Dynamic graph execution method and device for neural network calculation

Abstract

The invention discloses a dynamic graph execution method and device for neural network calculation. The method comprises the following steps: S1, constructing and distributing operators and tensors; s2, deriving an operator execution process by an operator interpreter; s3, the operator interpreter constructs an instruction of the virtual machine during operation; s4, the operator interpreter sends the instruction to the runtime virtual machine; s5, a virtual machine dispatch instruction; and S6, releasing the executed instruction by the virtual machine. According to the dynamic graph execution method and device for neural network calculation provided by the invention, the runtime is abstracted as the virtual machine, and the virtual machine obtains the sub-graph scheduling of each step established by the user through the interpreter in real time and issues and executes each sub-graph, so that the instant debugging requirement of the user is met, local adjustment and optimization can be realized, and the user experience is improved. And obtaining an optimal local model. And the requirements of algorithm researchers for instantly verifying the algorithm correctness and the local performance of the model in the model development process are met.

Description

technical field [0001] The invention relates to the field of computer systems based on a specific calculation model, in particular to a dynamic graph execution method and device for neural network calculation. Background technique [0002] With the rapid development of artificial intelligence industrial applications, the dynamic graph calculation method for deep neural networks is still a hot issue explored by deep learning framework researchers during the development of algorithm models by researchers and engineering users. Most of the existing graph computing technologies perform runtime computing tasks based on a whole static graph, so the algorithm model must be fully constructed before debugging and tuning can be performed. The disadvantage of this is that in the process of building the model, researchers cannot meet the needs of real-time verification of the local subgraph of the model, and can not meet the needs of real-time adjustment of the model structure. In order...

AI Technical Summary

Problems solved by technology

The disadvantage of this is that in the process of building the model, researchers cannot meet the needs of real-time verification of the local subgraph of the model, and can not meet the needs of real-time adjustment of the model structure. In order to solve the real-time debugging of the dynamic graph calculated by the neural network The present invention discloses a dynamic graph execution method and device for neural network calculation, and provides a dynamic graph execution method and device for neural network model calculation in a deep learning training system

Images