Welcome to the official offline AI setup repository for our train hackathon!
Since internet connectivity on a moving train is often unstable, this repository provides everything you need to run Google's Gemma open models locally on your laptop. By following these guides, you will have a powerful, fully offline AI coding assistant integrated directly into your editor before we board.
We have prepared dedicated setup guides for all major operating systems. Choose the file that matches your machine:
- Windows Setup: Setup instructions for Windows utilizing PowerShell or the official installer.
- macOS Setup: Optimized instructions for Apple Silicon (M-series) using Homebrew.
- Linux Setup: Terminal-based instructions tailored for Ubuntu/Debian users.
We are exclusively using Google's open-weight models, scaled for different laptop hardware constraints:
- Gemma 4 (
e4b&12b): Google's latest highly efficient models for general logic, reasoning, and web development queries. - Gemma 3 (
4b&12b): Excellent, stable all-rounder models for complex architectural questions. - CodeGemma (
2b&7b): Lightning-fast models specifically trained and optimized for inline code completion.
To make this seamless, participants will be using:
- Ollama: A lightweight runner that operates quietly in the background to serve the models locally.
- Continue.dev: A VS Code extension that connects directly to Ollama, providing an offline chat and autocomplete experience right in your IDE.
Please ensure you complete the setup for your respective operating system before arriving at the station. Downloading the runner, the IDE extension, and pulling the actual model weights requires a stable Wi-Fi connection. Once downloaded, everything runs 100% offline.
Happy hacking!