At Huawei Connect 2023, iFLYTEK and Huawei Data Center Network (DCN) initiated a joint innovation project to build a super-connectivity AI base.
The overarching aim is to jointly create a DCN solution best suited to foundation model scenarios. At the signing ceremony were Wang Jinyang, Vice President of iFLYTEK Engineering Institute; Arthur Wang, President of Data Center Network Domain, Huawei Data Communication Product Line; and Ma Ye, President of Security Product Domain, Huawei Data Communication Product Line.
Continuous Evolution of Foundation Models, Requiring a Large-Scale, High-Throughput, High-Reliability, and High-Security AI Model-Oriented High-Performance Network
Since 2020, AI has taken the entire technology field by storm. The emergence of ChatGPT continues to accelerate AI development, while foundation models are driving AI toward a new development phase. Training foundation models involves complex system engineering, and network infrastructure is one of the keys to long-term stable training.
To improve the networking scale, deployment efficiency, AI training efficiency, and AI training network reliability of foundation models, a large-scale, high-throughput, high-reliability, and high-security AI model-oriented high-performance DCN is urgently needed. With such a network, it will then be possible to ensure efficient, stable, and reliable training of foundation models.
Xinghe + Spark: Huawei Xinghe Network Solution Accelerates iFLYTEK Spark Model Training
iFLYTEK SparkDesk cognitive model is iFLYTEK's next-generation large cognitive intelligence model that has seven core capabilities: text generation, language understanding, knowledge Q&A, logical inference, mathematical capability, coding capability, and multi-modal capability. Leveraging cross-domain knowledge and language understanding capabilities, the model can understand and execute tasks based on natural dialogues.
iFLYTEK and Huawei DCN have made tremendous achievements together. For example, iFLYTEK has used Huawei data center switches to build an ultra-large network and uses Huawei's exclusive AI accelerator to improve network throughput and increase training efficiency by 20% (Internal test result). In addition, Huawei's exclusive network digital map enables unified operations and maintenance (O&M) of computing and network resources and one-click diagnosis of communication exceptions, improving in-training troubleshooting efficiency by 90%.
To better seize the opportunities presented by AI development, iFLYTEK and Huawei DCN, as leaders in the AI industry, will continue to deepen cooperation and jointly explore new paths for industry development. To this end, the two parties officially launched the "Gemini Star Program" joint innovation project.
Win-Win Cooperation, Leading AI Model-Oriented High-Performance, High-Security Network Innovation Breakthroughs
Looking to the future, iFLYTEK, Huawei DCN, and Huawei Network Security will continue to combine forces in order to make breakthroughs in terms of fast deployment, ultimate throughput, efficient O&M, and security and reliability of AI model-oriented high-performance networks. In this way, they aim to jointly promote the innovation and transformation of foundation models, safeguard the access of such models, accelerate the implementation of AI industrialization through industry practices and verification, and lead social development.