README.md

Chain of Draft: Thinking Faster by Writing Less

[Paper]

Large Language Models (LLMs) have demonstrated remarkable performance in solving complex reasoning tasks through mechanisms like Chain-of-Thought (CoT) prompting, which emphasizes verbose, step-by-step reasoning. However, humans typically employ a more efficient strategy: drafting concise intermediate thoughts that capture only essential information. In this work, we propose Chain of Draft (CoD), a novel paradigm inspired by human cognitive processes, where LLMs generate minimalistic yet informative intermediate reasoning outputs while solving tasks. By reducing verbosity and focusing on critical insights, CoD matches or surpasses CoT in accuracy while using as little as only 7.6% of the tokens, significantly reducing cost and latency across various reasoning tasks.

Usage

To run evaluation:

python evaluate.py \
    --task gsm8k \      # Task to evaluate (options: gsm8k, date, sports, coin_flip)
    --model gpt-4o \    # Model to be evaluated
    --prompt cod \      # Prompting strategy (options: baseline, cod, cot)
    --shot 5 \          # [Optional] Number of few-shot examples to include in the prompt (uses all available examples by default if omitted)
    --url $BASE_URL \   # [Optional] Base URL for an OpenAI-compatible interface (e.g., locally hosted models)
    --api-key $KEY \    # [Optional] API key for model access (automatically loads from environment variables for Claude and OpenAI models if not provided)

Currently, the script supports Claude models, OpenAI models, as well as any model that uses an OpenAI-compatible interface. The evaluation results will be stored under ./results/.

All prompts and fewshot examples are stored under ./configs/{task}-{prompt}.yaml.

Citation

@article{xu2025cod,
    title={Chain of Draft: Thinking Faster by Writing Less},
    author={Xu, Silei and Xie, Wenhao and Zhao, Lingxiao and He, Pengcheng},
    journal={arXiv preprint arXiv:2502.18600},
    year={2025}
}