snorkelai/Tau2-Bench-Verified-Airline-With-Code-Agents

该数据集包含了代码代理与AI助手之间多轮交互的样本轨迹及相关元数据，基于Sierra.ai的Tau^2 Bench中的Airline环境，并由Amazon AGI团队验证。数据集旨在研究代码代理与工具导向代理之间的差异，特别是在需要数据库更新的场景中。数据集包含多个字段，如任务ID、模型、版本、用户场景、是否需要数据库更新、交互轨迹、奖励等。数据集由Snorkel AI整理，采用Apache-2.0许可。

This dataset includes sample traces and associated metadata from multi-turn interactions between a code agent and AI assistant, based on the Airline environment from Sierra.ais Tau^2 Bench and verified by Amazon AGI group. The dataset is designed to investigate the differences between code agents and tool-oriented agents, particularly in scenarios requiring database updates. It features various fields like task_id, model, version, user_scenario, db_update_required, trace, reward, and db_diff, among others. The dataset is curated by Snorkel AI and licensed under Apache-2.0.