Reproduce “Deliberate Reasoning in Language Models as Structure-Aware Planning with an Accurate World Model” 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, Yuan Yang, Faramarz Fekri

SWAP represents language-model reasoning as an entailment graph whose statement nodes are linked to their premises. Its policy proposes candidate updates, a world model predicts next graphs, and symbolic checks and contrastive ranking select continuations. With LLaMA3-8B-Instruct, it answers 43.2 ± 0.3% of MATH500 questions correctly versus RAP’s 26.2 ± 0.3%, but graph checks do not prove every claim true.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.