env-gpu

# Test RAPIDS (cuDF/cuML):
python -c "import cudf, cuml; print('cuDF', cudf.__version__, 'cuML', cuml.__version__)"