This is a fork of llama.cpp with a custom ggml backend that offloads matrix multiplication to the AMD XDNA2 NPU found in Ryzen AI MAX processors (e.g. Ryzen AI MAX 385). The NPU backend accelerates ...