Jarrodbarnes/tau2-sft-final
收藏Hugging Face2025-12-15 更新2025-12-20 收录
下载链接：
https://hf-mirror.com/datasets/Jarrodbarnes/tau2-sft-final
下载链接
链接失效反馈官方服务：
资源简介：
Tau2 SFT数据集是一个多领域监督微调数据集，用于在tau2-bench双控制环境中训练使用工具的代理。该数据集包含416个轨迹，覆盖航空、零售和电信三个领域，采用<think> + [ACTION]的格式。数据集仅用于训练，提供了基线性能数据，并包含完整数据集和过滤后的数据集。数据集的选择标准是桥接对齐选择，优先考虑成功案例，并填充高质量失败案例。
The Tau2 SFT Dataset is a multi-domain supervised fine-tuning dataset for training tool-use agents on tau2-bench dual-control environments. It contains 416 trajectories across airline, retail, and telecom domains in a `<think>` + `[ACTION]` format. Train-only with baseline performance metrics provided. Includes both full and filtered (10+ word reasoning) dataset versions. Uses bridge-aligned selection prioritizing successes supplemented with high-quality failures.
提供机构：
Jarrodbarnes
