On September 25th, 2024, the Baidu Cloud Intelligence Conference was held in Beijing. Robin Lee, the founder of Baidu, said that agent is his most optimistic development direction of AI applications
Generating intelligent agents requires AI infrastructure. At this conference, Baidu AI Cloud comprehensively upgraded two AI infrastructures, Baige AI heterogeneous computing platform 4.0 and Qianfan big model platform 3.0, and upgraded three AI native application products, namely, code assistant, intelligent customer service and digital human, for computing power, models and AI applications.
Shen Shuan, Executive Vice President of Baidu AI Cloud Group and President of Baidu Intelligent Cloud Business Group, introduced the specific effect of the upgrade and the technical principle of its implementation in detail. For example, during the model training phase, stability and efficiency are the "hard indicators" for measuring the level of GPU clusters. If a GPU fails, the entire cluster will shut down, and a lot of time and cost will be wasted on fault recovery and data rollback. As a result, Baige AI Heterogeneous Computing Platform 4.0 has overcome this challenge and achieved an effective training time ratio of over 99.5% on a 10000 card cluster. Its technical principle is that Baige 4.0 can automatically screen the cluster status, predict GPU failures in advance, and transfer workloads in a timely manner, thereby reducing the frequency of failures.
Shen Dou stated that large models, along with supporting computing power management platforms, model and application development platforms, are rapidly becoming the new generation of infrastructure.