Model conversion method and system between deep learning frameworks based on minimum execution cost

Abstract

The invention provides a model conversion method and system between deep learning frameworks based on minimum execution cost, and the method comprises the steps: adding an operation conversion cost value on the basis of the original technology, considering the fusion condition of a plurality of independent operations, and supplementing fusion mapping. The specific implementation of the model is embodied in the operation conversion of forming the model, and in the stage, a converted model structure with the lowest execution cost is obtained through a dynamic programming algorithm according to amodel conversion mapping table. According to the method, the read-write process of the intermediate results of a plurality of operation rooms can be reduced through operation fusion, so that the calculation performance and the storage space are optimized, and the execution cost of the converted model is reduced. Meanwhile, when multiple fusion options exist, a model conversion method with the minimum execution cost is obtained through a dynamic programming algorithm.

Description

technical field [0001] The present invention relates to the field of deep learning, and in particular to a method and system for model conversion between deep learning frameworks based on the minimum execution cost, which is specifically applied to the model conversion of different frameworks. Background technique [0002] Current deep learning models can be implemented with different frameworks. After a model is trained on framework A, the user can convert it to a model under framework B so that it can directly make inferences under framework B without retraining the model. like figure 1 As shown, each model Model is composed of a variety of different operations. In general, each operation OP (operator) has operation type, input (supports zero to multiple), output (supports 1 to multiple), attributes, etc. content. like figure 2 As shown, each operation is associated with its own input and output, figure 2 The model structure contains 5 OPs, namely Conv, SpatialBN, R...

AI Technical Summary

Problems solved by technology

[0010] The purpose of the present invention is to solve the problem in the prior art that the deep learning model is converted between different frameworks, and the execution cost of the converted model is too high, and a minimum execution cost-based method that can generate a minimum execution cost model is proposed Method and system for model conversion between deep learning frameworks

Images