🧾 Dataset Card for CAC‑CoT Dataset

Dataset Summary

Name : CAC‑CoT (Connector‑Aware Compact Chain‑of‑Thought) Dataset

Based on paper: CAC‑CoT: Connector‑Aware Compact Chain‑of‑Thought for Efficient Reasoning Data Synthesis Across Dual‑System Cognitive Tasks" by Sunguk Choi, Yonghoon Kwon, Heondeuk Lee (2025). arXiv:2508.18743

Purpose / Task(s):
This dataset is for training and evaluating models on reasoning tasks with chain‑of‑thought traces that emphasize compactness and connector awareness. It supports learning reasoning trajectories, improving efficiency (shorter traces) while maintaining performance, particularly on tasks that require both intuitive (fast) and analytical (slow) reasoning.

Artifacts:

• Model: datumo/CAC-CoT: model_link
• Paper: CAC-CoT: Connector-Aware Compact Chain-of-Thought for Efficient Reasoning Data Synthesis Across Dual-System Cognitive Tasks: paper_link
• Git_link: selectstar-ai/CAC-CoT: git_link

Language(s)

• English (Only)

---

Data Splits

💡 Only the train split is available in this release. If you'd like to contribute validation or test sets, please contact us.

---

Dataset Creation

Generation Process:

• The dataset was created via synthetic generation using the Gemini-2.0-Flash model.
• Prompts were designed to enforce connector-aware and compact chain-of-thought reasoning.
• Generation examples were filtered based on formatting constraints and logical coherence.
• Duplicates with previous datasets (e.g., LIMO, s1) were removed.

Motivation:
The dataset was developed as part of the CAC‑CoT project to explore the efficiency-accuracy trade-off in reasoning traces.
Traditional CoT datasets tend to be verbose, which leads to increased cost and slower inference. This dataset provides compact, interpretable reasoning traces that preserve performance across both System-1 (intuitive) and System-2 (analytical) tasks.

Licensing Information

The dataset is available under the Apache License, Version 2.0.

Citation

Please consider citing the following paper if our method and resources were helpful to your work.

@misc{choi2025caccotconnectorawarecompactchainofthought,
      title={CAC-CoT: Connector-Aware Compact Chain-of-Thought for Efficient Reasoning Data Synthesis Across Dual-System Cognitive Tasks}, 
      author={Sunguk Choi and Yonghoon Kwon and Heondeuk Lee},
      year={2025},
      eprint={2508.18743},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2508.18743}, 
}