Data processing system and method for heterogeneous architecture

Abstract

The present disclosure discloses a data processing system for a heterogeneous architecture, comprising: a job decomposition component for decomposing a job to be completed into a series of tasks executed by an executor in the heterogeneous architecture; a task topological graph generation assembly whch is used for generating a task relation topological graph based on the inherent relationship among the decomposed tasks while the operation decomposition assembly is used for performing operation decomposition, wherein task nodes of the task topological graph contain all node attributes requiredfor executing the corresponding tasks; an executor creating component which is used for creating a corresponding executor for each task in the computing resources based on the task relationship topological graph; and an executor network component which comprises one or more data processing paths containing various created executors and is used for fragmenting the actual operation data into task data when receiving the actual operation data, wherein the task data is continuously input into the data processing paths so as to complete the processing of the task data.

Description

technical field [0001] The present disclosure relates to a data processing technology. More specifically, the present disclosure relates to a data processing system for a heterogeneous architecture and a method thereof. Background technique [0002] With the development of machine learning and the gradual deepening of artificial neural network research, the concept of deep learning has been widely concerned and applied. Deep learning is a special kind of machine learning. It uses a network hierarchical structure to express the learning objects, combines simple concepts into abstract concepts, and realizes abstract concept expression through simple concept calculations. At present, deep learning has made great progress in the fields of image recognition, speech recognition and natural language processing. Deep learning involves many model parameters, resulting in a huge amount of calculation, and the large scale of training data, so it needs to consume more computing resour...

AI Technical Summary

Problems solved by technology

However, the model parameters calculated by each GPU in hybrid parallelism need to interact, resulting in a very high order of interaction overhead, which will seriously affect the learning efficiency of the model

[0009] Therefore, judging from the data parallel, model parallel, and hybrid parallel technologies currently proposed in this field, most developers and users of dedicated AI chips usually only focus on the power consumption and efficiency of the computing part, such as how to design A

Images