Reproduce “EDU-Jev Training System-One Decision Models for Education” 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
Jiazheng Li, Yulan He
EDU-Jev adapts typed decisions to educational assessment, returning probabilities over supplied answer choices rather than generating a rationale and verdict. Its Qwen3.5-based models use education-focused training and a pointer readout that restricts outputs to those choices. EDU-Jev 4B achieves 67.6% accuracy across 12 education benchmarks, eight from represented task families, but ties the strongest API on four cross-domain sets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.