AFlow: Automating Agentic Workflow Generation

Jiayi Zhang, Jinyu Xiang, Zhaoyang Yu, Fengwei Teng, Xionghui Chen, Jiaqi Chen, Mingchen Zhuge, Xin Cheng, Sirui Hong, Jinlin Wang, Bingnan Zheng, Bang Liu, Yuyu Luo, Chenglin Wu·October 14, 2024

Summary

AFLOW is an automated framework that optimizes agentic workflows for large language models. It uses Monte Carlo Tree Search to efficiently explore a space of code-represented workflows, iteratively refining them through code modification, tree-structured experience, and execution feedback. AFLOW demonstrates superior performance across six benchmark datasets, outperforming state-of-the-art baselines by 5.7% on average. It enables smaller models to outperform GPT-4 on specific tasks at 4.55% of its inference cost. The code is available at <https://github.com/geekan/MetaGPT>.

Key findings

17

Tables

3

Advanced features