Weak-for-Strong: Training Weak Meta-Agent to Harness Strong Executors

Fan Nie, Lan Feng, Haotian Ye, Weixin Liang, Pan Lu, Huaxiu Yao, Alexandre Alahi, James Zou

Paper

---

Framework Overview

Our W4S framework operates as an iterative process of workflow generation, execution, and refinement:

1. Workflow Generation: The weak meta-agent design a new workflow to leverage the given strong model, represented as executable Python code.
2. Execution and Feedback.: The generated workflow is executed by a strong model on validation samples, producing performance feedback.
3. Refinement: The meta-agent uses feedback to iteratively improve the workflow.

Install

Store your API keys in key.env:

OPENAI_API_KEY=""
ANTHROPIC_API_KEY=""

Basic Installation (default)

conda create -n w4s python=3.11
conda activate w4s
pip install .

Installation with vLLM + training support

conda create -n w4s python=3.11
conda activate w4s
pip install .[vllm]

---

📖 Citation

If you find this work useful, please cite our paper:

@misc{nie2025weakforstrong,
      title={Weak-for-Strong: Training Weak Meta-Agent to Harness Strong Executors}, 
      author={Fan Nie and Lan Feng and Haotian Ye and Weixin Liang and Pan Lu and Huaxiu Yao and Alexandre Alahi and James Zou},
      year={2025},
      eprint={2504.04785},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2504.04785}, 
}

Acknowledgment

We thank ADAS and AFlow for their codebase and prompts.