Reproduce “Enhancing Language Model Reasoning with Structured Multi-Level Modeling” 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
Siheng Xiong, Ali Payani, Faramarz Fekri
MLR alternates high-level descriptors of subgoals with detailed execution, and its planner revises descriptors using summaries of steps. Its training compares candidate steps by estimating which continuations are more likely to reach a correct answer. On a four-task average Pass@1, full MLR scores 66.1 versus 59.1 for a low-level policy with Step-DPO, supporting the full configuration, not adaptive replanning alone.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.