NEURAL NETWORK MODEL COMPRESSING METHOD, CORPUS TRANSLATION METHOD AND DEVICE THEREOF

摘要

The present disclosure relates to a neural network model compressing method, a corpus translation method and a device thereof. The neural network model compressing method comprises: acquiring a training sample set including multiple training sample pairs, each training sample pair including source data and target data corresponding to the source data; using the source data as input and the target data as verification to train an original teacher model; training an intermediate teacher model on the basis of the training sample set and the original teacher model, one or more of the intermediate teacher models forming a teacher model set; training multiple candidate student models on the basis of the training sample set, the original teacher model and the teacher model set, the multiple candidate student models forming a student model set; and according to the training result of the multiple candidate student models, selecting one candidate student model as a target student model. By introducing multiple teacher models, a multi-level guidance on the training of a student model is achieved, thus improving the quality of the student model.