Graph execution method and device for neural network model calculation

Abstract

The invention discloses a graph execution method and device for neural network model calculation, and the method comprises the steps: creating task execution bodies on a local machine according to a physical calculation graph generated through the compiling of a deep learning framework, and distributing a plurality of free memory blocks for each task execution body through design. According to the graph execution method and device for neural network model calculation, the execution body of the operator kernel function is used as a basic unit, the tensor of production and consumption is used as flowing data in the whole calculation graph, and therefore the whole calculation graph can participate in deep learning training tasks of different batches of data at the same time in a pipeline parallel mode. And the execution body realizes the training process of the model in a pipeline parallel mode. In a distributed application scene of a large-scale deep neural network, the use threshold of a user is relatively low, and the model can learn internal association of a large amount of data flowing into the neural network in batches, so that intelligent perception and judgment capability in a corresponding scene is obtained.

Description

technical field [0001] The invention relates to the technical field of deep learning, in particular to a graph execution method and device for neural network model calculation. Background technique [0002] With the rapid development of artificial intelligence industrial applications, the demand for large models in practical application scenarios has become more and more urgent. Most of the existing deep learning frameworks provide efficient interfaces for the expression of neural network model computation and the training of neural network models on a single device. However, existing deep learning operating systems may not be flexible and effective when targeting new distributed devices for large-scale deep neural network model training, because distributed devices require more complex parallelism than single devices. In addition, the distributed training interface that has been developed strengthens the parallelism of the models of the existing deep learning framework, bu...

AI Technical Summary

Problems solved by technology

However, existing deep learning operating systems may not be flexible and effective when targeting new distributed devices for large-scale deep neural network model training, because distributed devices require more complex parallelism than single devices.

In addition, the distributed training interface that has been developed strengthens the parallelism of the models of the existing deep learning framework, but complicates the use and implementation of distributed deep learning

Images