Skip to content

Customized development container environment for consistent and efficient execution of machine learning projects.

License

Notifications You must be signed in to change notification settings

SkywardAI/bundoora

Repository files navigation

BUNDOORA

Building Checking 🚀 Releasing Image 🚀 Dependabot Updates Release Drafter 🚀

🚀 Overview

A tailored development container environment for consistent and efficient machine learning project execution. Encapsulates all necessary dependencies (Python, PyTorch, CUDA, Ubuntu 22.04) to resolve compatibility issues and enhance reproducibility.

Includes automated installation scripts for setting up NVIDIA GPU drivers, Docker rootless mode, and NVIDIA Container Toolkit on Ubuntu systems.

📦 Installation Scripts

Quick setup scripts for Ubuntu systems:

# Install NVIDIA GPU driver (requires reboot)
curl -fsSL https://raw.githubusercontent.com/SkywardAI/bundoora/refs/heads/main/scripts/nvidia-gpu-driver.sh | sudo bash

# Install Docker rootless mode
# For Ubuntu 22.04:
curl -fsSL https://raw.githubusercontent.com/SkywardAI/bundoora/refs/heads/main/scripts/install_rootless_docker_22_04.sh | bash
# For Ubuntu 24.04+:
curl -fsSL https://raw.githubusercontent.com/SkywardAI/bundoora/refs/heads/main/scripts/install_rootless_docker_24_03.sh | bash

# Install NVIDIA Container Toolkit
curl -fsSL https://raw.githubusercontent.com/SkywardAI/bundoora/refs/heads/main/scripts/nvidia-container-tool.sh | sudo bash

See scripts/README.md for detailed instructions.

🐳 Docker Images

This repository builds specialized container images for Python development:

Image Description Base Package Manager
bundoora:uv Fast Python development Ubuntu 22.04 uv
bundoora:pytorch NVIDIA PyTorch with flash-attention2 NVIDIA PyTorch 25.10 pip

The images include:

  • NVIDIA GPU support with CUDA
  • Python 3.x
  • PyTorch with CUDA support
  • Zsh with Oh My Zsh
  • Git and development tools

Note: The uv image is based on Ubuntu 22.04 but runs seamlessly on Ubuntu 24.03 host systems. The pytorch image includes flash-attention2 support.

🛠️ Environment Specifications

Component Version
Python 3.10
PyTorch 2.5.1
OS AWS Ubuntu 24.03
CUDA 12.1

🔨 Build Commands

Build Options
Command Description
make build Build for default platform (linux/amd64)
make build-multi Build for multiple platforms (local)
make push-multi IMAGE_NAME=yourusername/test Build and push to registry
make create-builder Create new builder instance

📚 Research Projects

📄 Citation

@software{Li_Bundoora_2024,
  author = {Li, Bowen},
  doi = {<>},
  month = dec,
  title = {{Bundoora}},
  url = {https://github.com/SkywardAI/bundoora},
  version = {1.0.0},
  year = {2024}
}

About

Customized development container environment for consistent and efficient execution of machine learning projects.

Topics

Resources

License

Stars

Watchers

Forks

Packages