TaiChuYuanQi Partners with Shanghai AI Lab to Launch AI4S Operator Optimization Competition

The KernelSwift Operator Innovation Competition, organized by the Shanghai Artificial Intelligence Laboratory, has been officially announced. The competition focuses on high-performance operator optimization and domestic computing ecosystem development, with multiple tracks open to the public.

Domestic AI computing company TaiChu (Hangzhou) Integrated Circuit Co., Ltd. (TaiChuYuanQi), as a major partner, has launched the AI4S and Novel Model Architecture Operator Optimization track, targeting the AI for Science field where AI and scientific computing converge.

The KernelSwift Intelligent Operator Migration System will serve as the competition's operator generation tool, providing powerful underlying tool support for participants and significantly lowering the barrier to operator development and optimization.

According to competition organizers, the AI4S track revolves around OneGenomeRice, an AI model for rice genomics. Participants must optimize the reshape_and_cache_flash and flash_attn_varlen_func operators and integrate them into the Teco-vLLM framework. Teams of 2–5 members can register freely.

The total prize pool reaches the million RMB scale across all tracks. For the TaiChu track, the prize pool is 180,000 RMB, with individual teams eligible for up to 30,000 RMB in cash prizes. Every team that completes the challenge receives 1,000 RMB worth of token rewards via the TaiChuYuanQi TecoAPI platform, deployed on the Henan Airport AI Computing Center.

Registration is open until August 21, with code submissions due by August 31. All participating teams will be provided with cloud servers equipped with TaiChu AI accelerator cards, along with detailed training videos and development documentation.