动态/静态计算图的混合训练测试数据及SOTA模型

数据集依托项目课题一《图像感知人工智能开放创新平台》建设。课题完成面向图像感知的开放式分层框架设计，实现深度学习模型的静态图和动态图统一，并支持SOTA模型。数据集基于对应指标测试用例运行日志采集，包含MegEngine支持动态/静态计算图混合训练的运行日志；MegEngine 开源的SOTA模型的训练代码或官网链接、预训练权重下载链接。This dataset is developed based on Project Topic 1: "Open Innovation Platform for Image Perception Artificial Intelligence". This topic completed the design of an open layered framework for image perception, realized the unification of static and dynamic computation graphs for deep learning models, and supports SOTA models. The dataset is collected from the operation logs generated by running corresponding indicator test cases, including the operation logs of mixed dynamic/static computation graph training supported by MegEngine, as well as the training codes or official website links of SOTA models open-sourced by MegEngine, and the download links of their pre-trained weights.

搜集汇总

背景与挑战

背景概述

该数据集依托“图像感知人工智能开放创新平台”项目，旨在支持深度学习模型的静态图和动态图统一，并包含SOTA模型。数据集提供了MegEngine框架的混合训练测试日志、SOTA模型的训练资源及预训练权重链接，数据量为17.13MB，包含2个docx格式文件。