Neural network caching method, system and device and storage medium

Abstract

The invention discloses a caching method, system and device of a neural network and a storage medium. The method comprises the following steps: acquiring configuration information including a neural network dimension; setting a working mode of the cache according to the configuration information; obtaining target to-be-processed data through the set cache; and according to the configuration information, processing the target to-be-processed data through the set cache. According to the method, the cache is configured to determine the target to-be-processed data, and different to-be-processed data are processed by adopting different data processing schemes according to the configuration information, so that cache mapping aiming at neural networks of different dimensions is realized, the cache can be prevented from being jammed, and the data processing efficiency is improved. Therefore, writing and outputting with high concurrency and high throughput rate are carried out. The method can be widely applied to the technical field of neural network algorithms.

Description

technical field [0001] The invention relates to the technical field of neural network algorithms, in particular to a neural network cache method, system, device and storage medium. Background technique [0002] There are differences in neural networks of different dimensions and sizes. Networks of different dimensions need to allocate additional resources due to differences in computing dimensions, resulting in a waste of computing resource distribution; networks of different dimensions will make the cache that cannot be configured flexibly unable to meet the needs of high-performance computing and become a bottleneck for success. The implementation and deployment of convolutional neural networks in hardware platforms are increasingly differentiated, and hardware design lacks the flexibility to support multiple network dimensions and sizes. [0003] To sum up, the problems existing in the related technologies need to be solved urgently. SUMMARY OF THE INVENTION [0004] ...

AI Technical Summary

Problems solved by technology

Networks of different dimensions need to allocate additional resources to calculate the difference in dimensions, resulting in a waste of computing resource distribution; networks of different sizes will make the cache that cannot be flexibly configured unable to meet the needs of high-performance computing and become the bottleneck of success

The implementation and deployment of convolutional neural networks on hardware platforms are increasingly differentiated, and hardware design lacks the flexibility to support multiple network dimensions and sizes

Images