Lightweight network model training method, device and equipment

The invention is suitable for the technical field of deep learning, and provides a lightweight network model training method, which comprises the steps of obtaining an initial lightweight network model and a teacher network model; obtaining a training sample set, wherein the training sample set comprises first data provided with a first label and second data not provided with a label; inputting the second data into the teacher network model to obtain a sample feature, and determining the sample feature as a second label corresponding to the second data; and training the initial lightweight network model according to the first data, the first label, the second data and the second label to obtain a target lightweight network model. The structure of the lightweight network model obtained through the method is not limited by an original model any more, and the lightweight network model can be compatible with multiple platforms and has universality. 本申请适用于深度学习技术领域，提供了一种轻量化网络模型的训练方法，包括：获取初始轻量化网络模型，以及教师网络模型；获取训练样本集，