Hongbin Zhang
Postdoctoral Researcher at the Institute of Software, Chinese Academy of Sciences, and Member of the RISC-V International Technical Steering Committee
Hongbin Zhang, Postdoctoral Researcher at the Institute of Software, Chinese Academy of Sciences, and a member of the Technical Steering Committee (TSC) of RISC-V International. His main research focuses on multi-level compilation technologies. He has participated in major national research projects in the field of foundational software and initiated and leads the open-source community of the Ruyi Buddy Compiler for AI compilation.
Topic
Design and Implementation of an AI Compiler for Large Model Inference on RISC-V
This talk focuses on the application requirements of the RISC-V architecture in large model inference and introduces the design and implementation of the Ruyi Buddy Compiler for end-to-end AI inference compilation, targeting the RISC-V Vector extension and custom acceleration extensions. The compilation pipeline follows the PyTorch–MLIR–RISC-V workflow, performing multi-stage compilation with optimizations such as operator fusion, memory access transformation, and vectorization. This enables a complete compilation stack and performance optimization path for running large model inference efficiently on the RISC-V platform.
