Mojo24x7/rk-llama.cpp

llama.cpp on the Rockchip RK3588 NPU - RKNPU2 backend by @invisiofficial, rebased onto current upstream, with MoE experts on the NPU, cross-board tensor parallelism and a measurement study