Huawei Cloud, in collaboration with East China Normal University, has developed a native time-series pre-trained model. The joint research team includes members from Huawei Cloud industry model teams, the 2012 Application Scenario Innovation Laboratory, and the Xiaoqiaoling team.
The model, named Qiyao, has secured the top position on two international time-series evaluation benchmarks: TIME-leaderboard and GIFT-Eval in the Pretrained category. The development was based on the Huawei Ascend ecosystem, focusing on optimizing training and inference processes.
The model is designed for zero-shot time-series forecasting tasks, including power grid load assessment, industrial equipment anomaly detection, and financial trend analysis. It also supports secondary fine-tuning for specific industry requirements.
According to Huawei, the project aims to provide a foundation for AI applications in meteorology, energy scheduling, industrial operations, and financial analysis. The achievement marks the entry of domestic time-series foundation models into the global top tier.
