Model training method, reply information determination method and device and storage medium

The invention provides a model training method, a reply information determination method and device and a storage medium, and relates to the technical field of artificial intelligence. The model training method comprises the following steps: generating a control signal according to a reply sample in training samples; according to the control signal, the reply sample and the previous dialogue text in the training sample, co-estrus reply information for the previous dialogue text is obtained through a conditional diffusion model with a control range mask strategy, and the reply sample is a reply for the previous dialogue text; and adjusting the parameters of the conditional diffusion model through the minimization variation lower bound until the training is completed. 本公开提出一种模型的训练方法、回复信息确定方法及装置和存储介质，涉及人工智能技术领域。本公开的一种模型训练方法包括：根据训练样本中的回复样本生成控制信号；根据控制信号、回复样本和训练样本中的对话上文，通过带控制范围掩码策略的条件扩散模型获取针对对话上文的共情回复信息，其中，回复样本为针对对话上文的回复；通过最小化变分下界调整条件扩散模型的参数，直至完成训练。