Project

Pytorch

Pytorch2 is an ML compiler framework for dynamic deep learning workloads. It features Dynamo and Inductor as the compiler frontend and backend to optimize deep learning codes. I have contributed to both components during my PhD study, which deepens my understanding to ML compilers.

Triton

Triton is a GPU compiler for writing customized high-performance kernels. I have contributed to the interpreter runtime and an argsort kernel for the Top-K operation during my free time.