Compression system and method for molecular attribute prediction model based on parameter sharing and knowledge distillation

The invention relates to a compression system and method for a molecular attribute prediction model based on parameter sharing and knowledge distillation. The compression system comprises: an acquisition module configured to acquire a training set, a verification set and a test set of molecular structure data for pre-training and fine tuning, respectively; the parameter sharing module is configured to construct a parameter-shared molecular attribute prediction model based on an original molecular attribute prediction model and pre-train the parameter-shared molecular attribute prediction model; the knowledge distillation module is configured to construct a molecular attribute prediction model of knowledge distillation by utilizing the pre-trained parameter-shared molecular attribute prediction model and pre-train the molecular attribute prediction model; the fine tuning module is configured to perform fine tuning on the pre-trained knowledge distillation model so as to obtain a final molecular attribute predi