We present a large, tunable neural conversational response generation model,
DialoGPT (dialogue generative pre-trained transformer). Trained on 147M
conversation-like exchanges extracted from Reddit comment chains over a period
spanning from 2005 through 2017, DialoGPT extends the Hugging Face PyTorch
transformer to attain a performance close to human both in terms of automatic
and human evaluation in single-turn dialogue settings. We show that
conversational systems that leverage DialoGPT generate more relevant,
contentful and context-consistent responses than strong baseline systems. The
pre-trained model and training pipeline are publicly released to facilitate
research into neural response generation and the development of more
intelligent open-domain dialogue systems.

本文介绍了 DialoGPT，一种大型的、可调整的神经对话响应生成模型，它在 Reddit 上 2005 年至 2017 年的时间跨度内提取的 1.47 亿条对话样本上进行训练。在单轮对话设置中，DialoGPT 能够达到接近人类的性能水平，能够生成比强基线系统更相关、更富有内涵和更具上下文一致性的响应。该预训练模型和训练流程已公开发布，以促进神经响应生成研究和更智能的开放域对话系统的开发。