https://pytorch.org/blog/accelerating-moes-with-a-triton-persistent-cache-aware-grouped-gemm-kernel/
https://pytorch.org/blog/accelerating-moes-with-a-triton-persistent-cache-aware-grouped-gemm-kernel/