Data processing method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a data processing method and device, electronic equipment and a storage medium, and the method comprises the steps: after receiving a data processing request of a target client, allocating target application computing resources to the target client, and enabling the application computing resources of different clients at the same moment to be different; Pre-processing the data processing request of the target client through the target application computing resource to obtain target pre-processed data; And processing the target preprocessing data throughthe GPU to obtain a target processing result. The CPU logic is divided into a plurality of application computing resources, different application computing resources process data processing requestsof different clients, and parallel processing of the data processing requests of the different clients is achieved. The method has the advantages that the programs of the CPU and the GPU are parallelized, the computing time of the CPU is hidden, the idle time of the GPU is eliminated, the effective computing time ratio of the GPU is increased, and the overall reasoning service performance is improved.

Description

technical field [0001] The present application relates to the field of computer technology, especially data processing methods, devices, electronic equipment and storage media. Background technique [0002] With the rapid development of machine learning technology, deep learning technology plays an irreplaceable role in more and more business scenarios. The amount of calculation for deep learning is relatively large. As the demand grows, using a GPU (Graphics Processing Unit, Graphics Processing Unit) that is good at parallel computing as a computing device is currently the mainstream solution. The process of performing calculations using a trained deep model is called "inference". In order to enable the model to be used more flexibly, first start the program loaded with the deep learning model on a machine equipped with a GPU, and then send a remote request to it through the client to obtain the calculation result, which becomes the inference business of deep learning gen...

AI Technical Summary

Problems solved by technology

[0004] However, in the above method, when a client initiates a request to the server, the server is occupied by the client and cannot process requests from other clients before returning a response

Since it takes time for the receiver to send data, the CPU and GPU of the server are idle. When the number of clients is large, the server processes each client's request in order, and the processing efficiency is low.

Images