The ZhiYuan Research Institute and Zhongzhi FlagOS Make Their Debut at the Zhongguancun Forum

To trade stocks, just look at the Golden Unicorn Analyst Reports, authoritative, professional, timely, and comprehensive, helping you uncover potential thematic opportunities!

(Source: Global Network Information)

Source: Global Network

【Global Network Technology Comprehensive Report】The 2026 Zhongguancun Forum Annual Conference was recently held in Beijing. This year’s forum conference is themed “Deep Integration of Technological Innovation and Industrial Innovation,” featuring five major activities: forum meetings, achievement releases, technology transactions, cutting-edge competitions, and supporting activities, to discuss cutting-edge technology and future industrial development trends. As an invited institution for this forum, the Zhiyuan Research Institute showcased its exploration progress in the AI field.

At the exhibition site, the Zhiyuan Research Institute demonstrated key technological breakthroughs achieved with its ecological partners in building the open-source unified software stack, Zhongzhi FlagOS. It is reported that under the framework of Zhongzhi FlagOS, six manufacturers’ AI chips, three types of AI large models, and five homogeneous + heterogeneous kilocalorie clusters have completed end-to-end training with full element verification, becoming the industry’s first technical system to achieve this result with a unified system software stack. This breakthrough not only lays a solid technological foundation for the development of China’s AI multi-dimensional computing power ecosystem but also promotes a substantial step towards making AI computing power more inclusive, injecting new momentum into the high-quality development of China’s AI industry.

In terms of hardware adaptation, Zhongzhi FlagOS successfully completed end-to-end training verification for six mainstream domestic AI chips: TianShu Zhixin, MuXi, HanWuJi, HaiGuang, MoEr Threads, and KunLunXin. In the specified language model training, the training effects of each chip are highly aligned with international mainstream platforms, allowing users to obtain equally high-quality training experiences across diverse hardware platforms, effectively reducing the industry’s reliance on a single hardware system and building an important bridge for the large-scale application of domestic AI chips.

Among them, the HaiGuang homogeneous kilocalorie cluster completed the training of a 32 billion parameter multimodal large model, demonstrating excellent system scalability and stability; the MuXi homogeneous kilocalorie cluster achieved a balance of high performance and high precision in training multiple large models, reaching an internationally advanced level; the MoEr Threads homogeneous kilocalorie cluster completed the full-process training and optimization of an embodied intelligent large model.

Currently, Zhongzhi FlagOS has established a complete technical system, featuring open-source core libraries, open-source tools, and other core components, compatible with mainstream deep learning frameworks like PyTorch and PaddlePaddle, supporting various application scenarios such as robotics, edge computing, and data centers. It can adapt to various chip architectures such as NPU and GPGPU, and 25 AI chips from 16 manufacturers, forming a technical support capability covering the entire process of AI training and inference.

A relevant person in charge of the Zhiyuan Research Institute stated that in order to respond to the fluctuations brought by computing power chips on industry development trends, the Zhiyuan Research Institute will continue to work closely with industrial ecological partners, delve into technological innovation and ecological construction, continuously improve the technological capabilities and ecological layout of Zhongzhi FlagOS, promote the popularization and application of diversified computing power, and ensure that high-quality AI computing resources take root in more industries and scenarios, providing solid computing power support for the self-reliance and strength of China’s artificial intelligence industry. (ChunJun)

Massive news, precise interpretations, all in the Sina Finance APP.