Neural network compiler configuration method, device, computer equipment and storage medium

Abstract

The present application relates to a neural network compiler configuration method, device, computer equipment and storage medium. The method includes: obtaining a target neural network model and an initial calculation graph corresponding to the target neural network model; the initial calculation graph includes a plurality of operators; dividing the plurality of operators to obtain a plurality of operator sets; determining each operator Combinations of multiple operator types corresponding to each set, and obtain the operator running time corresponding to each operator type combination, and obtain the multiple operator running times corresponding to each operator set; combine multiple operator types, and calculate The operator type combination with the smallest running time is used as the target operator type combination corresponding to each operator set; the target calculation graph of the target neural network model is generated according to the target operator type combination corresponding to each operator set, and the target calculation graph is generated based on the target calculation graph. Compiler configuration information for the target neural network model. The method can improve the configuration optimization effect of the neural network model compiler.

Description

technical field [0001] The present application relates to the field of computer technology, in particular to a neural network compiler configuration method, device, computer equipment and storage medium. Background technique [0002] With the development of computer technology, neural network models have been used more and more widely, but neural network models still face huge challenges in computing performance. At present, in order to realize the efficient calculation of the neural network model, the calculation performance of the neural network model is usually improved by optimizing the configuration of the compiler of the neural network model. [0003] In traditional technology, the optimization configuration of the compiler of the neural network model is usually based on the handwritten rules of experts. By splitting the calculation graph of the neural network model into multiple sub-graphs according to the splitting rules experienced by experts, and then separately M...

AI Technical Summary

Problems solved by technology

However, the current neural network compiler configuration method needs to rely on experts' handwritten rules to optimize the calculation graph, so the compiler configuration optimization effect for neural network models with emerging operators or model structures is poor

Images