llama.cpp

12b5900d - ggml : sync ggml (add GPT-NeoX RoPE implementation)