Reproduce “CDW-CoT: Clustered Distance-Weighted Chain-of-Thoughts Reasoning” 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

Yuanheng Fang, Guoqing Chao, Wenqiang Lei, Shaobo Li, Dianhui Chu

CDW-CoT proposes a Clustered Distance-Weighted Chain-of-Thoughts method that dynamically tailors prompts to individual data instances, overcoming the "one-size-fits-all" limitation of prior CoT approaches. It improved average reasoning accuracy by 25.34% on LLaMA2 (13B) and 15.72% on LLaMA3 (8B) compared to Manual-CoT across six diverse benchmarks.

Key insights

  • Dynamically tailoring prompts to individual data instances, rather than using a uniform approach, is crucial for enhancing Large Language Model (LLM) reasoning across diverse datasets.
  • Integrating cluster-based prompt candidate initialization, cluster-specific probability distribution optimization via Black-Box Prompt Learning, and a distance-weighted selection mechanism enables adaptive and robust prompting.
  • Each component of the CDW-CoT framework (clustering, optimized distributions, and distance weighting) contributes critically to the overall performance gains.

Summary by alphaXiv.

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