en.Wedoany.com Reported - Shanghai Artificial Intelligence Laboratory, together with multiple research institutions, telecom operators, and large model enterprises, jointly launched an AI full-chain software and hardware verification cooperation plan during the second Pujiang AI Academic Annual Conference, with plans to release the verification platform this year. The platform will build scenario-specific verification environments for autonomous AI software and hardware capabilities, aiming to address issues such as complex software-hardware combinations, difficult selection, and inconsistent evaluation standards. On May 14, the DeepLink team of Shanghai Artificial Intelligence Laboratory held the first promotion exchange meeting for the verification platform and took the lead in establishing a verification platform working group. Ascend, as one of the first co-building member units, deeply participated and joined the working group.
The verification platform covers the entire adaptation chain of "chip-middleware-framework-model-application," accelerating the compatibility and adaptation of autonomous software and hardware with core applications by establishing unified verification standards and flexible solutions. The platform aims to provide autonomous, innovative, and out-of-the-box solutions for intelligent computing centers, third-party institutions, and "Artificial Intelligence+" related enterprises, helping AI empower various industries.
Ascend has recently successfully completed the Day-0 full-process adaptation and complete operation verification of large models such as DeepSeek-V4, GLM-5.1, and MiniMax 2.7. This validates that, based on its super-node architecture and open CANN software ecosystem, it can instantly respond to and support the latest global large model demands.
Facing the industry pain points of complex software-hardware combinations and fragmented adaptation testing, Ascend will integrate its practical experience in heterogeneous adaptation and joint debugging, automatic generation of core operators, and end-to-end deployment verification into the upcoming AI full-chain software and hardware verification platform. By creating a reproducible, comparable, and traceable engineering closed loop, Ascend aims to help upstream and downstream industry chains establish an effective "chip-application" two-way selection mechanism in real application scenarios.
Leveraging its accumulation in ten-thousand-card intelligent computing cluster solutions and the CANN compatible ecosystem, Ascend will collaborate with developers, research teams, and enterprise users. By sharing measured data, optimization experience, and implementation cases, it supports vertical domain users in adopting autonomous innovative computing power, forming a virtuous cycle of promoting use through evaluation and improving quality through use.
The construction of the verification platform and the autonomous ecosystem community is a systematic project. Ascend stated that it will work with partners to closely follow the requirements of the "Artificial Intelligence+" action and continuously strengthen the platform's service capabilities. Ascend will continue to base itself on high-performance computing power products, supported by a full-chain verification system, to reduce developers' deployment and migration costs, provide a computing power foundation for the intelligent upgrade of various industries, and accelerate the large-scale implementation process of China's AI industry.
