性能分析方法、装置、系统、计算设备及存储介质

By receiving configuration information from the cloud platform and using containers to obtain performance data, the problem of AI models being unable to be analyzed automatically and in a standardized manner on graphics processors is solved, enabling a unified display of performance analysis across multiple frameworks and simplifying the performance optimization process.

CN116701145BActive Publication Date: 2026-07-17MOORE THREADS TECH CO LTD

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
MOORE THREADS TECH CO LTD
Filing Date
2023-06-07
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing technologies cannot achieve automated and standardized performance analysis of AI models on graphics processors, nor can they simultaneously display performance analysis results from multiple frameworks, making performance optimization difficult.

Method used

By receiving configuration information from the cloud platform, copying the code repository, and using containers to obtain raw performance data, the system analyzes and transforms the data into structured performance visualization data, supporting performance analysis across multiple frameworks.

Benefits of technology

It enables one-click, automated performance analysis, generates standardized performance data, supports performance analysis results from multiple frameworks, and facilitates the determination of optimization methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116701145B_ABST
    Figure CN116701145B_ABST
Patent Text Reader

Abstract

本公开涉及一种性能分析方法、装置、系统、计算设备及存储介质,所述方法应用于计算设备,计算设备上设置有图形处理器和容器,所述方法包括:从云端平台接收配置信息、拷贝代码仓库;根据代码仓库指示从云端平台下载运行模型所需的资源,基于资源使用图形处理器运行模型,使用容器获取模型运行的性能原始数据并分析得到性能展示数据;上传性能展示数据到云端平台,用户设备从云端平台下载性能展示数据并展示给用户;性能原始数据包括图形处理器以多种框架分别运行模型时获取到的多种数据格式的痕迹数据。根据本公开实施例的性能分析方法,一次可展示图形处理器以多种框架运行模型时的性能分析结果,使基于性能分析结果确定性能优化方式更便捷。
Need to check novelty before this filing date? Find Prior Art