LifelongAlignment/aifgen-long-piecewise

这是一个为终身强化学习在大型语言模型上的基准测试而创建的数据集，包含数学、社会科学、物理学和化学等领域的提示性答案问题和直接答案问题。数据集通过AIF-Gen框架使用gpt-4o-mini模型生成，旨在帮助模型在各个领域与人类偏好保持一致。

This dataset is created for benchmarking lifelong reinforcement learning on large language models, containing hint-based and direct answer questions in domains such as math, social sciences, physics, and chemistry. The dataset is generated using the AIF-Gen framework with the gpt-4o-mini model, aiming to help models align with human preferences across various domains.