A recent collaboration involving Shenzhen Hetao College's AI training platform team, Harbin Institute of Technology (Shenzhen), Shenzhen Institute of Big Data, and Huawei has successfully completed the training of a 1.6 trillion parameter model, DeepSeek-V4-Pro, on a domestic AI computing platform. According to Odaily, this project utilized the Ascend 910C domestic AI computing cluster.
This achievement marks a significant milestone for third-party institutions globally, demonstrating that domestic AI chips can support world-class large parameter model training. The project highlights the technical feasibility of using domestic AI computing power for top-tier model training tasks.
The experiment underscores the distinction between leveraging existing models for business inference and conducting full parameter training from scratch, which differ in technical complexity and hardware requirements. The results indicate that domestic AI computing capabilities are now sufficient for handling large-scale model training.