arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2405.14751cs.LG
收藏

AGILE:一种新颖的LLM智能体强化学习框架

AGILE: A Novel Reinforcement Learning Framework of LLM Agents

ByteDance Research(字节跳动研究院);University of Science and Technology of China(中国科学技术大学);Shanghai Jiao Tong University(上海交通大学)

Peiyuan Feng, Yichen He, Guanhua Huang, Yuan Lin, Hanchong Zhang, Yuchen Zhang, Hang Li

AI总结:

本文提出AGILE框架,将LLM智能体构建视为强化学习问题,结合记忆、工具、专家咨询与反思,通过PPO训练7B和13B模型,在多个问答数据集上超越GPT-4,并验证各组件不可或缺。

摘要:

我们介绍了一种名为AGILE(与环境交互并从环境中学习的智能体)的新型LLM智能体强化学习框架,旨在利用LLM、记忆、工具以及与专家的交互来执行复杂的对话任务。该智能体具备超越对话的能力,包括反思、工具使用和专家咨询。我们将此类LLM智能体的构建形式化为一个强化学习(RL)问题,其中LLM作为策略模型。我们使用标注的动作数据和PPO算法对LLM进行微调。我们专注于问答任务,并发布了一个名为ProductQA的智能体数据集,其中包含在线购物中的挑战性问题。我们在ProductQA、MedMCQA和HotPotQA上的大量实验表明,基于7B和13B LLM并使用PPO训练的AGILE智能体可以胜过GPT-4智能体。我们的消融研究强调了记忆、工具、咨询、反思和强化学习在实现智能体强大性能中的不可或缺性。数据集和代码可在https://github.com/bytarnish/AGILE获取。

英文摘要:

We introduce a novel reinforcement learning framework of LLM agents named AGILE (AGent that Interacts and Learns from Environments) designed to perform complex conversational tasks with users, leveraging LLMs, memory, tools, and interactions with experts. The agent possesses capabilities beyond conversation, including reflection, tool usage, and expert consultation. We formulate the construction of such an LLM agent as a reinforcement learning (RL) problem, in which the LLM serves as the policy model. We fine-tune the LLM using labeled data of actions and the PPO algorithm. We focus on question answering and release a dataset for agents called ProductQA, comprising challenging questions in online shopping. Our extensive experiments on ProductQA, MedMCQA and HotPotQA show that AGILE agents based on 7B and 13B LLMs trained with PPO can outperform GPT-4 agents. Our ablation study highlights the indispensability of memory, tools, consultation, reflection, and reinforcement learning in achieving the agent's strong performance. Datasets and code are available at https://github.com/bytarnish/AGILE.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑