This repository contains small, reproducible experiment scripts around the Agent Data Protocol (ADP) data release.
Current contents:
openhands_sdk_training/: reproduce the original Qwen3.5 0.8B LLaMA-Factory run and track the current Qwen3.5 4B/9B OpenHands SDK condenser SFT experiments.
Keep source code and generated experiment artifacts separate:
- Code checkouts and experiment scripts:
~/work/adp/ - Data, caches, logs, model outputs, and generated artifacts:
~/exp/adp/
Recommended experiment layout:
~/work/adp/
├── adp-experiments/
└── agent-data-protocol/
~/exp/adp/
├── datasets/
├── runs/
├── cache/
└── tmp/
The current experiment ledger is
openhands_sdk_training/CURRENT_EXPERIMENTS.md.