RM新时代有限公司

On October 24, the 2023 Huawei Ascend AI Innovation Competition National Finals were held at the Beijing National Convention Center. CloudWalk Technology won the Gold Award in the Application Track with the highest overall score for its CongRong Large Model Training and Inference Integrated Solution.

The Ascend AI Innovation Competition is a top-level competition for AI developers, jointly organized by national Ascend ecosystem innovation centers, Huawei, the New Generation Artificial Intelligence Industry Technology Innovation Strategic Alliance, and the OpenI Qizhi Community. It aims to work with developers to promote the large-scale development of AI industry applications and lead the future of the new generation of AI industry.

The total prize pool for this competition exceeded 10 million yuan. After five months of fierce competition, over 6,300 developers nationwide registered for the competition. Ultimately, 78 teams stood out in the Application Track and Developer Kit Innovation Track, successfully advancing to the national finals.

CloudWalk Technology has been working closely with Huawei to jointly advance the development of domestic computing power and large model training and inference technologies. In July this year, the two companies launched the "CongRong Large Model Ascend Training and Inference Integrated Solution" at the World Artificial Intelligence Conference (WAIC), laying a solid foundation for the further advancement of the domestic large model ecosystem.

The CongRong Large Model Ascend Distributed Training and Inference Integrated Solution, based on the AI2 professional large model training platform and the MaaS large model inference and application service platform, brings higher production efficiency for large model applications in the industry. It achieves multiple advantages such as customizing exclusive large models, improving training and inference efficiency and accuracy, and lowering the barrier to deployment.

The CongRong Large Model Training and Inference Appliance is based on Huawei's latest Atlas 900 series training servers and comes pre-installed with the CongRong Large Model training and inference integrated platform. The platform is adapted to domestic chips and, through engineering optimization, achieves accuracy alignment with NVIDIA chips. It supports fine-tuning, optimization, inference deployment, and validation of CongRong pre-trained large models of different parameter sizes, as well as open-source large models, including language large models, vision large models, and multimodal large models. This enables large models to be quickly applied to users' business scenarios through knowledge learning and skill learning.

CloudWalk Technology has received strong support from Huawei Ascend, MindSpore, and the Chongqing AICC (AICC Intelligent Computing Center) in R&D and domestic adaptation, becoming a model of cooperation between upstream and downstream enterprises in the domestic AI ecosystem.

As a leader in China's AI industry, CloudWalk Technology will continue to invest in and promote the localization of domestic large models and the AI ecosystem, contributing more to the development of China's AI industry.