TapTechNews reported on May 11 that the Kunpeng-Ascend Developer Conference took place yesterday at the Zhongguancun International Innovation Center in Beijing, under the theme 'With Passion, Let Us Shine Together'. The conference introduced the Native Enabling Program, inaugurated the Kunpeng-Ascend Science and Education Innovation Excellence Center, and launched the Native Innovation Gathering initiative.
Zhang Xiwei, President of Huawei's Computing Product Line, stated that the company has fully launched native development based on Kunpeng & Ascend this year, aiming to make application development more efficient and improve performance. The efforts encompass development kits, enabling programs, and talent development, rapidly building the native ecosystem.
He also revealed that Huawei currently has over 6.1 million developers and more than 6,700 partners, with over 18,700 solution certifications.
For native development, Huawei plans to invest 1 billion RMB (approximately 150 million USD) annually over the next three years to nurture the ecosystem vigorously, supporting the popularization and advancement of native development. The goal is to develop over 1,500 native application partners and cover more than 80% of computing scenarios, ensuring that applications based on Kunpeng & Ascend native development constitute 80% of the total.
Regarding Kunpeng native development, Huawei integrates the full-stack native development capabilities of the Kunpeng Development Kit DevKit, the Kunpeng Application Enablement Kit BoostKit, and the EulerOS into the existing development pipeline modularly as required. This integration allows for a single set of code and pipeline to build multiple platform versions.
In the coding phase, the BoostKit's seven scenario-based SDKs provide a rich programming framework, significantly reducing code development time;
During debugging and tuning diagnostics, DevKit's scenario-based debugging features and eight major Kunpeng affinity scenario scanning capabilities quickly diagnose complex issues and achieve synchronized performance optimization.
For Ascend native development, Huawei plans to use CANN and the inference engine MindIE as the foundation, based on a unified development system for operators, models, and application scenarios, to ignite developers' native creativity, as summarized by TapTechNews.
In operator development, we are further opening CANN, fully upgrading AscendC programming language to support developers in quickly developing operators and building unique competitive advantages;
In model development, the MindSpore AI framework supports both static and dynamic graph modes, balancing model development efficiency and execution performance, and efficiently managing large model debugging and tuning through MindStudio toolchain;
In inference application development, based on the Ascend inference engine MindIE, the framework meets various inference application development needs, enhances out-of-the-box inference performance, reduces application deployment barriers, and accelerates large model innovation implementation.