Plan-over-Graph: Towards Parallelable LLM Agent Schedule

TL;DR

This paper presents a novel plan-over-graph paradigm that enables large language models to decompose tasks into graphs and generate parallel execution plans, significantly improving task performance and efficiency.

Contribution

The paper introduces a new plan-over-graph framework with an automated pipeline and two-stage training, advancing LLM task planning for scalable, parallelizable execution.

Findings

Significant performance improvements on API-based and open-source LLMs.

Effective task decomposition into executable graphs.

Enhanced parallel execution and efficiency.

Abstract

Large Language Models (LLMs) have demonstrated exceptional abilities in reasoning for task planning. However, challenges remain under-explored for parallel schedules. This paper introduces a novel paradigm, plan-over-graph, in which the model first decomposes a real-life textual task into executable subtasks and constructs an abstract task graph. The model then understands this task graph as input and generates a plan for parallel execution. To enhance the planning capability of complex, scalable graphs, we design an automated and controllable pipeline to generate synthetic graphs and propose a two-stage training scheme. Experimental results show that our plan-over-graph method significantly improves task performance on both API-based LLMs and trainable open-sourced LLMs. By normalizing complex tasks as graphs, our method naturally supports parallel execution, demonstrating global efficiency. The code and data are available at https://github.com/zsq259/Plan-over-Graph.