Reproduce “AutoMaster: Autonomous Post-Training by Making the Same Model a Better Teacher” 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
Guoxin Chen, Shuang Sun, Fanzhe Meng, Xinyu Tang, Ruihua Song, Wayne Xin Zhao
Skill-FT uses an LLM agent, AutoMaster, to guide a fixed teacher with natural-language skills and select its trajectories for student supervised fine-tuning; evaluation omits generated skills. Across six benchmark–model settings, student mean scores exceed Vanilla SFT by 1.42–8.99 percentage points. For Qwen3.5-9B on Terminal-Bench 2.1, accuracy is 35.58% versus 26.59%; 4B uncertainty intervals include zero in two comparisons.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.