Data processing method and device, computer system and storage medium

Abstract

The invention relates to a data processing method and device, a computer system and a storage medium, which can run an original network according to a connection relationship and an execution sequenceof each equivalent node and a model data set and a model structure parameter of the original network so as to obtain an instruction corresponding to each computing node in each original network. According to the data processing method and device, the computer system and the storage medium, the execution sequence of the computing nodes is adjusted, so that the switching times between different devices and the data copying times between different devices are reduced, and the data processing speed and efficiency of the computer system can be improved.

Description

technical field [0001] The present application relates to the technical field of deep learning, in particular to a data processing method, device, computer system and storage medium. Background technique [0002] With the development of artificial intelligence technology, deep learning has become ubiquitous and indispensable, and many scalable deep learning systems have emerged, such as TensorFlow, MXNet, Caffe, and PyTorch. The above deep learning systems can It is used to provide various neural network models that can run on processors such as CPU (Central Processing Unit, central processing unit) or GPU (Graphics Processing Unit, graphics processing unit). [0003] Generally, when a processor runs a neural network model, such as a Caffe network model, it needs to compile and analyze each computing node in the neural network model separately, and then, according to the structure of the neural network model, follow A certain form executes each computing node. Wherein, the...

AI Technical Summary

Problems solved by technology

For example, a neural network has three computing nodes, which are the first computing node, the second computing node, and the third computing node, where the first computing node and the third computing node need to be executed on the CPU, and the The two computing nodes need to be executed on the GPU, and the second computing node and the third computing node are connected in parallel to the output of the first computing node. In this way, when processing the neural network, the system may sequentially execute the first One computing node, the second computing node, and the third computing node, so that the system will frequently switch between different processors, and the number of communications between different processors is large and multiple data copies are required.

Images