arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2204.05862cs.CLcs.LG
收藏

利用基于人类反馈的强化学习训练有益无害的助手

Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback

Anthropic(Anthropic公司)
Yuntao Bai, Andy Jones, Kamal Ndousse, Amanda Askell, Anna Chen, Nova DasSarma, Dawn Drain, Stanislav Fort, Deep Ganguli, Tom Henighan, Nicholas Joseph, Saurav … 展开作者

Yuntao Bai, Andy Jones, Kamal Ndousse, Amanda Askell, Anna Chen, Nova DasSarma, Dawn Drain, Stanislav Fort, Deep Ganguli, Tom Henighan, Nicholas Joseph, Saurav Kadavath, Jackson Kernion, Tom Conerly, Sheer El-Showk, Nelson Elhage, Zac Hatfield-Dodds, Danny Hernandez, Tristan Hume, Scott Johnston, Shauna Kravec, Liane Lovitt, Neel Nanda, Catherine Olsson, Dario Amodei, Tom Brown, Jack Clark, Sam McCandlish, Chris Olah, Ben Mann, Jared Kaplan

摘要:

我们应用偏好建模和基于人类反馈的强化学习(RLHF)来微调语言模型,以充当有用和无害的助手。我们发现这种对齐训练提高了几乎所有NLP评估的性能,并且与Python编码和总结等专业技能的训练完全兼容。我们探索了一种迭代的在线训练模式,其中偏好模型和强化学习策略每周更新一次,使用新鲜的人类反馈数据,有效地改进了我们的数据集和模型。最后,我们研究了RLHF训练的鲁棒性,并确定了RL奖励和策略与其初始化之间KL偏差的平方根之间的大致线性关系。除了我们的主要结果,我们进行周边分析校准,竞争的目标,并使用OOD检测,比较我们的模型与人类作家,并提供样本,从我们的模型使用提示出现在最近的相关工作。

英文摘要:

We apply preference modeling and reinforcement learning from human feedback (RLHF) to finetune language models to act as helpful and harmless assistants. We find this alignment training improves performance on almost all NLP evaluations, and is fully compatible with training for specialized skills such as python coding and summarization. We explore an iterated online mode of training, where preference models and RL policies are updated on a weekly cadence with fresh human feedback data, efficiently improving our datasets and models. Finally, we investigate the robustness of RLHF training, and identify a roughly linear relation between the RL reward and the square root of the KL divergence between the policy and its initialization. Alongside our main results, we perform peripheral analyses on calibration, competing objectives, and the use of OOD detection, compare our models with human writers, and provide samples from our models using prompts appearing in recent related work.

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。

↑