Reproduce “Learning Composable Chains-of-Thought” 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
Fangcong Yin, Zeyu Leo Liu, Liu Leqi, Xi Ye, Greg Durrett
Composable Chain-of-Thought trains language models to continue reasoning after a placeholder prefix, then combines models trained on separate skills. The authors evaluate multitask learning and model merging on unseen skill combinations, using atomic-task examples without compositional training traces. Qwen2.5-7B reaches 96.3% exact match on Next Letter plus ASCII Multiplication, but results do not establish length generalization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.