Performance result optimization method and system for deep learning model training

Abstract

The invention discloses a performance result optimization method and system for deep learning model training. The method comprises the following steps: obtaining a performance result index of the deep learning model training and obtaining a debugging parameter of the deep learning model training in response to the completion of the deep learning model training; the debugging parameters are searched, performance result indexes corresponding to the debugging parameters are obtained from search results, the performance result index corresponding to current search is compared with the performance result index corresponding to last search, and the performance result index corresponding to the last search comprises the performance result index trained by the deep learning model; in response to the condition that the comparison result meets the preset condition, stopping searching the debugging parameters and determining final parameters of the debugging parameters; and carrying out deep learning model training again based on each determined final parameter to obtain an optimized performance result. According to the scheme, the performance evaluation period is shortened, and the performance evaluation efficiency is improved.

Description

Technical field [0001] The present invention relates to the field of deep learning techniques, and more particularly to a performance result optimization method and system for depth learning model training. Background technique [0002] As the depth learning model gradually developed from the study to the application deployment phase, various AI hardware and software frameworks for deep learning model models are endless, and in the application deployment process, with a large number of depth learning models and data sets, for each AI application software The performance evaluation of the framework and the AI ​​hardware architecture is becoming the object of the industry, the AI ​​performance evaluation tool is generated, and it is used for fair and effective assessment of the benchmark performance indicators for deep learning models (such as Samples / S / FLOPS, etc.). . [0003] The current AI performance evaluation tool is mainly used to perform real-time monitoring of key perf...

AI Technical Summary

Problems solved by technology

[0006] In view of this, the present invention proposes a performance result optimization method and system for deep learning model training. During the basic performance evaluation process of the deep learning model, the performance-related parameter analysis and parameter analysis can be automatically performed according to the last round of evaluation results. Tuning, so as to ensure that the deep learning model can exert the optimal performance results during the evaluation process, and solve the problem that if the evaluation results of the current round of evaluation tools are not satisfactory during the evaluation process of the existing performance evaluation tools, it is necessary to manually adjust the parameters to perform new One round of evaluation, resulting in long evaluation cycle and low efficiency

Images