Yi Huimin
Senior Technical Expert at Alibaba and Project Lead of RecIS
Yi Huimin is the Technical Team Lead of the Alibaba Search & Advertising Model Training Team. He is responsible for the sparse training framework that supports core business lines across the group, including Taotian, International Commerce, and Amap. He has been deeply involved in—and has led—multiple major iterations of Alibaba’s sparse training framework.
Topic
Recis — PyTorch-based Recommendation System Training Performance Optimization
Recis is an extremely optimized training framework designed to address the performance challenges encountered when migrating recommendation systems to the PyTorch/large-model ecosystem. It unifies the “recommendation + large-model” technology stack, incorporates optimizations aligned with emerging hardware trends, and has already delivered significant improvements in both effectiveness and large-scale performance across Alibaba’s core business lines.
