AgentBank: Towards Generalized LLM Agents via Fine-Tuning on 50000+ Interaction Trajectories

TL;DR

AgentBank is a large-scale dataset of over 50,000 diverse interaction trajectories used to fine-tune LLMs, significantly enhancing their generalized agent capabilities across multiple tasks.

Contribution

This work introduces AgentBank, the largest trajectory dataset for fine-tuning LLMs, and demonstrates its effectiveness in improving generalized agent skills.

Findings

Scaling trajectory data improves agent capabilities.

Fine-tuned models outperform baselines.

Key insights into trajectory tuning and skill generalization.

Abstract

Fine-tuning on agent-environment interaction trajectory data holds significant promise for surfacing generalized agent capabilities in open-source large language models (LLMs). In this work, we introduce AgentBank, by far the largest trajectory tuning data collection featuring more than 50k diverse high-quality interaction trajectories which comprises 16 tasks covering five distinct agent skill dimensions. Leveraging a novel annotation pipeline, we are able to scale the annotated trajectories and generate a trajectory dataset with minimized difficulty bias. Furthermore, we fine-tune LLMs on AgentBank to get a series of agent models, Samoyed. Our comparative experiments demonstrate the effectiveness of scaling the interaction trajectory data to acquire generalized agent capabilities. Additional studies also reveal some key observations regarding trajectory tuning and agent skill generalization.