Neural network model evaluation method and device, electronic equipment and storage medium

• Main Authors: ZHANG AIFEI, LIU HANJIE, MAO YAPENG, LIU BAOSEN
• Format: Patent
• Language: Chinese; English
• Published: 10.05.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; DATA PROCESSING SYSTEMS OR METHODS, SPECIALLY ADAPTED FORADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORYOR FORECASTING PURPOSES; PHYSICS; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE,COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTINGPURPOSES, NOT OTHERWISE PROVIDED FOR

The invention discloses a neural network model evaluation method and device, electronic equipment and a storage medium. The method comprises the following steps: determining a model structure and parameter information of a neural network model; if it is determined that the neural network model is matched with the storage and calculation integrated chip according to the parameter information, scheduling each operator of the neural network model according to the model structure and the parameter information to obtain a pipeline parallel window and a calculation kernel corresponding to each operator; and performing performance evaluation on the scheduled neural network model according to the pipeline parallel window and the calculation kernel corresponding to each operator. According to the technical scheme, quantitative evaluation of the adaptation degree of the deep neural network model and the storage and calculation integrated chip can be realized, so that a foundation is laid for subsequent model optimizati