Reproduce “BIRD-RL: Scaling Agentic Reinforcement Learning over Stateful Data-Centric Environments” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ge Qu, Jinyang Li, Xiaohan Xu, Xiaolong Li, Bowen Qin, Nan Huo, Ziwei Tang, Chenhao Ma and 1 more
BIRD-RL trains SQL-debugging agents through four stages, from supervised imitation to reinforcement learning with execution feedback. It assigns each rollout a database replica and resets modified copies only when reused. On BIRD-CRITIC-SQLite, its 14B model reaches 48.00% success rate, near Claude-Opus-4.6 Agent at 48.20%; reusable replicas use 12.8× less measured online lifecycle I/O than per-rollout containers in the evaluated SQLite setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.