microsoft/DialoGPT-medium

DialoGPT is a SOTA large-scale pretrained dialogue response generation model for multiturn conversations . The model is trained on 147M multi-turn dialogue from Reddit discussion thread . The response quality is comparable to human response quality under a single-turn conversation Turing test . The full details of the model can be found in the original DialOGPT repository: https://arxiv.org/abs/19111111.00536/DuoGPT-medium. The model was trained on a series of examples from an interactive environment: Multi-turn generation examples: Do money buy happiness? Do you have no idea how hard it is to be a millionaire and happy,