Data fixed-point acceleration method and device, electronic equipment and storage medium

Abstract

The invention discloses a fixed-point acceleration method for data. The method comprises the following steps: acquiring test data and first model parameters; inputting the test data and the first model parameters into a fixed-point algorithm model needing fixed-point parameter adjustment to obtain a first characteristic value; Converting the data format of the first characteristic value into a floating point format from a fixed point format, and obtaining a second characteristic value represented by the floating point format; and performing fixed-point operation on the second characteristic value by using a GPU acceleration mechanism to obtain output data. The invention further provides a fixed-point acceleration device, electronic equipment and a computer storage medium. According to theinvention, the calculation of the fixed-point algorithm model can be accelerated, and the fixed-point iteration period is accelerated.

Description

technical field [0001] The present invention relates to the technical field of intelligent terminals, in particular to a data fixed-point acceleration method, device, electronic equipment and storage medium. Background technique [0002] Deep learning neural network algorithms require high computing power and a large amount of storage space, which limits the application of deep learning neural network algorithms on embedded terminal devices. By fixed-pointing the deep learning neural network algorithm, the dependence on storage space can be reduced, so that the deep learning neural network algorithm can be expanded on the terminal device. [0003] Due to the high complexity of the deep learning neural network algorithm, usually only a relatively low quantization bit width can be used to ensure the operating efficiency on the terminal device, which will cause a relatively large loss of precision. In order to reduce the loss of precision, in the process of fixed-pointing the ...

AI Technical Summary

Problems solved by technology

[0005] Usually, this simulation is very slow, and it takes a lot of time to test the fixed-point algorithm model, which makes the iteration cycle of the fixed-point algorithm of the deep learning neural network algorithm longer

Images