Reproduce “Training LLM Agents to Automate LLM Post-Training with RSI-Dojo” 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
Shuzheng Gao, Zhaojian Yu, Penghao Yin, Yuling Shi, Kai Cai, Shilin He, Michael R. Lyu
RSI-Dojo trains language models to carry out post-training work on smaller target models. It builds executable practice environments from research tasks, collects agent trajectories, filters them for behavior and outcomes, and fine-tunes models on selected examples. On PostTrainBench, Qwen3.6-35B-A3B rises from 13.08 to 29.95; this measures benchmarked work, not recursive self-improvement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.