This website presents a collection of research projects focused on the performance and portability of computations on WebGPUs, with an emphasis on key operations such as MatMul (Matrix Multiplication), MatVec (Matrix-Vector Multiplication), and Flash Attention.
A tool for testing, tuning, benchmarking, and characterizing GEMM operations.
A repository of results where users can input their requirements to identify the optimal configuration and method.
A platform to learn and play with GEMM kernels.
Running a kernel means you agree that the platform may store and use your data and outputs.