BeagleMind

Chat with BeagleBoard docs using Qwen2.5-Instruct and RAG. Ask about BeagleBone, BeaglePlay, or anything in `docs.beagleboard.io`.

Setup

1. Clone the Docs

   git clone https://github.com/beagleboard/docs.beagleboard.io

2. Install uv

   • macOS/Linux: curl -LsSf https://astral.sh/uv/install.sh | sh
   • Windows: powershell -c "irm https://astral.sh/uv/install.ps1 | iex"
   • More options: uv docs

3. Sync Dependencies

   uv sync

4. Get a Model Download Qwen2.5-Instruct (e.g., Qwen/Qwen2.5-7B-Instruct) from Hugging Face and note its path.

Usage

Run the chat:

uv run python inference.py -c path/to/qwen2.5/model

• Ask away: What’s the BeagleBone AI?
• Commands:
  • :h - Help
  • :q - Quit
  • :cl - Clear screen
  • :clh - Clear history
  • :his - Show history
  • :seed <N> - Set seed
  • :conf - View config
  • :conf <key>=<value> - Tweak config
  • :reset-conf - Reset config

Options:

• -c <path> - Model path (default: model)
• -s <number> - Seed (default: 1234)
• --cpu-only - Run on CPU

Tips

• First run embeds docs.beagleboard.io—takes a sec.
• Adjust k in inference.py’s retrieve(query, k=2) for more context.

License

Follows Qwen2.5’s license—keep it legit. See model/LICENSE.