NEURAL NETWORK MODEL CONVERSION METHOD, MODEL PRECISION POSITIONING METHOD, AND DEVICES

摘要

A neural network model conversion method, a model precision positioning method, and devices. In the scheme of neural network model conversion, a model conversion device can automatically obtain a precision verification result of a second neural network model according to inference results obtained by respectively performing processing, by means of a first neural network model under a source framework and a second neural network model under a target framework, on same model input data, so that the precision verification efficiency of neural network model conversion is improved. In the scheme of model precision positioning, a model precision positioning device can position, according to output data of each network layer obtained by respectively performing processing, by means of the first neural network model under the source framework and the second neural network model under the target framework, on the same model input data, a target network layer, in the second neural network model, in which precision is not aligned, so that the accurate positioning of the network layer when the precision is not aligned is implemented.