Grove: Distributed ML Training over AirDrop

swar_ja

Python library for training models across MacBooks over wireless. Uses AWDL for discovery and fallback transport, upgrades to WiFi when available. Works on networks with client isolation (eduroam etc.). grove start train.py -n 4 on one Mac, grove join on the others. DiLoCo for communication-efficient sync, optional SparseLoCo compression (~30x) for slower links. Swift sidecar for the AWDL layer since Network.framework isn't accessible from Python. macOS/Apple Silicon only, built on MLX.