Batch processing method and device for non-fixed-length input data, terminal and storage medium

Abstract

The embodiment of the invention discloses a batch processing method and device for non-fixed-length input data, a terminal and a storage medium. The method comprises the steps of generating a mask matrix according to input data, and the input data comprises at least two word sequences with different lengths; determining index information of the input data according to the mask matrix; and performing batch processing on effective data in the input data according to the index information. Through the method of the embodiment of the invention, invalid operation aiming at the filling data can be reduced, and the operation speed and the processing efficiency of the deep learning model during batch processing of the non-fixed-length input data are improved.

Description

technical field [0001] Embodiments of the present disclosure relate to computer technology, and in particular to a batch processing method, device, terminal, and storage medium for non-fixed-length input data. Background technique [0002] With the improvement of the computing power of hardware devices, in the process of training deep learning models or using deep learning models for prediction, operations can be performed on multiple input data at the same time. [0003] In scenarios such as natural language processing or speech recognition, the length of input data is generally not fixed. However, limited by the deep learning framework or some computing operations, it is required that all input data have the same length for batch operations. At present, the length of all input data is unified by filling the mark. However, this method will inevitably increase the amount of calculation of the deep learning model and affect the calculation speed and processing efficiency of ...

AI Technical Summary

Problems solved by technology

However, limited by the deep learning framework or some computing operations, it is required that all input data have the same length in order to perform batch operations

At present, the length of all input data is unified by filling the mark. However, this method will inevitably increase the amount of calculation of the deep learning model and affect the calculation speed and processing efficiency of the model.

Images